Short-term power load prediction method and system considering comprehensive climate and load, and medium
By extracting comprehensive climate characteristics and load data for dual evaluation and screening, and combining the BiLSTM model and Holt-Winters smoothing method, the problem of inaccurate power load forecasting under extreme weather conditions is solved, and higher forecast accuracy is achieved.
Patent Information
- Application Number
- CN202511706965.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-02-27
AI Technical Summary
Existing technologies fail to adequately consider the impact of climate factors on power load during periods of frequent extreme weather, resulting in inaccurate short-term power load forecasts. Furthermore, considering only load characteristics leads to large forecast errors.
By extracting comprehensive climate features such as temperature and humidity index and perceived temperature, and combining them with power load data for dual evaluation and screening, load forecasting is performed using the BiLSTM model. Feature decomposition is then performed using the Holt-Winters triple exponential smoothing method, and the results are input into the BiLSTM model for prediction.
It improves the accuracy and precision of short-term power load forecasting, and significantly reduces forecasting errors, especially under extreme weather conditions.
Smart Images
Figure CN121584545A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of power load prediction, in particular to a short-term power load prediction method and system considering comprehensive climate and load, and a medium. BACKGROUND
[0002] In the key stage of new power system construction, short-term power load prediction, as the core technology support for power grid dispatching optimization, power generation planning and power supply reliability guarantee, its prediction accuracy is directly related to the operation efficiency and safety resilience of the power system. In recent years, the frequent occurrence of extreme weather has intensified the fluctuation of power load, bringing new challenges to short-term power load prediction. To cope with these difficulties, the existing technology has proposed a power load prediction method based on an improved federated learning algorithm in response to the requirements of power users in data privacy protection, which can reduce the impact of malicious attacks while ensuring high prediction accuracy. In view of the fact that most existing prediction methods use the time correlation of historical load sequences to construct models, there is a limitation that spatial coupling and external factors are not considered. Accordingly, the existing technology proposes a power system multi-node load prediction method based on attribute-enhanced multi-scale spatio-temporal graph convolutional neural network and Transformer fusion. In view of the problem of poor short-term load prediction accuracy of a single model, the existing technology proposes a combination model based on a deep learning algorithm, which decomposes, predicts and combines the load sequence to obtain higher prediction accuracy. In view of the problem of less data and low load prediction accuracy under extreme high temperature weather, the existing technology proposes a load prediction model based on the combination of TimeGAN and BiLSTM, TimeGAN increases extreme weather scenario data, and BiLSTM performs load prediction, the prediction error of the proposed model is significantly reduced compared with traditional models. To solve the problem of increasing power load volatility and complexity in new power systems and the difficulty of accurate prediction, the existing technology proposes a prediction model based on load secondary decomposition and feature processing fusion, which significantly reduces the error compared with mainstream deep learning prediction models, and the existing technology also proposes a short-term power load prediction model based on similar days and bidirectional long short-term memory neural network combination. The actual data of a certain place in New England is simulated and verified, and the proposed model is superior to the control model in terms of prediction accuracy, improving the load prediction accuracy. However, the above models have two limitations, first, the frequent occurrence of extreme weather leads to abnormal fluctuations in cooling and heating loads in the power system, and the influence of climate factors on load is not fully considered; second, most existing models only consider power load sequence features, and fail to construct climate factors and load features as inputs for deep learning model prediction. SUMMARY
[0003] In order to solve the problem that the prior art cannot fully consider the influence of climate factors on load when dealing with abnormal fluctuation of cooling and heating load in power system caused by frequent end weather, and the problem of inaccurate prediction caused by single consideration of load characteristics in load prediction process, the present application provides a short-term power load prediction method and system considering comprehensive climate and load, and a medium, which improves the method on the basis of the prior art, processes the basic meteorological data into comprehensive climate characteristics with stronger correlation with load, extracts the horizontal component, trend component and seasonal component of power load data, combines the two as the input of BiLSTM model for load prediction, and effectively improves the prediction accuracy of short-term power load.
[0004] The above technical purpose of the present application is realized by the following technical scheme:
[0005] The present application provides a short-term power load prediction method considering comprehensive climate and load, which comprises the following steps:
[0006] Collecting meteorological data and power load data of a target area;
[0007] Extracting comprehensive climate characteristics affecting power load based on the meteorological data;
[0008] Conducting double evaluation and screening of the comprehensive climate characteristics in combination with the power load data to obtain target climate characteristics;
[0009] Conducting multi-component update processing on the power load data;
[0010] Inputting the target load characteristics and target climate characteristics after multi-component update processing into the BiLSTM power load prediction model constructed to predict the short-term power load of the target area.
[0011] Further optimization scheme is that the extraction of comprehensive climate characteristics based on the meteorological data comprises the following steps:
[0012] Calculating the temperature and humidity index THI, effective temperature ET, apparent temperature AT and human comfort index SSD of the target area based on the meteorological data;
[0013] ;
[0014] ;
[0015] ;
[0016] ;
[0017] Wherein, RH represents the relative humidity of the target area; v represents the wind speed of the target area; T represents the temperature of the target area; Q represents the solar radiation intensity of the target area; and exp() represents a natural exponential function.
[0018] Further optimization scheme is that the target climate characteristics are obtained by double evaluation and screening of the comprehensive climate characteristics combined with the power load data; the method comprises the following steps:
[0019] The linear correlation strength and the nonlinear correlation strength of each input variable and output variable are determined by taking the comprehensive climate characteristics as input variables and taking the power load data as output variables;
[0020] The comprehensive importance score of each input variable is calculated combined with the linear correlation strength and the nonlinear correlation strength, and the input variable with the comprehensive importance score reaching a first threshold value is taken as a candidate factor;
[0021] The multiple collinearity of all candidate factors is tested based on a variance inflation factor, and the candidate factor with the variance inflation factor reaching a second threshold value is taken as the target climate characteristic.
[0022] Further optimization scheme is that the calculation method of the variance inflation factor comprises the following steps: taking each climate factor as a dependent variable and taking all remaining candidate factors as independent variables, a multiple linear regression model is constructed, and a determination coefficient R 2 is calculated; and the variance inflation factor VIF is calculated according to the following formula:
[0023] .
[0024] Further optimization scheme is that the power load data is subjected to multi-component update processing; the method comprises the following steps:
[0025] The power load data is subjected to characteristic decomposition based on a Holt-Winters triple exponential smoothing method to obtain a level update component of the power load data at time t, a trend update component and a seasonal update component :
[0026] ;
[0027] ;
[0028] ;
[0029] Wherein, represents an observation value, represents a level at time t, represents a seasonal component of a previous seasonal period, denotes trend component, denotes a horizontal smoothing constant; denotes a trend smoothing constant; denotes a seasonal smoothing constant.
[0030] Further optimization scheme is that the BiLSTM power load prediction model is a bidirectional structure network composed of LSTM basic units, the BiLSTM power load prediction model includes a forward LSTM layer and a backward LSTM layer; the forward LSTM layer processes from the sequence starting position to the rear, for capturing information from the past to the present; the backward LSTM layer processes from the sequence end position to the front, capturing information from the future to the present.
[0031] Further optimization scheme is that the LSTM basic unit includes a forgetting gate, an input gate and an output gate;
[0032] The state of the forgetting gate is:
[0033] ;
[0034] wherein, denotes time input; denotes a bias vector of the forgetting gate; denotes a weight matrix of the forgetting gate; denotes a hidden state at t-1 moment; ) denotes a Sigmoid function;
[0035] The state of the output gate is:
[0036] ;
[0037] wherein, denotes a bias vector of the output gate; denotes a weight matrix of the output gate.
[0038] Further optimization scheme is that the evaluation index of the BiLSTM power load prediction model is mean square error, mean absolute percentage error, determination coefficient and root mean square error.
[0039] The scheme also provides a short-term power load prediction system considering comprehensive climate and load, which is used to realize the short-term power load prediction method considering comprehensive climate and load, and the system includes:
[0040] The acquisition module is used to acquire meteorological data and power load data of a target area;
[0041] The feature extraction module is used to extract comprehensive climate features affecting power load based on the meteorological data;
[0042] An evaluation module is configured to perform double evaluation screening on the comprehensive climate features in combination with the power load data to obtain target climate features.
[0043] An updating module is configured to perform multi-component updating processing on the power load data.
[0044] A prediction module is configured to input the target load features and the target climate features after the multi-component updating processing into a built BiLSTM power load prediction model to predict short-term power load of a target region.
[0045] The present application also provides a computer readable medium having a computer program stored thereon, and the computer program is executable by a processor to implement the short-term power load prediction method considering comprehensive climate and load as described above.
[0046] Compared with the prior art, the present application has the following beneficial effects:
[0047] 1. The present application provides a short-term power load prediction method, system and medium considering comprehensive climate and load, which improves the method on the basis of the prior art, processes the basic meteorological data into comprehensive climate features with stronger correlation with the load, extracts the horizontal component, trend component and seasonal component of the power load data, combines the two as the input of the BiLSTM model for load prediction, and effectively improves the prediction accuracy of the short-term power load.
[0048] 2. The present application provides a short-term power load prediction method, system and medium considering comprehensive climate and load, which improves the accuracy of load prediction through a deep learning model in the case of difficult and low-precision power load prediction caused by frequent extreme weather, combines basic meteorological data such as temperature, light intensity, wind speed and other basic factors to form comprehensive climate features in the processing of meteorological data, analyzes the correlation through Pearson correlation coefficient and MIC correlation coefficient, and selects key features as the input of the BiLSTM model; in the processing of the power load data, the load data is decomposed into features based on the Holt-Winters triple exponential smoothing method, the horizontal, trend and seasonal components are input into the BiLSTM model, and multi-dimensional feature data is provided for the prediction model, and the above data processing process provides high-correlation features for the BiLSTM model to improve the calculation efficiency and prediction accuracy.
[0049] 3.The application provides a short-term power load forecasting method, system and medium considering comprehensive climate and load; the forecasting model adopts a BiLSTM model, the specific bidirectional time sequence information capturing capability of which can more accurately cope with the complex change rule of power load data, and simulation verification is carried out on the climate and power load data set disclosed in Texas, USA, and the results show that the BiLSTM model has higher prediction accuracy in prediction accuracy compared with traditional models. BRIEF DESCRIPTION OF DRAWINGS
[0050] In order to more clearly illustrate the technical solutions of the example embodiments of the application, the drawings needed to be used in the embodiments will be briefly introduced as follows, and it should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings. In the drawings:
[0051] Figure 1 The short-term power load forecasting method considering comprehensive climate and load is shown in the schematic diagram;
[0052] Figure 2 The short-term power load forecasting method considering comprehensive climate and load is shown in the schematic diagram;
[0053] Figure 3 The BiLSTM power load forecasting model architecture is shown in the schematic diagram;
[0054] Figure 4 The Pearson correlation coefficient matrix heat map of the comprehensive climate characteristics and power load data is shown in the schematic diagram;
[0055] Figure 5 The MIC correlation coefficient matrix heat map of the comprehensive climate characteristics and power load data is shown in the schematic diagram;
[0056] Figure 6 The power load data multi-component update processing data schematic diagram is shown in the schematic diagram;
[0057] Figure 7 The power load structure predicted by the method of the application and the real power load are compared in the schematic diagram;
[0058] Figure 8 The loss function effect of the forecasting model of the application is shown in the schematic diagram;
[0059] Figure 9 The prediction results of the three kinds of forecasting models and the real load are compared in the schematic diagram;
[0060] Figure 10 The evaluation index results of the method of the application and the traditional model are compared in the schematic diagram;
[0061] Figure 11A structure diagram of a short-term power load forecasting system considering comprehensive climate and load. DETAILED DESCRIPTION
[0062] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be given to the present application in combination with embodiments and drawings, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application.
[0063] The prior art fails to fully consider the influence of climate factors on load when dealing with abnormal fluctuation of cooling and heating load in power system caused by frequent end weather, and the single consideration of load characteristics in the load forecasting process leads to inaccurate prediction. In view of this, the present application provides the following embodiments to solve the above technical problems.
[0064] Embodiment 1
[0065] The embodiment provides a short-term power load forecasting method considering comprehensive climate and load, as shown in Figure 1 and Figure 2 The method comprises the following steps:
[0066] Step 1: collecting meteorological data and power load data of a target area;
[0067] Step 2: extracting comprehensive climate characteristics affecting power load based on the meteorological data; the step specifically comprises the following method:
[0068] The power consumption behavior of users is significantly affected by meteorological factors, and high-temperature weather in summer will especially cause the load of air conditioning and other refrigeration equipment to increase significantly. First, a plurality of comprehensive climate characteristics are analyzed, and then a multi-dimensional screening method is used to screen the optimal input features for the power consumption forecasting model. In processing meteorological data, in addition to the basic meteorological indexes such as temperature, relative humidity, wind speed and solar radiation, the comprehensive meteorological factors reflecting the interaction between these basic indexes are further considered; specifically, the temperature-humidity index THI, effective temperature ET, apparent temperature AT and human comfort index SSD of the target area are calculated based on the meteorological data;
[0069] ;
[0070] ;
[0071] ;
[0072] ;
[0073] wherein RH represents the relative humidity of the target area; v represents the wind speed of the target area; T represents the temperature of the target area; Q represents the solar radiation intensity of the target area; and exp() represents a natural exponential function.
[0074] Step three, double evaluation screening of target climate characteristics is carried out by combining the power load data; the step specifically includes the method:
[0075] The linear correlation strength and nonlinear correlation strength of each input variable and output variable are determined by taking the comprehensive climate characteristics as the input variable and taking the power load data as the output variable; in view of the complex correlation between the output power load and the input meteorological factors, it is difficult to realize quantification through a single formula. Therefore, the present scheme constructs a double evaluation framework from the linear and nonlinear dimensions; specifically, on the one hand, the Pearson correlation coefficient is used to measure the linear correlation between the meteorological input and the power load output, and on the other hand, the MIC method is used to capture the nonlinear correlation between the two, and finally rely on this multi-dimensional means to screen out the key input meteorological factors. The Pearson correlation coefficient is used to quantify the linear relationship between the input variables in the power load prediction; the specific calculation method is:
[0076] ;
[0077] In the formula, represents the linear correlation coefficient between the input variables, , represents and the average value of each; The closer to ±1, the stronger the correlation.
[0078] MIC is a statistical quantity based on mutual information to evaluate the nonlinear correlation strength between input variables, and adopts a grid division method to quantify the correlation degree between variables; the specific calculation method is:
[0079] ; ;
[0080] In the formula, represents the power load data sequence as the output variable, represents the th comprehensive climate characteristic sequence as the input variable, , respectively represent the specific values of sequences and . represents the joint probability distribution of and , and represent and Marginal probability distribution; the MIC value is between 0 and 1, the greater the value, the stronger the correlation; log represents the logarithmic function; |*| represents taking the absolute value; the specific constructed Pearson correlation coefficient matrix heat map and MIC correlation coefficient matrix heat map are as shown in Figure 4 and Figure 5 .
[0081] The comprehensive importance score of each input variable is calculated by combining the linear correlation strength and the nonlinear correlation strength, and the input variable whose comprehensive importance score reaches a first threshold value is taken as a candidate factor;
[0082] The multicollinearity of all candidate factors is tested based on the variance inflation factor, and the candidate factor whose variance inflation factor reaches a second threshold value is taken as a target climate feature.
[0083] Step four, performing multi-component updating processing on the power load data; this step specifically includes the method:
[0084] Power load data often contains complex dynamic characteristics, specifically covering horizontal, trend and seasonal components, and in-depth mining of the internal characteristics of load data is the core of improving prediction accuracy and depicting load variation law; this scheme uses Holt-Winters triple exponential smoothing method to perform feature decomposition on the load data, which can adaptively estimate and separate the three core components of the load sequence; Holt-Winters method is suitable for power load data with obvious multiple seasonality, and its core idea is to describe the current level, trend and seasonal components through three smoothing equations, and to complete prediction and decomposition work with the help of these components. Considering that the seasonal fluctuation amplitude of power load often changes significantly with the overall load level, this scheme selects a multiplication model for processing; specifically, based on the Holt-Winters triple exponential smoothing method, the power load data is decomposed to obtain the horizontal update component , the trend update component and the seasonal update component of the power load data at time t:
[0085] ;
[0086] ;
[0087] ;
[0088] wherein, represents the observed value, represents the horizontal component at time t, represents the seasonal component of the previous seasonal period, represents the trend component at time t, denotes a horizontal smoothing constant, controlling the trend updating speed; denotes a trend smoothing constant, controlling the trend updating speed; denotes a seasonal smoothing constant, controlling the seasonal component updating speed. The data of the power load after the multi-component updating process is shown in Figure 6
[0089] Step five, input the target load characteristics and target climate characteristics after the multi-component updating process into the built BiLSTM power load prediction model to predict the short-term power load of the target area.
[0090] As shown in Figure 3 , the BiLSTM power load prediction model is a bidirectional structure network composed of LSTM basic units, and the BiLSTM power load prediction model includes a forward LSTM layer and a backward LSTM layer; the forward LSTM layer processes from the sequence starting position to the back, capturing information from the past to the present; the backward LSTM layer processes from the sequence end position to the front, capturing information from the future to the present.
[0091] The BiLSTM power load prediction model in this scheme is an improved model that introduces a bidirectional structure based on the LSTM model, and the core is to capture past and future information of sequence data through hidden layers in the front and back directions, solving the limitation of traditional LSTM models that can only process sequences in one direction; its structure can be disassembled into basic units and bidirectional architecture, which are combined to achieve comprehensive extraction of sequence information. Each LSTM cell in the BiLSTM power load prediction model controls the forgetting, updating and output of information through three core components: the forgetting gate, the input gate and the output gate, which can effectively alleviate the problem of gradient disappearance in long sequence training. The BiLSTM power load prediction model contains two parallel LSTM layers, the forward LSTM layer processes from the sequence starting position to the back, capturing information from the past to the present, and the backward LSTM layer processes from the sequence end position to the front, capturing information from the future to the present, and finally the hidden states of the two directions are spliced as the final feature output at the current time.
[0092] The LSTM basic unit includes a forgetting gate, an input gate and an output gate;
[0093] The state of the forgetting gate is:
[0094] ;
[0095] wherein, denotes the input at time t; denotes the input at time t; denotes the bias vector of the forgetting gate; denotes the weight matrix of the forgetting gate; denotes the hidden state at time t-1; ) represents a sigmoid function;
[0096] The state of the output gate is:
[0097] ;
[0098] wherein, represents a bias vector of the output gate; represents a weight matrix of the output gate.
[0099] ;
[0100] ;
[0101] ;
[0102] ;
[0103] wherein, represents a forward hidden state; represents a backward hidden state; , represent weight matrices of the input gate and the memory cell, respectively; , represent bias vectors of the input gate and the memory cell, respectively; represents a Hadamard product; represents a forget gate state; represents a new memory cell; represents a current memory cell; represents an output gate state; tanh() represents a hyperbolic tangent function.
[0104] The evaluation indexes of the BiLSTM power load prediction model are mean square error, mean absolute percentage error, determination coefficient, and root mean square error.
[0105] Reasonable selection of evaluation indexes is a key step to test the performance of the BiLSTM power load prediction model. A proper standard can not only reflect the prediction accuracy, but also provide detailed reference information for model optimization. Systematic investigation of errors helps to accurately evaluate the difference between the prediction results and the true data, thereby improving the performance of the algorithm. Four evaluation indexes are selected in the present scheme: mean square error (MSE), mean absolute percentage error (MAPE), determination coefficient (R²), and root mean square error (RMSE). Among them, the mean square error reflects the average level of the square difference between the predicted value and the actual value; the mean absolute percentage error reflects the percentage size of the prediction error relative to the actual value; the determination coefficient evaluates the explanatory power of the model to the data fluctuation; and the root mean square error is used to convert the error scale back to the order of magnitude of the original data.
[0106] ;
[0107] ;
[0108] ;
[0109] ;
[0110] in, Indicates the number of samples in the test set. Indicates the first The predicted value for each sample, Indicates the first The true value of each sample;
[0111] This embodiment provides a schematic diagram comparing the predicted power load results with the actual power load using the method of the present invention, as shown below. Figure 7 As shown in the figure, the effect of the loss function of the prediction model of the present invention is illustrated in the diagram. Figure 8 As shown in the figure, this embodiment compares the proposed scheme with two other traditional prediction models. The prediction results of the three prediction models (BiLSTM model, GRU model, and M model in this scheme) are compared with the actual power load in the diagram. Figure 9 As shown in the diagram, the evaluation index results of the method of the present invention and the traditional model are compared. Figure 10 As shown, this scheme improves the accuracy of load forecasting through a deep learning model, which has significant advantages over traditional forecasting models.
[0112] In the processing of meteorological data, this scheme combines basic meteorological data such as temperature, light intensity, and wind speed to form comprehensive climate characteristics. The correlation between these characteristics is analyzed using Pearson correlation coefficient and MIC correlation coefficient, and key features are selected as inputs for the prediction model. In the processing of power load feature extraction, the Holt-Winters triple exponential smoothing method is used to decompose the load data into features, obtaining horizontal, trend, and seasonal components to input into the prediction model. The prediction model uses a BiLSTM model, whose unique bidirectional time-series information capture capability can more accurately address the complex changing patterns of power load data. Simulation verification is performed on a publicly available climate and power load dataset from Texas, USA. The results show that the method of this invention has higher prediction accuracy compared to traditional models.
[0113] Example 2
[0114] This embodiment provides a short-term power load forecasting system that takes into account both climate and load factors, such as... Figure 11 As shown, the system for implementing the short-term power load forecasting method considering integrated climate and load as described in Example 1 includes:
[0115] Collecting a meteorological data and a power load data of a target area;
[0116] Extracting a comprehensive climate feature affecting the power load based on the meteorological data;
[0117] Conducting double evaluation screening on the comprehensive climate feature in combination with the power load data to obtain a target climate feature;
[0118] Updating the power load data in a multi-component manner;
[0119] Inputting the target load feature and the target climate feature after the multi-component update into a BiLSTM power load prediction model to predict a short-term power load of the target area.
[0120] Embodiment 3
[0121] The embodiment provides a computer readable medium, which stores a computer program, and the computer program can realize the short-term power load prediction method considering comprehensive climate and load as described in the embodiment 1 when executed by a processor; and the following steps are specifically executed:
[0122] Step one, collecting a meteorological data and a power load data of a target area;
[0123] Step two, extracting a comprehensive climate feature affecting the power load based on the meteorological data;
[0124] Step three, conducting double evaluation screening on the comprehensive climate feature in combination with the power load data to obtain a target climate feature;
[0125] Step four, updating the power load data in a multi-component manner;
[0126] Step five, inputting the target load feature and the target climate feature after the multi-component update into a BiLSTM power load prediction model to predict a short-term power load of the target area.
[0127] The example uses the climate and load data disclosed in Texas, USA, as an example implementation, on the basis of analyzing the correlation between meteorological factors and load data, combining the decomposition of the characteristics of the Holt-Winters method for power load data as the input of the BiLSTM model; in the data set, 60% is the training set, which is used for the training and learning of the BiLSTM power load prediction model, 25% is the verification set, which is used for adjusting parameters in the training process, and 15% is the test set, which is used to evaluate the prediction performance of the BiLSTM power load prediction model; then compared with the traditional load prediction model, the comparison results show that the BiLSTM power load prediction model provided in the scheme has superior performance in evaluation indexes.
[0128] The specific embodiments described above further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A short-term power load forecasting method considering both climate and load factors, characterized in that, The methods include: Collect meteorological data and power load data for the target area; Based on the meteorological data, comprehensive climate characteristics affecting power load are extracted; The target climate characteristics are obtained by combining the power load data with a dual evaluation and screening of the comprehensive climate characteristics. The power load data is then processed using a multi-component update method. The target load characteristics and target climate characteristics after multi-component update are input into the constructed BiLSTM power load prediction model to predict the short-term power load of the target area.
2. The short-term power load forecasting method considering integrated climate and load as described in claim 1, characterized in that, The method for extracting comprehensive climate characteristics based on the meteorological data includes: Based on the meteorological data, the temperature and humidity index (THI), effective temperature (ET), perceived temperature (AT), and human comfort index (SSD) of the target area were calculated. ； ； ； ； Where RH represents the relative humidity of the target area; v represents the wind speed of the target area; T represents the temperature of the target area; Q represents the solar radiation intensity of the target area; and exp() represents the natural exponential function.
3. The short-term power load forecasting method considering integrated climate and load as described in claim 2, characterized in that, The target climate characteristics are obtained by combining the power load data with a dual evaluation and screening of the comprehensive climate characteristics. Including methods: Using comprehensive climate characteristics as input variables and electricity load data as output variables, the linear and nonlinear correlation strengths between each input variable and the output variable are determined. The overall importance score of each input variable is calculated by combining the linear correlation strength and the nonlinear correlation strength, and the input variable whose overall importance score reaches the first threshold is used as the candidate factor; Based on the variance expansion factor test, the multicollinearity of all candidate factors is tested, and the candidate factors whose variance expansion factor reaches the second threshold are used as the target climate features.
4. The short-term power load forecasting method considering integrated climate and load as described in claim 3, characterized in that, The method for calculating the variance inflation factor includes: By treating each climate factor as the dependent variable and all other candidate factors as independent variables, a multiple linear regression model is constructed, and the coefficient of determination R is calculated. 2 The variance inflation factor (VIF) is calculated using the following formula: 。 5. The short-term power load forecasting method considering integrated climate and load as described in claim 1, characterized in that, The power load data is then processed using a multi-component update method. Including methods: Based on the Holt-Winters triple exponential smoothing method, eigenvalue decomposition is performed on the power load data to obtain the horizontal update component of the power load data at time t. Trend update components and seasonally updated ingredients : ； ； ； in, Represents the observed value. express Time-level components, This indicates the seasonal component of the previous seasonal cycle. express Momentary trend components, Represents the horizontal smoothing constant; Represents the trend smoothing constant; This represents the seasonal smoothing constant.
6. The short-term power load forecasting method considering integrated climate and load as described in claim 1, characterized in that, The BiLSTM power load forecasting model is a bidirectional network composed of LSTM basic units. The BiLSTM power load forecasting model includes a forward LSTM layer and a backward LSTM layer. The forward LSTM layer processes information from the beginning of the sequence to the present, and is used to capture information from the past to the present. The backward LSTM layer processes information from the end of the sequence to the present, and is used to capture information from the future to the present.
7. The short-term power load forecasting method considering integrated climate and load as described in claim 6, characterized in that, The basic LSTM unit consists of a forget gate, an input gate, and an output gate. The state of the forget gate is: ； in, express Input at any time; The bias vector representing the forget gate; The weight matrix representing the forget gate; This represents the hidden state at time t-1; () represents the Sigmoid function; The state of the output gate is: ； in, This represents the bias vector of the output gate; This represents the weight matrix of the output gate.
8. The short-term power load forecasting method considering integrated climate and load as described in claim 6, characterized in that, The evaluation metrics for the BiLSTM power load forecasting model are mean square error, mean absolute percentage error, coefficient of determination, and root mean square error.
9. A short-term power load forecasting system considering integrated climate and load factors, characterized in that, For implementing the short-term power load forecasting method considering integrated climate and load as described in any one of claims 1-8, the system comprises: The data acquisition module is used to collect meteorological data and power load data for the target area. The feature extraction module is used to extract comprehensive climate features affecting power load based on the meteorological data; The evaluation module is used to perform a dual evaluation and screening of the comprehensive climate characteristics based on the power load data to obtain the target climate characteristics; The update module is used to perform multi-component update processing on the power load data; The prediction module is used to input the target load characteristics and target climate characteristics after multi-component update processing into the pre-built BiLSTM power load prediction model to predict the short-term power load of the target area.
10. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, executed by a processor, can implement the short-term power load forecasting method that takes into account integrated climate and load as described in any one of claims 1-8.