Real-time electricity price prediction system and method based on multi-density clustering and multi-kernel SVM

Abstract

The application provides a real-time electricity price prediction system based on multi-density clustering and multi-core SVM, which is characterized by comprising a database management module connected with a data acquisition module, a power generation and energy consumption statistical module, a real-time electricity price prediction module and a data visualization module in signal connection. The system is mainly applied to analyzing the time-space distribution characteristics of real-time electricity price, power load, main energy power generation and power generation cost and summarizing the rules, predicting the real-time electricity price of a selected area, comprehensively considering the nonlinear, sparsity and volatility characteristics of the real-time electricity price, taking into account various factors such as power load, main energy power generation and power generation cost affecting the real-time electricity price, improving the prediction accuracy and adaptability of the system, avoiding overfitting of the prediction model, improving the distributed processing capacity, reducing the computational complexity and time complexity. And a scientific and reasonable, highly applicable real-time electricity price prediction method based on multi-density clustering and multi-core SVM is provided.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power price prediction, and is a real-time price prediction system and method based on multi-density clustering and multi-core SVM. BACKGROUND

[0002] Real-time price refers to the marginal cost of providing power to users within a limited extremely short period of time, considering the operation and basic investment of the power system, which directly reflects the relationship between market price and real-time market purchase cost, and is one of the most ideal price mechanisms. Accurate prediction of real-time price can provide reliable value basis for power purchase users, so as to formulate scientific power utilization strategies, and can provide important reference for power market supervision departments, so as to formulate reasonable market rules and promote the healthy, stable and orderly development of the power market. However, due to the influence of various factors on real-time price, the real-time price shows strong volatility and sparsity, which leads to difficulty in effective prediction. Therefore, the prediction of real-time price has become one of the important topics in the field of market operation of the power system.

[0003] At present, there are mainly two types of real-time price prediction methods. One type is a prediction method based on time series, including autoregressive conditional heteroscedasticity model and autoregressive moving average model, which are mainly used to represent the linear relationship between real-time price and time and determine the similarity correlation between sample data and the price to be predicted. The other type is an intelligent prediction method based on machine learning, including support vector machine (SVM) and artificial neural network prediction methods. The support vector machine has good non-linear mapping ability and improves the generalization performance of the system. The artificial neural network has good parallel distributed processing ability and high fault tolerance performance to price noise.

[0004] The limitations of the above-mentioned real-time price prediction methods are as follows: with the integration of new energy and new equipment into power grids at all levels, the price time series shows more complex nonlinear characteristics, which makes it difficult to select the number of input variables for the time series prediction method; the real-time price prediction method using artificial neural network is prone to overfitting of the prediction model, which affects the prediction performance of the model; although the prediction method based on support vector machine overcomes the shortcomings of poor generalization ability and slow convergence in the artificial neural network prediction method, large-scale training sample data will greatly increase the computational complexity. Therefore, it is difficult to achieve ideal results by using the existing real-time price prediction methods. SUMMARY

[0005] The present application aims at overcoming the shortcomings of the prior art, and provides a real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM, which is reasonable in structure, accurate in prediction and good in effect, and a real-time electricity price prediction method based on multi-density clustering and multi-kernel SVM, which is scientific, reasonable and applicable.

[0006] The technical scheme adopted by one of the objectives of the present application is: a real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM, characterized in that it comprises: a data acquisition module for acquiring real-time electricity prices and corresponding electricity load data of a power market, and coal, oil, solar, hydrogen, nuclear and wind power generation capacity and corresponding energy price data; a database management module for classifying and managing data and constructing a system database; a power generation and consumption statistical module for real-time statistics and calculation of coal, oil, solar, hydrogen, nuclear and wind power generation capacity, power generation cost and power generation energy consumption ratio; a real-time electricity price prediction module for predicting and analyzing real-time electricity prices and corresponding electricity load by using multi-density clustering and time series pattern recognition method, and for predicting coal, oil, solar, hydrogen, nuclear and wind power generation consumption by using optimal multi-kernel SVM combination, and for calculating the weight of each influencing factor by using entropy weight method; a real-time electricity price prediction evaluation module for evaluating the accuracy of the system prediction value, calculating the prediction error and adjusting the prediction model parameters, thereby proving the accuracy and rationality of real-time electricity price prediction; a data visualization module for displaying real-time electricity price prediction value, real-time electricity price prediction evaluation scale, coal, oil, solar, hydrogen, nuclear and wind power generation capacity, power generation cost and power generation energy consumption ratio; the database management module is signal connected with the data acquisition module, the power generation and consumption statistical module, the real-time electricity price prediction module and the data visualization module.

[0007] The data acquisition module is used for online acquisition of real-time electricity prices, electricity load, coal power generation capacity, coal power generation cost, oil power generation capacity, oil power generation cost, hydrogen power generation capacity, hydrogen power generation cost, nuclear power generation capacity, nuclear power generation cost, wind power generation capacity, wind power generation cost, solar power generation capacity and solar power generation cost of the selected regional power market.

[0008] The database management module is used for classification, storage, management and transmission of the data collected by the data acquisition module, wherein the coal, oil, solar, hydrogen, nuclear and wind power generation capacity and power generation cost are inputted into the power generation and consumption statistical module for statistics, and the statistical energy generation and consumption ratio of each type of energy is stored in the database management module; the real-time electricity price and electricity load are inputted into the real-time electricity price prediction module for calculation, and the real-time electricity price prediction value is stored in the database management module.

[0009] The function of the power generation energy consumption statistical module is to statistically collect the power generation capacity and power generation cost of six power generation energies, namely coal, oil, hydrogen, nuclear energy, solar energy and wind energy, and to calculate the power generation energy consumption proportion of each type of energy, and to use the power generation energy consumption proportion of each type of energy as the input of the real-time electricity price prediction module.

[0010] The function of the real-time electricity price prediction module is to predict the real-time electricity price of the selected regional electricity market in real time. Firstly, a multi-density clustering algorithm is applied to cluster analyze the real-time electricity price and the corresponding electricity load data, and a time series pattern recognition method is used to search for similar days to preliminarily predict the real-time electricity price and the corresponding electricity load. Secondly, a multi-kernel SVM is selected to predict the consumption proportion of each type of energy, and the optimal multi-kernel function combination is selected according to the prediction result. Finally, the weights of each influencing factor are calculated by using the entropy weight method, and the real-time electricity price is predicted by using the weighted calculation method of the influencing factors.

[0011] The function of the real-time electricity price prediction evaluation module is to evaluate the accuracy of the system prediction value, by calculating the absolute percentage error between the real value of the electricity price and the system prediction value, and adjusting the corresponding prediction model parameters, so as to prove the accuracy and rationality of the real-time electricity price prediction.

[0012] The function of the data visualization module is to display the corresponding real-time electricity price prediction value, electricity price prediction evaluation scale, energy generation energy consumption ratio, main power generation energy generation capacity and power generation cost, including coal power generation capacity and cost, oil power generation capacity and cost, hydrogen power generation capacity and cost, nuclear power generation capacity and cost, wind power generation capacity and cost, and solar power generation capacity and cost.

[0013] The technical solution adopted by the second purpose of the application is: a real-time electricity price prediction method based on multi-density clustering and multi-kernel SVM, characterized in that it comprises the following steps:

[0014] Step 1: a multi-density clustering algorithm is proposed to cluster analyze the collected real-time electricity price and the corresponding load data;

[0015] (a) KNN algorithm is selected to cluster analyze the real-time electricity price dimension of the original data and the corresponding electricity load, and the reference radius is calculated according to formula (1):

[0016]

[0017] Wherein, r is the reference radius length, k is the core object threshold, Γ(n) is the gamma function, m is the sample point dimension, n is the sample point number, is the sample point product, x i is the i th sample point;

[0018] (b) According to the within-class Mahalanobis distance, arrange the clusters from small to large, extract the cluster where the quartile, median and three-quarters are located in the cluster array, assign the three within-class mean Mahalanobis distances with corresponding weights, and calculate the weighted distance as the distance between adjacent scales of the time axis, and re-encode the time axis;

[0019] (c) According to the reference radius r, select the point with the highest density in the sample points as the center point, take the average distance between the center point and its k-nearest neighbors as the density radius of this time, and perform density clustering; when the cluster clustering is completed, stop clustering, and mark the sample points as clustered points; according to the reference radius, select the point with the highest density in the unclustered sample points as the center point again, take the average distance between the center point and its k-nearest neighbors as the density radius of this time, and when the cluster clustering is completed, stop clustering, and mark the sample points as clustered points; repeat the process until all sample points are clustered;

[0020] Step 2: Using time series pattern recognition algorithm, according to the cluster time series distribution characteristics of real-time electricity price and corresponding electricity load, search for the day with the highest similarity in historical data as the similar day, and preliminarily predict the real-time electricity price and corresponding electricity load;

[0021] (d) Determine the day-ahead cluster time series distribution of the real-time electricity price and electricity load to be predicted;

[0022] (e) Search for the day with the highest similarity in historical data as the similar day, and calculate the similarity of real-time electricity price and corresponding electricity load according to formula (2):

[0023]

[0024] Wherein, Cp is the time series pattern to be processed, which is composed of the clusters to which each sample point belongs in 24 hours of the day, C is the time series pattern to be calculated for similarity with Cp, Cp i is the cluster to which the ith sample point belongs, n * is the number of sample points in the day, C i is the cluster to which the ith sample point belongs. Similarity is the similarity of two time series patterns;

[0025] (f) According to the real-time electricity price and corresponding electricity load change trend of the next day of the similar day, predict the real-time electricity price and corresponding electricity load of the day;

[0026] Step 3: The multi-kernel SVM algorithm is used to predict the proportion of power generation energy consumption of main energy, and the optimal multi-kernel function combination is determined according to the prediction results of different SVM kernel function combinations; according to the characteristics of different influencing factors, the appropriate single kernel SVM function is selected to predict the energy consumption proportion of various main power generation energy, and the expression (3) of the multi-group multi-kernel SVM function is constructed:

[0027]

[0028] Wherein, K d (x i , x j ) is a single kernel function, x i is a time point, x j is the sample value of the dth influencing factor at the time point x i , m is the number of influencing factors, K MKL (x i , x j ) is a multi-kernel SVM function;

[0029] Step 4: The entropy weight method is used to calculate the weight of each influencing factor, a regression prediction model is established, and the final real-time electricity price prediction value is calculated by the weighted method; the entropy weight method is used to calculate the standard value of each influencing factor in formula (4);

[0030] (g) Standardize each influencing factor data according to the standardization equation:

[0031]

[0032] Wherein, Y ij is the standard value of the influencing factor, x ij is the actual value of the influencing factor, min(x i ) is the minimum value of the influencing factor, and max(x i ) is the maximum value of the influencing factor;

[0033] (h) Calculate the information entropy of each influencing factor according to the information entropy formula (5):

[0034]

[0035] Wherein, E j is the information entropy of the influencing factor, n' is the sample set size of the influencing factor, Standardized information factor

[0036] Calculate the weight of each influencing factor according to formula (6):

[0037]

[0038] Wherein, λ i is the weight of the influencing factor, Ei is the information entropy of the influencing factor, k is the number of influencing factors;

[0039] (i) obtaining a regression prediction function expression (7):

[0040]

[0041] wherein, λ i is the weight of each single kernel SVM function, m is the number of influencing factor types, λ p is the weight of real-time electricity price, λ l is the weight of power load, is the kernel function of the influencing factor, P f is the real-time electricity price prediction value, P is the preliminary prediction result of real-time electricity price, and L is the preliminary prediction result of power load.

[0042] The real-time electricity price prediction system and method based on multi-density clustering and multi-kernel SVM have the following advantages:

[0043] 1) The real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM is mainly applied to analyze the time-space distribution characteristics of real-time electricity price, power load, coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation capacity and generation cost, and summarize the rules, so as to accurately predict the real-time electricity price of the selected area.

[0044] 2) The real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM comprehensively considers the nonlinearity, sparsity, and volatility characteristics of real-time electricity price due to the integration of new energy and new equipment into power grids at all levels, and takes into account the power load, coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation capacity and generation cost of real-time electricity price, which improves the prediction accuracy of the system.

[0045] 3) The real-time electricity price prediction method based on multi-density clustering and multi-kernel SVM proposes a multi-density clustering algorithm with time axis adaptability according to the nonlinearity, sparsity, and volatility of real-time electricity price and corresponding power load data; according to the different nonlinear distribution characteristics of coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation capacity, a multi-kernel SVM optimal combination algorithm is proposed; different methods are used to process data with different characteristics, which avoids overfitting of the prediction model, improves distributed processing capability, and reduces computational complexity and time complexity.

[0046] 4) The method is scientific and reasonable, and has strong applicability. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1A structure block diagram of a real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM of the present application;

[0048] Figure 2 A flow chart of a real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM of the present application;

[0049] Figure 3 A real-time electricity price clustering schematic diagram based on multi-density clustering of the embodiment of the present application;

[0050] Figure 4 A comparison schematic diagram of the prediction method and the neural network prediction of the embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the object, technical scheme and advantages of the present application clearer, the present application is further described in detail below with specific embodiments combined with the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0052] REFERENCE Figure 1 In an embodiment of the present application, a real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM is provided to predict at least the real-time electricity price at a future time point, and the time range can be at least two time nodes. The real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM of the present application comprises a data acquisition module for acquiring real-time electricity price and corresponding electric load data, coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation and corresponding energy price data; a database management module for classifying and managing data, and further constructing a system database; a power generation and energy consumption statistical module for real-time statistics and calculation of coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation, power generation cost, and power generation energy consumption ratio; a real-time electricity price prediction module for predicting and analyzing real-time electricity price and corresponding electric load by using multi-density clustering and time series pattern recognition method, and predicting coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation consumption by using optimal multi-kernel SVM combination, and calculating the weight of each influencing factor by using entropy weight method; a real-time electricity price prediction evaluation module for evaluating the accuracy of the system prediction value, calculating the prediction error, and adjusting the prediction model parameters, thereby proving the accuracy and rationality of the real-time electricity price prediction; a data visualization module for displaying real-time electricity price prediction value, real-time electricity price prediction evaluation scale, coal, oil, solar energy, hydrogen, nuclear energy, and wind power generation and power generation cost, and power generation energy consumption ratio; and the database management module is signal connected with the data acquisition module, the power generation and energy consumption statistical module, the real-time electricity price prediction module, and the data visualization module.

[0053] In this embodiment, the selected region is the PJM power market in the United States, and the data collection period is from 2015 to 2017. The data collected by the data collection module includes: real-time electricity price data p and corresponding electricity load data l, coal power generation X1, coal power generation cost Xc1, oil power generation X2, oil power generation cost Xc2, hydrogen power generation X3, hydrogen power generation cost Xc3, nuclear power generation X4, nuclear power generation cost Xc4, wind power generation X5, wind power generation cost Xc5, solar power generation X6, and solar power generation cost Xc6.

[0054] Data list one:

[0055] Real-time electricity price p Hydrogen power generation cost Xc3 Electricity load l Nuclear power generation X4 Coal power generation X1 Nuclear power generation cost Xc4 Coal power generation cost Xc1 Wind power generation X5 Oil power generation X2 Wind power generation cost Xc5 Oil power generation cost Xc2 Solar power generation X6 Hydrogen power generation X3 Solar power generation cost Xc6

[0056] The function of the database management module is to classify, store, manage and transmit the data collected by the data collection module. The coal, oil, solar, hydrogen, nuclear and wind power generation and power generation cost are inputted into the power generation energy consumption statistical module for statistical calculation, and the energy generation energy consumption ratio calculated by the statistical calculation is stored in the database management module. The real-time electricity price and the electricity load are inputted into the real-time electricity price prediction module for prediction value calculation, and the real-time electricity price prediction value is stored in the database management module.

[0057] The function of the power generation energy consumption statistical module is to real-time statistical collect the coal, oil, hydrogen, nuclear, solar and wind power generation energy and power generation cost, calculate the power generation energy consumption ratio of coal, oil, solar, hydrogen, nuclear and wind power, and input the power generation energy consumption ratio as the input of the real-time electricity price prediction module.

[0058] The function of the real-time electricity price prediction module is to real-time predict the real-time electricity price of the selected regional power market. First, the multi-density clustering algorithm is applied to cluster analyze the real-time electricity price and the corresponding electricity load data, and the time series pattern recognition method is used to search for similar days to preliminarily predict the real-time electricity price and the corresponding electricity load. Second, the multi-kernel SVM is selected to predict the consumption proportion of each type of energy, and the optimal multi-kernel function combination is selected according to the prediction result. Finally, the entropy weight method is used to calculate the weight of each influencing factor, and the real-time electricity price is predicted by the weighted calculation of the influencing factors.

[0059] The function of the real-time electricity price prediction evaluation module is to evaluate the accuracy of the system prediction value. By calculating the absolute percentage error between the real electricity price and the system prediction value, and adjusting the corresponding prediction model parameters, the accuracy and rationality of the real-time electricity price prediction are proved.

[0060] The data visualization module is used to display real-time electricity price prediction value, electricity price prediction evaluation scale, power generation energy consumption ratio, power generation capacity and cost of power generation energy, and power generation capacity and cost of power generation energy, including coal power generation capacity and cost, oil power generation capacity and cost, hydrogen power generation capacity and cost, nuclear power generation capacity and cost, wind power generation capacity and cost, and solar power generation capacity and cost.

[0061] Referring to Figure 2 In the embodiment, a real-time electricity price prediction method based on multi-density clustering and multi-core SVM includes the following steps:

[0062] Step 1: A multi-density clustering algorithm is proposed to perform clustering analysis on the collected real-time electricity price and corresponding load data.

[0063] (a) The KNN algorithm is selected to perform clustering analysis on the real-time electricity price dimension and corresponding load of the original data, and the reference radius is calculated according to formula (1):

[0064]

[0065] Where r is the reference radius length, k is the core object threshold, Γ(n) is the gamma function, m is the sample point dimension, n is the sample point number, is the sample point product, x i is the i-th sample point.

[0066] (b) The intra-class Mahalanobis distance is used to arrange each cluster from small to large, and the fourth quantile, median and three-quarters of the cluster in the cluster array are extracted. The three intra-class mean Mahalanobis distances are assigned with corresponding weights, and in the embodiment, the weight values are 0.25, 0.5 and 0.25 respectively. The weighted distance is calculated as the distance between adjacent scales on the time axis, and the time axis is re-encoded.

[0067] (c) According to the reference radius r, the point with the maximum density in the sample points is selected as the center point, and the average distance between the center point and its k nearest neighbors is taken as the density radius of this time, and the density clustering is performed. When the cluster clustering is completed, the clustering is stopped, and the sample points are marked as clustered points. In the unclustered sample points, the point with the maximum density is selected as the center point according to the reference radius again, and the average distance between the center point and its k nearest neighbors is taken as the density radius of this time. When the cluster clustering is completed, the clustering is stopped, and the sample points are marked as clustered points. Repeat the process until all sample points are clustered. (Refer to Figure 3 )

[0068] Step 2: Using time series pattern recognition algorithm, according to the real-time electricity price and the cluster time series distribution characteristics of the similar day, search the highest similarity day in the historical data as the similar day, the preliminary real-time electricity price P and the similar electricity load L;

[0069] (d) Determine the day-ahead cluster time series distribution of the real-time electricity price and the similar electricity load to be predicted;

[0070] (e) Search the highest similarity day in the historical data as the similar day S d The similarity calculation of real-time electricity price and similar electricity load is based on formula (2):

[0071]

[0072] Wherein, Cp is the time series pattern to be processed, which is composed of the class cluster to which each sample point of the day belongs, C is the time series pattern to be calculated for similarity with Cp, Cp i is the class cluster to which the ith sample point belongs, n * is the number of sample points of the day, C i is the class cluster to which the ith sample point belongs. Similarity is the similarity of two time series patterns;

[0073] (f) According to the change trend of real-time electricity price and similar electricity load of the next day of the similar day, predict the real-time electricity price and the corresponding electricity load of the day;

[0074] Step 3: Propose a multi-kernel SVM algorithm to predict the energy consumption proportion of main energy, determine the optimal multi-kernel function combination according to the prediction results of different SVM kernel function combinations;

[0075] (g) According to the characteristics of different influencing factors, select the appropriate single kernel SVM function to predict the energy consumption proportion of various main power generation energy, and constitute the expression (3) of multi-kernel SVM function:

[0076]

[0077] Wherein, K d (x i , x j ) is a single kernel function, x i is a time point, x j is the sample value of the dth influencing factor at time point x i , m is the number of influencing factor types, K MKL (x i , x j ) is a multi-kernel SVM function;

[0078] (h) Select the optimal multi-kernel SVM function K according to the prediction results of different multi-kernel SVM functions b ;

[0079] Step 4: Calculate the weight of each influencing factor by using the entropy weight method, establish the regression prediction model, and calculate the final real-time electricity price prediction value by using the weighted method;

[0080] (i) Standardize each influencing factor data according to the standardization equation (4):

[0081]

[0082] Where Y ij is the standard value of the influencing factor, x ij is the actual value of the influencing factor, min(x i ) is the minimum value of the influencing factor, and max(x i ) is the maximum value of the influencing factor;

[0083] (j) Calculate the information entropy of each influencing factor according to the information entropy equation (5):

[0084]

[0085] Where E j is the information entropy of the influencing factor, n' is the sample set size of the influencing factor, standardized information factor;

[0086] (k) Calculate the weight of each influencing factor according to equation (6):

[0087]

[0088] Where λ i is the weight of the influencing factor, E i is the information entropy of the influencing factor, and k is the number of influencing factors;

[0089] (l) Obtain the regression prediction function expression (7):

[0090]

[0091] Where λ i is the weight of each single kernel SVM function, m is the number of influencing factor types, λ p is the weight of the real-time electricity price, λ l is the weight of the electricity load, is the kernel function of the influencing factor, P f is the real-time electricity price prediction value, P is the preliminary prediction result of the real-time electricity price, and L is the preliminary prediction result of the electricity load.

[0092] Referring toFigure 4 The market of April 1, 2016 is selected as a prediction day, data from January 1 to March 30 is used as a training sample set, and the real-time electricity price prediction system based on multi-density clustering and multi-kernel SVM is trained and tested; the following table is a comparison table of the prediction value of the BP neural network and the prediction value of the system:

[0093]

[0094] The following table is an average error table of the prediction value of the neural network, the prediction value of the system and the real-time electricity price true value:

[0095] Average value Average real value Average error value BP neural network 18.706 19.382 3.54％ The system 19.106 19.382 2.45％

[0096] It can be seen from the above two tables that the real-time electricity price prediction method of the system has higher prediction accuracy compared with the BP neural network method.

[0097] The software program involved in the application is prepared according to the Internet and natural language processing technology, which is a technology familiar to those skilled in the art.

[0098] Obviously, the above embodiments are only examples for clearly illustrating, and are not limited to the embodiments. For those skilled in the art, other different forms of changes or variations can be made on the basis of the above description. Here, all the embodiments cannot be exhausted, and the changes or variations deduced therefrom should be covered in the protection scope of the application, therefore, the protection scope of the application should be the protection scope of the claims.

Claims

1. A real-time electricity price forecasting system based on multi-density clustering and multi-kernel SVM, characterized in that it The utility model relates to a kind of real-time electricity price prediction system based on multi-energy power generation, comprising: Real-time electricity price and corresponding electricity load data are collected for electricity market; Coal, oil, solar energy, hydrogen, nuclear energy, wind power generation and corresponding energy price data are collected by data collection module; Database management module is used for classified management of data, and then system database is constructed; Power generation and energy consumption statistics module is used for real-time statistics and calculation of coal, oil, solar energy, hydrogen, nuclear energy, wind power generation, power generation cost and power generation energy consumption ratio; Real-time electricity price and corresponding electricity load are predicted and analyzed by using multi-density clustering and time series pattern recognition method, and the power consumption of coal, oil, solar energy, hydrogen, nuclear energy and wind energy is predicted by using optimal multi-kernel SVM combination; Entropy weight method is used to calculate the weight of each influencing factor for predicting real-time electricity price; Real-time electricity price prediction evaluation module is used to evaluate the accuracy of system prediction value, calculate prediction error and adjust prediction model parameters, so as to prove the accuracy and rationality of real-time electricity price prediction; Data visualization module is used to display real-time electricity price prediction value, real-time electricity price prediction evaluation scale, coal, oil, solar energy, hydrogen, nuclear energy, wind power generation, power generation cost and power generation energy consumption ratio; The database management module is signal connected with data collection module, power generation and energy consumption statistics module, real-time electricity price prediction module and data visualization module; The function of real-time electricity price prediction module is to predict the real-time electricity price of selected regional electricity market in real time; Firstly, multi-density clustering algorithm is used for clustering analysis of real-time electricity price and corresponding electricity load data, and time series pattern recognition method is used to search similar days to preliminarily predict real-time electricity price and corresponding electricity load; Secondly, multi-kernel SVM is selected to predict the consumption ratio of each type of energy, and the optimal multi-kernel function combination is selected according to the prediction result; Finally, the weight of each influencing factor is calculated by using entropy weight method, and real-time electricity price is predicted by using influencing factor weighting calculation method.

2. The multi-density clustering and multi-kernel SVM based real-time electricity price prediction system of claim 1, wherein, The function of data collection module is to collect real-time electricity price, electricity load, coal power generation, coal power generation cost, oil power generation, oil power generation cost, hydrogen power generation, hydrogen power generation cost, nuclear power generation, nuclear power generation cost, wind power generation, wind power generation cost, solar power generation, solar power generation cost data of selected regional electricity market online.

3. The multi-density clustering and multi-kernel SVM based real-time electricity price forecasting system according to claim 1, characterized in that, The function of database management module is to classify, store, manage and transmit the data collected by data collection module; Coal, oil, solar energy, hydrogen, nuclear energy and wind power generation and power generation cost are used as input for power generation and energy consumption statistics module to calculate the energy consumption ratio of each type of energy, and the energy consumption ratio of each type of energy is stored in database management module; Real-time electricity price and electricity load are used as input for real-time electricity price prediction module to calculate, and real-time electricity price prediction value is stored in database management module.

4. The multi-density clustering and multi-kernel SVM based real-time electricity price forecasting system of claim 1, wherein, The function of the power generation energy consumption statistical module is to statistically collect the power generation capacity and cost of coal, oil, hydrogen, nuclear energy, solar energy and wind energy, and to calculate the energy consumption ratio of each type of energy, which is used as the input of the real-time electricity price prediction module.

5. The multi-density clustering and multi-kernel SVM based real-time electricity price forecasting system according to claim 1, characterized in that, The function of the real-time electricity price prediction and evaluation module is to evaluate the accuracy of the system prediction value by calculating the absolute percentage error between the real value and the system prediction value, and adjusting the corresponding prediction model parameters, thereby proving the accuracy and rationality of the real-time electricity price prediction.

6. The multi-density clustering and multi-kernel SVM based real-time electricity price forecasting system according to claim 1, characterized in that, The function of the data visualization module is to display the real-time electricity price prediction value, electricity price prediction evaluation scale, energy consumption ratio of each type of energy, and the power generation capacity and cost of the main power generation energy, including coal, oil, hydrogen, nuclear energy, wind energy and solar energy.

7. A real-time electricity price prediction method based on multi-density clustering and multi-core SVM, characterized in that, It comprises the following steps: Step 1: A multi-density clustering algorithm is proposed to cluster the collected real-time electricity price and corresponding load data; (a) Select KNN algorithm to cluster the real-time electricity price dimension and corresponding load of the original data, and calculate the reference radius according to formula (1): wherein r is a reference radius length, k is a core object threshold value, Γ(n) is a gamma function, m is a sample point dimension, and n is a sample point number, is a sample point product, x i is the i-th sample point; (b) Arrange the clusters from small to large according to the within-class Mahalanobis distance, extract the quartiles, median and three-quarters of the cluster array, assign the three within-class mean Mahalanobis distances with corresponding weights, and calculate the weighted distance as the distance between adjacent scales on the time axis, and re-encode the time axis; (c) According to the reference radius r, select the point with the maximum density in the sample points as the center point, and take the average distance between the center point and its k nearest neighbors as the density radius of this time, and perform density clustering; when the cluster is completed, stop clustering and mark the sample points as clustered points; In the unclustered sample points, select the point with the maximum density as the center point according to the reference radius again, and take the average distance between the center point and its k nearest neighbors as the density radius of this time, and when the cluster is completed, stop clustering and mark the sample points as clustered points; Repeat this process until all sample points are clustered; Step 2: Adopt time series pattern recognition algorithm, according to the cluster time series distribution characteristics of real-time electricity price and corresponding load, search the day with the highest similarity in historical data as the similar day, and preliminarily predict the real-time electricity price and corresponding load; (d) Determine the day-ahead cluster time series distribution of the real-time electricity price and load to be predicted; (e) Search the day with the highest similarity in historical data as the similar day, and calculate the similarity of real-time electricity price and corresponding load according to formula (2): Wherein, Cp is the time series pattern to be processed, which is composed of the class cluster to which each sample point of 24 hours of the day belongs, C is the time series pattern to be calculated for similarity with Cp, Cp i is the class cluster to which the i th sample point belongs, n * is the number of sample points of the day; C i is the class cluster to which the i th sample point belongs; Similarity is the similarity of the two time series patterns; (f) According to the change trend of real-time electricity price and corresponding load of the next day of the similar day, predict the real-time electricity price and corresponding load of the day; Step 3: The multi-kernel SVM algorithm is used to predict the proportion of power generation energy consumption of main energy, and the optimal multi-kernel function combination is determined according to the prediction results of different SVM kernel function combinations; according to the characteristics of different influencing factors, the appropriate single kernel SVM function is selected to predict the energy consumption proportion of various main power generation energy, and the expression (3) of the multi-group multi-kernel SVM function is constructed: where K d (x i ,x j ) is a single kernel function, x i is a time point, x j is a sample value of the dth influencing factor at the time point x i , m is the number of influencing factor types, K MKL (x i ,x j ) is a multi-kernel SVM function; Step 4: The weight of each influencing factor is calculated by entropy weight method, a regression prediction model is established, and the final real-time electricity price prediction value is calculated by weighted method; the standard value of each influencing factor is calculated by formula (4) by selecting entropy weight method: (g) Standardize the data of each influencing factor according to the standardization equation: where Y ij is the standard value of the influencing factor, x ij is the actual value of the influencing factor, min(x i ) is the minimum value of the influencing factor, and max(x i ) is the maximum value of the influencing factor. (h) Calculate the information entropy of each influencing factor according to the information entropy formula (5): wherein E j is the information entropy of the influencing factor, n' is the sample set size of the influencing factor, standardized information factor Calculate the weight of each influencing factor according to formula (6): wherein λ i is the weight of the influencing factor, E i is the information entropy of the influencing factor, and k is the number of influencing factors. (i) Get the regression prediction function expression (7): where λ i is the weight of each single kernel SVM function, m is the number of influence factors, λ p is the weight of real-time electricity price, λ l is the weight of electricity load, is the kernel function of influence factors, P f is the real-time electricity price prediction value, P is the preliminary prediction result of real-time electricity price, and L is the preliminary prediction result of electricity load.

Images