Joint power station operation optimization method and system based on high-precision prediction of wind and light power output

Abstract

The present application proposes a method and system for optimizing the operation of a combined power station based on high-precision prediction of wind and light output. The method includes: collecting historical wind and light output data of a new energy and energy storage combined power station, and preprocessing the historical wind and light output data to obtain prediction data; constructing a wind and light output prediction model with multi-model fusion, and predicting the upper and lower limit thresholds of wind and light output through the wind and light output prediction model based on the prediction data; determining an energy storage power scheduling strategy according to the energy storage capacity of the energy storage system, and generating a power generation index tracking strategy for the combined power station by combining the upper and lower limit thresholds and the energy storage power scheduling strategy; constructing an energy storage system power optimization allocation model that simultaneously satisfies grid frequency modulation and the power generation index tracking strategy, and generating and executing an optimization control instruction for the combined power station. This method improves the flexibility and coordination of charge and discharge regulation of the energy storage and new energy combined power station, can meet the grid frequency modulation requirements, and improves the operation efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of new energy power generation, and particularly to a combined power station operation optimization method and system based on high-precision wind and light output prediction. Background Art

[0002] With the strengthening of people's environmental awareness and the increasing energy demand, new energy power generation technologies such as wind power generation and solar power generation are developing towards large-scale and high-quality directions. However, due to the volatility and randomness of new energy, the current power grid system is still unable to fully absorb the large-scale and high-proportion access of new energy. Therefore, a combined power station including an energy storage system and a new energy power generation system has been designed and gradually popularized.

[0003] However, the integrated operation ability of the combined power station in the related technology is poor. In some scenarios where the power grid needs to perform frequency modulation, the adjustment method is single, and its operation mode cannot fully take into account the energy storage and the output of new energy, reducing the operation efficiency of the combined power station. Therefore, there is an urgent need for a solution that can optimize the operation of the combined power station. Summary of the Invention

[0004] The present application aims to solve at least one of the technical problems in the related technologies to some extent.

[0005] To this end, the first object of the present application is to propose a combined power station operation optimization method based on high-precision wind and light output prediction. This method improves the flexible adjustment ability of the source-network-load-storage integration, performs operation optimization control of the energy storage and the new energy combined power station on the basis of meeting the power grid frequency modulation requirements, improves the flexibility and coordination of the charge and discharge adjustment of the energy storage and the new energy combined power station, and is beneficial to improving the operation efficiency.

[0006] The second object of the present application is to propose a combined power station operation optimization system based on high-precision wind and light output prediction;

[0007] The third object of the present application is to propose a non-transitory computer-readable storage medium.

[0008] To achieve the above object, the first aspect embodiment of the present application is to propose a combined power station operation optimization method based on high-precision wind and light output prediction. The method includes the following steps:

[0009] Collect historical wind and light output data of the new energy and energy storage combined power station, and preprocess the historical wind and light output data to obtain prediction data;

[0010] Construct a wind and light output prediction model with multi-model fusion, and predict the upper and lower limit thresholds of the wind and light output through the wind and light output prediction model based on the prediction data;

[0011] Generate a storage power dispatch strategy based on the storage capacity of the energy storage system, and generate a power generation index tracking strategy for the combined power station by combining the upper and lower limit thresholds and the storage power dispatch strategy;

[0012] Construct a power optimization allocation model for the energy storage system that simultaneously satisfies grid frequency modulation and the power generation index tracking strategy, generate an optimized control instruction for the combined power station according to the power optimization allocation model for the energy storage system, and execute the optimized control instruction.

[0013] Optionally, in an embodiment of the present application, preprocess the historical wind and light output data, including: removing outliers in the historical wind and light output data based on the K-means clustering algorithm or the chi-square test algorithm; filling the missing data generated after removing the outliers based on the mean filling method or the Mahalanobis distance method.

[0014] Optionally, in an embodiment of the present application, removing outliers in the historical wind and light output data based on the K-means clustering algorithm includes: determining the value of the clustering number K through the elbow method; based on the value of the clustering number K, clustering the clustering sample data through the K-means clustering algorithm to generate K clusters; for the clustering result data in each cluster, estimating the standard deviation through the Bessel formula, and discriminating outliers through the 3-sigma rule according to the standard deviation estimate value, and removing the outliers from the clustering result data.

[0015] Optionally, in an embodiment of the present application, filling the missing data generated after removing the outliers based on the Mahalanobis distance method includes: calculating the Mahalanobis distance value between the missing data and other data except the missing data; determining a preset number of nearest neighbor data of the missing data according to all the Mahalanobis distance values, and performing normalization calculation on the Mahalanobis distances of the preset number of nearest neighbor data to obtain normalized distance values; calculating the entropy value of each neighbor data according to the normalized distance value; calculating the variation degree coefficient of the corresponding neighbor data according to each entropy value; calculating the weighted coefficient of the corresponding neighbor data according to each variation degree coefficient; calculating the estimated value of the missing data according to the weighted coefficient of each neighbor data and the value of each neighbor data.

[0016] Optionally, in an embodiment of the present application, the integrated wind-solar power output prediction model includes a combined wind-solar power output prediction model and a wind-solar power output prediction model based on similar days. Building the combined wind-solar power output prediction model includes: building a plurality of single wind-solar power output prediction models, where the single wind-solar power output prediction model includes a differencing integrated moving average autoregressive model ARIMA, a backpropagation BP neural network model, and an XGBoost regression prediction model; taking the weighted sum of the predicted power outputs of all the single wind-solar power output prediction models and the minimum absolute deviation from the actual wind-solar power output as the objective function, and solving the objective function under the constraint condition that the weights satisfy normalization to obtain the weight values of each single wind-solar power output prediction model; building the combined wind-solar power output prediction model according to each single wind-solar power output prediction model and its weight value.

[0017] Optionally, in an embodiment of the present application, building the wind-solar power output prediction model based on similar days includes: performing a correlation analysis on the power output of the new energy system and multiple meteorological elements within a preset time period through the Pearson correlation coefficient method to determine the meteorological variables whose correlation with the power output within the preset time period is greater than a preset correlation threshold; clustering multiple historical days and the prediction day through the transitive closure method, and determining similar days related to the meteorological variables of the prediction day from the multiple historical days according to the clustering result; using the power output data and meteorological data of the similar days as training data to train any one of the single wind-solar power output prediction models to obtain the wind-solar power output prediction model based on similar days.

[0018] Optionally, in an embodiment of the present application, clustering multiple historical days and the prediction day through the transitive closure method includes: obtaining multiple characteristic indicators of each historical day and building a characteristic indicator matrix corresponding to all the historical days; performing data standardization processing on the characteristic indicator matrix; calculating the similarity between every two pieces of data for the standardized data to obtain a fuzzy similarity matrix; calculating the transitive closure corresponding to the fuzzy similarity matrix; calculating the cut matrix corresponding to all the elements in the transitive closure and classifying according to the similarity values between the data in the cut matrix.

[0019] Optionally, in an embodiment of the present application, the energy storage power scheduling strategy includes: judging the operation mode of the energy storage system according to the current energy storage capacity of the energy storage system; performing corresponding charge and discharge operations in different operation modes and calculating the real-time energy storage capacity after responding to the charge and discharge operations; based on the real-time energy storage capacity, repeatedly execute the judgment of the operation mode of the energy storage system according to the scheduling requirements until the scheduling task is completed.

[0020] To achieve the above object, an embodiment of the second aspect of the present application further provides a combined power station operation optimization system based on high-precision prediction of wind and light output, including the following modules:

[0021] A preprocessing module, configured to collect historical wind and light output data of a new energy and energy storage combined power station, and preprocess the historical wind and light output data to obtain prediction data;

[0022] A prediction module, configured to construct a wind and light output prediction model with multi-model fusion, and predict the upper and lower limit thresholds of wind and light output through the wind and light output prediction model based on the prediction data;

[0023] A generation module, configured to generate a storage power scheduling strategy based on the energy storage capacity of the energy storage system, and generate a power generation index tracking strategy for the combined power station by combining the upper and lower limit thresholds and the storage power scheduling strategy;

[0024] An execution module, configured to construct a storage system power optimization allocation model that simultaneously satisfies grid frequency modulation and the power generation index tracking strategy, generate an optimization control instruction for the combined power station according to the storage system power optimization allocation model, and execute the optimization control instruction.

[0025] The technical solution provided by the embodiment of the present application at least brings the following beneficial effects: The present application first analyzes and evaluates the multi-time and multi-space scale operation characteristics of renewable energy power generation, including collecting wind and light output data of the power station, establishing a standardized database containing wind and light resources of the power station and unit operation data, and using statistical methods to analyze the wind and light output data to obtain the spatio-temporal distribution law of wind and light output. Then, intelligent tracking of the power generation index of the energy storage and renewable energy combined power station is carried out, including studying the dynamic distribution law of the power generation index of renewable energy power stations under channel or peak shaving and power limiting conditions, establishing wind and light power prediction models at different time scales and estimating the upper and lower limits of wind and light output, calculating the state of charge and operation interval of the energy storage system in the combined power station under different conditions, and proposing an intelligent tracking strategy for the power generation index of the energy storage and renewable energy combined power station. Finally, an optimization control strategy for the energy storage and renewable energy combined power station that meets the grid frequency modulation requirements is studied, including establishing a storage system power optimization allocation model that meets the technical requirements of the energy storage for different operation scenarios of the energy storage and renewable energy combined power station and participates in grid secondary frequency modulation and tracks the power generation index at the same time, proposing a power optimization control strategy for the energy storage system considering the maximization of comprehensive benefits, and optimizing the operation of the power station according to this strategy. Thus, the present application improves the flexible adjustment ability of the source-network-load-storage integration, optimizes the operation control of the energy storage and new energy combined power station on the basis of meeting the grid frequency modulation requirements, improves the flexibility and coordination of the charge and discharge regulation of the energy storage and new energy combined power station, and is beneficial to improving the operation efficiency.

[0026] To implement the above embodiments, a third aspect embodiment of the present application further proposes a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the joint power station operation optimization method based on high-precision wind and light output prediction in the above embodiments is implemented.

[0027] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0028] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where:

[0029] Figure 1 is a flowchart of a joint power station operation optimization method based on high-precision wind and light output prediction proposed by an embodiment of the present application;

[0030] Figure 2 is a flowchart of an outlier removal method based on the K-means clustering algorithm proposed by an embodiment of the present application;

[0031] Figure 3 is a flowchart of a missing data filling method based on the Mahalanobis distance method proposed by an embodiment of the present application;

[0032] Figure 4 is a flowchart of a construction method of a wind and light output combined prediction model proposed by an embodiment of the present application;

[0033] Figure 5 is a flowchart of a construction method of a wind and light output prediction model based on similar days proposed by an embodiment of the present application;

[0034] Figure 6 is a flowchart of a clustering method based on the transitive closure method proposed by an embodiment of the present application;

[0035] Figure 7 is a flowchart of a storage power scheduling strategy proposed by an embodiment of the present application;

[0036] Figure 8 is a schematic structural diagram of a joint power station operation optimization system based on high-precision wind and light output prediction proposed by an embodiment of the present application. Detailed Embodiments

[0037] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0038] A combined power station operation optimization method and system based on high-precision wind and light output prediction proposed in an embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0039] Figure 1 The flowchart of a combined power station operation optimization method based on high-precision wind and light output prediction proposed in an embodiment of the present application is as Figure 1 shown, and the method includes the following steps:

[0040] Step S101: Collect historical wind and light output data of the new energy and energy storage combined power station, and preprocess the historical wind and light output data to obtain prediction data.

[0041] Among them, the combined power station targeted by the operation optimization method of the present application refers to a power station that includes a new energy power generation subsystem and an energy storage subsystem, and combines the two systems for power generation. The new energy power generation subsystem can be a renewable energy power generation system such as wind power generation or photovoltaic power generation. The energy storage subsystem includes multiple energy storage devices such as lithium batteries and corresponding controller devices, which are used to store the excess power of the new energy power generation subsystem and participate in tasks such as peak shaving and frequency modulation.

[0042] Among them, the historical wind and light output data is the historical data of the output power of the new energy power generation subsystem during previous wind power generation and / or photovoltaic power generation.

[0043] Specifically, the present application first collects the historical wind and light output data of the combined power station in different ways. For example, during the operation of the combined power station, the real-time output data is added with time information and stored in the database, and then the pre-stored historical wind and light output data is retrieved from the operation data record database when obtaining the data. Then, the output of the new energy subsystem at a future moment is predicted based on the historical wind and light output data. It can be understood that in actual applications, due to the influence of various unexpected factors, there may be abnormal data in the obtained historical wind and light output data. Therefore, to improve the prediction accuracy, the present application preprocesses the historical wind and light output data, and the processed data is used as the prediction data for predicting the output power of the new energy subsystem.

[0044] In an embodiment of the present application, preprocessing of historical wind and solar power output data includes removing abnormal data therefrom and filling in missing data. Specifically, during implementation, preprocessing can be performed through different abnormal data screening methods and data filling methods. As one possible implementation, the present application can remove outliers from the historical wind and solar power output data based on the K-means clustering algorithm or the chi-square test algorithm, and then fill in the missing data generated after removing the outliers based on the mean filling method or the Mahalanobis distance method. The process of outlier removal and data filling by the present application in the above manner will be described in detail below.

[0045] To more clearly illustrate the specific implementation of outlier removal based on the K-means clustering algorithm in the present application, an outlier removal method proposed in the embodiment of the present application will be used for exemplary illustration below. Figure 2 The flowchart of an outlier removal method based on the K-means clustering algorithm proposed in the embodiment of the present application is as Figure 2 shown, and the method includes the following steps:

[0046] Step S201, determining the value of the clustering number K through the elbow method.

[0047] Specifically, the core idea of the K-means clustering algorithm is to perform clustering with k points in space as the centers, classify the objects closest to them, and update the values of each clustering center successively through an iterative method until the best clustering result is obtained. The core step in this algorithm is to determine the value of K.

[0048] In this embodiment, the value of K is determined through the elbow method. The core index of the elbow method is the sum of squared errors (SSE for short). The calculation formula of SSE is as follows:

[0049]

[0050] where C i is the i-th cluster, p is the sample point in C i m i is the centroid of C i (i.e., the mean of all samples in C i ), and SSE is the clustering error of all samples, representing the quality of the clustering effect.

[0051] Specifically, when using the elbow method to determine the value of K, as the number of clusters K increases, the sample division will be more refined, the aggregation degree of each cluster will gradually increase, and the sum of squared errors (SSE) will gradually decrease. Moreover, when K is less than the true number of clusters, the increase in K leads to a significant increase in the aggregation degree of each cluster, and the decline rate of SSE also increases; when K reaches the true number of clusters, the return of the aggregation degree obtained by increasing the value of K will rapidly decrease, and the decline rate of SSE will suddenly decrease and finally level off as the value of K continues to increase. Therefore, the relationship graph between SSE and K is in the shape of an elbow, and the present application obtains the true number of clusters of the data by determining the value of K corresponding to the elbow.

[0052] Step S202: Based on the value of the number of clusters K, perform clustering on the clustering sample data through the K-means clustering algorithm to generate K clusters.

[0053] Specifically, use the historical wind and light output data as the clustering sample data to be clustered, and perform clustering on the clustering sample data according to the determined value of K through the K-means clustering algorithm. When performing clustering through the K-means clustering algorithm, the following steps are included:

[0054] Step S2021: Use the value of k determined by the elbow method as the number of clusters, and then randomly select k sample data from the clustering sample data as the initial clustering centers {m1, m2,... mk}.

[0055] Step S2022: Calculate the Euclidean distance from each sample to each initial clustering center, select the clustering center with the closest distance to form k clusters, and update the k clusters according to the distance formula. The Euclidean distance calculation formula is as follows:

[0056]

[0057] Step S2023: Recalculate the clustering centers for the newly generated k clusters. The calculation formula for the clustering centers is as follows:

[0058]

[0059] Step S2024: Repeat Step S2022 and Step S2023 until the termination condition |mn+1 - mn| ≤ ε or the preset number of iterations is reached.

[0060] Thus, the operation of the K-means algorithm ends, realizing the clustering of similar data and dividing the historical wind and light output data into multiple clusters.

[0061] Step S203: For the clustering result data in each cluster, perform standard deviation estimation through the Bessel formula, and perform outlier discrimination according to the standard deviation estimation value through the 3-sigma rule to remove outliers from the clustering result data.

[0062] Specifically, in the actual application of outlier screening, in this embodiment, Bessel's formula can be used to estimate the standard deviation of the finite data samples in the cluster. The standard deviation estimate is represented by s, and the specific calculation formula is as follows:

[0063]

[0064] where l i is the sample element to be estimated, and is the sample mean.

[0065] Furthermore, when sorting out the clustering result data, it is usually found that some data have large deviations or are relatively special. Then these data may be outliers. In this embodiment, the Pauta criterion is used to discriminate such data at a relatively large distance to determine whether they are outliers. Among them, the calculation formula for discrimination by the Pauta criterion is as follows:

[0066] |x v - x| ≥ 3s

[0067] where x is the average value under a certain criterion such as the closest distance to the cluster center, and x v is the distance of the detected data from the cluster center.

[0068] Thus, among all the data at a relatively large distance for discrimination, if the above relationship can be met, it is determined as an outlier point, and the determined outliers can be removed.

[0069] In an embodiment of the present application, outliers in historical wind and light output data can also be removed through the chi-square test algorithm. Specifically, the chi-square test is a hypothesis testing method for count data. It belongs to the category of non-parametric tests and is mainly used for the association analysis of two or more sample sizes (constituent ratios) and two categorical variables. In this embodiment, the basic idea of the test is as follows: First, assume that H0 holds. Based on the previously calculated χ 2 value, which represents the degree of deviation between the observed value and the theoretical value. According to the χ 2 distribution and degrees of freedom, the probability P of obtaining the current statistic and more extreme cases under the condition that the H0 hypothesis holds can be determined. If the P value is very small, it indicates that the deviation degree between the observed value and the theoretical value is large, and the null hypothesis should be rejected, indicating that there are significant differences between the comparison data. Otherwise, the null hypothesis cannot be rejected, and it cannot be considered that there is a difference between the actual situation represented by the sample and the theoretical hypothesis. Among them, the calculation formula for the χ 2 statistic is as follows:

[0070]

[0071] where A iis the observed frequency at level i, E i is the expected frequency at level i, n is the total frequency, p i is the expected frequency at level i. i is the expected frequency E at level i equals the total frequency n times the expected probability p at level i i , and K is the number of cells.

[0072] Specifically, when n is relatively large, χ 2 statistic approximately follows a chi-square distribution with k - 1 degrees of freedom. Since each sample distance approximately follows a chi-square distribution with 1 degree of freedom, X can be calculated under the condition of a confidence level of 99.5% 2 p，α , when D M (x) > X 2 p,α , then the sample is judged as an outlier.

[0073] It should be noted that in actual data elimination, any of the above methods can be used to eliminate outliers, or multiple methods can be combined for data elimination. For example, the above two elimination methods can be performed sequentially to ensure the comprehensiveness of outlier elimination.

[0074] Furthermore, after eliminating outliers, there will be missing values null in the historical wind power output data. To improve the integrity of the data and the accuracy of subsequent predictions, the generated missing data is filled.

[0075] In an embodiment of the present application, for data filling of missing values and outliers, the mean filling method can be selected for data filling. Specifically, for the missing values in each column, the mean of the current column is filled. The calculation formula is as follows:

[0076]

[0077] where a is the mean to be filled, x i is the normal value in the current column data, and m is the number of normal values in the current column data.

[0078] In an embodiment of the present application, data filling can also be performed based on the Mahalanobis distance method. In this embodiment, data filling based on the Mahalanobis distance method selects the nearest neighbor data through the Mahalanobis distance between data, and applies the obtained estimated value to the subsequent estimation result process. Then, the concept of entropy value in information theory is used to calculate the weighted coefficient of the nearest neighbor to obtain the filling value of the missing data.

[0079] To more clearly illustrate the specific implementation method of data filling based on the Mahalanobis distance method in the present application, the following is an exemplary description of a data filling method proposed in this embodiment. Figure 3The following is a flowchart of a method for filling missing data based on the Mahalanobis distance method proposed in an embodiment of this application. As Figure 3 shown, the method includes the following steps:

[0080] Step S301: Calculate the Mahalanobis distance value between the missing data and other data except the missing data.

[0081] It should be noted that when filling data in this embodiment, it can be for filling all historical wind and light output data, or after clustering in the above steps, in each cluster in turn, filling the missing data for the data in that cluster.

[0082] Specifically, the Mahalanobis distance represents the covariance distance of data and is a method for detecting outliers in multivariate data. The calculation method of the Mahalanobis distance is as follows:

[0083] For a multi-dimensional data: x = (x1, x2, x3,... xp) T , a set of averages u = (u1, u2, u3,... un) T , and the covariance matrix is ∑, then the distance from point x to the data set can be calculated by the following formula:

[0084]

[0085] For two given variables, x = (x1, x2,... xn) and y = (y1, y2... yn), the Mahalanobis distance between the two points can be calculated by the following formula:

[0086]

[0087] In this embodiment, through the above calculation method of the Mahalanobis distance value, the Mahalanobis distance value between the missing data and other data except the missing data can be calculated. Among them, when filling all historical wind and light output data, other data except the missing data are other non-missing data, and when filling the missing data for the data in each cluster, other data except the missing data are other non-missing data in the current cluster.

[0088] Step S302: Determine a preset number of nearest neighbor data of the missing data according to all the Mahalanobis distance values, and perform normalization calculation on the Mahalanobis distances of the preset number of nearest neighbor data to obtain a normalized distance value.

[0089] Specifically, for each missing data, all the Mahalanobis distance values corresponding to the current missing data are compared pairwise to determine the magnitudes of the respective Mahalanobis distance values, which are then sorted in ascending order. A preset number of the closest data are selected as the nearest neighbor data, where the preset number k can be determined based on actual factors such as prediction accuracy.

[0090] Then, the k nearest neighbor Mahalanobis distances obtained are normalized, and the normalization calculation formula is as follows:

[0091]

[0092] where d i represents different nearest neighbor Mahalanobis distances, and i = 1, 2, 3... k.

[0093] Step S303: Calculate the entropy value of each neighbor data based on the normalized distance value.

[0094] Specifically, for any neighbor data, the entropy value of the i-th neighbor gene is calculated through the following formula:

[0095] h i = -mp i lnp i

[0096] The entropy values of each neighbor data can be calculated sequentially through the above formula.

[0097] Step S304: Calculate the coefficient of variation degree of the corresponding neighbor data based on each entropy value.

[0098] Specifically, calculate the coefficient of variation degree of the i-th neighbor value. Since 0 ≤ hi ≤ 1, according to the principle that the magnitude of the entropy value is opposite to its degree of variation, the coefficient of variation degree of the i-th similar value is defined as:

[0099] v i = 1 - h i , i = 1, 2,... k

[0100] The coefficients of variation degree of each neighbor data can be calculated sequentially through the above formula.

[0101] Step S305: Calculate the weighting coefficient of the corresponding neighbor data based on each coefficient of variation degree.

[0102] Specifically, for any neighbor data, the weighting coefficient of the i-th neighbor value is calculated through the following formula:

[0103]

[0104] where i = 1, 2, 3... k.

[0105] The weighted coefficients of each neighbor data can be calculated successively through the above formula. The smaller the variation degree of a certain similar gene, the greater the deterministic information it contains, and the greater the corresponding weighted coefficient in the prediction, and vice versa.

[0106] Step S306: Calculate the estimated value of the missing data according to the weighted coefficient of each neighbor data and the value of each neighbor data.

[0107] Specifically, calculate the predicted value of the current missing data, where the missing value can be obtained by the following formula;

[0108]

[0109] where x i is the expression level value at the position corresponding to the missing value in the similar gene, and the calculated value is the estimated value of the missing data in the target gene.

[0110] Thus, the present application realizes the collection and preprocessing of historical wind and light output data, and a standardized database containing the wind and light resources of the power station and the operation data of the units can be established according to the processed prediction data.

[0111] Step S102: Construct a wind and light output prediction model with multi-model fusion, and predict the upper and lower limit thresholds of the wind and light output through the wind and light output prediction model based on the prediction data.

[0112] Among them, the wind and light output prediction model with multi-model fusion refers to a model that fuses multiple single prediction models or multiple prediction methods, and is used to predict the output situation of the combined power station on the prediction day.

[0113] Specifically, first construct a wind and light output prediction model with multi-model fusion, then input the prediction data obtained in the previous step into the model, and estimate the upper limit threshold and the lower limit threshold of the real-time wind and light output of the new energy subsystem at a future moment through relevant estimation methods.

[0114] The process of constructing a wind and light output prediction model with multi-model combination will be described below. As a possible implementation manner, the wind and light output prediction model with multi-model fusion includes: a wind and light output combined prediction model and a wind and light output prediction model based on similar days.

[0115] To more clearly illustrate the specific implementation manner of constructing the wind and light output combined prediction model in the present application, an exemplary description will be given below by taking a model construction method proposed in this embodiment. Figure 4 For the flowchart of a method for constructing a wind and light output combined prediction model proposed in an embodiment of the present application, as Figure 4 shown, the method includes the following steps:

[0116] Step S401, construct multiple single wind and light output prediction models, where the single wind and light output prediction models include: Autoregressive Integrated Moving Average model (ARIMA), Back Propagation (BP) neural network model, and XGBoost regression prediction model.

[0117] Specifically, first construct models for single wind and light output prediction methods based on intelligent algorithms. The single wind and light output prediction models include: Autoregressive Integrated Moving Average model (ARIMA), Back Propagation (BP) neural network model, and XGBoost regression prediction model.

[0118] Among them, the Autoregressive Integrated Moving Average model (ARIMA) is one of the implementation methods of time series prediction analysis. The time series method is a method for predicting the future change trend of a variable based on its long historical sequence values. In this embodiment, since this method only requires historical data of wind and light output, it can effectively avoid the amplification of errors by the power curve, the model is simple and the prediction accuracy is good, so it can be used for ultra-short-term and short-term predictions. Subject to the influence of meteorological factors, the wind and light power generation shows non-stationary characteristics. For non-stationary time series, the commonly used time series model is the ARIMA(p,d,q) model, where AR is the autoregressive model; p represents the autoregressive term; MA is the moving average model; q represents the moving average term; d is the number of differences required to convert the non-stationary time series into a stationary time series. The specific expression is as follows:

[0119] Φ(B)(1 - B) d x t =Θ(B)ε t

[0120] Among them, Φ(B) is the autoregressive operator, p is the order of the autoregressive model; (1 - B) d is the difference operator, d is the order of differences; x t is the short-term photovoltaic output time series; Θ(B) is the moving average operator, Θ(B) = 1 - θ1B - θ2B 2 -…θ q B q ; q is the order of the moving average model; ε t is the residual sequence.

[0121] When constructing the ARIMA model, it includes steps such as the stationarity test of the time series, stationary processing, model order determination, parameter estimation, and model evaluation. First, the unit root and white noise are used to test the stationarity of the time series, the differencing method is used for stationary processing, and the optimal order is determined through the Bayesian Information Criterion (BIC); then, parameter estimation is carried out by the least squares method; finally, residual tests are performed to complete the prediction of wind and light output based on the ARIMA model.

[0122] The Back Propagation (BP) neural network model is a multi-layer feedforward neural network model based on the error backpropagation algorithm, including an input layer, a hidden layer, and an output layer. It has the characteristics of simple structure, convenience, and high efficiency, and can be applied to the prediction of new energy output. The use of the BP model includes two processes: the forward propagation of signals and the backpropagation of errors. Among them, forward propagation is to apply the input signal to the output layer through the calculation of the hidden layer to generate an output signal; when the difference between the actual output and the predicted output is too large, it enters the error backpropagation process. By means of iterative processing, the connection weights between neurons are gradually optimized, and finally, the error between the actual output and the predicted output tends to be stable and minimized. In the embodiment of the present application, a 3-layer BP neural network is constructed. Among them, the number of nodes in the input layer mainly depends on the number of main meteorological elements selected by the PCC method, the output layer has 1 node corresponding to the photovoltaic power generation, and the number of nodes in the hidden layer is determined by the following formula:

[0123]

[0124] where a is the number of nodes in the hidden layer; b is the number of nodes in the input layer; c is the number of nodes in the output layer.

[0125] The XGBoost regression prediction model uses the XGBoost algorithm for prediction. The full name of the XGBoost algorithm is extreme Gradient Boosting, which is improved on the basis of the gradient boosting decision tree (GBDT) algorithm. It is an ensemble algorithm formed by combining its base functions and weights according to the idea of the ensemble algorithm, and has the characteristics of fast speed, high efficiency, and strong generalization ability, significantly improving the accuracy of regression prediction. XGBoost is composed of multiple classification and regression trees (CART) integrated. Starting from initially building a tree (weak classifier), and then iterating, adding a tree in each iteration process, and finally obtaining a strong classifier integrated by multiple tree models. Assuming that the ensemble model has a total of K decision trees, the final prediction result obtained by the sample i through the model is as follows:

[0126]

[0127] The XGBoost algorithm consists of three parts: objective function design, optimal tree structure search, and optimal branch search. By introducing model complexity to measure the computational efficiency of the algorithm, a balance between model performance and computational speed is achieved. The objective function is composed of the loss function l of the model and the regularization term Ω that suppresses model complexity, and its objective function is expressed as follows:

[0128]

[0129] Among them, i is the sample type; m is the total amount of data of the k-th number imported, and k is the total amount of trees established; is the predicted value; y i is the true value.

[0130] The idea of the XGBoost algorithm is that the new decision tree formed in each iteration will be superimposed on the model formed previously. Therefore, the first term loss function in the objective function has a certain correlation with all the tree structures that have been established, and the relevant formula is as follows:

[0131]

[0132] The complexity of the second term tree of the objective function is obtained by adding the number of leaves and the regularization term, and its calculation formula is as follows:

[0133]

[0134] Among them, T is the number of leaf nodes; j is the number of copies of each leaf node; γ and λ are the coefficients that control the number and number of copies of leaf nodes respectively.

[0135] In this embodiment, XGBoost finds the optimal tree structure by performing a second-order Taylor expansion on the loss function to obtain higher solution accuracy. The process of finding the optimal branch is as follows: (1) Define the quality of the structure using the objective function; (2) Calculate the difference in information entropy before and after branching; (3) Select the feature branch point with the largest Gain for branching; (4) When Gain is less than a certain threshold, the tree stops growing. As an ensemble model of trees, the solution process of XGBoost usually uses the most common greedy algorithm in decision tree calculation for calculation. By calculating the value after splitting minus the value before splitting, the gain is obtained. Define IL and IR as the sets of left and right leaf nodes after partitioning, that is, I is determined according to IL and IR, and the gain formula is as follows:

[0136]

[0137] Among them, G L is the first-order derivative of the left sub-expression; H Lis the second derivative of the left minor; G R is the first derivative of the right minor; H R is the second derivative of the right minor.

[0138] Step S402: Taking the weighted sum of the predicted outputs of all single wind-solar power output prediction models and the minimum absolute deviation between the weighted sum and the actual wind-solar power output as the objective function, and solving the objective function under the constraint condition that the weights satisfy normalization to obtain the weight values of each single wind-solar power output prediction model.

[0139] It can be understood that although each of the above single prediction models has a certain accuracy, it is still impossible to fully identify the variation law and information of new energy output. Therefore, in this application, by organically synthesizing the prediction results of multiple methods, a combined prediction model is constructed, which can comprehensively reflect the variation law of the system.

[0140] Specifically, based on the optimization theory, taking the minimum absolute deviation between the weighted sum of the output predictions of the above three single prediction models and the actual output as the objective function, and solving the optimization model under the constraint condition that the weights satisfy normalization to obtain the optimal weight values of each single model. The objective function is shown in the following formula:

[0141]

[0142] Among them, the constraint conditions are shown as follows:

[0143] ω2 > 0

[0144] ω2 > 0

[0145] ω3 > 0

[0146] ω1 + ω2 + ω3 = 1

[0147] Among them, f is the absolute deviation between the predicted value of the combined model and the actual output; a(i) is the prediction result of the time series analysis model; ω1 is the weight of the time series analysis model; b(i) is the prediction result of the BP neural network model; ω2 is the weight of the BP neural network model; c(i) is the output result of the XGBoost model, ω3 is the weight of the XGBoost model; e(i) is the actual output.

[0148] Furthermore, solve the above optimization model, that is, solve the objective function under the above constraint conditions to obtain the optimal weight values of each single prediction model.

[0149] Step S403: Construct a combined wind-solar power output prediction model according to each single wind-solar power output prediction model and its weight value.

[0150] Specifically, each single prediction model is multiplied by the corresponding weight value and then added together, finally constructing a combined prediction model that comprehensively reflects the characteristics of the fluctuations in the wind and light output and has a higher prediction accuracy.

[0151] Thus, a combined prediction model for the wind and light output is constructed.

[0152] Based on the above embodiments, the process of constructing a wind and light output prediction model based on similar days will be described below. To more clearly illustrate the specific implementation method of constructing this model in the present application, the construction method of another model proposed in this embodiment will be used for exemplary illustration. Figure 5 The flowchart of a method for constructing a wind and light output prediction model based on similar days proposed in an embodiment of the present application is as Figure 5 shown, and this method includes the following steps:

[0153] Step S501, perform a correlation analysis on the output of the new energy system and multiple meteorological elements within a preset time period through the Pearson correlation coefficient method, and determine the meteorological variables whose correlation with the output within the preset time period is greater than the preset correlation threshold.

[0154] Specifically, this embodiment constructs the model and predicts the output according to the process of "element identification - similar day determination - output prediction - comparative analysis". First, the Pearson correlation coefficient method is used to perform a correlation analysis on the meteorological elements and the output of the new energy system within a preset time period, and the factors that have a greater impact on the output are extracted. Among them, the output within the preset time period is short-term output, and this preset time period represents a relatively short time length of the output, for example, it can be within one day, etc.

[0155] Among them, the Pearson correlation coefficient (Pearson Correlation Coefficient, abbreviated as PCC), also called the linear correlation coefficient, can be used to measure the linear correlation degree between variables. The multiple meteorological elements in the present application include multiple factors related to the meteorological environment such as wind speed, wind direction, temperature, and light intensity.

[0156] In the embodiment of the present application, the formula for calculating the correlation is as follows:

[0157]

[0158] Among them, r is the correlation coefficient between variables; Y is the short-term output of wind and light; X is the meteorological factor affecting the short-term output of wind and light; N is the number of data included in X and Y.

[0159] In this embodiment, the value range of the correlation coefficient r obtained by the PCC method is [-1, 1]. Positive and negative values respectively indicate that the relationship between two variables is positive correlation or negative correlation. The larger the absolute value, the stronger the correlation. Furthermore, in this embodiment, the correlation coefficients between the above-mentioned various meteorological elements and the short-term output are calculated and compared with a preset correlation threshold in sequence. The correlation threshold is a threshold for determining the degree of correlation between meteorological elements and output. If the correlation coefficient is greater than the correlation threshold, it is determined that the meteorological element is a meteorological variable with a greater impact on the output.

[0160] Step S502: Cluster multiple historical days and the prediction day by the transitive closure method, and determine similar days related to the meteorological variables of the prediction day from the multiple historical days according to the clustering result.

[0161] Among them, the transitive closure method is a method that constructs a fuzzy similarity matrix according to the attributes of the research object itself, and determines the classification relationship based on different membership degrees on this basis. The historical day is the date on which the output of historical wind and light has occurred in the historical wind and light output data, and the prediction day is a future day on which the wind and light output is predicted on that day.

[0162] Specifically, this application clusters historical days and the prediction day by the transitive closure method, and finds similar days highly correlated with the meteorological variables of the prediction day according to the clustering result.

[0163] Figure 6 As shown in the flowchart of a clustering method based on the transitive closure method proposed in the embodiment of this application, Figure 6 the method includes the following steps:

[0164] Step S601: Obtain multiple characteristic indicators of each historical day, and construct a characteristic indicator matrix corresponding to all historical days.

[0165] Specifically, let X = {x1, x2,..., x n} be the historical days to be classified, n represents the number of historical days, and each historical day x i corresponds to m characteristic indicators, that is, x i can be represented by the m-dimensional characteristic indicator vector (x i1 , x i2 ,..., x im ), (i = 1, 2,..., n). x ij represents the jth characteristic indicator of the ith historical day, and j belongs to 1 to m. Finally, the following characteristic indicator matrix is obtained:

[0166]

[0167] Step S602: Perform data standardization processing on the characteristic indicator matrix.

[0168] Specifically, since the orders of magnitude and dimensions of the m characteristic indicators are different, it may affect the classification effect. Therefore, this application adopts the maximum normalization method to standardize the data of the indicators to ensure that each indicator value falls within the range of [0, 1]. The specific operation process is to determine the maximum value M of the j-th column of X*. j = max{x 1j , x 2j , …, x nj}; calculate x' ij = x ij / M j to ensure that the indicator value falls within the range of [0, 1].

[0169] Step S603: For the data after normalization processing, calculate the similarity between every two data to obtain a fuzzy similarity matrix.

[0170] Specifically, based on the standardized data, the similarity between x i = (x i1 , x i2 , …, x im ) and x j = (x j1 , x j2 , …, x jm ) is denoted as r ij ∈[0, 1], and the fuzzy similarity matrix R = (r ij ) n×n is obtained. In this embodiment, the calculation methods that can be adopted for r ij include the dot product method, the cosine of the included angle method, the correlation coefficient method, the maximum-minimum method, the arithmetic mean minimum method, and the geometric mean minimum method, etc. As an example, this application adopts the maximum-minimum method to calculate the similarity, and the specific calculation formula is as follows:[[]]

[0171]

[0172] Step S604: Calculate the transitive closure corresponding to the fuzzy similarity matrix.

[0173] Specifically, since the fuzzy similarity matrix R does not necessarily have transitivity, a transitive closure needs to be constructed from R as the fuzzy equivalence matrix of R for dynamic clustering. In specific implementation, on the basis of successively calculating R 2 , R 4 , …, when R 2k = R 2k+1 appears for the first time, R 2k is the transitive closure corresponding to the fuzzy similarity matrix, and the elements of the transitive closure are sorted from large to small, that is, λ1 > λ2 > ….

[0174] Step S605: Calculate the cut matrices corresponding to all elements in the transitive closure, and classify them according to the similarity values between the data in the cut matrices.

[0175] Specifically, calculate the cut matrix R corresponding to all λ. λ Preset a value λ. k For the elements greater than or equal to λ in the transitive closure t(R), take 1, and for the elements less than λ, k take 0, then the cut matrix corresponding to λ can be obtained. In the cut matrix, if r k = 1, then the corresponding x k and x ij are classified into the same category, and thus the classification result at the λ i level is obtained. j k level is obtained.

[0176] Thus, through classification by the transitive closure method, similar days related to the meteorological variables on the prediction date are determined according to the classification results.

[0177] Step S503: Use the output data and meteorological data of the similar days as training data to train any single wind-solar power output prediction model, and obtain a wind-solar power output prediction model based on the similar days.

[0178] Specifically, use the meteorological and output data of the similar days as training samples to train any single wind-solar power output prediction model in the above embodiments, and construct a combined wind-solar power output prediction model based on the similar days, thereby significantly improving the prediction accuracy.

[0179] In an embodiment of the present application, two or three of the above single wind-solar power output prediction models can also be combined. After obtaining a combined wind-solar power output prediction model, use the above training data to train the combined wind-solar power output prediction model to obtain a combined wind-solar power output prediction model for the similar days.

[0180] Furthermore, after establishing wind-solar power prediction models at different time scales based on statistical machine learning methods in the present application, methods such as quantile regression and Gaussian process can be used to estimate the upper and lower limit thresholds of the real-time wind-solar power output, so as to realize the prediction of the output power of the new energy subsystem at a certain time.

[0181] Thus, the present application uses statistical methods to analyze historical wind-solar power output data, obtains the spatio-temporal distribution law of wind-solar power output, and based on the analysis of the wind-solar power output characteristics in typical periods, establishes a fluctuation process model of wind-solar power output, thereby realizing the prediction of wind-solar power output.

[0182] Step S103: Generate a storage power scheduling strategy based on the energy storage capacity of the energy storage system, and generate a power generation index tracking strategy for the combined power station in combination with the upper and lower limit thresholds and the storage power scheduling strategy.​

[0183] Specifically, the power generation index of the combined power station is determined by combining the wind and light output prediction results of the new energy substation and the energy storage output capacity, so as to determine the energy storage power generation index under multiple objectives. When determining the power generation index, after predicting the upper and lower limit thresholds of the wind and light output, it is also necessary to combine the operation mode of the energy storage power station to realize the intelligent tracking of the power generation index of the energy storage and renewable energy combined power station.

[0184] Among them, the operation of the energy storage system is divided into the following three modes: First, the medium energy mode Mod-A. In this mode, the upward regulation ability and downward regulation ability of the energy storage system are close to each other, which is suitable for participating in the frequency modulation auxiliary service market. Second, the high energy mode Mod-B. In this mode, the energy storage system in the high power mode has a long discharge potential, which is suitable for participating in the energy market to provide energy services. Third, the low energy mode Mod-C. In this mode, it is difficult for the energy storage to participate in the energy market to provide sufficient electrical energy, and it is also difficult to respond to the frequency modulation requirements of the frequency modulation auxiliary service market, and the upward regulation capacity is seriously insufficient.

[0185] In an embodiment of the present application, the operation mode of the energy storage of the combined power station is determined according to the size of the energy storage capacity of the energy storage system. Thus, the following energy storage power scheduling strategy of the photovoltaic and energy storage combined power station can be obtained. The energy storage power scheduling strategies in the three modes are are the upper and lower limits of the energy storage capacity state in the frequency modulation mode. Within this range, the energy storage is allowed to enter the frequency modulation mode, η b is the charge-discharge efficiency of the energy storage, is the rated energy storage capacity.

[0186] It should be noted that there are differences in the time scales in the actual operation processes of the energy market and the frequency modulation market. The scheduling time scale of the energy market is generally 1 h, while the time scale of the frequency modulation signal in the frequency modulation market is in seconds or minutes, and the scheduling moment of the frequency modulation capacity is 1 h. Therefore, in this embodiment, 1 h is used as the scheduling period for the energy storage capacity scheduling, which is consistent with the energy market, and multiple groups of frequency modulation signals δ f ∈[-1,1] are continuously received within 1 h. After all the frequency modulation signals are executed, the final energy storage capacity state is obtained.

[0187] In an embodiment of the present application, the generated energy storage power scheduling strategy includes: judging the operation mode of the energy storage system according to the current energy storage capacity of the energy storage system; performing corresponding charge and discharge operations in different operation modes, and calculating the real-time energy storage capacity after responding to the charge and discharge operations; based on the real-time energy storage capacity, repeating the judgment of the operation mode of the energy storage system according to the scheduling requirements until the scheduling task is completed.

[0188] To more clearly illustrate the energy storage power dispatching strategy of the present application, the following uses a specific embodiment of the dispatching strategy for exemplary illustration. Figure 7 The flowchart of an energy storage power dispatching strategy proposed in an embodiment of the present application is as Figure 7 shown, including the following steps:

[0189] Step S701: Obtain the current energy storage capacity of the energy storage system, and determine whether the current energy storage capacity is within the energy storage capacity range in the frequency regulation mode. If it is, execute Step S702. If not, further determine whether the current energy storage capacity is greater than the upper limit of the energy storage capacity range. If so, execute Step S703. If not, execute Step S704.

[0190] Specifically, input the energy storage capacity where the subscript i represents the i-th energy storage controller, and t represents the t-th moment, and determine whether the energy storage capacity satisfies If it is satisfied, enter Step S702. Otherwise, determine whether the energy storage capacity satisfies If it is satisfied, enter Step S703. Otherwise, enter Step S704.

[0191] Step S702: Control the energy mode during the operation of the energy storage system, respond to the frequency regulation market signal, and calculate the energy storage capacity state after the response.

[0192] Specifically, the energy storage receives the frequency regulation market signal δ f ∈[-1,1], and the change in the energy storage capacity in response to the frequency regulation signal under the unit charge-discharge power is (δ f+ η b +δ f- η b )Δt. Then the energy storage capacity state after the response is shown in the following formula: After completion, enter Step 705.

[0193] Step S703: Control the energy storage system to operate in the high energy mode, determine the target discharge mode by comparing the deviation assessment cost reduction benefit and the electricity energy market benefit, and calculate the energy storage capacity state after discharge.

[0194] Specifically, by comparing the deviation assessment cost reduction benefit and the electricity energy market benefit, select the discharge mode with higher benefit. The energy storage capacity state after the response is After completion, enter Step 705.

[0195] Step S704: Control the energy storage system to operate in the low energy mode, control the energy storage system to absorb the positive deviation between the day-ahead declared electricity quantity of the combined power station and the actual output, and calculate the energy storage capacity state after charging.

[0196] Specifically, the energy storage system intelligently absorbs the positive deviation between the declared power generation of the PV power station for the day ahead and the actual power output, and the change in the energy storage capacity can be determined by the following formula: Proceed to step 705 after completion.

[0197] Step S705: Obtain the capacity status after the charge and discharge behavior of the energy storage system under each operating mode, and repeatedly execute step S701 according to the requirements of the total dispatching period until the energy storage dispatching task is completed.

[0198] Specifically, output the capacity status after the charge and discharge behavior of the energy storage system under the corresponding mode According to the requirements of the total dispatching period, repeat step S701 to perform a new round of operating mode judgment and execute the corresponding operating mode. Furthermore, iterate the above charge and discharge process until the energy storage dispatching task ends.

[0199] Thus, after estimating the upper and lower limits of the wind and light power output through the wind and light power prediction model, the state of charge of the energy storage system in the combined power station under different working conditions is also calculated and the corresponding operating range is determined. The upper and lower limit thresholds of the predicted wind and light power output are used as the index limit values for the power generation of the new energy subsystem, and the energy storage power scheduling strategy is used to control the operation of the energy storage system during the power generation process, obtaining an intelligent tracking strategy for the power generation index of the energy storage and renewable energy combined power station.

[0200] Step S104: Construct an energy storage system power optimization allocation model that simultaneously meets the requirements of power grid frequency regulation and power generation index tracking strategy, generate an optimized control instruction for the combined power station according to the energy storage system power optimization allocation model, and execute the optimized control instruction.

[0201] Specifically, analyze the technical requirements of the energy storage system for different operating scenarios of the energy storage and renewable energy combined power station, and establish an energy storage system power optimization allocation model that meets the requirements of simultaneously participating in the secondary frequency regulation of the power grid and tracking the power generation index.

[0202] As an example, the energy storage system power optimization allocation model can be a spot market revenue model of the combined power station under multiple constraint conditions, that is, a multi-objective optimal revenue model of the combined power station. In this example, the spot market revenue model of the combined power station is shown in formula (1), and the multiple constraint conditions are shown in the following formulas (1) to (11):

[0203]

[0204]

[0205]

[0206]

[0207]

[0208]

[0209]

[0210]

[0211]

[0212]

[0213] s min ≤s i,t ≤s max (11)

[0214] Among them, in formula (1) represents the revenue from providing frequency regulation capacity at the i-th measurement device in the photovoltaic-storage integrated power station at the t-th moment under Mod-A mode; represents the revenue from providing electrical energy in the photovoltaic-storage integrated power station under Mod-B mode or the revenue from reducing the deviation assessment cost by discharging to make up for the deviation of new energy output; represents the revenue from providing electrical energy in the photovoltaic-storage integrated power station under Mod-C mode or the revenue from reducing the deviation assessment cost by discharging to make up for the deviation of new energy output; represents the revenue from reducing the deviation assessment cost by charging to make up for the deviation of new energy output in the photovoltaic-storage integrated power station under Mod-C mode; C i,t is the operating cost of the energy storage system. Since the operating mode of the energy storage is discrete, unknown, and time-varying, discrete integer variables κ i,t are introduced in equations (2)-(9). The source of this variable is the binary coding of the energy storage operating mode. Assuming the introduction of auxiliary variables then the discrete variable κ i,t has the following logical relationship with the auxiliary variable Among them, the auxiliary variables respectively determine the selection of energy storage operating modes Mod-A, Mod-B, and Mod-C. The three are mutually exclusive, and only one variable can be 1, that is, it satisfies Subsequently, the auxiliary variable is the dependent variable of the variable that is, it satisfies Its physical meaning is that when it means that the energy storage enters Mod-B mode, and at this time the energy storage control module has two options: 1) Use high capacity to participate in the spot electrical energy market to provide electrical energy services; 2) Use high capacity to discharge to make up for the deviation of new energy output. That is, it corresponds to the above two options. For example, assume that the energy storage capacity status satisfies at this time According to the energy storage operation strategy described in Section 1, the energy storage enters the Mod-B mode at this time. The energy storage module selects to participate in the spot electricity energy market behavior with higher benefits. According to the logical relationship, κ i,t = 5.

[0215] In constraint (2), represents the number of energy storage control modules entering the Mod-A mode. Each energy storage control module controls an energy storage device with a maximum capacity of p max . That is, it means that there are energy storage control modules executing the Mod-A mode at time t, and each energy storage control module selects to provide capacity of energy storage to participate in the Mod-A mode, providing a total of frequency modulation capacity. Φ per represents the frequency modulation performance parameter of the PV and energy storage power station in the frequency modulation market. λ t is the regulation mileage ratio, ρ rp is the clearing price of the frequency modulation market performance, ρ rc is the clearing price of the frequency modulation market capacity, ρ AoD is the penalty cost for the deviation assessment of new energy output. and represent the day-ahead declared output and real-time output of the photovoltaic. δ is the new energy deviation assessment threshold. Deviations below this threshold are not assessed, and deviations exceeding this threshold are subject to assessment and penalty.

[0216] Similarly, in constraint (3), represents the number of energy storage control modules entering the Mod-B mode. represents the capacity of the energy storage control module to provide electricity energy services. represents the clearing price of the electricity energy market. Constraint (4) represents the calculation formula for the benefits brought by the energy storage choosing to discharge to make up for the new energy deviation and thus reduce the deviation assessment cost. Constraint (5) represents the calculation formula for the benefits brought by the energy storage choosing to provide charging capacity to make up for the new energy deviation and thus reduce the deviation assessment cost. Constraint (6) is the energy storage operation cost. and represent the charging and discharging costs of the energy storage at the i-th measurement device. Note that there is a 1 / 2 coefficient in front of it. The reason is that there are multiple frequency modulation signals adjusting the capacity within time t. Assuming that the positive and negative frequency modulation signals are roughly the same in probability distribution, the energy storage device discharges for half of the time and charges for half of the time.

[0217] Constraint (7) is the change in the state of charge (SOC) of the energy storage device after participating in the frequency regulation market at time t, where and represent the positive and negative values of the frequency regulation signal, represents the charge-discharge efficiency of the energy storage device, represents the number of segments of the frequency regulation signal within time t, the duration of each segment of the signal is, and is the rated capacity of the energy storage device. Similarly, constraints (8) and (9) are the SOC changes of the energy storage device after executing Mod-B mode and Mod-C mode. Constraint (10) represents the maximum and minimum capacity limits provided by the energy storage device under each mode, and constraint (11) represents the change limit of the energy storage SOC.

[0218] Furthermore, an optimized control instruction for the combined power station is generated according to the above-mentioned energy storage system power optimization allocation model, and the combined power station is controlled to execute the optimized control instruction.

[0219] For example, the optimized control instruction includes the charge-discharge instruction of the energy storage system and the Automatic Generation Control (AGC) instruction of the new energy power generation subsystem, and each type of instruction is sent to the corresponding subsystem, so that the subsystems in the power station execute the received instructions. Thus, the power optimization control of the energy storage system with the goal of maximizing the comprehensive benefit is realized, that is, the operation optimization control of the energy storage and new energy combined power station that meets the grid frequency regulation requirements.

[0220] In summary, the joint power station operation optimization method based on high-precision wind and light output prediction in the embodiments of the present application first analyzes and evaluates the multi-time and multi-space scale operation characteristics of renewable energy power generation, including collecting the wind and light output data of the power station, establishing a standardized database containing the wind and light resources of the power station and the operation data of the units, and using statistical methods to analyze the wind and light output data to obtain the spatio-temporal distribution law of the wind and light output. Then, intelligent tracking of the power generation indicators of the energy storage and renewable energy joint power station is carried out, including studying the dynamic distribution law of the power generation indicators of renewable energy power stations under channel or peak shaving and power limit conditions, establishing wind and light power prediction models at different time scales and estimating the upper and lower limits of the wind and light output, calculating the state of charge and operation interval of the energy storage system in the joint power station under different conditions, and proposing an intelligent tracking strategy for the power generation indicators of the energy storage and renewable energy joint power station. Finally. The optimization control strategy of the energy storage and renewable energy joint power station that meets the grid frequency modulation requirements is studied, including establishing a power optimization allocation model of the energy storage system that meets the requirements of participating in the secondary frequency modulation of the grid and tracking the power generation indicators according to the technical requirements of the energy storage for different operation scenarios of the energy storage and renewable energy joint power station, proposing a power optimization control strategy for the energy storage system considering the maximization of comprehensive benefits, and optimizing the operation of the power station according to this strategy. Thus, this method improves the flexible regulation ability of the source-network-load-storage integration, optimizes the operation control of the energy storage and new energy joint power station on the basis of meeting the grid frequency modulation requirements, improves the flexibility and coordination of the charge and discharge regulation of the energy storage and new energy joint power station, and is conducive to improving the operation efficiency.

[0221] To implement the above embodiments, the present application also proposes a joint power station operation optimization system based on high-precision wind and light output prediction, Figure 8 which is a structural schematic diagram of a joint power station operation optimization system based on high-precision wind and light output prediction proposed in the embodiments of the present application, as Figure 8 shown. The system includes a preprocessing module 100, a prediction module 200, a generation module 300, and an execution module 400.

[0222] Among them, the preprocessing module 100 is used to collect the historical wind and light output data of the new energy and energy storage joint power station, and preprocess the historical wind and light output data to obtain prediction data.

[0223] The prediction module 200 is used to construct a wind and light output prediction model with multi-model fusion, and predict the upper and lower limit thresholds of the wind and light output through the wind and light output prediction model based on the prediction data.

[0224] The generation module 300 is used to generate an energy storage power scheduling strategy based on the energy storage capacity of the energy storage system, and generate a power generation index tracking strategy for the joint power station in combination with the upper and lower limit thresholds and the energy storage power scheduling strategy.

[0225] The execution module 400 is configured to build a power optimization allocation model for an energy storage system that simultaneously meets the power grid frequency regulation and power generation index tracking strategies, generate an optimized control instruction for the combined power station according to the power optimization allocation model of the energy storage system, and execute the optimized control instruction.

[0226] Optionally, in an embodiment of the present application, the preprocessing module 100 is specifically configured to: eliminate outliers in the historical wind and light output data based on the K-means clustering algorithm or the chi-square test algorithm; fill the missing data generated after eliminating the outliers based on the mean filling method or the Mahalanobis distance method.

[0227] Optionally, in an embodiment of the present application, the preprocessing module 100 is specifically configured to: determine the value of the clustering number K through the elbow method; based on the value of the clustering number K, cluster the clustering sample data through the K-means clustering algorithm to generate K clusters; for the clustering result data in each cluster, estimate the standard deviation through the Bessel formula, and discriminate outliers based on the standard deviation estimate value through the 3-sigma rule, and eliminate the outliers from the clustering result data.

[0228] Optionally, in an embodiment of the present application, the preprocessing module 100 is further configured to: calculate the Mahalanobis distance value between the missing data and other data except the missing data; determine a preset number of nearest neighbor data of the missing data according to all the Mahalanobis distance values, and perform normalization calculation on the Mahalanobis distances of the preset number of nearest neighbor data to obtain the normalized distance value; calculate the entropy value of each neighbor data according to the normalized distance value; calculate the variation degree coefficient of the corresponding neighbor data according to each entropy value; calculate the weighted coefficient of the corresponding neighbor data according to each variation degree coefficient; calculate the estimated value of the missing data according to the weighted coefficient of each neighbor data and the value of each neighbor data.

[0229] Optionally, in an embodiment of the present application, the wind and light output prediction model with multi-model fusion includes: a wind and light output combined prediction model and a wind and light output prediction model based on similar days. The prediction module 200 is specifically configured to: build multiple single wind and light output prediction models. The single wind and light output prediction model includes: autoregressive integrated moving average model ARIMA, backpropagation BP neural network model, and XGBoost regression prediction model; take the weighted sum of the predicted outputs of all single wind and light output prediction models and the minimum absolute deviation from the actual wind and light output as the objective function, and solve the objective function under the constraint condition that the weights satisfy normalization to obtain the weight value of each single wind and light output prediction model; build the wind and light output combined prediction model according to each single wind and light output prediction model and its weight value.

[0230] Optionally, in an embodiment of the present application, the prediction module 200 is further configured to: perform a correlation analysis on the output power of the new energy system and multiple meteorological elements within a preset time period by using the Pearson correlation coefficient method, and determine meteorological variables whose correlation with the output power within the preset time period is greater than a preset correlation threshold; cluster multiple historical days and the prediction day by using the transitive closure method, and determine similar days related to the meteorological variables of the prediction day from the multiple historical days according to the clustering result; use the output power data and meteorological data of the similar days as training data to train any single wind-solar output power prediction model, and obtain a wind-solar output power prediction model based on the similar days.

[0231] Optionally, in an embodiment of the present application, the prediction module 200 is specifically configured to: obtain multiple characteristic indicators of each historical day, and construct a characteristic indicator matrix corresponding to all historical days; perform data standardization processing on the characteristic indicator matrix; for the data after the standardization processing, calculate the similarity between every two data to obtain a fuzzy similarity matrix; calculate the transitive closure corresponding to the fuzzy similarity matrix; calculate the cut matrix corresponding to all elements in the transitive closure, and classify according to the numerical value of the similarity between the data in the cut matrix.

[0232] Optionally, in an embodiment of the present application, the generation module 300 is specifically configured to: judge the operation mode of the energy storage system according to the current energy storage capacity of the energy storage system; perform corresponding charge and discharge operations in different operation modes, and calculate the real-time energy storage capacity after responding to the charge and discharge operations; based on the real-time energy storage capacity, repeatedly judge the operation mode of the energy storage system according to the scheduling requirements until the scheduling task is completed.

[0233] It should be noted that the foregoing explanation of the embodiments of the joint power station operation optimization method based on high-precision wind-solar output power prediction also applies to the system of this embodiment, and will not be repeated here.

[0234] In summary, the joint power station operation optimization system based on high-precision wind-solar output power prediction in the embodiments of the present application improves the flexible adjustment ability of the source-network-load-storage integration, performs operation optimization control of the energy storage and the new energy joint power station on the basis of meeting the grid frequency modulation requirements, improves the flexibility and coordination of the charge and discharge regulation of the energy storage and the new energy joint power station, and is beneficial to improving the operation efficiency.

[0235] To implement the above embodiments, the present application also proposes a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the joint power station operation optimization method based on high-precision wind-solar output power prediction as described in any one of the above embodiments.

[0236] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0237] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0238] Any process or method description depicted in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of this application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of this application pertain.

[0239] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing as necessary, and then storing it in a computer memory.

[0240] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0241] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0242] In addition, each functional unit in various embodiments of the present application may be integrated into one processing module, may exist separately as individual physical units, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0243] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A combined power station operation optimization method based on high-precision prediction of wind and light power output, characterized in that The following steps are involved: Collect historical wind and solar power output data of the new energy and energy storage combined power station, and pre-process the historical wind and solar power output data to obtain forecast data; Constructing a wind-solar output prediction model integrating multiple models, and predicting the upper and lower thresholds of the wind-solar output through the wind-solar output prediction model based on the prediction data; Generate an energy storage power dispatching strategy based on the energy storage capacity of the energy storage system, and generate a power generation index tracking strategy for the combined power station by combining the upper and lower thresholds and the energy storage power dispatching strategy; Constructing a power optimization allocation model for the energy storage system that satisfies both the grid frequency regulation and the power generation index tracking strategy, generating an optimization control instruction for the combined power station according to the power optimization allocation model for the energy storage system, and executing the optimization control instruction; The multi-model fusion wind and solar output prediction model includes: a wind and solar output combination prediction model and a wind and solar output prediction model based on similar days. The wind and solar output combination prediction model is constructed, including: Constructing multiple single wind and solar power output prediction models, wherein the single wind and solar power output prediction models include: a differential integrated moving average autoregressive model ARIMA, a back propagation BP neural network model and an XGBoost regression prediction model; Taking the weighted sum of the predicted outputs of all the single wind-solar output prediction models and the minimum absolute deviation from the actual wind-solar output as the objective function, solving the objective function under the constraint condition that the weights satisfy the normalization, and obtaining the weight value of each single wind-solar output prediction model; The wind-solar output combined prediction model is constructed according to each of the single wind-solar output prediction models and their weight values.

2. The method according to claim 1, characterized in that, The preprocessing of the historical wind and solar power output data includes: Eliminate outliers in the historical wind and solar power output data based on a K-means clustering algorithm or a chi-square test algorithm; The missing data generated after removing the outliers are filled based on the mean value filling method or the Mahalanobis distance method.

3. The method according to claim 2, wherein Eliminating abnormal values ​​in the historical wind and solar power output data based on the K-means clustering algorithm includes: The value of the cluster number K is determined by the elbow method; Based on the value of the cluster number K, clustering the cluster sample data is performed using the K-means clustering algorithm to generate K clusters; For the clustering result data in each of the clusters, the standard deviation is estimated by using the Bessel formula, and the outliers are identified by using the Raida criterion according to the estimated standard deviation, and the outliers are removed from the clustering result data.

4. The method according to claim 2, wherein Filling the missing data generated after removing the outliers based on the Mahalanobis distance method includes: Calculate the Mahalanobis distance between the missing data and other data except the missing data; Determine a preset number of nearest neighbor data of the missing data according to all the Mahalanobis distance values, and perform normalized calculation on the Mahalanobis distances of the preset number of nearest neighbor data to obtain a normalized distance value; Calculate the entropy value of each of the neighbor data according to the normalized distance value; Calculate the coefficient of variation of the corresponding neighbor data according to each entropy value; Calculate the weight coefficient of the corresponding neighbor data according to each coefficient of variation; Calculate the estimated value of the missing data according to the weighted coefficient of each piece of the neighbor data and the value of each piece of the neighbor data.

5. The method according to claim 1, characterized in that, Construct the prediction model of the wind-solar power output based on similar days, including: Perform a correlation analysis on the power output of the new energy system and multiple meteorological elements within a preset time period by the Pearson correlation coefficient method, and determine the meteorological variables whose correlation with the power output within the preset time period is greater than a preset correlation threshold; Cluster multiple historical days and the prediction day by the transitive closure method, and determine similar days related to the meteorological variables of the prediction day from the multiple historical days according to the clustering result; Use the power output data and meteorological data of the similar days as training data to train any one of the single wind-solar power output prediction models to obtain the prediction model of the wind-solar power output based on similar days.

6. The method according to claim 5, characterized in that, The clustering of multiple historical days and the prediction day by the transitive closure method includes: Obtain multiple characteristic indexes of each historical day, and construct a characteristic index matrix corresponding to all the historical days; Perform data standardization processing on the characteristic index matrix; For the data after the standardization processing, calculate the similarity between every two pieces of data to obtain a fuzzy similarity matrix; Calculate the transitive closure corresponding to the fuzzy similarity matrix; Calculate the cut matrix corresponding to all the elements in the transitive closure, and classify according to the numerical value of the similarity between the data in the cut matrix.

7. The method according to claim 1, characterized in that, The energy storage power scheduling strategy includes: Judge the operation mode of the energy storage system according to the current energy storage capacity of the energy storage system; Perform corresponding charge and discharge operations in different operation modes, and calculate the real-time energy storage capacity after responding to the charge and discharge operations; Based on the real-time energy storage capacity, repeatedly execute the judgment of the operation mode of the energy storage system according to the scheduling requirements until the scheduling task is completed.

8. A combined power station operation optimization system based on high-precision wind and light output prediction, characterized in that, It includes the following modules: A preprocessing module, which is used to collect the historical wind-solar power output data of the new energy and energy storage combined power station, and preprocess the historical wind-solar power output data to obtain prediction data; A prediction module, which is used to construct a wind-solar power output prediction model with multi-model fusion, and predict the upper and lower limit thresholds of the wind-solar power output through the wind-solar power output prediction model based on the prediction data; A generation module, which is used to generate an energy storage power scheduling strategy based on the energy storage capacity of the energy storage system, and generate a power generation index tracking strategy for the combined power station by combining the upper and lower limit thresholds and the energy storage power scheduling strategy; An execution module, which is used to construct a power optimization allocation model of the energy storage system that simultaneously satisfies grid frequency modulation and the power generation index tracking strategy, generate an optimized control instruction for the combined power station according to the power optimization allocation model of the energy storage system, and execute the optimized control instruction; The wind-solar power output prediction model with multi-model fusion includes: a wind-solar power output combined prediction model and a wind-solar power output prediction model based on similar days. Constructing the wind-solar power output combined prediction model includes: Construct multiple single wind-solar power output prediction models. The single wind-solar power output prediction model includes: an autoregressive integrated moving average model ARIMA, a backpropagation BP neural network model, and an XGBoost regression prediction model; Taking the weighted sum of the predicted outputs of all the single wind-solar power output prediction models and minimizing the absolute deviation from the actual wind-solar power output as the objective function, solve the objective function under the constraint condition that the weights satisfy normalization to obtain the weight values of each single wind-solar power output prediction model; Construct the combined wind-solar power output prediction model according to each single wind-solar power output prediction model and its weight value.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the combined power station operation optimization method based on high-precision wind-solar power output prediction as described in any one of claims 1-7.

Images