Temperature and photovoltaic scene generation method based on time series correlation feedback correction

Abstract

The present invention discloses a method for generating temperature and photovoltaic scenes based on time-series correlation feedback correction, including: 1. Generate a temperature set and a corresponding photovoltaic scene related to the illumination scene obtained by the Monte Carlo method; 2. Take the illumination intensity scene value and the temperature prediction value as input, use Kalman filtering to obtain the temperature output under each scene, and use it as the reference value of each scene; 3. Substitute the illumination intensity scene value and the corresponding temperature reference value as the mean into the multivariate normal distribution at each moment, and generate multiple groups of random variables (E, T) for each scene and moment; 4. Find the temperature closest to the reference value within the range of the illumination intensity scene value floating up and down ε, as the final temperature scene value. The present invention can focus on the mutual correlation between temperature and illumination intensity and the temporal autocorrelation of temperature when generating an uncertainty scene set, which is more in line with the actual situation and avoids the influence of some unreasonable scenes on the scheduling results.

Description

Technical Field

[0001] Aiming at the correlation between photovoltaic scenes and temperature, the present invention proposes a method for generating temperature and photovoltaic output correlation scenes taking into account the temporal sequence based on covariance analysis and Kalman filtering. Background Art

[0002] In recent years, stochastic robust methods have begun to be applied to problems such as unit commitment and grid planning. Stochastic robust optimization can be regarded as a robust version of stochastic optimization, which represents uncertainty as different sets and optimizes the worst-case probability expected cost in different sets, thus obtaining a more economical result while ensuring reliability.

[0003] In stochastic robust optimization, scenario methods are often used to generate uncertainty sets. There have been a lot of studies on scenario generation methods, including Monte Carlo and Latin hypercube sampling. These methods have good results when sampling a certain time section, but they do not consider the relationship between the current moment and the adjacent moment variables. The generated scenario sequence usually cannot represent the actual fluctuation characteristics of the variables. Another type of method includes autoregressive moving average model (ARMA), Kalman filtering, scenario tree, etc. The scenarios generated by these methods cannot represent the correlation between different spaces or multiple variables.

[0004] At present, the research on correlation mainly includes the joint probability distribution method of multivariate variables and the Copula function analysis method. For example, the covariance matrix of the multivariate normal distribution is used to associate the distribution of wind power fluctuations with the distribution of wind power forecast errors to generate short-term wind power scenarios; the Copula function is used to analyze the dependencies of the multivariate distributions that affect wind speed and establish a wind speed correlation model. The traditional covariance analysis method is difficult to consider the temporal nature of temperature. The scenario generated by this method cannot take into account the relationship between the temperature reference value at the current moment and the adjacent moment, and cannot characterize the actual fluctuation characteristics of the temperature reference value. Summary of the invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art. Based on covariance analysis and Kalman filtering, a method for generating scenarios related to temperature and photovoltaic output taking into account the temporal nature is proposed, in order to focus on the mutual correlation between temperature and light intensity and the temporal autocorrelation of temperature when generating an uncertain scenario set, so as to be more in line with the actual situation and avoid the impact of certain unreasonable scenarios on the scheduling results.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] The temperature and photovoltaic scene generation method based on time series correlation feedback correction of the present invention is characterized in that it is carried out according to the following steps:

[0008] Step 1: Obtain the illumination scene according to the Monte Carlo method, and generate the relevant temperature set and the corresponding photovoltaic scene:

[0009] Step 1.1: Based on the probability statistics of light intensity, use Monte Carlo simulation to generate m light intensity scene values. Each scene is divided into n moments. Then, use formula (1) to calculate the photovoltaic output value P corresponding to the jth moment under the i-th light intensity scene: i.j :

[0010] P i.j =E i.j Aτ (1)

[0011] In formula (1), E i.j is the light intensity of the photovoltaic power station at the jth moment in the i-th scenario; A is the effective area of ​​the photovoltaic panel; τ is the photoelectric conversion efficiency;

[0012] Step 2: Take each light intensity scene value and temperature prediction value as input, use Kalman filtering to obtain the temperature output for each scene, and use it as the reference value for the corresponding scene:

[0013] Step 2.1, according to the historical data of light intensity E and temperature of the photovoltaic power station, the least square method is used to fit the A, B, and H transfer matrices, thereby establishing a dynamic model, where the transfer matrix A represents the temperature drop in the absence of light, the transfer matrix B represents the effect of light on temperature, and the transfer matrix H represents the relationship between the state estimate and the observed value, and the elements on the diagonal are 1, and the remaining elements are 0;

[0014] Define the time variable k and initialize k=2;

[0015] The temperature prediction value and light intensity scene value at the k-1th moment in the i-th scene are recorded as X i.k-1 and U i.k-1 , and initialized to the temperature prediction value and light intensity scene value at the initial moment in the i-th scene;

[0016] The state covariance of the temperature at the k-1th moment in the i-th scenario is recorded as P i.k-1 , and initialize P i.k-1 is a diagonal matrix with only non-diagonal elements being 0 and diagonal elements being any value between 0 and 1;

[0017] Step 2.2: Use formula (2) to predict the prior state estimate of the temperature at the kth moment in the i-th scenario:

[0018]

[0019] In formula (2), U i.k-1represents the control variable of temperature at time k in the i-th scenario;

[0020] Step 2.3, use formula (3) to calculate the prior estimated covariance of the temperature at the kth moment in the i-th scenario:

[0021]

[0022] In formula (3), Q is the covariance of the temperature error that follows the Gaussian distribution during the estimation process;

[0023] Step 2.4: Use formula (4) to calculate the Kalman gain K at the kth moment in the i-th scenario: i.k :

[0024]

[0025] In formula (4), R is the covariance of the temperature error that follows the Gaussian distribution during the measurement process;

[0026] Step 2.5: Use formula (5) to calculate the optimal estimated value X of the temperature at time k in the i-th scenario: i.k :

[0027]

[0028] In formula (5), Z i.k is the observed value of temperature at time k in the i-th scenario;

[0029] Step 2.6: Use formula (6) to calculate the state covariance P of the temperature at time k in the i-th scenario: i.k :

[0030]

[0031] Step 2.7, after assigning k+1 to k, return to step 2.2 and execute sequentially until k reaches n, thereby obtaining a temperature reference sequence with time sequence;

[0032] Step 3: Take the scene value of light intensity in each scene and its corresponding temperature reference value as the mean and substitute them into the multivariate normal distribution of each time period, so as to generate multiple sets of random variables (E, T) from each scene and time:

[0033] Step 3.1: Take the illumination intensity scene value and its temperature prediction value at the jth moment in the i-th scene as the mean μ i.j , using formula (7) to establish the multivariate normal distribution Z at the jth moment in the i-th scenario i.j ~N(μ i.j ,Σ i.j ):

[0034]

[0035] In formula (7), Σ is the covariance matrix; σ 1.1 is the covariance of the scene illumination intensity value, σ 1.2 is the covariance between the scene value of light intensity and the predicted value of temperature, σ 2.1 is the covariance between the temperature prediction value and the light intensity scene value, and σ 1.2 =σ 2.1 , σ 2.2 is the covariance of the temperature prediction value;

[0036] Step 3.2: Use the multivariate Gaussian random number generator to generate multiple sets of random variables Z based on the multivariate normal distribution at the jth moment in the i-th scenario. i.j ;

[0037] Step 4: Float up and down within the range of ε of the light intensity scene value, find the temperature closest to the reference value, and use it as the final temperature scene value, thereby obtaining a set of uncertainty scenarios that consider the mutual correlation between temperature and light intensity and the temporal autocorrelation of temperature, where ε represents the floating threshold.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] 1. Taking into account the strong correlation between air temperature and light intensity, the present invention regards the light intensity E and air temperature T in each time period as a set of variables Z=(E,T), introduces the multivariate normal distribution Z~N(μ,Σ) to study the correlation between light intensity and temperature in each time period, thereby obtaining multiple groups of scenes and ensuring the mutual correlation between light intensity and temperature in each group of scenes.

[0040] 2. The present invention establishes a Kalman filter dynamic equation based on the historical data of light intensity and temperature of the photovoltaic power station, regards the light intensity as the control variable, the temperature as the system state variable, and the day-ahead temperature prediction value as the system observation value. Through Kalman filter recursion, the temperature estimation value at the previous moment is used to correct the temperature prediction value at the current moment to obtain the output value at the current moment, and then the prediction value at the next moment is corrected according to the output value to obtain the output value at the next moment, thereby obtaining a temperature reference sequence with time series, fully considering the relationship between the temperature reference values ​​at the current moment and the adjacent moments, and can characterize the actual fluctuation characteristics of the temperature reference value.

[0041] 3. The present invention combines the time-series Kalman filter and covariance analysis, and proposes a new scene generation method. This method solves the impact of certain unreasonable scenes on the scheduling results in actual situations. In the first stage, the lighting scene value is generated, and the temperature reference value is obtained by correcting the temperature under each lighting scene. In the second stage, the multivariate normal distribution of temperature and light is established to generate a temperature scene set. Finally, the temperature and photovoltaic scenes are obtained by screening, thereby solving the problem that the traditional covariance analysis method is difficult to consider the time series of temperature, and then generating an uncertainty scene set that considers the mutual correlation between temperature and light intensity and the time series autocorrelation of temperature, so that the scene set is more in line with the actual situation and avoids the impact of unreasonable scenes on the scheduling results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the method for generating temperature and photovoltaic related scenes proposed by the present invention. DETAILED DESCRIPTION

[0043] In this embodiment, Figure 1 As shown, a method for generating temperature and photovoltaic scenarios based on time series correlation feedback correction is that the generated uncertainty scenario set can characterize the mutual correlation between temperature and light intensity and the time series autocorrelation of temperature, and is carried out in the following steps:

[0044] Step 1: Obtain the illumination scene according to the Monte Carlo method, and generate the relevant temperature set and the corresponding photovoltaic scene:

[0045] Step 1.1: According to the probability and statistical characteristics of light intensity, Monte Carlo simulation is used to generate m light intensity scene values. Each scene is divided into n moments. 50 light intensity scenes are set, and each scene is divided into 24 time periods. Then, the photovoltaic output value P corresponding to the jth moment under the i-th light intensity scene is calculated using formula (1): i.j :

[0046] P i.j =E i.j Aτ (1)

[0047] In formula (1), E i.j is the light intensity of the photovoltaic power station at the jth moment in the i-th scenario; A is the effective area of ​​the photovoltaic panel; τ is the photoelectric conversion efficiency;

[0048] Step 2: Take each light intensity scene value and temperature prediction value as input, use Kalman filtering to obtain the temperature output for each scene, and use it as the reference value for the corresponding scene:

[0049] Step 2.1, according to the historical data of light intensity E and temperature of the photovoltaic power station, the least square method is used to fit the A, B, and H transfer matrices, thereby establishing a dynamic model, where the transfer matrix A represents the temperature drop in the absence of light, the transfer matrix B represents the effect of light on temperature, and the transfer matrix H represents the relationship between the state estimate and the observed value, and the elements on the diagonal are 1, and the remaining elements are 0;

[0050] Define the time variable k and initialize k=2;

[0051] The temperature prediction value and light intensity scene value at the k-1th moment in the i-th scene are recorded as X i.k-1 and U i.k-1 , and initialized to the temperature prediction value and light intensity scene value at the first moment in the i-th scene;

[0052] The state covariance of the temperature at the k-1th moment in the i-th scenario is recorded as P i.k-1 , and initialize P i.k-1 is a diagonal matrix with only non-diagonal elements being 0 and diagonal elements being any value between 0 and 1. Set the matrix to a diagonal matrix with diagonal elements being 0.5;

[0053] Step 2.2: Use formula (2) to predict the prior state estimate of the temperature at the kth moment in the i-th scenario:

[0054]

[0055] In formula (2), U i.k-1 represents the control variable of temperature at time k in the i-th scenario;

[0056] Step 2.3, use formula (3) to calculate the prior estimated covariance of the temperature at the kth moment in the i-th scenario:

[0057]

[0058] In formula (3), Q is the covariance of the temperature error that follows the Gaussian distribution during the estimation process;

[0059] Step 2.4: Use formula (4) to calculate the Kalman gain K at the kth moment in the i-th scenario: i.k :

[0060]

[0061] In formula (4), R is the covariance of the temperature error that follows the Gaussian distribution during the measurement process;

[0062] Step 2.5: Use formula (5) to calculate the optimal estimated value X of the temperature at time k in the i-th scenario:i.k :

[0063]

[0064] In formula (5), Z i.k is the observed value of temperature at time k in the i-th scenario;

[0065] Step 2.6: Use formula (6) to calculate the state covariance P of the temperature at time k in the i-th scenario: i.k :

[0066]

[0067] Step 2.7, after assigning k+1 to k, return to step 2.2 and execute sequentially until k reaches n, thereby obtaining a temperature reference sequence with time sequence;

[0068] Step 3: Take the scene value of light intensity in each scene and its corresponding temperature reference value as the mean and substitute them into the multivariate normal distribution of each time period, so as to generate multiple sets of random variables (E, T) from each scene and time:

[0069] Step 3.1: Take the illumination intensity scene value and its temperature prediction value at the jth moment in the i-th scene as the mean μ i.j , using formula (7) to establish the multivariate normal distribution Z at the jth moment in the i-th scenario i.j ~N(μ i.j ,Σ i.j ):

[0070]

[0071] In formula (7), Σ is the covariance matrix; σ 1.1 is the covariance of the scene illumination intensity value, σ 1.2 is the covariance between the scene value of light intensity and the predicted value of temperature, σ 2.1 is the covariance between the temperature prediction value and the light intensity scene value, and σ 1.2 =σ 2.1 , σ 2.2 is the covariance of the temperature prediction value;

[0072] Step 3.2: Use the multivariate Gaussian random number generator to generate multiple sets of random variables Z based on the multivariate normal distribution at the jth moment in the i-th scenario. i.j ;

[0073] Step 4: Float up and down within the range of ε of the light intensity scene value, find the temperature closest to the reference value, and use it as the final temperature scene value, thereby obtaining a set of uncertainty scenarios that consider the mutual correlation between temperature and light intensity and the temporal autocorrelation of temperature, where ε represents the floating threshold, and the floating threshold ε is set to 10%.

[0074] In summary, the temperature and photovoltaic scene generation method based on time-series correlation feedback correction proposed in the present invention combines the time-series Kalman filtering and covariance analysis, which can fully consider the time-series of temperature, is more in line with the actual situation, and avoids the impact of certain unreasonable scenarios on the scheduling results.

Claims

1. A method for generating temperature and photovoltaic scenes based on time series correlation feedback correction, characterized in that: The steps are as follows: Step 1: Obtain the illumination scene according to the Monte Carlo method, and generate the relevant temperature set and the corresponding photovoltaic scene: Step 1.1: Based on the probability statistics of light intensity, use Monte Carlo simulation to generate m light intensity scene values. Each scene is divided into n moments. Then, use formula (1) to calculate the photovoltaic output value P corresponding to the jth moment under the i-th light intensity scene: i.j : (1) In formula (1), E i.j is the light intensity of the photovoltaic power station at the jth moment in the i-th scenario; A is the effective area of ​​the photovoltaic panel; τ is the photoelectric conversion efficiency; Step 2: Take each light intensity scene value and temperature prediction value as input, use Kalman filtering to obtain the temperature output for each scene, and use it as the reference value for the corresponding scene: Step 2.1, according to the historical data of light intensity E and temperature of the photovoltaic power station, the least square method is used to fit the A, B, and H transfer matrices, thereby establishing a dynamic model, where the transfer matrix A represents the temperature drop in the absence of light, the transfer matrix B represents the effect of light on temperature, and the transfer matrix H represents the relationship between the state estimate and the observed value, and the elements on the diagonal are 1, and the remaining elements are 0; Define the time variable k and initialize k=2; The temperature prediction value and light intensity scene value at the k-1th moment in the i-th scene are recorded as and , and initialized to the temperature prediction value and light intensity scene value at the initial moment in the i-th scene; The state covariance of the temperature at the k-1th moment in the i-th scenario is recorded as , and initialize is a diagonal matrix with only non-diagonal elements being 0 and diagonal elements being any value between 0 and 1; Step 2.2: Use formula (2) to predict the prior state estimate of the temperature at the kth moment in the i-th scenario: : (2) Step 2.3: Use formula (3) to calculate the prior estimated covariance of the temperature at the kth moment in the i-th scenario: : (3) In formula (3), is the covariance of the temperature error that follows a Gaussian distribution during the estimation process; Step 2.4: Use formula (4) to calculate the Kalman gain at the kth moment in the i-th scenario: : (4) In formula (4), is the covariance of the temperature error that follows a Gaussian distribution during the measurement process; Step 2.5: Use formula (5) to calculate the optimal estimated value of the temperature at time k in the i-th scenario: : (5) In formula (5), is the observed value of temperature at time k in the i-th scenario; Step 2.6: Use formula (6) to calculate the state covariance of the temperature at time k in the i-th scenario: : (6) Step 2.7, after assigning k+1 to k, return to step 2.2 and execute sequentially until k reaches n, thereby obtaining a temperature reference sequence with time sequence; Step 3: Take the scene value of light intensity in each scene and its corresponding temperature reference value as the mean and substitute them into the multivariate normal distribution of each time period, so as to generate multiple sets of random variables (E, T) from each scene and time: Step 3.1: Take the illumination intensity scene value and its temperature prediction value at the jth moment in the i-th scene as the mean , using formula (7) to establish the multivariate normal distribution of the jth moment in the i-th scenario : (7) In formula (7), Σ is the covariance matrix; is the covariance of the light intensity scene values, is the covariance between the scene value of light intensity and the predicted value of temperature, is the covariance between the temperature prediction value and the light intensity scene value, and = , is the covariance of the temperature prediction value; Step 3.2: Use the multivariate Gaussian random number generator to generate multiple sets of random variables Z based on the multivariate normal distribution at the jth moment in the i-th scenario. i.j ; Step 4: Float up and down within the range of ε of the light intensity scene value, find the temperature closest to the reference value, and use it as the final temperature scene value, thereby obtaining a set of uncertainty scenarios that consider the mutual correlation between temperature and light intensity and the temporal autocorrelation of temperature, where ε represents the floating threshold.

Images