CN118246351A - A deep learning method for solving unit commitment problems considering unit confidence - Google Patents

Abstract

The invention relates to a solution method for a deep learning unit combination problem considering unit confidence, and belongs to the technical field of power system planning. Aiming at the problem that the solving time is overlong in solving the unit combination problem by using mixed integer linear programming at present, a deep learning unit combination problem model considering the unit confidence is provided. The model is divided into two stages, firstly, a solution of a binary decision variable of the start-stop state of the unit is obtained by training a long-short-time memory network; and secondly, setting a confidence threshold on the basis of determining the confidence unit, setting a non-confidence unit decision meeting the confidence threshold condition as a hot start initial value of the model, and carrying the non-confidence unit decision into a solver for solving. The result shows that the method remarkably improves the solving efficiency of the unit combination problem. The method is beneficial to reducing the waste of power resources and maintaining the stability of the power system, and has important significance for the development of the unit combination problem.

Description

A deep learning method for solving unit commitment problems considering unit confidence

Technical Field

The present invention relates to the technical field of power system planning, and in particular to a method for solving a deep learning unit combination problem taking unit confidence into consideration.

Background technique

Unit combination is one of the important daily tasks of power systems. With the continuous improvement of mixed-integer linear programming (MILP) theory and the continuous development of commercial solvers, more and more power companies are turning to MILP to solve this problem. Although the solution of the MILP model is stable and the distance from the optimal value is observable, its calculation time is sensitive to the increase in the number of units.

With the development of machine learning, the use of machine learning, especially deep learning represented by deep neural networks, to solve unit combination problems has become a frontier research area.

In order to solve the problem of long calculation time, the present invention proposes a deep learning unit combination problem model that takes into account the unit confidence. The core of the model is that it not only ensures the success rate of the solution, but also improves the speed of solving the unit combination problem.

Summary of the invention

In view of the defect that the current solution time of the unit commitment problem is too long, the purpose of the present invention is to propose a deep learning method for solving the unit commitment problem taking into account the unit confidence. The method can effectively shorten the solution time while ensuring the accuracy of the solution.

This method is mainly divided into two stages:

The first stage is to train the Long Short-Term Memory (LSTM) network to obtain the binary solution (0-1) corresponding to the unit combination start-stop decision.

The second stage is to determine whether the unit decision of the untrusted unit meets the confidence threshold requirement by setting the unit confidence threshold under the premise of determining the trusted-generators set (TGS). If it meets the requirement, it is set as the initial value of the hot start and brought into the commercial solver for solution.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1. The process of solving the deep learning unit commitment problem considering unit confidence;

Figure 2: Neural network structure diagram.

Detailed ways

First, the binary solution (0-1) corresponding to the start-stop decision of the unit combination is obtained by training the long short-term memory network. The present invention selects the power load as the input of the neural network. After the power load is input into the long short-term memory network, the nodes therein are randomly removed through a dropout layer with a dropout rate of 0.5. It is then connected to the fully connected layer, and the fully connected layer in the network uses the Sigmoid function as the activation function. Next, the output result is judged, and the result greater than or equal to 0.5 is output as 1, and the result less than 0.5 is output as 0. Finally, the solution of the start-stop decision corresponding to the unit at each moment is obtained.

The next step is to determine the confidence group and set the confidence threshold of the group. The determination of the confidence group is as follows:

Solving the mathematical programming model for unit start and stop decision making;/> The start and stop decision of the unit is output by the neural network; /> is the absolute difference between the two.

, that is, the sum of the absolute differences between the start-stop decision of a unit solved by the mathematical programming model at all times and the output of the neural network is equal to 0. Such a unit is identified as a trusted unit, that is, TGS=1. /> , that is, there is an error between the start-stop decision solved by the mathematical programming model and the output result of the neural network. Such a unit is set as an untrusted unit, that is, TGS = 0. When it is brought into the solver for solution calculation, the binary variable of the trusted unit will be set to a fixed value.

For non-confident units that do not meet the confidence unit conditions, their binary decision variables (obtained from the neural network) are not discarded and set as unknown quantities in the unit combination mathematical programming model. Instead, the confidence condition is set to determine whether the non-confident unit decision variables meet this condition. For non-confident unit decision variables that meet the condition, they are set as the hot start initial value and brought into the unit combination mathematical programming model, and solved using the solver. The confidence threshold formula is as follows:

N represents the number of samples; T represents the number of moments; is the confidence level of the setting.

Finally, the unit combination mathematical programming model is brought in and solved using the solver. The unit combination mathematical programming model is as follows:

The objective function of the model includes startup type cost, startup phase cost and production cost:

Where: It is the collection of all thermal power units in the system; /> is the set of all moments; /> is the startup type cost of unit g at time t; /> is the startup cost of unit g at time t; /> is the production cost of unit g at time t; /> is the hot start cost of unit g; /> is the cold start cost of unit g; /> is the time until time t when unit g has been shut down; /> is the minimum shutdown time of unit g; /> is the time required for cold start of unit g; /> is the cost coefficient of unit g; /> is the power size of the i-th stage during the startup process of unit g; /> is the index of the unit startup method; /> This is the startup process of unit g; /> is the startup type s of unit g at time t (a binary variable, 1 indicates that unit g adopts the startup method of type s; 0 indicates that unit g does not adopt the startup method of type s);/> is the actual power of unit g at time t; /> is the state of unit g at time t (a binary variable, 1 represents that unit g is online at time t; 0 represents that unit g is offline at time t).

Load constraints of thermal power units:

Where: is the load size at each moment t.

Power constraints of thermal power units:

Where: is the maximum output power of unit g; /> is the minimum output power of unit g; /> is the power level of unit g at time t when it is above the minimum output power of the unit; /> is the shutdown state of unit g at time t (binary variable, 1 means unit g is in shutdown state; 0 means unit g is in non-shutdown state);/> is the power size of the i-th stage during the shutdown process of unit g; /> It is the shutdown process of the unit.

Thermal power unit climbing constraints:

Where: is the upper spinning reserve capacity of unit g at time t; /> is the lower spinning reserve capacity of unit g at time t; /> is the gradeability of unit g; /> is the downhill capability of unit g; /> is the startup state of unit g at time t (a binary variable, 1 indicates that unit g is in the startup state; 0 indicates that unit g is in the non-startup state).

The climbing constraint for the first moment is considered separately:

Where: It is the power of unit g at the initial moment (i.e. the last moment of the previous day).

Spinning reserve constraints for thermal power units:

Where: is the upper spinning reserve capacity required at time t; /> is the lower spinning reserve capacity required at time t.

Minimum start/shutdown time constraints for thermal power units:

Where: is the minimum start-up time of unit g; /> is the initial state of unit g (binary variable, 1 means that unit g is online at the initial moment; 0 means that unit g is offline at the initial moment);/> is the time that unit g must remain in operation before scheduling begins; /> is the time that unit g must remain shut down before scheduling begins; /> is the time that unit g has been turned on before being dispatched; /> is the time that unit g has remained shut down before being dispatched.

Enable/disable logic constraints:

Unit start type selection:

In order to clearly illustrate the advantages of the method proposed in this patent, Table 1 compares the solution time of the unit combination problem in this patent with the solution time of the pure mathematical programming model.

Table 1 Comparison of model results

Test system 3 units 10 units 20 units 32 units Solution 1 (seconds) 27.0 798.1 1214.3 7948.1 Solution 2 (seconds) 27.6 2144.8 2489.7 11527.0

As shown in Table 1, Scheme 1 is the time required for a deep learning unit commitment problem solving method considering unit confidence proposed in this patent. Scheme 2 is the solution time of a pure mathematical programming model. The results show that Scheme 1 has significantly improved the solution time.

The above embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or exchanges made by technicians in the technical field based on the present invention are all within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims (4)

1. A deep learning method for solving unit commitment problems considering unit confidence, characterized in that the method comprises the following operations: first, a deep learning model based on a long short-term memory network is constructed, the power load is used as the input of the deep learning model, and the output is the solution of the unit start-stop decision; secondly, when a confident unit is determined, a confidence threshold is set for the error of the start-stop decision solution of the unconfident unit, and when the confidence threshold is met, this part of the decomposition is set as the hot start initial value of the unit combination mathematical programming model, and is brought into the solver for solving, so as to obtain the unit power size and the start-stop decision solution that meet the constraints of the unit combination problem. 2. According to a deep learning unit commitment problem solving method considering unit confidence according to claim 1, it is characterized in that the method also includes the construction of a neural network model: taking the power load as the input of the neural network, after the load is input into the long short-term memory network, the nodes therein are removed through a dropout layer with a dropout rate of 0.5; then connected with the fully connected layer, the fully connected layer in the network uses the sigmoid function as the activation function; then the output result is judged, and the result greater than or equal to 0.5 is output as 1, and the result less than 0.5 is output as 0; finally, the solution of the start and stop decision corresponding to the unit at each time is obtained. 3. According to a deep learning unit commitment problem solving method considering unit confidence according to claim 1, it is characterized in that the method further comprises setting a unit confidence threshold. For non-confident units that do not meet the confidence unit conditions, their start-stop decision variables are not discarded and set as unknown quantities in the mathematical programming model; instead, the confidence threshold is set to determine whether the start-stop decision variables of the non-confident units meet the confidence threshold conditions. For the start-stop decision variables of the non-confident units that meet the conditions, they are set as hot start initial values and brought into the unit combination mathematical programming model, and solved using a solver; the confidence threshold requirements are as follows: In the above formula: N represents the number of samples; T represents the number of moments; is the confidence level of the setting; g is the unit /> The index of ; t is the time /> The index of; /> It is the absolute difference between the solution of the binary decision variables output by the neural network and the solution of the binary decision variables solved by the mathematical programming model. 4. According to a deep learning unit commitment problem solving method considering unit confidence level in claim 1, the method further comprises constructing a unit commitment mathematical programming model; the objective function of the mathematical programming model is expressed as: Where: It is the collection of all thermal power units in the system; /> is the set of all moments; /> is the startup type cost of unit g at time t; /> is the startup cost of unit g at time t; /> is the production cost of unit g at time t; /> is the hot start cost of unit g; /> is the cold start cost of unit g; /> is the time until time t when unit g has been shut down; /> is the minimum shutdown time of unit g; /> is the time required for cold start of unit g; /> is the cost coefficient of unit g; /> is the power size of the i-th stage during the startup process of unit g; /> is the index of the unit startup method; /> This is the startup process of unit g; /> is the startup type s selected by unit g at time t: a binary variable, 1 means that unit g adopts the startup method of type s at time t; 0 means that unit g does not adopt the startup method of type s at time t; /> is the actual power of unit g at time t; /> is the state of unit g at time t: a binary variable, 1 represents that unit g is online at time t; 0 represents that unit g is offline at time t.