CN117522066A - Combined optimization method and system based on peak shaving power supply equipment combination prediction - Google Patents

Abstract

The invention discloses a joint optimization method and system based on peak-shaving power equipment combination prediction, and relates to the technical field of hydropower unit status monitoring. According to the different characteristics of hydropower unit parameters, ARIMA models, random forest models, and LSTM neural network models are established to predict parameters and weighted averages to obtain predicted parameters; then the predicted parameters are input into the support vector machine SVM model and XGBoost model for hydropower unit fault classification and early warning. , the weighted evidence theory method is used to fuse the information of the early warning results to obtain the final fault classification early warning results. Through the calculation and fusion of multiple models, the problems of insufficient classification and early warning accuracy and low reliability of a single model are overcome, and accurate and reliable parameter prediction and status early warning of peak-shaving power equipment, especially hydropower equipment, are achieved. It provides strong support for the safe and stable operation of power equipment.

Description

A joint optimization method and system based on peak-shaving power equipment combination prediction

Technical field

The invention belongs to the technical field of hydropower unit status monitoring, and specifically relates to a joint optimization method and system based on peak-shaving power equipment combination prediction.

Background technique

Hydropower units are one of the key equipment of the power system, and their operating status is related to whether the power grid can provide reliable power supply. Therefore, reliable condition monitoring technology is used to analyze and predict its future operating status, obtain fault information in a timely manner, prevent further spread of accidents, and avoid major power accidents, thereby ensuring the normal operation of hydropower units and for grid peak shaving power supply. of great significance.

However, traditional hydropower unit diagnosis methods rely on theoretical knowledge of equipment and human experience, have poor universality and low efficiency, and cannot effectively predict the status of hydropower units and provide timely warnings. There are three main categories of condition monitoring methods based on data mining. Among them, statistical analysis methods are limited by statistical analysis theory and only have higher accuracy for specific data sets and poor generalization; signal processing methods extract time-frequency domain features Parameters characterize the system state, but time-frequency transformation is prone to information loss; there are many equipment monitoring methods based on machine learning, but there are problems such as poor versatility of single prediction model monitoring classification and insufficient prediction reliability. Therefore, the above-mentioned manual or data mining-based detection methods cannot be widely used for accurate and reliable prediction of the operating status of hydropower units.

Therefore, how to propose a multi-model combination method for predicting the operating status of hydropower units, thereby improving the reliability and credibility of hydropower unit operating status predictions, is a technical problem that technicians in the field urgently need to solve.

Contents of the invention

In view of this, the present invention provides a joint optimization method and system based on combination prediction of peaking power supply equipment, which uses data mining to realize parameter prediction and status warning of peaking power supply equipment, especially hydroelectric power equipment, thereby improving Reliability and credibility of hydropower unit operating status prediction.

In order to achieve the above objects, the present invention provides the following technical solutions:

The invention discloses a joint optimization method based on peak-shaving power supply equipment combination prediction. The specific steps are as follows:

Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit and construct a data table;

According to the data table, determine the initial data set;

According to the linear characteristics, nonlinear characteristics and complex characteristics of the data in the initial data set, construct the ARIMA model, random forest model, and LSTM neural network model to predict parameters; calculate the combination weight, perform a weighted average of the predicted parameters, and obtain the final output result as prediction data set;

Define multiple different types of hydropower unit faults as a fault space, and combine them with the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use an optimization algorithm to optimize the parameters of the two models; according to the prediction data set and In the fault space, the faults of the hydropower unit are classified and early-warned, and the initial early-warning results are obtained respectively; the initial early-warning results are information fused to output the final state early-warning results.

Further, the internal working parameters include: the water inlet pressure, vibration, and ferry coefficient data of the turbine, the AC impedance and magnetic pole coefficient data of the generator rotor, and the oil temperature and gas concentration data in the transformer; the environmental parameters include: Air temperature, ground temperature, relative humidity, and average wind speed data; the power grid load parameters include: operating voltage, active power, and reactive power; then, a data table is constructed based on the acquisition time of each parameter.

Further, the determination of the initial data set includes: first, judging the proportion of missing values in the data table to the total sample size. When the proportion is greater than the threshold, the interpolation method is used to complete the missing values in the data set. When the proportion is less than the threshold, When , the deletion method is used to directly delete missing values; then, outlier processing is performed; then, the Z-score method is used for data standardization; finally, the out-of-bag data feature replacement method is used for feature extraction to obtain the initial data set.

Further, the construction of ARIMA model, random forest model, and LSTM neural network model for parameter prediction includes:

Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters; select parameters with nonlinear characteristics in the initial data set, build a random forest model and train it, and use the trained random forest The model performs parameter prediction; select parameters with complex characteristics in the initial data set, construct an LSTM neural network model and train it, and use the trained LSTM neural network model to perform parameter prediction;

Calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output result, including:

Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model. First, calculate the determination coefficient p of the goodness-of-fit algorithm. The normal value range of p is [0,1]. Its calculation The formula is:

in, Predict the value for the i-th parameter,/> is the average of the real values, z _i is the real value of the i-th parameter, and n is the total number of predicted parameters;

Then, use the tangent function to optimize the determination coefficient p, and finally obtain the optimized combination weight calculation formula:

Among them, w _j represents the combined weight of the j-th sub-model, z _j represents the p-value of the j-th sub-model, and h(z _j ) represents the p-value of the j-th sub-model transformed using the tangent function;

According to the weight value w _j , calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model to obtain the final output result as a prediction data set.

Further, the ARIMA model is composed of an autoregressive model and a moving average model while introducing a differential operation, and its initialization coefficient includes the number of autoregressive terms and the number of moving average terms; the random forest model selects Bootstrap sampling to construct each basis Decision tree; the LSTM neural network model is optimized using the stochastic gradient descent Adam algorithm, and the loss function is the cross entropy function.

Further, the fault space is Θ{F1, F2,...,F10} as the hydropower unit fault identification framework;

To construct the support vector machine SVM model, the grid optimization method is used to roughly select the penalty parameter c and the kernel function parameter g; then the K-fold cross-validation method is used to select the optimal c value and g value as the support vector machine Optimal parameters of SVM model;

In the construction of the XGBoost model, the adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model to determine the optimization parameters;

The initial warning results include: using the optimized support vector machine SVM model and XGBoost model, using the final output results of the prediction data set parameter prediction model as input, to classify and warn the faults of the hydropower unit, and obtain two initial warning results respectively. ;

The final status early warning result is: the hydropower unit status early warning result obtained by information fusion of the two initial early warning results through the weighted evidence method.

Further, the K-fold cross-validation method is specifically: divide the data in the initial data set into k subsets, of which k-1 subsets are used as training subsets, 1 subsets are used as test subsets, and a group of c is selected. and g values for support vector machine SVM model training, and calculate the mean square error MSE of the test subset test results, calculate and select the MSE of multiple groups of different c and g values, and select the optimal c value and g value.

Further, the adaptive particle swarm optimization algorithm is specifically:

Determine the inertia weight w value of the adaptive particle swarm optimization algorithm, and its formula is:

Among them, w _max and w _win represent the maximum and minimum values of inertia weight respectively, k is the current number of iterations, and k _max is the maximum number of iterations;

According to the degree of evolution and aggregation of the particle swarm, the inertia weight of the adaptive particle swarm optimization algorithm is dynamically updated. The improved adaptive inertia weight update formula is:

w ^* =w _in -(w _in -w _min )*evol+(w _max -w _in )*aggr

Among them, w ^* represents the adaptive inertia weight, w _in represents the initial inertia weight value, evol and aggr are the degree of evolution and the degree of aggregation respectively.

Further, the weighted evidence method is specifically:

It is defined that each type of hydropower unit failure in the two initial warning results has two evidence bodies to be combined E1 and E2, the corresponding basic probability distribution functions are m1 and m2 respectively, the corresponding focal elements are Ap and Bq respectively, and the common focal points of Ap and Bq The element is C, the conflict degree coefficient between m1 and m2 is H, then the fusion formula is:

in, It is the average support degree of the focal element of the evidence source.

Through the fusion formula, the early warning results of different types of hydropower unit failures are obtained through fusion.

Further, it also includes evaluating the performance of hydropower unit status early warning by calculating average accuracy indicators, which include: average accuracy, precision rate, recall rate and harmonic mean F1-score.

The invention also discloses a joint optimization system based on peak-shaving power equipment combination prediction, including:

Data acquisition module: Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit, and construct a data table;

Data preprocessing module: determine the initial data set according to the data table;

Parameter prediction module: Based on the linear characteristics, non-linear characteristics and complex characteristics of the data in the initial data set, construct ARIMA models, random forest models, and LSTM neural network models for parameter prediction; calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output The results, as a prediction data set;

Status early warning module: Define multiple different types of hydropower unit faults as a fault space, and combine them with the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use an optimization algorithm to optimize the parameters of the two models; according to Predict the data set and fault space, classify and early-warn the faults of the hydropower unit, and obtain the initial early-warning results respectively; perform information fusion on the initial early-warning results, and output the final state early-warning results.

Further, the parameter prediction module includes:

ARIMA model unit: Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters;

Random forest model unit: select parameters with non-linear characteristics in the initial data set, build a random forest model and train it, and use the trained random forest model to predict parameters;

LSTM neural network model unit: Select parameters with complex characteristics in the initial data set, build and train an LSTM neural network model, and use the trained LSTM neural network model to predict parameters;

Combined weight unit: Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model, and calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model based on the weight values. The final output result is used as a prediction data set.

Further, the status warning module includes:

SVM model unit: Use the grid optimization method to roughly select the penalty parameter c and kernel function parameter g; then use the K-fold cross-validation method to select the optimal c value and g value as the optimal parameters of the support vector machine SVM model , using the prediction data set as input to classify and warn hydropower unit failures;

XGBoost model unit: The adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model, determine the optimization parameters, and use the prediction data set as input to classify and warn hydropower unit failures;

Early warning result fusion unit: Through the weighted evidence method, the initial early warning results obtained by the SVM model unit and the XGBoost model unit are information fused to obtain the final status early warning result of the hydropower unit.

It can be seen from the above technical solutions that compared with the existing technology, the present invention provides a joint optimization method and system based on peak-shaving power equipment combination prediction. By preprocessing and feature extraction of historical parameter data of hydropower units, According to the classification of linear features, nonlinear features and complex features, different parameter prediction models are established for parameter prediction and weighted fusion, which can more accurately predict future data of relevant parameters of hydropower units. By inputting the prediction parameters into the support vector machine SVM model and the XGBoost model for hydropower unit fault early warning, the weighted evidence theory is used to fuse the two early warning results, further providing the accuracy and credibility of hydropower unit early warning, and multiple models The integration of this method overcomes the insufficient reliability of a single prediction model and the low accuracy and credibility of classification warnings. It uses data mining to achieve parameter prediction and status warning for peaking power supply equipment, especially hydropower equipment. It provides strong support for the safe and stable operation of hydropower units and power systems.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without exerting creative efforts.

Figure 1 is a schematic diagram of the overall process of the present invention.

Figure 2 is a schematic diagram of the status warning process of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

The embodiment of the present invention discloses a joint optimization method based on peak-shaving power equipment combination prediction. The overall process is shown in Figure 1. The specific steps are as follows:

Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit and construct a data table.

Perform preprocessing and feature extraction on the data in the data table to obtain the initial data set.

Construct a parameter prediction model. Based on the linear characteristics, nonlinear characteristics and complex characteristics of the data in the initial data set, construct an ARIMA model, a random forest model, and an LSTM neural network model to predict parameters; calculate the combination weight and perform a weighted average of the prediction parameters to obtain The final output result is used as a prediction data set.

Construct a state early warning model, define multiple different types of hydropower unit faults as a fault space, and combine the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use optimization algorithms to optimize the parameters of the two models; According to the prediction data set and fault space, the faults of hydropower units are classified and early-warned, and the initial early-warning results are obtained respectively; the initial early-warning results of the two models are information fused to output the final state early-warning results.

In one embodiment of the present invention, the internal working parameters of the hydropower unit are obtained, including: the inlet water pressure, vibration, and ferry coefficient data of the hydraulic turbine, the AC impedance and magnetic pole coefficient data of the generator rotor, and the oil temperature and gas concentration data in the transformer; obtain The external environmental parameters of the hydropower unit include: air temperature, ground temperature, relative humidity, and average wind speed data; the grid load parameters are obtained, including: operating voltage, active power, and reactive power. Then, a data table is constructed based on the acquisition time of each parameter. The data table is shown in Table 1.

Table 1 Data Sheet

In one embodiment of the present invention, determining the initial data set further includes:

First, determine the proportion of missing values in the data table to the total sample size. When the proportion is greater than the threshold, the statistical method in the interpolation method is used, that is, the mean value of the corresponding data in the same category is used to complete, thereby completing the data set missing values; when less than the threshold, the deletion method is used to directly delete the missing values. When there are many missing samples, the sample size is small. The interpolation method can ensure a certain number of samples, ensure the number of parameters in the data set, and ensure the accuracy of the training model. When there are few missing samples, there is a certain redundancy in the number of samples. , the direct deletion method can reduce the additional calculation amount caused by the interpolation method.

Then perform outlier processing. For samples without hydropower unit failure, the outliers are directly deleted. For samples with hydropower unit failure, the outliers are retained.

Then, the Z-score method is used for data standardization, and its transformation function is:

Among them, μ is the mean of all sample data; σ is the standard deviation of all sample data, x ^* is the standardized value, and x is the value before processing.

Finally, the out-of-bag data feature replacement method is used for feature extraction to obtain the initial data set.

In one embodiment of the present invention, an ARIMA model, a random forest model, and an LSTM neural network model are constructed for parameter prediction, including:

Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters; for example, the output power of a hydraulic turbine is proportional to the inlet water pressure and usually has linear characteristics within the normal operating range. Select parameters with nonlinear characteristics in the initial data set, build and train a random forest model, and use the trained random forest model to predict parameters; for example, gas concentration in a fuel tank usually has nonlinear characteristics, and this parameter is affected by temperature, pressure, and oil quality. and many other factors. Select parameters with complex characteristics in the initial data set, build and train an LSTM neural network model, and use the trained LSTM neural network model to predict parameters; for example, the swing coefficient is related to the structure and operating conditions of equipment such as hydraulic turbines, and it is difficult to use linear or Nonlinear model description,with complex characteristics. A single prediction model often focuses on processing the linear or nonlinear characteristics of parameters, making it difficult to identify all relationships. Combining multiple models to build a combined state prediction model can achieve better results than a single model.

The ARIMA model is composed of an autoregressive model and a moving average model while introducing a difference operation. Its initialization coefficient includes the number of autoregressive terms, that is, the order of the free regression model, and the number of moving average terms.

The random forest model averages the predicted values of multiple CART regression trees before outputting them. The number of regression trees selected in this invention is between 100 and 200. The sampling method selects Bootstrap sampling to build each basic decision tree. The minimum number of samples is Set to 1% of data size.

The input dimension of the LSTM neural network model is 9, the output dimension is 1, and the number of hidden layer neurons is 15. It is optimized using the stochastic gradient descent Adam algorithm, and the loss function is the cross-entropy function. This invention uses the deep learning library TensorFlow to build an LSTM prediction model, and the backend uses the TensorFlow-GPU open source framework developed based on C++. The data is divided according to a certain proportion, with 4000 sets of feature data as training set samples and 120 sets of data as test set samples. When constructing the LSTM model, compare the prediction effects under different numbers of layers, select the appropriate number of hidden layers, and determine the relevant parameters of the model. The number of neurons is 15, the step size is 10, the learning rate is 0.01, and the Dropout ratio is 0.5.

Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model. First, calculate the determination coefficient p of the goodness-of-fit algorithm. The normal value range of p is [0,1]. Its calculation The formula is:

in, Predict the value for the i-th parameter,/> is the average of the real values, z _i is the real value of the i-th parameter, and n is the total number of predicted parameters.

Then, use the tangent function to optimize the determination coefficient p, thereby amplifying the difference in the prediction effect, and finally obtain the optimized combination weight calculation formula:

Among them, w _j represents the combined weight of the j-th sub-model, z _j represents the p-value of the j-th sub-model, and h(z _j ) represents the p-value of the j-th sub-model transformed using the tangent function.

According to the weight value w _j , calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model to obtain the final output result as a prediction data set.

As shown in Figure 2, one embodiment of this application also includes:

Multiple different types of hydropower unit faults are defined as a space Θ{F1, F2,...,F10} as the identification framework.

To construct the support vector machine SVM model, the grid optimization method is used to roughly select the penalty parameter c and the kernel function parameter g; then the K-fold cross-validation method is used to select the optimal c value and g value as the support vector machine SVM model. the optimal parameters;

In the construction of the XGBoost model, the adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model to determine the optimization parameters;

The initial warning results include: using the optimized support vector machine SVM model and XGBoost model, using the prediction data set as input, to classify and warn the faults of the hydropower unit, and obtain two initial warning results respectively;

The final status warning result is: using the weighted evidence method to fuse the two initial warning results to obtain the final hydropower unit status warning result.

Among them, the K-fold cross-validation method is specifically:

Use K-fold cross-validation method to select the optimal c value and g value, specifically: divide the data in the initial data set into k subsets, of which k-1 subsets are used as training subsets and 1 subset is used as test subsets Set, select a set of c and g values for support vector machine SVM model training, and calculate the mean square error MSE of the test subset test results. Calculate and select the MSE of multiple sets of different c and g values, and select based on the MSE minimum value principle. Optimal c and g values. In this invention, K = 4, that is, the 4-fold cross-validation method, first classifies the data samples of the hydropower unit into 4 groups of subsets, uses the first 3 groups of samples as training sets, and the last group of samples is used for testing, and then cycles multiple times , each time a mean square error MSE is obtained; the data is trained 10 times, and the data obtained is sorted and analyzed after each training, and then the 10 MSEs are averaged; finally, the c value is selected based on the MSE minimum principle and g value as the optimal parameters for SVM modeling.

Adaptive particle swarm optimization algorithm, specifically:

Determine the inertia weight w value of the adaptive particle swarm optimization algorithm, and its formula is:

Among them, w _max and w _win represent the maximum and minimum values of inertia weight respectively, k is the current number of iterations, and k _max is the maximum number of iterations.

The inertia weight ω is a key parameter for balancing the local and global search of the algorithm. A larger ω has a strong global search ability, but it is easy to cross the optimal solution, while a smaller ω has a strong local search ability, but it is easy to fall into the local optimum. . According to the degree of evolution and aggregation of the particle swarm, the inertia weight of the adaptive particle swarm optimization algorithm is dynamically updated to optimize the movement process of particles in the search space. The improved adaptive inertia weight update formula is:

w ^* =w _in -(w _in -w _min )*evol+(w _max -w _in )*aggr

Among them, w ^* represents the adaptive inertia weight, w _in represents the initial inertia weight value, evol and aggr are the degree of evolution and the degree of aggregation respectively.

The formula for the degree of evolution is:

In the formula, evol represents the degree of evolution of the particle swarm in the range of [0, 1]; e is a natural constant; c ^k and c ^k-1 represent the fitness changes brought about by the k-th and k-1-th iterations respectively. quantity. When the amount of change this time is larger than the last time, it indicates that the global optimal solution still has a large room for improvement, and the degree of evolution evol will obtain a smaller value, and it is stipulated that when c ^k-1 = 0, evol obtains the minimum value 0; otherwise, evol obtains the minimum value 1; specifically, when c ^k-1 = 0 and c ^k = 0, the global optimal solution position does not change for many consecutive iterations, and the evol at this time is 1, and the particle The degree of evolution of the group reaches its maximum.

The formula for the degree of aggregation is:

Among them, aggr represents the aggregation degree of the particle swarm in the range of [0, 1]; N represents the number of particles contained in each subspace (N1, N2, N3...Nn), and the base of the log function is equal to the number of subspaces n. When all particles are evenly distributed into each subspace, that is, N1=N2=N3=...Nn, aggr obtains the minimum value 0, and the degree of aggregation of the particle swarm is the lowest at this time; when all particles are in the same subspace, that is, there is When Ni=N, aggr obtains the maximum value 1, and the particle swarm has the highest degree of aggregation at this time.

Weighted evidence method, specifically:

It is defined that each type of hydropower unit failure in the two initial warning results has two evidence bodies to be combined E1 and E2, the corresponding basic probability distribution functions are m1 and m2 respectively, the corresponding focal elements are Ap and Bq respectively, and the common focal points of Ap and Bq The element is C, the conflict degree coefficient between m1 and m2 is H, then the fusion formula is:

in, is the average support degree of the focus element of the evidence source. When it is equal to 0, it means that one of the diagnostic model output results does not belong to the hydropower unit fault space.

Through the fusion formula, the early warning results of different types of hydropower unit failures are obtained through fusion.

One embodiment of this application also includes evaluating the performance of hydropower unit status early warning by calculating average accuracy indicators. The average accuracy indicators include: average accuracy rate, precision rate, recall rate and harmonic mean F1-score, where:

Average accuracy = (true examples + true negative examples)/total number of samples;

The precision rate refers to the proportion of the number of samples correctly predicted as positive categories by the model to the number of samples predicted as positive categories. The precision rate measures the accuracy of the model in all positive category samples. The calculation formula is as follows:

Precision rate = true examples/(true examples + false positive examples);

The recall rate refers to the proportion of the number of samples correctly predicted as positive categories by the model to the number of all true positive category samples. The recall rate measures the proportion of positive category samples identified by the model to the total positive category samples. The calculation formula is as follows:

Recall rate = true examples/(true examples + false negative examples);

F1-score is the harmonic mean of precision rate and recall rate. It takes into account the precision and recall rate of the model and is used to balance the relationship between the two. The calculation formula is as follows:

F1-score=2*(precision rate*recall rate)/(precision rate+recall rate).

One embodiment of the present invention discloses a joint optimization system based on peak-shaving power equipment combination prediction, including:

Data acquisition module: Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit, and construct a data table;

Data preprocessing module: determine the initial data set according to the data table;

Parameter prediction module: Based on the linear characteristics, non-linear characteristics and complex characteristics of the data in the initial data set, construct ARIMA models, random forest models, and LSTM neural network models for parameter prediction; calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output The results, as a prediction data set;

Status early warning module: Define multiple different types of hydropower unit faults as a fault space, and combine them with the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use an optimization algorithm to optimize the parameters of the two models; according to Predict the data set and fault space to classify and early-warn the faults of hydropower units, and obtain the initial early-warning results respectively; conduct information fusion on the initial early-warning results, and output the final status early-warning results.

In one embodiment of the present invention, the parameter prediction module includes:

ARIMA model unit: Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters;

Random forest model unit: select parameters with non-linear characteristics in the initial data set, build a random forest model and train it, and use the trained random forest model to predict parameters;

LSTM neural network model unit: Select parameters with complex characteristics in the initial data set, build and train an LSTM neural network model, and use the trained LSTM neural network model to predict parameters;

Combined weight unit: Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model, and calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model based on the weight values. The final output result is used as a prediction data set.

In one embodiment of the present invention, the status warning module includes:

SVM model unit: Use the grid optimization method to roughly select the penalty parameter c and kernel function parameter g; then use the K-fold cross-validation method to select the optimal c value and g value as the optimal parameters of the support vector machine SVM model , using the prediction data set as input to classify and warn hydropower unit failures;

XGBoost model unit: The adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model, determine the optimization parameters, and use the prediction data set as input to classify and warn hydropower unit failures;

Early warning result fusion unit: Through the weighted evidence method, the initial early warning results obtained by the SVM model unit and the XGBoost model unit are information fused to obtain the final status early warning result of the hydropower unit.

Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on its differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant details, please refer to the description in the method section.

The above description of the disclosed embodiments enables those skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be practiced in other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A joint optimization method based on peak-shaving power equipment combination prediction, which is characterized by the following specific steps: Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit and construct a data table; According to the data table, determine the initial data set; According to the linear characteristics, nonlinear characteristics and complex characteristics of the data in the initial data set, construct the ARIMA model, random forest model, and LSTM neural network model for parameter prediction; calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output result as the prediction data set; Define multiple different types of hydropower unit faults as a fault space, and combine them with the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use an optimization algorithm to optimize the parameters of the two models; according to the prediction data set and In the fault space, the faults of the hydropower unit are classified and early-warned, and the initial early-warning results are obtained respectively; the initial early-warning results are information fused to output the final state early-warning results. 2. A joint optimization method based on peak-shaving power equipment combination prediction according to claim 1, characterized in that the internal working parameters include: water turbine inlet pressure, vibration, ferry coefficient data, generator rotor AC impedance, magnetic pole coefficient data, oil temperature and gas concentration data in the transformer; the environmental parameters include: air temperature, ground temperature, relative humidity, average wind speed data; the power grid load parameters include: operating voltage, active power, Reactive power; Build a data table based on the acquisition time of each parameter. 3. A joint optimization method based on peak-shaving power equipment combination prediction according to claim 1, characterized in that the determination of the initial data set includes: First, determine the proportion of missing values in the data table to the total sample size. When the proportion is greater than the threshold, the interpolation method is used to complete the missing values in the data set. When it is less than the threshold, the deletion method is used to directly delete the missing values; then, Perform outlier processing; then, use the Z-score method for data standardization; finally, use the out-of-bag data feature replacement method for feature extraction to obtain the initial data set. 4. A joint optimization method based on combination prediction of peaking power supply equipment according to claim 1, characterized in that the construction of an ARIMA model, a random forest model, and an LSTM neural network model for parameter prediction includes: Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters; select parameters with nonlinear characteristics in the initial data set, build a random forest model and train it, and use the trained random forest The model performs parameter prediction; select parameters with complex characteristics in the initial data set, construct an LSTM neural network model and train it, and use the trained LSTM neural network model to perform parameter prediction; Calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output result, including: Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model. First, calculate the determination coefficient p of the goodness-of-fit algorithm. The normal value range of p is [0,1]. Its calculation The formula is: in, Predict the value for the i-th parameter,/> is the average of the real values, zi is the real value of the i-th parameter, and n is the total number of predicted parameters; Then, use the tangent function to optimize the determination coefficient p, and finally obtain the optimized combination weight calculation formula: Among them, w _j represents the combined weight of the j-th sub-model, z _j represents the p-value of the j-th sub-model, and h(z _j ) represents the p-value of the j-th sub-model transformed using the tangent function; According to the weight value w _j , calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model to obtain the final output result as a prediction data set. 5. A joint optimization method based on peak-shaving power equipment combination prediction according to claim 4, characterized in that, The ARIMA model is composed of an autoregressive model and a moving average model while introducing a difference operation, and its initialization coefficient includes the number of autoregressive terms and the number of moving average terms; The random forest model selects Bootstrap sampling to build each basic decision tree; The LSTM neural network model is optimized using the stochastic gradient descent Adam algorithm, and the loss function is the cross-entropy function. 6. A joint optimization method based on peak-shaving power equipment combination prediction according to claim 1, characterized in that, The fault space is Θ{F1, F2,...,F10} as the hydropower unit fault identification framework; To construct the support vector machine SVM model, the grid optimization method is used to roughly select the penalty parameter c and the kernel function parameter g; then the K-fold cross-validation method is used to select the optimal c value and g value as the support vector machine Optimal parameters of SVM model; In the construction of the XGBoost model, the adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model to determine the optimization parameters; The initial warning results include: using the optimized support vector machine SVM model and XGBoost model, using the prediction data set as input, to classify and warn the faults of the hydropower unit, and obtain the initial warning results respectively; The final status early warning result is: the hydropower unit status early warning result obtained by fusing the information of the two initial early warning results through the weighted evidence method The K-fold cross-validation method is specifically: Divide the data in the initial data set into k subsets, of which k-1 subsets are used as training subsets, and 1 subset is used as the test subset. Select a set of c and g values to train the support vector machine SVM model, and calculate the test For the mean square error MSE of the subset test results, calculate and select the MSE of multiple sets of different c and g values, and select the optimal c and g values based on the MSE minimum principle; The adaptive particle swarm optimization algorithm is specifically: Determine the inertia weight w value of the adaptive particle swarm optimization algorithm, and its formula is: Among them, w _max and w _win represent the maximum and minimum values of inertia weight respectively, k is the current number of iterations, and k _max is the maximum number of iterations; According to the degree of evolution and aggregation of the particle swarm, the inertia weight of the adaptive particle swarm optimization algorithm is dynamically updated. The improved adaptive inertia weight update formula is: w ^* =w _in -(w _in -w _min )*evol+(w _max -w _in )*aggr Among them, w ^* represents the adaptive inertia weight, w _in represents the initial inertia weight value, evol and aggr are the degree of evolution and the degree of aggregation respectively; The weighted evidence method is specifically: It is defined that each hydropower unit failure in the two initial warning results has two evidence bodies to be combined E1 and E2, the corresponding basic probability distribution functions are m1 and m2 respectively, and the corresponding focal elements are Ap, Bq, Ap and Bq respectively. The common focal element is C, and the conflict degree coefficient between m1 and m2 is H. The fusion formula is: in, The average support level for the focal element of the evidence source; Through the fusion formula, the early warning results of different types of hydropower unit failures are obtained through fusion. 7. A joint optimization method based on peak-shaving power equipment combination prediction according to claim 1, characterized in that it also includes evaluating the status early warning performance of hydropower units by calculating an average accuracy index, and the average accuracy index includes: average Accuracy, precision, recall and harmonic mean F1-score. 8. A joint optimization system based on peak-shaving power equipment combination prediction, which is characterized by including: Data acquisition module: Obtain the internal working parameters, external environmental parameters and power grid load parameters of the hydropower unit, and construct a data table; Data preprocessing module: determine the initial data set according to the data table; Parameter prediction module: Based on the linear characteristics, non-linear characteristics and complex characteristics of the data in the initial data set, construct ARIMA models, random forest models, and LSTM neural network models for parameter prediction; calculate the combination weight, perform a weighted average of the prediction parameters, and obtain the final output The results, as a prediction data set; Status early warning module: Define multiple different types of hydropower unit faults as a fault space, and combine them with the initial data set to build a support vector machine SVM model and an XGBoost model respectively, and use an optimization algorithm to optimize the parameters of the two models; according to Predict the data set and fault space, classify and early-warn the faults of the hydropower unit, and obtain the initial early-warning results respectively; perform information fusion on the initial early-warning results, and output the final state early-warning results. 9. A joint optimization system based on peak-shaving power equipment combination prediction according to claim 8, characterized in that the parameter prediction module includes: ARIMA model unit: Select parameters with linear characteristics in the initial data set, build an ARIMA model and train it, and use the trained ARIMA model to predict parameters; Random forest model unit: select parameters with non-linear characteristics in the initial data set, build a random forest model and train it, and use the trained random forest model to predict parameters; LSTM neural network model unit: Select parameters with complex characteristics in the initial data set, build and train an LSTM neural network model, and use the trained LSTM neural network model to predict parameters; Combined weight unit: Use the goodness-of-fit algorithm to allocate the weight values of the ARIMA model, random forest model, and LSTM neural network model, and calculate the weighted average of the prediction parameters of the ARIMA model, random forest model, and LSTM neural network model based on the weight values. The final output result is used as a prediction data set. 10. A joint optimization system based on peak-shaving power equipment combination prediction according to claim 8, characterized in that the status warning module includes: SVM model unit: Use the grid optimization method to roughly select the penalty parameter c and kernel function parameter g; then use the K-fold cross-validation method to select the optimal c value and g value as the optimal parameters of the support vector machine SVM model , using the prediction data set as input to classify and warn hydropower unit failures; XGBoost model unit: The adaptive particle swarm optimization algorithm is used to continuously update the number of base learners, learning rate, and maximum tree depth of the XGBoost model, determine the optimization parameters, and use the prediction data set as input to classify and warn hydropower unit failures; Early warning result fusion unit: Through the weighted evidence method, the initial early warning results obtained by the SVM model unit and the XGBoost model unit are information fused to obtain the final status early warning result of the hydropower unit.