CN105139080A - Improved photovoltaic power sequence prediction method based on Markov chain - Google Patents

Abstract

The invention discloses an improved photovoltaic power sequence prediction method based on the Markov chain, which is characterized in that data preprocessing is performed on photovoltaic power sample data, and the sample data is classified according to seasons, time periods, and weather characteristics, and is established based on the Markov chain Different transition matrices; establish a weather state transition matrix based on weather history data; calculate the basic part of power data, and describe the fluctuation characteristics of power data by using the first-order difference between the time points before and after the sample data. The beneficial effects achieved by the present invention: consider the weather characteristics of photovoltaic output in the Markov chain model, and use the weather state transition matrix to describe the change of weather state; add the fluctuation characteristics to the calculation process from the photovoltaic output state to the photovoltaic output value, reflecting Photovoltaic characteristics: consider the seasonal characteristics, daily characteristics and weather characteristics of photovoltaic power generation to generate multiple state transition matrices, select the corresponding matrix through the discrimination of climate characteristics and time attributes, and generate the state quantity at the target time.

Description

An Improved Photovoltaic Power Sequence Prediction Method Based on Markov Chain

technical field

The invention relates to a photovoltaic power forecasting method, which belongs to the technical field of new energy photovoltaic forecasting.

Background technique

The rapid development of photovoltaic power generation technology has promoted the large-scale and grid-connected photovoltaic power plants. Due to the randomness and volatility of photovoltaic power generation, large-scale photovoltaic grid-connection has a negative impact on the power quality, stability and reliability of the grid. At present, the research methods of photovoltaic output mainly include artificial neural network, least square vector machine, combination method and so on. It is of great significance to study the characteristics of photovoltaic power generation and generate its simulation sequence for evaluating the impact of photovoltaic grid-connected on the grid and the scheduling planning of the grid.

Compared with the historical series, the simulated series has the following advantages: (1) Data missing and data errors are very likely to occur in the historical series, and the evaluation results using such historical data will be biased. (2) Due to the late start of the domestic photovoltaic industry, For some areas, there are few historical data of photovoltaic output, and the length of the data is not enough for the evaluation requirements. However, the use of simulation sequences can generate data sequences of any length, which is more convenient for evaluation. (3) The simulation sequences are generated after characteristic extraction research and can reflect the photovoltaic Therefore, the results of grid connection evaluation are credible.

The modeling methods of photovoltaic power simulation in existing research are generally divided into two categories. The first type is to first establish a model of radiation intensity based on statistical methods based on historical radiation data, and then use the radiation intensity to obtain photovoltaic power simulation data according to the energy conversion relationship of the photovoltaic array. Since the energy conversion of photovoltaic arrays is affected by various factors such as temperature and chemical composition of photovoltaic cells, it is difficult for this type of method to comprehensively consider these factors, which will lead to inaccurate data results. At the same time, it is difficult to take into account the timing of the generated sequence when modeling with a probabilistic statistical model. The second type of method is to directly use the historical data of photovoltaic output to simulate and generate photovoltaic output data regardless of the conversion process from irradiation intensity to photovoltaic power. This kind of method saves the photoelectric conversion process, does not need to consider various factors of the conversion process, improves the accuracy of the data, and simplifies the modeling process at the same time.

Contents of the invention

In order to solve the deficiencies in the prior art, the object of the present invention is to provide a Markov chain-based improved photovoltaic power sequence prediction method, based on the original first-order Markov chain method, considering the seasonal characteristics of photovoltaic power generation, period Characteristics, weather characteristics generate multiple state transition matrices, select the corresponding matrix through characteristic discrimination to generate the state quantity at the target time, and consider the difference characteristics to superimpose the difference, so as to establish an improved Markov chain model for photovoltaic power prediction.

In order to achieve the above object, the present invention adopts the following technical solutions:

A kind of improved photovoltaic power sequence prediction method based on Markov chain, it is characterized in that, comprises the steps:

1) Perform data preprocessing on historical photovoltaic power data, eliminate erroneous data, and replace missing data with the average value of before and after time points to form a sample data set S for prediction;

2) According to the sample data, classify the sample data according to seasons, time periods, and weather characteristics; classify the sample data into thirty-two data sets {s ₁ , s ₂ ,...s ₃₂ };

3) Carry out state classification to the data set in step 2), then respectively establish Markov state transition matrix P, calculate cumulative state transition matrix Q;

4) Suppose the current moment is t, the current output state is β _t , the PV output value is α _t , the season is γ _t , the period is θ _t , the weather state is ε _t , and the fluctuation is f _t ; according to the current γ t _t , θ _t , ε _t , select the corresponding state transition matrix Its elements are p _i,j , and the corresponding cumulative state transition matrix is calculated Its elements are

5) Suppose the state of photovoltaic output at the next moment is β _t+1 ; generate a random number μ _t that obeys uniform distribution, and judge the value range of μ _t ; if Then β _t+1 = 1, if Then β _t+1 = m+1;

6) Determine the fluctuation amount σ _t by sampling, superimpose the fluctuation amount σ _t on the current output value α _t to obtain η, and judge whether η is within the output value range corresponding to β _t+1 obtained in step 5); if so, then Output value at the next moment α _t+1 = η, otherwise regenerate the fluctuation amount for judgment; if t is not the end time of the period, then ε _t+1 = ε _t , otherwise, after selecting the weather state according to the weather state transition matrix Repeat the steps above.

The aforementioned improved photovoltaic power sequence prediction method based on Markov chain is characterized in that, in the step 2), the seasonal characteristics are divided according to the four seasons, the period characteristics are divided into morning and afternoon periods, and the weather characteristics are divided into heavy rain, rain, Cloudy and sunny four types.

The beneficial effects achieved by the present invention: consider the weather characteristics of photovoltaic output in the Markov chain model, and use the weather state transition matrix to describe the change of weather state; add the fluctuation characteristics to the calculation process from the photovoltaic output state to the photovoltaic output value, reflecting Photovoltaic characteristics: consider the seasonal characteristics, daily characteristics and weather characteristics of photovoltaic power generation to generate multiple state transition matrices, select the corresponding matrix through the discrimination of climate characteristics and time attributes, and generate the state quantity at the target time.

Detailed ways

The present invention will be further described below. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention.

The present invention relates to a kind of photovoltaic power prediction method based on Markov chain, specifically comprises the following steps:

Step 1) Perform data preprocessing on the historical photovoltaic power data, remove the erroneous data, and replace the missing data with the average value of the previous and subsequent time points to form a sample data set S for prediction.

The specific formula for calculating the substitute data is:

Where d _t is the missing power data, d _t-1 and d _t+1 are the photovoltaic power values at the previous moment and the next moment of the actual data, respectively.

Step 2) Classify the sample data according to seasons, time periods, and weather characteristics according to the historical data of the sample data. The seasonal characteristics are the four seasons of the year, the time period characteristics are divided into morning and afternoon periods, and the weather characteristics are divided into four types: heavy rain, rain, cloudy and sunny. The sample data is classified into 32 data sets {s ₁ , s ₂ , ...s ₃₂ }. The sunny diary is F, the cloudy diary is C, the showery diary is S, and the heavy rain diary is R. The conversion between weather types can be represented by the probability transition matrix P _w , which is a 4×4 square matrix, w represents the weather type, and w∈{F, C, S, R}.

The element P(F|F) represents the conditional probability that today is also a sunny day when the previous day was a sunny day, and the meanings of other elements can be deduced by analogy, and the weather characteristics are judged according to the weather state transition matrix.

Step 3) Classify the state of the data set, and then establish the Markov state transition matrix respectively, and calculate the cumulative state transition matrix.

The Markov chain state transition matrix is a matrix composed of transition probabilities between different states called a state transition matrix P, which is an N×N square matrix with the sum of each row being 1.

This matrix remains unchanged during state transitions.

For the Markov process of photovoltaic output, its elements can be estimated by the following formula: n _ij represents the frequency of transition from state i to state j after one step, x _t and x _t+1 represent the state at time t and t+1 respectively, and i and j are elements in the state space.

The cumulative state transition matrix Q is calculated based on the state transition matrix.

Step 4) Assume that the current moment is t, the current output state is β _t , the photovoltaic output value is α _t , the month is γ _t , the time period is θ _t , the weather state is ε _t , and the fluctuation is f _t .

According to the current γ _t , θ _t , ε _t , select the corresponding state transition matrix Its elements are p _i,j , and the corresponding cumulative state transition matrix is calculated where the elements are

Step 5) Let the state of photovoltaic output at the next moment be β _t+1 . Generate a random number μ _t that obeys the uniform distribution, and judge the value range of μ _t , if Then β _t+1 = 1, if Then β _t+1 =m+1.

Step 6) Sampling and determining the fluctuation amount σ _t , superimposing the fluctuation amount on the current output value α _t to obtain η, judging whether η is within the value range of β _t+1 , and if so, the next moment output value α _t+1 = η, otherwise regenerate the fluctuation amount and then make a judgment. If t is not the starting time of the period, then ε _t+1 =ε _t , otherwise, select the weather state according to the weather state transition matrix and then proceed to the above steps.

The present invention considers the weather characteristics of photovoltaic output in the Markov chain model, and uses the weather state transition matrix to describe the weather state change; adds the fluctuation characteristics to the calculation process from the photovoltaic output state to the photovoltaic output value to reflect the photovoltaic characteristics; considers photovoltaic power generation The seasonal characteristics, daily characteristics and weather characteristics of power generate multiple state transition matrices, and the corresponding matrix is selected through the discrimination of climate characteristics and time attributes to generate the state quantity at the target time.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (2)

1., based on an improvement photovoltaic power sequence prediction method for Markov chain, it is characterized in that, comprise the steps:

1) data prediction is done to history photovoltaic power data, reject misdata, missing data surrounding time point mean value being replaced, forming the sample data collection S for predicting;

2) according to sample data data according to season, period, weather features classification samples data; Sample data is categorized as 32 data set { s ₁, s ₂... s ₃₂;

3) to step 2) in data set carry out state classification, then set up Markovian state transition matrix P respectively, calculate accumulation state transition matrix Q;

4) suppose that current time is t, the current state of exerting oneself is β _t, photovoltaic numerical value of exerting oneself is α _t, place season is γ _t, the period is , state of weather is ε _t, undulate quantity is f _t; According to current γ _t, , ε _t, selected corresponding state-transition matrix its element is p _i,j, and calculate corresponding accumulation state transition matrix its element is $q_{i,j}={\Σ}_{\mathrm{\δ}=1}^j{p}_{i,\mathrm{\δ}};$

5) photovoltaic is set to exert oneself subsequent time state as β _t+1; Generate and obey equally distributed random number μ _t, and judge μ _tspan; If then β _t+1=1, if then β _t+1=m+1;

6) undulate quantity σ is determined in sampling _t, at the current numerical value α that exerts oneself _ton superpose undulate quantity σ _tobtain η, judge that whether η is in step 5) middle gained β _t+1state corresponding to exert oneself in span; If so, then subsequent time goes out force value α _t+1=η, otherwise regenerate undulate quantity and judge; If t is not the finish time of this period, then ε _t+1=ε _t, otherwise carry out above-mentioned steps again carry out the selection of state of weather according to state of weather transition matrix after.

2. a kind of improvement photovoltaic power sequence prediction method based on Markov chain according to claim 1, it is characterized in that, described step 2) middle Seasonal Characteristics is according to delimiting four seasons, and period characteristic is divided into the morning and afternoon hours, and weather features is divided into heavy rain, rain, cloudy and fine Four types.