Virtual power plant short-term load prediction method, device and equipment

Abstract

The invention discloses a short-term load prediction method, device and equipment for a virtual power plant, and relates to the field of load prediction. According to the method, firstly, a multi-objective optimization function covering prediction precision and model complexity is constructed, then, a composite method of a radial basis function (RBF) neural network and an improved non-dominated sorting genetic algorithm (NSGA-II) is set, and by combining the strong nonlinear mapping capability of the RBF neural network and the multi-objective optimization advantage of the NSGA-II algorithm, the prediction accuracy and the model complexity are improved. The multi-variable and multi-target problems in the virtual power plant load prediction are effectively processed, and the prediction precision and the calculation efficiency of the short-term load prediction of the virtual power plant are improved.

Description

Technical Field

[0001] The present application relates to the field of load forecasting, and in particular to a short-term load forecasting method, device and equipment for a virtual power plant. Background Art

[0002] With the large-scale access of renewable energy, the virtual power plant (VPP, Virtual Power Plant), as a flexible scheduling system integrating various distributed energy sources, loads and energy storage resources, has become an important part of modern power systems. The virtual power plant can not only optimize power dispatching and enhance the reliability of the power grid, but also improve the economy of the power market. However, due to the wide distribution and uncertainty of the resources of the virtual power plant, how to accurately predict the short-term load demand of the virtual power plant has become an urgent problem to be solved.

[0003] Short-term load forecasting is one of the key technologies for the dispatching and market trading of virtual power plants, and it can provide a basis for load balancing in the power market and grid operation. The existing short-term load forecasting methods mainly include time series analysis methods, machine learning methods, and hybrid modeling methods, etc. Among them, traditional time series analysis methods, such as the Auto Regressive Moving Average Model (ARMA) and the AutoRegressive Integrated Moving Average Model (ARIMA), perform poorly in dealing with complex non-linear relationships and cannot make full use of a large amount of historical data and external variables. In recent years, prediction methods based on machine learning, such as the Support Vector Machine (SVM), Neural Network (NN), etc., have been widely used in load forecasting due to their strong non-linear modeling ability and self-learning characteristics. However, these methods are prone to problems such as high computational complexity and insufficient optimization in large-scale data training, and still cannot meet the dual requirements of accuracy and efficiency, especially when facing various complex resources and scheduling strategies of virtual power plants. Summary of the Invention

[0004] The purpose of the present application is to provide a short-term load forecasting method, device and equipment for a virtual power plant, which can improve the forecasting accuracy and calculation efficiency of short-term load forecasting for virtual power plants.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In the first aspect, the present application provides a short-term load forecasting method for a virtual power plant, including:

[0007] Obtain the historical load data sequence of the virtual power plant and multiple load influencing factor sequences;

[0008] Use the load influencing factors in the multiple load influencing factor sequences as the input data of the samples, and use the load data in the historical load data sequence as the short-term load label values of the samples to construct a training set and a test set;

[0009] Construct a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function;

[0010] Use the training set, the test set and the multi-objective optimization function, and adopt the NSGA-II algorithm to train and test the RBF neural network model to obtain a trained RBF neural network model;

[0011] Use the trained RBF neural network model to perform short-term load forecasting on the virtual power plant.

[0012] In a second aspect, the present application provides a virtual power plant short-term load forecasting device. The virtual power plant short-term load forecasting device applies the above-mentioned virtual power plant short-term load forecasting method. The virtual power plant short-term load forecasting device includes:

[0013] A historical load data sequence acquisition module, configured to obtain the historical load data sequence of the virtual power plant and multiple load influencing factor sequences;

[0014] A training set and test set construction module, configured to use the load influencing factors in the multiple load influencing factor sequences as the input data of the samples, and use the load data in the historical load data sequence as the short-term load label values of the samples to construct a training set and a test set;

[0015] A multi-objective optimization function construction module, configured to construct a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function;

[0016] An RBF neural network model training module, configured to use the training set, the test set and the multi-objective optimization function, and adopt the NSGA-II algorithm to train and test the RBF neural network model to obtain a trained RBF neural network model;

[0017] A short-term load forecasting module, configured to use the trained RBF neural network model to perform short-term load forecasting on the virtual power plant.

[0018] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement the steps of the above-mentioned virtual power plant short-term load forecasting method.

[0019] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application:

[0020] The present application provides a short-term load forecasting method, device and equipment for a virtual power plant. First, a multi-objective optimization function covering prediction accuracy and model complexity is constructed, and then a composite method of a radial basis function (RBF) neural network and an improved non-dominated sorting genetic algorithm (NSGA-II) is set. By utilizing the powerful non-linear mapping ability of the RBF neural network and the multi-objective optimization advantage of the NSGA-II algorithm, the multi-variable and multi-objective problems in the load forecasting of the virtual power plant are effectively processed, and the prediction accuracy and calculation efficiency of the short-term load forecasting of the virtual power plant are improved. Brief Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a schematic flowchart of a short-term load forecasting method for a virtual power plant provided by an embodiment of the present application;

[0023] Figure 2 It is a flowchart of the NSGA-II algorithm provided by an embodiment of the present application;

[0024] Figure 3 It is a comparison chart of the effects of different methods provided by an embodiment of the present application;

[0025] Figure 4 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Detailed Embodiments

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0027] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0028] In an exemplary embodiment, as Figure 1 and Figure 2 shown, a short-term load forecasting method for a virtual power plant is provided, including the following steps 101 to 105. Among them:

[0029] Step 101, obtain the historical load data sequence of the virtual power plant and multiple load influencing factor sequences.

[0030] Step 102, use the load influencing factors in the multiple load influencing factor sequences as the input data of the samples, and use the load data in the historical load data sequence as the short-term load label values of the samples to construct a training set and a test set.

[0031] Step 103, construct a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function.

[0032] Step 104, use the training set, the test set, and the multi-objective optimization function, and adopt the NSGA-II algorithm to train and test the RBF neural network model to obtain a trained RBF neural network model.

[0033] Step 105, use the trained RBF neural network model to perform short-term load forecasting on the virtual power plant.

[0034] Implementing the above steps 101 to 105 can improve the prediction accuracy and calculation efficiency of the short-term load forecasting of the virtual power plant.

[0035] In another exemplary embodiment, step 101 of the present application can be implemented based on the following steps 201 - 203.

[0036] Step 201, data collection.

[0037] Collect the historical load data and load influencing factor data of a certain park, including the power load data every 15 minutes, temperature, wind speed, and date type (weekday, weekend, holiday) data, etc., to form an original data set.

[0038] Collect the data related to forecasting, specifically as follows:

[0039] Historical load data: Collect the actual load data per hour of historical data. Let the collected historical load data sequence be L = {l 1 , l 2 , …, l H}, where l h represents the load data at the hth moment, h = 1, 2,..., H, and H is the total length of the data.

[0040] Historical wind speed data: Collect the actual wind speed data every hour for historical data. The collected wind speed data sequence is V = {v 1 , v 2 , …, v H}, where v h represents the wind speed data at the h-th moment, and h = 1, 2, …, H.

[0041] Temperature data: Obtain the corresponding temperature data every hour within the same time period. Let the collected temperature data sequence be T = {t 1 , t 2 , …, t H}, and t h represents the temperature data at the h-th moment.

[0042] Holiday data: Represent the information of whether it is a holiday with binary variables, also corresponding to every hour in the past week. Let the holiday data sequence be H' = {h 1 , h 2 , …, h H}. If h h = 1, it means the h-th moment is a holiday; if h h = 1, it means the h-th moment is not a holiday.

[0043] Step 202, data cleaning.

[0044] 1. Outlier handling.

[0045] For the load data, use the "3σ principle" based on the normal distribution to detect and remove outliers. First, calculate the mean and standard deviation σ p of the load data sequence L. The calculation formulas are as follows:

[0046]

[0047] If a data point l h satisfies , then determine l h as an outlier and remove it from the data set. In addition, it is also feasible to set a reasonable threshold range based on the common sense of the electricity market and past experience to judge outliers. For example, if the electricity price at a certain moment far exceeds or is far lower than the upper and lower limits of the normal fluctuation range of the same period in the history of this market, it can be identified as an outlier for processing.

[0048] For the temperature and wind speed data, outliers can be detected according to the reasonable fluctuation range of local historical temperature and wind speed and the judgment criteria of extreme values in meteorology.

[0049] Since holiday data is a binary variable generated based on clear calendar rules, there are generally no outlier situations. However, it is necessary to ensure its accuracy, which can be guaranteed by checking against the official holiday arrangements.

[0050] 2. Missing value handling.

[0051] When there are missing values in the load data l h fill them with the load values at the same time point in the previous week:

[0052] l h = l h-7×24 ;

[0053] l h-7×24 represents the load data at the same time point one week ago at the hth moment.

[0054] 3. Temperature and wind speed data filling: Use the K-Nearest Neighbors (KNN) filling method to fill in the missing values of the temperature data t h and the wind speed data v h . The core idea of the KNN algorithm is to find the k nearest neighbor samples that are most similar to the missing value sample based on the spatial distance of the sample data, and then predict the missing value according to the values of these k neighbor samples. The formula is:

[0055]

[0056] Where:

[0057] x h : represents the missing value, which is the missing temperature data or wind speed data;

[0058] K h : represents the number of known data of the nearest neighbors of x h ;

[0059] represents the kth h nearest neighbor known data of x h .

[0060] Step 203, data normalization.

[0061] In order to eliminate the adverse effects of data features of different magnitudes on the subsequent neural network training process, speed up the training speed and improve the training effect, it is necessary to perform normalization processing on various types of data collected and cleaned. This application uses the min-max normalization method to map the values of each data feature to a specified interval (usually the [0,1] interval). The calculation formula for the normalization process is as follows.

[0062]

[0063] Among them, is the data after normalization, n' is the original simulation data, and n min is the minimum value of the data, and n max is the maximum value of the data.

[0064] After such a normalization operation, all input data are within the same magnitude range, which is convenient for subsequent processing by the RBF neural network.

[0065] In another exemplary embodiment, in step 102 above, the training set is constructed using the data of the first 12 months of the data after normalization, and the test set is constructed using the data of the last 1 month of the data after normalization. The samples are constructed in a sliding window manner, and x h ={v' h , t' h , h' h} is used as the input, and y h =l' h is used as the output. v' h , t' h , h' h and l' h are the wind speed data, temperature data, holiday data, and load data after normalization, respectively.

[0066] In another exemplary embodiment, the prediction accuracy objective function in step 103 above is the mean absolute percentage error, and the model complexity objective function is the number of hidden layer nodes of the RBF neural network model.

[0067] The calculation formula of the mean absolute percentage error is:

[0068]

[0069] Among them, MAPE is the mean absolute percentage error, N is the number of samples in the test set, and y n is the short-term load label value of the nth sample in the test set, is the short-term load prediction value of the nth sample in the test set, and the short-term load prediction value is obtained based on the prediction of the RBF neural network model.

[0070] In each round of iteration process, the RBF neural network parameters corresponding to the population individuals are substituted into the network. After training the network using the training set, the mean square error between the predicted value and the true value is calculated on the validation set. The goal of algorithm optimization is to minimize this MAPE value, that is, to make the predicted value as close as possible to the true value, so as to improve the prediction accuracy of the model for short-term electricity price data.

[0071] To avoid constructing an overly complex RBF neural network model (a complex model may lead to problems such as overfitting and high computational resource consumption), the number of hidden layer nodes F in the RBF neural network model is used as an indicator to measure the model complexity. The goal is to make the value of F as small as possible while ensuring the prediction accuracy, that is, to simplify the model and improve its generalization ability and practicality. This can ensure accurate prediction of short-term electricity prices while preventing the model from being too complex to be applied and maintained.

[0072] In another exemplary embodiment, the RBF neural network model in step 104 above can be replaced by the following steps 301 - 302.

[0073] Step 301, determine the network structure.

[0074] 1. Input layer: Select three factors, namely temperature, wind speed, and holidays, as input variables, and set the number of input layer nodes to 3.

[0075] 2. Hidden layer: The hidden layer is the key part for the RBF neural network to achieve nonlinear mapping, and determining the number of its nodes is an important step. Initially, it can be roughly set according to empirical formulas or through simple experiments, and subsequent optimization will be carried out with the help of the NSGA-II algorithm. The neurons in the hidden layer use the Gaussian radial basis function as the activation function, and its expression is:

[0076]

[0077] In the above formula, represents the output of the f-th hidden layer neuron, x is the input vector, c f is the center vector of the f-th Gaussian radial basis function, whose dimension is the same as that of the input vector x, σ f is the width parameter of the f-th radial basis function, ‖·‖ represents the Euclidean norm of the vector. This function calculates the distance between the input vector x and the center vector c f and adjusts according to the width parameter σ f to output the corresponding function value, realizing the nonlinear transformation of the input data.

[0078] 3. Output layer: The number of output layer nodes is determined according to the dimension of the data to be predicted. Since this application aims to predict short-term load data at a single moment, the number of output layer nodes is set to 1. The output of the output layer neuron is a linear combination of the output of the hidden layer, and the calculation formula is as follows:

[0079]

[0080] Among them, y represents the output of the output layer (i.e., the predicted short-term load data), F is the number of hidden layer nodes, wf It is the weight coefficient connecting the f-th hidden layer node and the output layer node. Through this coefficient, the non-linear output of the hidden layer is linearly weighted and combined to obtain the final predicted output value.

[0081] Step 302, set the initial parameters.

[0082] First, for the structural design of the RBF neural network model, use the K-means clustering algorithm to classify the data. The number of neurons F in the hidden layer is the number of categories of the classified data, and the clustering center c can be obtained using the K-means algorithm f and the width σ of the radial basis function f .

[0083] σ f = min f' ||c f - c f' ||;

[0084] The quality of the number of categories in the data classification process will directly affect the performance of the RBF neural network model. Use the NSGA-II algorithm to obtain the optimal F value, that is, the optimal number of categories. The initial setting of the connection weight w f is usually randomly generated within a certain reasonable range. Initial values can be assigned to each w f in the interval [-1, 1] through a random number generator. In this way, in the subsequent training and optimization process, the optimal weight configuration can be gradually searched based on different initial weight states, and the weights are updated using the negative gradient descent method. Define the error function and performance function of the RBF neural network as follows:

[0085]

[0086] The weight adjustment strategy for the RBF output layer is:

[0087]

[0088] where, W m+1 is the weight matrix for the (m + 1)-th iterative training, W m is the weight matrix for the m-th iterative training, represents the gradient of the weight matrix for the m-th iterative training, η is the learning rate, 0 < η < 1, J m is the performance function of the RBF neural network model for the m-th iterative training, e m,i is the prediction error of the RBF neural network model for the i-th sample in the training set during the m-th iterative training, the short-term load label value of the i-th sample in the training set, is the short-term load prediction value of the i-th sample in the training set predicted by the RBF neural network model for the m-th iterative training.

[0089] In another exemplary embodiment, the above step 104 may be implemented by the following steps 401-410.

[0090] Step 401, initialize the value of the parameter optimization iteration count t to 0.

[0091] Step 402, take the initial parameters of the RBF neural network model as individuals of the NSGA-II algorithm, initialize the population, and use the initialized population as the population for the t-th parameter optimization iteration.

[0092] Step 403, use the individuals in the population for the t-th parameter optimization iteration to set the initial parameters of the RBF neural network model, and obtain the RBF neural network model corresponding to each individual.

[0093] Step 404, use the training set to train the RBF neural network model corresponding to each individual respectively, and obtain the trained RBF neural network model corresponding to each individual.

[0094] Step 405, use the test set to test the trained RBF neural network model corresponding to each individual, and obtain the predicted accuracy objective function value of each individual as the fitness value of each individual.

[0095] Step 406, according to the fitness value of each individual, perform non-dominated sorting on the individuals in the population for the t-th parameter optimization iteration, and calculate the crowding degree of each individual in the population for the t-th parameter optimization iteration;

[0096] Step 407, calculate the number of individuals to be discarded in each non-dominated layer.

[0097] Step 408, discard the individuals in each non-dominated layer according to the number of individuals to be discarded in each non-dominated layer and the crowding degree of each individual, and obtain the parental population.

[0098] Step 409, perform selection, crossover, and mutation on the individuals in the parental population to generate an offspring population;

[0099] Step 410: Combine the parent population and the offspring population to generate the population for the (t + 1)-th parameter optimization iteration. Increment the value of t by 1, and return to the step of "setting the initial parameters of the RBF neural network model using the individuals in the population for the t-th parameter optimization iteration to obtain the RBF neural network model corresponding to each individual", until the maximum number of iterations is reached or there exists an individual that meets the preset conditions. Output the trained RBF neural network model corresponding to the optimal individual as the trained RBF neural network model. The preset conditions are: an individual for which the prediction accuracy objective function is greater than the preset accuracy threshold and the model complexity objective function is less than the complexity threshold; the optimal individual is the individual with the maximum value of the prediction accuracy objective function or the individual with the maximum value of the prediction accuracy objective function among the individuals that meet the preset conditions.

[0100] In an exemplary embodiment, the above step 401 further includes determining parameters such as the population size, maximum number of iterations, crossover probability, and mutation probability of the NSGA-II algorithm.

[0101] In an exemplary embodiment, in the above step 402, the way to initialize the population is as follows: each individual in the population represents a set of parameters of the RBF neural network model (such as the number of hidden layer nodes, weights, etc.). The real number coding method is used to code the key parameters that need to be optimized in the RBF neural network model, so as to integrate these parameters into the representation form of the population individuals of the NSGA-II algorithm, which is convenient for subsequent genetic operations and parameter evolution.

[0102] Specific coding form: The coding of an individual can be expressed as I = [c 11 , c 12 , c 13 , c 14 , …, c Fe , σ 1 , σ 2 , …, σ F , w 1 , w 2 , …, w F , F], where represents the serial number of the hidden layer node, and F is the number of hidden layer nodes (the number of hidden layer nodes itself is also one of the parameters that need to be optimized, and its value range and change situation are reflected through coding). In the embodiment of the present application, for the RBF neural network with historical load, wind speed, temperature, and holidays as inputs, c fe (e = 1, 2, 3, 4, corresponding to the four input dimensions) represents the value of the center vector of the radial basis function of the f-th hidden layer node in each dimension, σ f is the width parameter of the f-th radial basis function, and w fIt is the weight coefficient for connecting the f-th hidden layer node and the output layer node. In this encoding form, f = 1, 2,..., F, and each individual represents a complete set of RBF neural network parameter configurations, facilitating subsequent various operations and optimizations within the framework of the genetic algorithm.

[0103] In an exemplary embodiment, as Figure 2 shown, in the above step 403, it is possible to first determine whether the scale of the population in the t-th parameter optimization iteration is greater than the scale of the parent population. If so, execute steps 403 - 408; otherwise, directly generate the parent population. Usually, the initialized population may have a scale not greater than that of the parent population, and the population during the iteration process will not have a scale not greater than that of the parent population.

[0104] In an exemplary embodiment, in the above step 406, the calculation method of crowding degree is improved. The Euclidean distance is used to represent the relative crowding degree of different individuals, and then the absolute distance is calculated based on the relative crowding degree as the crowding degree of the individual to represent its crowding degree. Then, the initial fitness is used to measure the dominance of the solution, and the crowding degree of the individual is comprehensively calculated. In particular, the crowding degrees of the maximum and minimum fitness values are set to infinity. The total crowding degree is equal to the sum of the crowding degrees of each individual. After the improved calculation of crowding degree, combined with the previously solved Pareto front, it is possible to more effectively find the offspring population with high fitness and high individual richness, which is beneficial to accelerating the optimization process of the algorithm. During the optimization process, it is necessary to select individuals with a larger crowding degree, so that the similarity between individuals is low, which is beneficial to ensuring the diversity of individuals in the population. Compared with the method of setting sharing parameters, it is not easy to fall into local optima.

[0105] The embodiment of the present application uses the Euclidean distance to represent the relative crowding degree of individuals. For individual k in the multi-objective optimization problem, assume that there are S objective functions in the multi-objective optimization function, and the coordinates of individual K in the objective space are (f 1k , f 2k , …, f Sk ). For any two individuals k and k', the Euclidean distance calculation formula between them in the objective space is:

[0106]

[0107] Assume that there are K k individuals in the neighborhood of individual k, and the relative crowding degree between individual k and individual k' in the neighborhood of individual k is d kk' , then the calculation formula for the crowding degree of individual k is as follows:

[0108]

[0109] Among them, D k is the crowding degree of individual k, dkk' is the relative crowding degree of individual k and individual k' within the neighborhood of individual k, K k is the number of individuals included in the neighborhood of individual k, d ab is the relative crowding degree of the a-th individual and the b-th individual, K is the number of individuals, f s,k is the value of the s-th objective function of individual k, f s,k' is the value of the s-th objective function of individual k', S is the number of objective functions.

[0110] The above-mentioned crowding degree calculation formula determines the crowding degree of an individual in a local area by comparing the distance relationship between the individual and its surrounding individuals.

[0111] In another exemplary embodiment, in the above steps 407 and 408, an improved elitist retention strategy is applied. The traditional elitist retention strategy first selects the part with a lower non-dominated level. Once the number of individuals in the population meets the requirements, it will no longer select individuals with a higher non-dominated level to enter the new population. When selecting in this way, the crowding degree only acts on the last selected non-dominated layer, resulting in a lack of diversity in the population. In the improved elitist retention strategy in the embodiment of the present application, the number of individuals to be deleted in each non-dominated layer is first determined, so that individuals in each non-dominated layer have the opportunity to enter the new parental population. Moreover, the lower the level of the non-dominated level, the smaller the proportion of individuals to be deleted and the more individuals are retained. This not only helps to retain the elite solutions but also increases the diversity of the population.

[0112] The formula for the number of individuals to be discarded in each non-dominated layer is:

[0113]

[0114] where Q g is the number of individuals to be discarded in non-dominated layer g, Q e is the number of individuals in the current population exceeding the parental population, G is the number of non-dominated layers is the floor function symbol, Q g' is the number of individuals to be discarded in non-dominated layer g'.

[0115] In another exemplary embodiment, the selection in the above step 409 is an important step of selecting some excellent individuals from the parental population to form a target population for generating the offspring population, so that the quality of the next generation of solutions is better than that of this generation and promotes the evolution of the solution set towards the region where the optimal solution exists. In this embodiment, the binary tournament selection method is adopted.

[0116] In another exemplary embodiment, in the above step 409, a strategy of improving the adaptive crossover and mutation probabilities is adopted. When the crossover probability is larger, the probability of changing the structure of the existing individual is also larger. However, if the original individual is relatively good, it will instead destroy the feasible solution. When the crossover probability is smaller, it is easy for the algorithm to fall into a local optimum. Therefore, the crossover probability of better individuals can be made smaller, and the crossover probability of worse individuals can be made larger. Mutation can change the value of one or more genes in an individual, and the execution of its operation depends on the setting of the mutation probability. The mutation probability is also adaptively adjusted.

[0117] In this embodiment, the calculation formula for the crossover probability is:

[0118]

[0119] where P k,c is the crossover probability of individual k, P c ∈[P c2 , P c1 , P c2 and P c1 are respectively the minimum and maximum values of the crossover probability, with the values being 0.9 and 0.5 respectively; f k is the fitness value of individual k, f k ′ is the maximum fitness value of all individuals obtained by crossing individual k with each individual in the target population except individual k; f max and f avg are respectively the maximum and average values of the fitness values of all individuals in the target population.

[0120] In this embodiment, the calculation formula for the mutation probability is:

[0121]

[0122] where P k,m is the mutation probability of individual k, P m1 and P m2 are respectively the maximum and minimum values of the crossover probability, with the values being 0.005 and 0.100 respectively, and f k ″ is the fitness value of the individual obtained by mutating individual k.

[0123] In another exemplary embodiment, in the above step 410, in order to find a better solution, a local search operation is introduced. After merging the above-mentioned parent population and offspring population to obtain a merged population, local search is performed around the target individuals in the merged population, and then the individuals after search are compared with the target individuals, and the individuals with better performance are retained, so as to find a better solution to ensure the uniform distribution and generality of the solution. Among them, the target individuals include the individuals with a non-dominated sorting rank of 1 in the population of the previous parameter optimization iteration and the top 5 individuals with the highest crowding degree except this individual.

[0124] In another exemplary embodiment, in the above step 105, the temperature, wind speed and holiday data of the period to be predicted are input into the trained RBF neural network model for short-term load prediction, and the prediction error is calculated by comparing the predicted value with the actual value.

[0125] In another exemplary embodiment, to illustrate the effect of the embodiments of the present application, first, relevant data of a certain park from January 1, 2018 to January 31, 2019 are collected, including historical load, temperature, wind speed and date type data, etc. Preprocess it, use the K-nearest neighbor filling method, the method of filling with the average value of historical data in the same period, etc. to process missing values, correct outliers according to the 3σ principle and related factors, and use the min-max normalization method to map the data to the [0,1] interval. Then perform feature selection and extraction, screen features strongly correlated with electricity price through Pearson correlation coefficient, and use principal component analysis to reduce the data dimension. Construct an RBF neural network model and train it with the data of the first 12 months. The training effect is measured by the mean absolute percentage error, and the early stopping method is used to prevent overfitting. Then, use the multi-objective optimization algorithm based on the improved NSGA-II to optimize the parameters of the RBF neural network. Encode parameters such as the number of hidden layer nodes as chromosomes. After initializing the population, perform non-dominated sorting and improved crowding distance calculation based on the mean absolute percentage error on the test set (the data of the last month) and the index that the F value is as small as possible on the premise of ensuring the prediction accuracy. After selection, improved adaptive crossover, and mutation multi-generation evolution operations, determine the optimal parameter combination to obtain an optimized model. Finally, after preprocessing and feature extraction of the data to be predicted, input it into the optimized model to obtain the load prediction result, use the mean absolute percentage error index to evaluate the prediction effect, and optimize the model accordingly. And compare the load data of the park on a certain day by the method of the present application and other methods. The comparison results are as Figure 3 shown to verify the effectiveness of the present method.

[0126] It can be seen that, compared with the prior art, the method of the embodiment of the present application has significant advantages. First, the RBF neural network model has a powerful non-linear mapping ability and can effectively capture the complex non-linear relationships in the virtual power plant load data without relying on traditional linear assumptions. Therefore, compared with time series methods such as ARMA or ARIMA, the prediction accuracy is significantly improved. Second, through multi-objective optimization, the NSGA-II algorithm effectively solves the contradiction between accuracy and computational efficiency in load forecasting. Compared with traditional methods with a single optimization objective (such as SVM or BP neural network), it can consider multiple objectives during the optimization process, ensuring a high convergence speed and low computational complexity. The method of the present application can adapt to the diverse load characteristics and complex scheduling requirements in the virtual power plant in a large-scale data training and multi-resource scheduling environment, has better generalization ability and application value, and is suitable for short-term load forecasting tasks with high requirements for accuracy and real-time performance in actual power systems. Moreover, the present application overcomes the subjectivity problem of existing crowding degree calculation, improves the load forecasting accuracy, ensures population diversity, enhances the stability and reliability of the algorithm, and has good adaptability and scalability.

[0127] Based on the same inventive concept, the embodiment of the present application also provides a virtual power plant short-term load forecasting device for implementing the above-mentioned virtual power plant short-term load forecasting method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the virtual power plant short-term load forecasting device provided below can refer to the limitations on the virtual power plant short-term load forecasting method in the above text and will not be repeated here.

[0128] In an exemplary embodiment, a virtual power plant short-term load forecasting device is provided, including:

[0129] A historical load data sequence acquisition module, configured to acquire the historical load data sequence of the virtual power plant and multiple load influencing factor sequences.

[0130] A training set and test set construction module, configured to use the load influencing factors in the multiple load influencing factor sequences as the input data of the samples, and use the load data in the historical load data sequence as the short-term load label values of the samples to construct a training set and a test set.

[0131] A multi-objective optimization function construction module, configured to construct a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function.

[0132] An RBF neural network model training module, configured to use the training set, the test set, and the multi-objective optimization function, and adopt the NSGA-II algorithm to train and test the RBF neural network model to obtain a trained RBF neural network model.

[0133] The short-term load forecasting module is used to perform short-term load forecasting on the virtual power plant by using the trained RBF neural network model.

[0134] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 4 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for short-term load forecasting of a virtual power plant.

[0135] Those skilled in the art can understand that Figure 4 the structure shown in

[0136] merely represents a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0137] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0138] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0139] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0140] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0141] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A short-term load forecasting method for a virtual power plant, characterized in that: The virtual power plant short-term load forecasting method comprises: Obtain the historical load data series and multiple load influencing factor series of the virtual power plant; The load influencing factors in multiple load influencing factor sequences are used as sample input data, and the load data in the historical load data sequence is used as the short-term load label value of the sample to construct a training set and a test set; Constructing a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function; Using the training set, the test set and the multi-objective optimization function, the NSGA-II algorithm is used to train and test the RBF neural network model to obtain a trained RBF neural network model; The trained RBF neural network model is used to perform short-term load forecasting for virtual power plants.

2. The short-term load forecasting method for a virtual power plant according to claim 1, characterized in that: The prediction accuracy objective function is the mean absolute percentage error, and the model complexity objective function is the number of hidden layer nodes of the RBF neural network model; The calculation formula of the mean absolute percentage error is: Where MAPE is the mean absolute percentage error, N is the number of samples in the test set, and y n is the short-term load label value of the nth sample in the test set, is the short-term load forecast value of the nth sample in the test set, and the short-term load forecast value is obtained based on the RBF neural network model prediction.

3. The short-term load forecasting method for a virtual power plant according to claim 1, characterized in that: Using the training set, the test set and the multi-objective optimization function, the NSGA-II algorithm is used to train and test the RBF neural network model to obtain a trained RBF neural network model, specifically including: Initialize the value of parameter optimization iteration number t to 0; The initial parameters of the RBF neural network model are used as individuals of the NSGA-II algorithm to initialize the population, and the initialized population is used as the population for the tth parameter optimization iteration; The initial parameters of the RBF neural network model are set using the individuals in the population of the tth parameter optimization iteration to obtain the RBF neural network model corresponding to each individual; The training set is used to train the RBF neural network model corresponding to each individual, and the trained RBF neural network model corresponding to each individual is obtained; The trained RBF neural network model corresponding to each individual is tested using the test set to obtain the prediction accuracy objective function value of each individual as the fitness value of each individual; According to the fitness value of each individual, perform non-dominated sorting on each individual in the population of the tth parameter optimization iteration, and calculate the crowding degree of each individual in the population of the tth parameter optimization iteration; Calculate the number of individuals that need to be discarded in each non-dominated layer; According to the number of individuals that need to be discarded in each non-dominated layer and the crowding degree of each individual, the individuals in each non-dominated layer are discarded to obtain the parent population; Selecting, crossing over and mutating the individuals in the parent population to generate a child population; The parent population and the child population are merged to generate a population of the t+1th parameter optimization iteration, the value of t is increased by 1, and the step of "using the individuals in the population of the tth parameter optimization iteration to set the initial parameters of the RBF neural network model and obtain the RBF neural network model corresponding to each individual" is returned until the maximum number of iterations is reached or an individual that meets the preset conditions exists, and the trained RBF neural network model corresponding to the optimal individual is output as the trained RBF neural network model, wherein the preset conditions are: the individual whose prediction accuracy objective function is greater than the preset precision threshold and whose model complexity objective function is less than the complexity threshold; the optimal individual is the individual with the largest prediction accuracy objective function value or the individual with the largest prediction accuracy objective function value among the individuals that meet the preset conditions.

4. The short-term load forecasting method for a virtual power plant according to claim 3, characterized in that: The training set is used to train the RBF neural network model corresponding to each individual, and the negative gradient descent method is used to update the weights during the training process. The specific formula is: Among them, W m+1 is the weight matrix for the m+1th iteration training, W m is the weight matrix for the mth iteration training, represents the weight matrix gradient of the mth iteration training, η is the learning rate, J m is the performance function of the RBF neural network model trained at the mth iteration, e m,i is the prediction error of the RBF neural network model trained for the mth iteration for the i-th sample in the training set, and the short-term load label value of the i-th sample in the training set, The short-term load forecast value of the i-th sample in the training set obtained by predicting the RBF neural network model trained for the m-th iteration.

5. The short-term load forecasting method for a virtual power plant according to claim 3, characterized in that: The calculation formula of congestion is: Among them, D k is the crowding degree of individual k, d kk' is the relative crowding degree between individual k and individual k' in its neighborhood, K k is the number of individuals in the neighborhood of individual k, d ab is the relative crowding degree between the ath individual and the bth individual, K is the number of individuals, f s,k is the value of the sth objective function of individual k, f s,k' is the value of the sth objective function of individual k', and S is the number of objective functions.

6. The short-term load forecasting method for a virtual power plant according to claim 3, characterized in that: The formula for calculating the number of individuals that need to be discarded in each non-dominated layer is: Among them, Q g is the number of individuals that need to be discarded in the non-dominated layer g, Q e is the number of individuals in the current population that exceeds the parent population, G is the number of non-dominated layers, is the floor function symbol, Q g' is the number of individuals that need to be discarded in the non-dominated layer g'.

7. The short-term load forecasting method for a virtual power plant according to claim 3, characterized in that: The individuals in the parent population are selected, crossed and mutated to generate a child population, specifically including: A binary tournament selection method is used to obtain a preset number of individuals from the parent population to form a target population; Calculate the crossover probability and mutation probability of each individual in the target population; Perform crossover on each individual in the target population according to the crossover probability of each individual in the target population; Each individual in the target population is mutated according to the mutation probability of each individual in the target population.

8. The short-term load forecasting method for a virtual power plant according to claim 7, characterized in that: The formula for calculating the crossover probability is: Among them, P k,c is the crossover probability of individual k, P c ∈[P c2 ,P c1 ], P c2 and P c1 are the minimum and maximum crossover probability respectively; f k is the fitness value of individual k, f k ′ is the maximum fitness of all individuals obtained by crossovering individual k with each individual in the target population except individual k; f max and f avg are the maximum and average fitness values ​​of all individuals in the target population, respectively; The calculation formula for mutation probability is: Among them, P k,m is the mutation probability of individual k, P m1 and P m2 are the maximum and minimum values ​​of the crossover probability, respectively, and f k ″ is the fitness value of the individual obtained by mutating individual k.

9. A short-term load forecasting device for a virtual power plant, characterized in that: The virtual power plant short-term load forecasting device applies the virtual power plant short-term load forecasting method according to any one of claims 1 to 8, and the virtual power plant short-term load forecasting device comprises: A historical load data sequence acquisition module is used to acquire the historical load data sequence of the virtual power plant and multiple load influencing factor sequences; A training set and a test set construction module is used to use the load influencing factors in multiple load influencing factor sequences as sample input data, and use the load data in the historical load data sequence as the short-term load label value of the sample to construct the training set and the test set; A multi-objective optimization function construction module is used to construct a multi-objective optimization function; the multi-objective optimization function includes: a prediction accuracy objective function and a model complexity objective function; An RBF neural network model training module is used to train and test the RBF neural network model using the training set, the test set and the multi-objective optimization function, using the NSGA-II algorithm to obtain a trained RBF neural network model; The short-term load forecasting module is used to perform short-term load forecasting on the virtual power plant using the trained RBF neural network model.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the virtual power plant short-term load forecasting method described in any one of claims 1-8.

Images