Multi-dimensional load electricity price prediction method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of electric power spot transaction prediction, and discloses a multi-dimensional load electricity price prediction method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining multi-scale historical data, the multi-scale historical data comprises multi-scale time sequence data, multi-dimensional meteorological data, historical load data and historical clearing price data; performing correlation analysis on the multi-scale historical data, and extracting strong correlation coupling data from the multi-scale historical data to obtain multi-dimensional feature data; sequentially carrying out weighted fusion according to the data features of the multi-dimensional feature data to obtain fused data; and taking the fusion data as input data of an Informer network, and predicting the load electricity price to obtain a load value and an electricity value. The problem that the spot transaction price is difficult to accurately predict is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power spot transaction prediction, and in particular to a multi-dimensional load electricity price prediction method, device, computer equipment and storage medium. Background Art

[0002] With the transformation of the global energy structure and the popularization of the concept of sustainable development, the power market is gradually becoming an increasingly important area in the global economy. The reform of the power market continues to deepen, the scale of market-oriented transactions continues to expand, and the electricity price system is also facing new adjustments. However, due to the particularity of the power market, electricity prices are affected by a variety of complex factors, such as seasonal demand changes, fuel costs, policy adjustments, etc. These factors make electricity price forecasting more complex and challenging. Therefore, the development and application of advanced electricity price forecasting technology is of great strategic significance for improving market efficiency, promoting energy structure optimization, and ensuring energy security.

[0003] The existing forecasting models based on the periodicity of electricity price series aim to improve the accuracy of electricity price forecasting by revealing the long-term dependencies in time series data. However, in actual application scenarios, especially in the short-term electricity price forecasting of the electricity spot market, its accuracy still cannot meet the demand. Summary of the invention

[0004] In view of this, the present invention provides a multi-dimensional load electricity price prediction method, device, computer equipment and storage medium to solve the problem of difficulty in accurately predicting spot transaction prices.

[0005] In the first aspect, the present invention provides a multi-dimensional load electricity price prediction method, which includes: obtaining multi-scale historical data, the multi-scale historical data includes multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data. Perform correlation analysis on the multi-scale historical data, extract strongly correlated coupling data from the multi-scale historical data, and obtain multi-dimensional feature data; perform weighted fusion in turn according to the data features of the multi-dimensional feature data to obtain fused data; use the fused data as input data of the Informer network, predict the load electricity price, and obtain the load value and electricity value.

[0006] In this implementation, by considering the impact of multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data on the prediction results, and extracting strongly correlated coupling data based on correlation analysis, the correlation between feature data is considered. Using strongly correlated coupling data as subsequent model input can improve the model prediction accuracy, and weighted fusion of multi-scale historical data in sequence can effectively integrate features of different feature types and complexities. Based on the Informer model, the processing capability of long sequence data is further optimized, and the load electricity price is predicted at the same time, which improves the accuracy of load electricity price prediction in spot trading.

[0007] In an optional implementation, correlation analysis is performed on multi-scale historical data, and strongly correlated coupling data is extracted from the multi-scale historical data to obtain multi-dimensional feature data, including: correlation analysis is performed on the multi-scale historical data, and strongly correlated coupling at least part of the time series data, at least part of the meteorological data, historical load data, and historical clearing price data are extracted from the multi-scale historical data to obtain multi-dimensional feature data.

[0008] In an optional implementation, correlation analysis is performed on multi-scale historical data, and at least part of the time series data, at least part of the meteorological data, historical load data, and historical clearing price data that are strongly correlated and coupled are extracted from the multi-scale historical data to obtain multi-dimensional feature data, including: multi-time sampling of the multi-scale historical data and construction of a high-dimensional data matrix with a time series; correlation analysis is performed on the data in the high-dimensional data matrix to calculate the correlation coefficient between each multi-scale historical data; and at least part of the data having a correlation coefficient greater than a correlation threshold is extracted from the multi-scale historical data to obtain multi-dimensional feature data.

[0009] In this implementation, based on the correlation analysis of multi-scale time windows and multi-dimensional meteorological factors in power spot trading, the strongly correlated feature matrix can be effectively screened out, which helps to improve the data processing capability of the network model in complex data analysis.

[0010] In an optional implementation, weighted fusion is performed in sequence according to data features of the multi-dimensional feature data to obtain fused data, including: obtaining the data type of the multi-dimensional feature data; performing a weighted fusion of data of the same type according to the data type of the multi-dimensional feature data; performing a second weighted fusion of data of the same type according to the data type of the fused data; and performing weighted fusion of data of the same type multiple times until the fusion of the multi-dimensional feature data is completed to obtain fused data.

[0011] In this implementation, the application of data fusion technology to integrate and process multi-dimensional data can effectively construct a coupled feature set, thereby providing a preprocessing basis for the input layer of the neural network model. This step helps to improve the data processing capabilities of the network model in complex data analysis.

[0012] In an optional embodiment, the Informer network includes an input layer, multiple encoding layers, multiple decoding layers and a double-layer linear output layer. The fused data is used as the input data of the Informer network to predict the load electricity price. Obtaining the load value and the electricity value includes: inputting the fused data into the input layer of the Informer network to obtain a feature vector sequence of the fused data; in the encoding layer, extracting features of the feature vector sequence based on the attention mechanism to obtain first feature data; in the decoding layer, extracting features of the first feature data based on the cross-attention mechanism to obtain second feature data; in the double-layer linear output layer, extracting the coupling dependency of the second feature data in the spatial dimension based on linear transformation, and simultaneously predicting the output load value and electricity value.

[0013] In an optional implementation, before performing correlation analysis on multi-scale historical data, the method also includes: performing outlier detection on the multi-scale historical data based on a lonely forest algorithm; for each outlier, obtaining K data values ​​closest to the outlier based on a KNN algorithm; and assigning a value to the outlier based on the K data values.

[0014] In this implementation, the lonely forest algorithm can handle the outlier problem in high-dimensional big data, and can achieve high processing accuracy and fast calculation speed. The KNN algorithm can effectively fill in the small number of missing values ​​in the data.

[0015] In the second aspect, the present invention provides a multi-dimensional load electricity price prediction device, which includes: an acquisition module for acquiring multi-scale historical data, the multi-scale historical data includes multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data. An extraction module is used to perform correlation analysis on the multi-scale historical data, extract strongly correlated coupling data from the multi-scale historical data, and obtain multi-dimensional feature data; a fusion module is used to perform weighted fusion in sequence according to the data features of the multi-dimensional feature data to obtain fused data; a prediction module is used to use the fused data as input data of the Informer network to predict the load electricity price and obtain the load value and electricity value.

[0016] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, computer instructions being stored in the memory, and the processor executing the multi-dimensional load electricity price forecasting method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the multi-dimensional load electricity price forecasting method of the first aspect or any corresponding embodiment thereof.

[0018] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which are used to enable a computer to execute the multi-dimensional load electricity price forecasting method of the above-mentioned first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 is a flow chart of a multi-dimensional load electricity price prediction method according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of another multi-dimensional load electricity price prediction method according to an embodiment of the present invention;

[0022] Figure 3 is a schematic diagram of correlation analysis results according to an embodiment of the present invention;

[0023] Figure 4 is a schematic diagram of a feature fusion method according to an embodiment of the present invention;

[0024] Figure 5 is a structural block diagram of a multi-dimensional load electricity price prediction device according to an embodiment of the present invention;

[0025] Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0027] Short-term electricity price forecasting is crucial for the effective operation of the power market. First, short-term electricity price forecasting can help market participants, including power generators, power users and power regulators, better understand market dynamics and make more informed business decisions and policy adjustments. Second, short-term electricity price forecasting has a positive effect on the operational stability of the power system. By predicting electricity prices, users can be guided to adjust their electricity consumption and electricity consumption time, reduce the peak-to-valley difference of the system, increase the system load rate, reduce the system operating costs, and ensure the stability of system operation. Furthermore, with the deepening of the market-oriented reform of the power industry, the impact of electricity price fluctuations on demand will be more significant, and the role of the electricity price adjustment mechanism will be more prominent. In power trading and bidding, the best benefits can be obtained by predicting electricity price information in advance and arranging production plans and bidding strategies in advance. Finally, short-term electricity price forecasting technology can provide a scientific basis for power market supervision, help regulators to promptly propose suggestions for adjusting electricity prices to government price departments, improve market transparency and fairness, maintain market order, and protect consumer interests.

[0028] The current electricity price prediction models are mainly divided into three categories: simulation modeling methods, statistical learning methods, and artificial intelligence methods. Among them, the simulation modeling method predicts electricity prices by establishing an accurate system model and solving it. However, the simulation method requires obtaining detailed system operation data, and the calculation cost is relatively high, so it is less feasible in practical applications. Statistical learning methods analyze historical electricity price data and predict future trends. The above methods include classic time series analysis techniques, such as ARIMA models, exponential smoothing methods, and other regression analysis methods. Since factors that affect electricity prices, such as load, are ignored, the prediction accuracy is average.

[0029] Current artificial intelligence methods are divided into two categories: one is traditional machine learning algorithms such as support vector regression (SVR), random forest (RF), etc. The other is deep learning algorithms, among which the common ones are recurrent neural network (RNN) and its variants long short-term memory network (LSTM) and gated recurrent unit (GRU).

[0030] Based on this, this application proposes a multi-dimensional load electricity price prediction method.

[0031] According to an embodiment of the present invention, an embodiment of a multi-dimensional load electricity price forecasting method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] In this embodiment, a multi-dimensional load electricity price prediction method is provided. Figure 1 is a flowchart of a multi-dimensional load electricity price prediction method according to an embodiment of the present invention. It should be noted that if there are substantially the same results, this embodiment does not use Figure 1 The process sequence shown is limited. Figure 1 As shown, the process includes the following steps:

[0033] Step S101, obtaining multi-scale historical data.

[0034] Among them, multi-scale historical data include multi-scale time data, multi-dimensional meteorological data, historical load data and historical clearing price data.

[0035] Specifically, multi-scale time data includes year, month, day, hour, and minute.

[0036] Multi-dimensional meteorological data includes dew point data, temperature data, humidity data, rainfall data, wind direction data, wind speed data, etc.

[0037] The historical load data is the load information of the historical power grid spot trading new energy power generation system, specifically the historical load curve data, wherein the historical load curve data represents the load data information of the new energy power generation system and the output data of the new energy power generation system, wherein the output data of the new energy power generation system includes the output data of the wind power generation system, the output data of the photovoltaic power generation system, etc.

[0038] The historical clearing price data refers to the historical grid spot transaction clearing price information.

[0039] Step S102 , performing correlation analysis on the multi-scale historical data, extracting strongly correlated coupling data from the multi-scale historical data, and obtaining multi-dimensional feature data.

[0040] Among them, correlation analysis is to analyze the degree of relationship between two or more variables. Correlation analysis is performed on multi-scale historical data to mine and extract the coupling characteristics between multi-scale historical data, extract the strongly correlated coupling data with a high degree of relationship in the multi-scale historical data, and obtain multi-dimensional feature data.

[0041] Specifically, the correlation coefficients between the multi-scale historical data are calculated, and the correlation coefficients are used to characterize the degree of relationship between the multi-scale historical data.

[0042] The correlation coefficients include Pearson correlation coefficient, Spearman correlation coefficient, Chebyshev correlation coefficient, etc. The present application performs correlation analysis on multi-scale historical data based on one or more correlation coefficients.

[0043] Step S103, weighted fusion is performed in sequence according to the data features of the multi-dimensional feature data to obtain fused data.

[0044] The multi-dimensional feature data includes data with multiple different data features. This application performs multiple weighted fusions on the multiple data to obtain fused data. Specifically, for data with different data features, the data with the same data features are first weighted fused, and then other data are weighted fused again based on the fusion.

[0045] It can be understood that the fusion method of the present application merges and fuses all data step by step according to the fusion order, and the number of fusion times is not limited.

[0046] Exemplarily, the multi-dimensional feature data includes feature data 1, feature data 2, feature data 3, and feature data 4. In one implementation, the first step fuses feature data 1 and feature data 2, the second step adds the fused feature data 3, and the third step adds the fused feature data 4. In another implementation, the first step fuses feature data 1 and feature data 2, feature data 3 and feature data 4, respectively, and the second step fuses the fusion results of the first step.

[0047] Because multi-dimensional feature data includes four types of data with different complexities, namely, multi-scale time data, multi-dimensional meteorological data, historical load data, and historical clearing price data, it is difficult to directly fuse data of different complexities, and the representation of the fused data types is uneven. This application fuses data in steps according to data characteristics, selectively weights different types of data, and can extract and retain the unique feature information of each data type, thereby improving the accuracy of subsequent prediction results.

[0048] Step S104, using the fused data as input data of the Informer network, predicting the load electricity price, and obtaining the load value and electricity value.

[0049] Informer network is an algorithm designed for long sequence time series prediction. It adopts ProbSparseSelf-Attention mechanism, which effectively reduces time complexity and memory usage, thereby improving the efficiency and accuracy of long sequence prediction.

[0050] This application uses the fused data as the input data of the Informer network, deeply mines the fused data, extracts the temporal and spatial features in the short-term time series of the fused data, captures the dependencies in the long-term time series, and jointly predicts the load and electricity price through its multi-layer encoder-decoder structure, outputting the load value and electricity value respectively.

[0051] The multi-dimensional load electricity price forecasting method provided in this embodiment considers the impact of multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data on the forecasting results, and extracts strongly correlated coupling data based on correlation analysis, taking into account the correlation between feature data. Using strongly correlated coupling data as subsequent model input can improve the model prediction accuracy, and weighted fusion of multi-scale historical data in sequence can effectively fuse features of different feature types and complexities. Based on the Informer model, the processing capability of long sequence data is further optimized, and the load electricity price is predicted at the same time, thereby improving the accuracy of load electricity price prediction in the spot trading process.

[0052] In this embodiment, a multi-dimensional load electricity price prediction method is provided, which can be used for the above-mentioned mobile terminals, such as mobile phones, tablet computers, etc. Figure 2 is a flowchart of another multi-dimensional load electricity price prediction method according to an embodiment of the present invention. It should be noted that if there are substantially the same results, this embodiment does not use Figure 2 The process sequence shown is limited. Figure 2 As shown, the process includes the following steps:

[0053] Step S201, obtaining multi-scale historical data.

[0054] Among them, multi-scale historical data includes multi-scale time data, multi-dimensional meteorological data, historical load data and historical clearing price data. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0055] Furthermore, the multi-scale historical data are preprocessed.

[0056] First, outlier detection is performed on multi-scale historical data based on the lonely forest algorithm. The lonely forest algorithm identifies and detects outliers in multi-scale time data, multi-dimensional meteorological data, historical load data, and historical clearing price data, and isolates and segments outliers from multi-scale historical data.

[0057] Specifically, the steps of detecting outliers on multi-scale historical data based on the lonely forest algorithm include:

[0058] Step a1, initializing model parameters. Initialize and set initial parameters such as the number of lonely trees and the maximum number of samples, and adjust the initial parameters according to the multi-scale historical data set and the current load and electricity price forecast requirements.

[0059] Step a2, generate a lonely tree. For each lonely tree, the algorithm randomly selects multi-scale historical data features and split points, and recursively splits the multi-scale historical data set into smaller and smaller subsets by randomly selecting a value between the maximum and minimum values ​​of the selected features to split the data point until the stopping condition is met.

[0060] Step a3, build a decision tree. The construction process of each decision tree is a recursive splitting process until all multi-scale historical data points are isolated or the preset tree height is reached. At each node, the algorithm is used to divide the multi-scale historical data into two parts according to the randomly selected feature values, and the process continues.

[0061] Step a4, anomaly score calculation. For each multi-scale historical data point, the path length from the point to the root node is calculated in each tree. It can be understood that anomalies are easier to be isolated than normal points, so the path length from the anomaly point to the root node is generally shorter.

[0062] Step a5, model integration. After all the lonely trees are constructed, the results of the lonely trees are integrated using an algorithm, and the anomaly score of each data point is evaluated. Generally, the anomaly score of each data point is evaluated by averaging the path length of the point in each isolated tree.

[0063] Step a6, anomaly detection. Identify the anomaly points according to the anomaly scores calculated in the above steps. Generally, the lower the anomaly score, the more likely the data point is an anomaly. And separate the anomaly points from the multi-scale historical data.

[0064] Secondly, for each outlier, the K data values ​​closest to the outlier are obtained based on the KNN algorithm, and the outlier is assigned a value based on the K data values. Among them, is a distance-based method that can be used to fill in missing values. It can be understood that the outliers are proposed using the above method, and the missing values ​​are filled using the KNN algorithm.

[0065] Specifically, the steps of supplementing missing values ​​based on the KNN algorithm include:

[0066] Step b1, select K value. The number of nearest neighbor data points K is predetermined, that is, the missing values ​​are filled using K nearest neighbor data points as references. The number of K values ​​depends on the size of the data set and the distribution of the features. It can be understood that the larger the data set, the larger the K value.

[0067] Step b2, determine missing data points. Identify which data points in the multi-scale historical dataset have missing values.

[0068] Step b3, find the nearest neighbor. For each missing value, find the K most similar data points in the data set. Specifically, for each missing value, find the K closest data points in the data set as reference data points based on the distance metric.

[0069] Step b4, calculate the mode, mean or median. For each missing value, fill the missing value with the mode, mean or median of the most similar data point.

[0070] Specifically, different calculation methods can be selected according to the data type. The data type includes categorical data and numerical data. For categorical data, the mode of the K nearest neighbors can be used to fill the missing values; for numerical data, the mean or median of the K nearest neighbor data points can be used to fill the missing values.

[0071] Step b5, fill in the missing values. Assign the calculated mode, mean or median to the position of the missing value.

[0072] Step b6, iterative processing. Repeat the above steps until all missing values ​​are filled.

[0073] Step S202 , performing correlation analysis on the multi-scale historical data, extracting strongly correlated coupling data from the multi-scale historical data, and obtaining multi-dimensional feature data.

[0074] Correlation analysis is performed on the multi-scale historical data, and at least part of the time series data, at least part of the meteorological data, historical load data and historical clearing price data with strong correlation coupling are extracted from the multi-scale historical data to obtain multi-dimensional feature data.

[0075] Specifically, the above step S202 includes:

[0076] Step S2021, perform multi-time sampling on the multi-scale historical data and construct a high-dimensional data matrix with time series.

[0077] Within a sampling time t, the N-dimensional multi-scale historical data is sampled to obtain the multi-scale historical data column vector at time t:

[0078] x t =[x 1,t x 2,t ……x N,t ] T .

[0079] Sampling is performed at multiple sampling times to obtain multi-scale historical data column vectors at multiple times, and the multi-scale historical data column vectors at multiple times are spliced ​​to obtain a high-dimensional data matrix with time series:

[0080] x=[x1,x2…x t …].

[0081] Furthermore, the high-dimensional data matrix is ​​standardized, that is, dedimensionalized and normalized. Specifically, the standardization formula is:

[0082] Among them, represents the data mean and represents the data standard deviation.

[0083] Step S2022, performing correlation analysis on the data in the high-dimensional data matrix, and calculating the correlation coefficients between the various multi-scale historical data.

[0084] Specifically, the correlation coefficients between the multi-scale historical data are calculated, and the correlation coefficients are used to characterize the degree of relationship between the multi-scale historical data.

[0085] The correlation coefficients include Pearson correlation coefficient, Spearman correlation coefficient, Chebyshev correlation coefficient, etc. The present application performs correlation analysis on multi-scale historical data based on one or more correlation coefficients.

[0086] In one implementation, a Pearson correlation analysis is performed on the data in the high-dimensional data matrix to calculate the Pearson correlation coefficient between each multi-scale historical data.

[0087] For example, the Pearson correlation coefficient between each two data in a high-dimensional data matrix is ​​calculated:

[0088]

[0089] Among them, ρ X,Y represents the Pearson correlation coefficient between data X and data Y, cov(X,Y) represents the covariance between data X and data Y, σ X represents the standard deviation of data X, σ Y Represents the standard deviation of the data Y.

[0090] Among them, when the Pearson correlation coefficient between data X and data Y is 0, it means that there is no relationship between the two variables of data X and data Y. When the Pearson correlation coefficient between data X and data Y is between 0.00 and 1.00, it means that the two variables of data X and data Y are positively correlated. When the value of data X increases (decreases), the value of data Y also increases (decreases) synchronously. When the Pearson correlation coefficient between data X and data Y is between -1.00 and 0.00, it means that the two variables of data X and data Y are negatively correlated. When the value of data X increases (decreases), the value of data Y decreases (increases).

[0091] Among them, the larger the absolute value of the Pearson correlation coefficient between data X and data Y, the stronger the correlation between data X and data Y, the closer the Pearson correlation coefficient between data X and data Y is to 1 or -1, the stronger the correlation between data X and data Y, and the closer the Pearson correlation coefficient between data X and data Y is to 0, the weaker the correlation between data X and data Y.

[0092] In a specific implementation, if the Pearson correlation coefficient between data X and data Y is between 0.8-1.0, it means that data X and data Y are extremely strongly correlated. If the Pearson correlation coefficient between data X and data Y is between 0.6-0.8, it means that data X and data Y are strongly correlated. If the Pearson correlation coefficient between data X and data Y is between 0.4-0.6, it means that data X and data Y are moderately correlated. If the Pearson correlation coefficient between data X and data Y is between 0.2-0.4, it means that data X and data Y are weakly correlated. If the Pearson correlation between data X and data Y is between 0.0-0.2, it means that data X and data Y are extremely weakly correlated or not correlated.

[0093] Among them, the Pearson correlation coefficient analysis method is used to perform correlation analysis on multi-scale historical data, which can quantify the correlation strength in the collected multi-scale historical data. By performing this analysis on data features at different scales, features with significant correlation with load and electricity price can be effectively screened out, thereby providing key features as input variables for the subsequent modeling and prediction process.

[0094] Step S2023, extracting at least part of the data whose correlation coefficient is greater than the correlation threshold from the multi-scale historical data to obtain multi-dimensional feature data.

[0095] A correlation threshold is obtained in advance, and according to the Pearson correlation coefficient, data having a Pearson correlation coefficient greater than the correlation threshold is determined from the multi-scale historical data to construct multi-dimensional feature data.

[0096] Specifically, at least part of the time series data, at least part of the meteorological data, historical load data and historical clearing price data are extracted.

[0097] In a specific implementation, the multi-dimensional feature data includes a time window one year before the forecast base date, meteorological data one year before the measurement base date, actual load data one year before the forecast base date, and clearing price data one year before the forecast base date. The granularity of the multi-dimensional feature data is 15 minutes.

[0098] Step S203, weighted fusion is performed in sequence according to the data features of the multi-dimensional feature data to obtain fused data.

[0099] Among them, the factors involved in at least part of the time series data, at least part of the meteorological data, the historical load data and the historical clearing price data include four types of data with different complexity levels: time, meteorology, load and electricity price. It is difficult to directly merge data of different complexity, and the representation of the merged data types is uneven.

[0100] This application targets data with four different data characteristics, namely time, weather, load and electricity price. The data with the same data characteristics are first weighted and fused. On the basis of the fusion, other data are weighted and fused again.

[0101] Specifically, the data type of the multi-dimensional feature data is obtained, and a weighted fusion of the same type of data is performed according to the data type of the multi-dimensional feature data. A second weighted fusion of the same type of data is performed according to the data type of the fused data. The above method is used to perform weighted fusion of the same type of data multiple times until the fusion of the multi-dimensional feature data is completed to obtain fused data.

[0102] Step S204, using the fused data as input data of the Informer network, predicting the load electricity price, and obtaining the load value and electricity value.

[0103] The Informer network is an algorithm designed for long-sequence time series prediction. The Informer network includes an input layer, multiple encoding layers, multiple decoding layers, and a two-layer linear output layer.

[0104] Specifically, the above step S204 includes:

[0105] Step S2041, input the fused data into the input layer of the Informer network to obtain a feature vector sequence of the fused data.

[0106] The fused data is input into the input (Input Embedding) layer of the Informer network. First, embedding encoding is performed in the input layer to map the discrete feature identifiers of the fused data into dense vector representations. Positional encoding is added to the vector, and positional encoding is used to give a unique identifier to the position of each vector sequence. After being processed by the Input Embedding layer, the feature vector sequence of the fused data is obtained, and the feature vector sequence is output to the next network unit, namely the encoding layer.

[0107] Step S2042: At the encoding layer, feature extraction is performed on the feature vector sequence based on the attention mechanism to obtain first feature data.

[0108] The feature vector sequence of the fused data is input into the encoder unit layer. In the encoder unit layer, Informer introduces the ProbSparse Attention mechanism to replace the traditional multi-head self-attention operation. The ProbSparse Attention mechanism reduces the complexity of attention calculation through the probability sparsification strategy. ProbSparse Attention uses the most representative keys for attention calculation, and the calculated results are processed through the Add & Normalize step and input into the Feed-Forward Network (FFN) for further nonlinear transformation. Finally, the Add & Normalize step is applied again to normalize the output features, and the output features are transmitted to the next encoder layer to continue feature extraction and information transfer. After multiple encoding layers, feature extraction is performed to obtain the first feature data, and the first feature data is output to the decoding layer.

[0109] Among them, ProbSparse Attention uses the most representative keys for attention calculation, which can greatly reduce the amount of calculation.

[0110] Step S2043: At the decoding layer, feature extraction is performed on the first feature data based on the cross-attention mechanism to obtain second feature data.

[0111] The first feature data is input into the decoder unit layer. Similarly, the decoder unit layer uses the ProbSparse Attention mechanism to reduce the complexity of attention calculation. When executing the multi-head self-attention mechanism, the query (Query, Q) vector comes from the decoder, and the key (Key, K) and value (Value, V) vectors come from the encoder's memory. The decoder of the present application also includes an additional cross attention module, which allows the decoder to focus on relevant parts of the encoder. Multiple components work together to generate a rich, hierarchical feature representation, and transmit the feature representation to the next decoding layer to continue feature extraction and information transfer. After multiple decoding layers, feature extraction is performed to obtain the second feature data, and the second feature data is output to the double-layer linear output layer.

[0112] Step S2044, in the double-layer linear output layer, the coupling dependency of the second feature data in the spatial dimension is extracted based on the linear transformation, and the output load value and the electricity value are predicted at the same time.

[0113] The second feature data is input into the double-layer linear output layer, and the high-dimensional feature space is mapped to the prediction space through the linear transformation layer. Among them, the Softmax activation function is applied to the double-layer linear output layer to extract the coupling dependency of the second feature data in the spatial dimension, and predict the output load value and power value at the same time.

[0114] Among them, the improved Informer model further optimizes the processing capability of long sequence data. Informer introduces a local self-attention mechanism, which only considers some elements in the sequence for attention calculation, and a probabilistic self-attention mechanism, which uses a probability matrix to approximate the complete attention weight matrix. These two mechanisms significantly reduce the computational complexity. In addition, Informer adopts a sparse attention mechanism to selectively focus on key points in the sequence. These improvements not only reduce computational costs and memory usage, but also maintain good prediction performance, making Informer an ideal choice for processing large-scale time series data. Therefore, in the joint prediction of load and electricity prices, the Informer algorithm has better efficiency and prediction accuracy.

[0115] The method provided in this embodiment considers the impact of multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data on the prediction results, and extracts strongly correlated coupling data based on correlation analysis, taking into account the correlation between feature data. Using strongly correlated coupling data as subsequent model input can improve the model prediction accuracy, and weighted fusion of multi-scale historical data in sequence can effectively fuse features of different feature types and complexities. The processing capability of long sequence data is further optimized based on the Informer model, and the load electricity price is predicted at the same time, thereby improving the accuracy of load electricity price prediction in the spot trading process.

[0116] Specifically, the lonely forest algorithm can handle outlier problems in high-dimensional big data, and has the characteristics of high processing accuracy and fast calculation speed. KNN can effectively fill in the problem of a small number of missing values ​​in the data. In spot electricity trading, Pearson correlation analysis of multi-scale time windows and multi-dimensional meteorological factors can effectively screen out strongly correlated feature matrices. The application of data fusion technology to integrate and process multi-dimensional data can effectively construct a coupled feature set, thereby providing a preprocessing basis for the input layer of the neural network model. This step is crucial for feature extraction and pattern recognition of subsequent models, and helps to improve the data processing capabilities of network models in complex data analysis.

[0117] In this embodiment, a specific embodiment of a multi-dimensional load electricity price prediction method is provided. The embodiment data comes from a data set of spot market transactions of a certain power grid. The historical data set records the measurement data of facilities from January 2023 to June 2023. The data set contains a variety of feature data. Some data of electricity spot transactions are shown in Table 1.

[0118] Table 1 Partial data of electricity spot trading

[0119]

[0120]

[0121] The first step is to obtain multi-scale historical data.

[0122] In the second step, outliers in the data matrix are removed based on the lonely forest algorithm, and the missing values ​​are filled using the KNN method.

[0123] In one implementation, the pre-made parameters of Lonely Forest include subsample size: 128, Tree height: 16, Number of trees: 100. In other embodiments, it can be flexibly adjusted according to the needs of a specific data set.

[0124] In one implementation, the samples are padded according to the size K=3 of this embodiment. In other embodiments, it can be flexibly adjusted according to the needs of a specific data set.

[0125] The third step is to construct a reasonable high-dimensional data matrix after completing the repair of the collected data, and standardize the high-dimensional data matrix, that is, de-dimensionalize and normalize it.

[0126] In one implementation, multi-scale time data includes five-dimensional time information, namely year, month, day, hour, and minute; multi-dimensional meteorological elements include three dimensions, namely temperature, dew point, and rainfall; and data sampling includes ten-dimensional information, namely historical load curves and historical clearing prices.

[0127] In one implementation, the data is dimensionally eliminated by using a 0-1 normalization method. Other normalization methods may also be used.

[0128] In the fourth step, we conducted in-depth feature extraction on multi-scale time windows, multi-dimensional meteorological factors, historical load curves, and historical clearing prices to construct a multi-dimensional input feature framework.

[0129] In one implementation, the Pearson correlation method is used to perform correlation analysis on multi-scale historical data. Figure 3 , Figure 3 Schematic diagram of the correlation analysis result according to an embodiment of the present invention. Figure 3 As shown, the actual clearing price, system load, daily characteristics, and hourly characteristics are highly correlated, while the correlations of other factors are relatively small. The reason why the annual and monthly characteristics have a relatively small correlation is mainly due to the short time range of the data volume.

[0130] The multi-dimensional feature data is constructed based on the correlation analysis structure:

[0131] Input1: The "day" time window one year before the forecast base date;

[0132] Input2: The time window of “hours” one year before the forecast base date;

[0133] Input3: "Temperature" data one year before the prediction base date;

[0134] Input4: "dew point" data one year before the forecast base date;

[0135] Input5: actual load data one year before the forecast base date;

[0136] Input6: Clearing price data one year before the forecast base date.

[0137] In this implementation, features with less correlation, such as the "year" characteristic, are not used as multi-dimensional feature data. Similarly, in the execution of other embodiments, feature selection is not entirely based on correlation analysis. According to experience, a manual correction step needs to be added to increase the data's executable degree and prediction accuracy.

[0138] The fourth step is to use feature fusion technology to effectively integrate data information from different scales and dimensions.

[0139] See also Figure 4 , Figure 4 FIG. 1 is a schematic diagram of a feature fusion method according to an embodiment of the present invention. Figure 4 As shown, the present application first fuses the first feature data and the second feature data, and then adds and fuses the third feature data to obtain fused data.

[0140] This application uses step-by-step feature fusion technology based on data types to effectively integrate data information from different scales and dimensions.

[0141] The fifth step is to predict the load electricity price based on the Informer network to obtain the load value and electricity value.

[0142] The sixth step is to evaluate the prediction performance after obtaining the prediction results.

[0143] In one implementation, the mean absolute percentage error (MAPE) is used as the main evaluation indicator of the prediction performance of each model, and the mean absolute error (MAE) is selected as an auxiliary evaluation indicator. The evaluation indicators shown in this embodiment are only used for this case. In other embodiments, performance indicators that are closer to the project requirements can also be selected for implementation according to task needs. The calculation formulas of MAPE and MAE in this embodiment are as follows:

[0144]

[0145] Among them, y t For actual data; For the predicted data; is the mean; n is the number of samples of data.

[0146] In this embodiment, a multi-dimensional load electricity price prediction device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0147] This embodiment provides a multi-dimensional load electricity price prediction device, such as Figure 5 As shown, including:

[0148] The acquisition module 501 is used to acquire multi-scale historical data, which includes multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data.

[0149] The analysis module 502 is used to perform correlation analysis on the multi-scale historical data, extract strongly correlated coupling data from the multi-scale historical data, and obtain multi-dimensional feature data.

[0150] The fusion module 503 is used to perform weighted fusion in sequence according to the data features of the multi-dimensional feature data to obtain fused data.

[0151] The prediction module 504 uses the fused data as input data of the Informer network to predict the load electricity price and obtain the load value and electricity value.

[0152] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0153] The multi-dimensional load electricity price prediction device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0154] The embodiment of the present invention also provides a computer device having the above Figure 6 The multi-dimensional load electricity price forecasting device shown.

[0155] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0156] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0157] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0158] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0159] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0160] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0161] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0162] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0163] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0164] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A multi-dimensional load electricity price prediction method, characterized in that: The method comprises: Acquire multi-scale historical data, wherein the multi-scale historical data includes multi-scale time series data, multi-dimensional meteorological data, historical load data, and historical clearing price data; Performing correlation analysis on the multi-scale historical data, extracting strongly correlated coupling data from the multi-scale historical data, and obtaining multi-dimensional feature data; Perform weighted fusion in sequence according to the data features of the multi-dimensional feature data to obtain fused data; The fused data is used as input data of the Informer network to predict the load electricity price and obtain the load value and electricity value.

2. The multi-dimensional load electricity price prediction method according to claim 1, characterized in that: The performing correlation analysis on the multi-scale historical data, extracting strongly correlated coupling data from the multi-scale historical data, and obtaining multi-dimensional feature data includes: A correlation analysis is performed on the multi-scale historical data, and at least a portion of the time series data, at least a portion of the meteorological data, the historical load data, and the historical clearing price data that are strongly correlated and coupled are extracted from the multi-scale historical data to obtain multi-dimensional feature data.

3. The multi-dimensional load electricity price prediction method according to claim 2, characterized in that: The performing of correlation analysis on the multi-scale historical data, extracting at least part of the time series data, at least part of the meteorological data, the historical load data and the historical clearing price data that are strongly correlated and coupled from the multi-scale historical data, and obtaining the multi-dimensional feature data includes: Performing multi-time sampling on the multi-scale historical data and constructing a high-dimensional data matrix with time series; Performing correlation analysis on the data in the high-dimensional data matrix to calculate the correlation coefficient between each of the multi-scale historical data; At least part of the data having a correlation coefficient greater than a correlation threshold is extracted from the multi-scale historical data to obtain the multi-dimensional feature data.

4. The multi-dimensional load electricity price prediction method according to claim 1, characterized in that: The weighted fusion is performed in sequence according to the data features of the multi-dimensional feature data to obtain fused data, including: Acquire the data type of the multi-dimensional feature data; Performing a weighted fusion of data of the same type according to the data type of the multi-dimensional feature data; Perform secondary weighted fusion of data of the same type according to the data type of the fused data; The weighted fusion of the same type of data is performed multiple times until the fusion of the multi-dimensional feature data is completed to obtain fused data.

5. The multi-dimensional load electricity price prediction method according to claim 4, characterized in that: The Informer network includes an input layer, multiple encoding layers, multiple decoding layers and a double-layer linear output layer. The fusion data is used as the input data of the Informer network to predict the load electricity price, and the load value and electricity value obtained include: Inputting the fused data into the input layer of the Informer network to obtain a feature vector sequence of the fused data; At the encoding layer, extracting features from the feature vector sequence based on an attention mechanism to obtain first feature data; At the decoding layer, feature extraction is performed on the first feature data based on a cross attention mechanism to obtain second feature data; In the double-layer linear output layer, the coupling dependency of the second feature data in the spatial dimension is extracted based on linear transformation, and the load value and the electricity value are predicted and output simultaneously.

6. The multi-dimensional load electricity price prediction method according to claim 1, characterized in that: Before the correlation analysis of the multi-scale historical data is performed, the following step is further included: Performing outlier detection on the multi-scale historical data based on the lonely forest algorithm; For each of the outliers, obtaining K data values ​​closest to the outlier based on the KNN algorithm; The outlier values ​​are assigned based on the K data values.

7. A multi-dimensional load electricity price prediction device, characterized in that: The device comprises: An acquisition module, used to acquire multi-scale historical data, wherein the multi-scale historical data includes multi-scale time series data, multi-dimensional meteorological data, historical load data and historical clearing price data; An extraction module, used to perform correlation analysis on the multi-scale historical data, extract strongly correlated coupling data from the multi-scale historical data, and obtain multi-dimensional feature data; A fusion module, used for performing weighted fusion in sequence according to the data features of the multi-dimensional feature data to obtain fused data; The prediction module is used to use the fused data as input data of the Informer network to predict the load electricity price and obtain the load value and the electricity value.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the multi-dimensional load electricity price forecasting method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the multi-dimensional load electricity price forecasting method according to any one of claims 1 to 6.

10. A computer program product, characterized in that It includes computer instructions, which are used to enable a computer to execute the multi-dimensional load electricity price forecasting method described in any one of claims 1 to 6.

Images