Industry load forecasting methods, systems, and equipment that integrate standard mutual information with improved bidirectional LSTM

Abstract

This invention discloses an industry load forecasting method, system, and device that integrates standard mutual information and an improved bidirectional LSTM, relating to the field of power system technology. Currently, monthly industry load forecasting is inaccurate. This invention includes the following steps: employing an improved fuzzy C-means clustering algorithm that considers kernel density estimation to cluster information including load sequences, consumption levels, temperature, and holidays; calculating standard mutual information based on the clustering results to quantitatively analyze the correlation between information factors and monthly industry load; allocating weights, constructing a bidirectional LSTM neural network to capture the temporal variation patterns of industry load, employing a multi-head attention mechanism to analyze the influence of external features, and connecting them through a convolutional neural network to output the monthly industry load forecasting results. This technical solution can effectively consider the impact of external influencing factors on monthly industry load, improving the accuracy of monthly industry load forecasting results.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to an industry load forecasting method, system and equipment that integrates standard mutual information and improved bidirectional LSTM. Background Technology

[0002] Accurate medium- and long-term load forecasting is a core technical support for ensuring the stable operation of the power system and the benefits of grid planning. Its forecasting results directly impact the construction progress of new power systems and the optimal allocation of electrical resources. With the development of smart grid construction, improving forecasting accuracy in complex and ever-changing operating environments has become a key research direction in the industry.

[0003] Currently, mainstream monthly load forecasting technologies in the industry mainly focus on two technical paths: extrapolation forecasting methods based on time series features and intelligent forecasting methods based on machine learning. Time series analysis methods mainly include exponential smoothing, grey forecasting, and similar-day methods, while the latter employs machine learning algorithms such as artificial neural networks and support vector regression. Existing technologies generally suffer from two limitations: firstly, they over-rely on the temporal continuity of historical load data, lacking effective modeling of the nonlinear correlations of external, multi-source, heterogeneous influencing factors such as temperature fluctuations and industrial policy adjustments; secondly, traditional correlation analysis methods struggle to adapt to the dynamic response characteristics of power loads. For example, linear correlation detection based on the Pearson coefficient cannot capture high-order nonlinear coupling relationships present in real-world scenarios, while Copula functions, although capable of constructing nonlinear correlations, rely excessively on prior parameterization assumptions of the correlation function, leading to generalization defects in the feature correlation models. Especially when sudden external disturbances trigger abrupt changes in power demand, existing technologies are prone to systematic biases. Furthermore, the correlation analysis results of the aforementioned studies only select a subset of external influencing factors strongly correlated with the load, failing to comprehensively quantify the impact of each factor on the power load.

[0004] Therefore, current technology has two key areas for improvement: 1. There is an urgent need to build a multimodal data fusion architecture to overcome the limitations of unidirectional time series analysis; 2. There is an urgent need to introduce nonparametric estimation methods that can quantify nonlinear relationships. Effectively combining temporal and spatial characteristics with various data information, while accurately capturing load change patterns, and constructing effective coupled models for dynamic characteristics such as weather changes and economic fluctuations, will become an important research direction for breaking through the technical bottlenecks of monthly load forecasting in the industry. Summary of the Invention

[0005] The technical problem to be solved and the technical task proposed by this invention is to improve and refine existing technical solutions, and to provide an industry load forecasting method, system, and device that integrates standard mutual information and improved bidirectional LSTM, with the aim of improving the accuracy of monthly load forecasting results for the industry. To this end, this invention adopts the following technical solution.

[0006] The first technical solution of this invention: an industry load forecasting method integrating standard mutual information and improved bidirectional LSTM, comprising the following steps:

[0007] 1) An improved fuzzy C-means clustering algorithm that takes kernel density estimation into account is used to cluster information including load sequences, consumption levels, temperature, and vacation time.

[0008] 2) Based on the clustering results, standard mutual information is calculated to quantitatively analyze the correlation between information factors, including consumption level, temperature, and vacation time, and the monthly load of the industry;

[0009] 3) Based on the correlation analysis results, weights are assigned, a bidirectional LSTM neural network is constructed to capture the temporal variation pattern of industry load, the Transformer multi-head attention mechanism is used to analyze the influence of external features, and the monthly load forecast results of the industry are output through convolutional neural network.

[0010] This technical solution addresses key bottlenecks in traditional industry monthly load forecasting, such as over-reliance on linear time series data, difficulty in effectively integrating multi-source heterogeneous nonlinear external factors, and poor adaptability to complex dynamic environments. It achieves this through "clustering preprocessing for noise reduction and efficiency improvement → precise quantification of nonlinear correlations using standard mutual information → dynamic fusion of correlation weights → bidirectional LSTM + multi-head attention spatiotemporal modeling." This significantly improves the robustness, accuracy, and adaptability to complex real-world scenarios (such as outliers, nonlinear relationships, and external abrupt changes) of the forecasting model, providing more reliable data support for energy planning and scheduling. Specifically:

[0011] An improved fuzzy C-means clustering algorithm incorporating kernel density estimation is employed to jointly cluster load sequences and external factors (consumption level, temperature, vacation time), effectively improving the efficiency and accuracy of subsequent standard mutual information calculation and enhancing robustness to outliers. By identifying the main patterns and inherent structures in the data, noise, redundancy, seasonal fluctuations, and interference from mixed patterns in the original data are reduced, providing a more reliable and homogeneous data foundation for mutual information calculation. Directly calculating mutual information from raw high-dimensional data is prone to distortion and computational complexity; clustering preprocessing significantly reduces computational complexity, especially in large-scale industry load data scenarios. Kernel density estimation is used to optimize the selection of initial cluster centers, overcoming the sensitivity of traditional fuzzy C-means to initial values ​​and reducing the risk of getting trapped in local optima, thereby improving clustering quality and laying a more solid foundation for subsequent analysis.

[0012] Standard mutual information effectively captures and quantifies complex nonlinear dependencies between variables. It overcomes the fundamental limitation of traditional linear correlation coefficients (such as the Pearson coefficient) in industry load analysis—linear methods may fail when load exhibits complex nonlinear relationships with factors such as temperature and consumption levels. Based on information theory, standard mutual information does not rely on specific data distribution assumptions, enabling a more scientific and comprehensive assessment of the true impact of various factors on load.

[0013] Bidirectional LSTM can simultaneously utilize forward and backward information from historical sequences to more comprehensively understand and capture the inherent long-term and short-term time dependencies and trends in monthly industry load. Employing a Transformer multi-head attention mechanism to analyze the impact of external features (consumption levels, temperature, holidays) allows for parallel and dynamic capture of complex cross-dimensional relationships between different external features and their correlation with the prediction target, particularly adept at identifying the heterogeneous impact of key events (such as sudden temperature changes and holidays). Finally, a convolutional neural network connects the bidirectional LSTM (temporal features) and the Transformer (external feature influences) to form a comprehensive prediction result. This architecture, which deeply integrates temporal patterns and external dynamic features, significantly improves the model's adaptability and prediction accuracy in complex and ever-changing environments.

[0014] An improved bidirectional LSTM neural network is used for load forecasting. Compared to traditional unidirectional LSTM or other forecasting models, bidirectional LSTM can simultaneously consider both forward and backward information from time series data, providing a more comprehensive understanding of the data's temporal dependencies and improving the accuracy of monthly load forecasting. Furthermore, considering the influence of external factors, a multi-head attention mechanism is employed to incorporate clustered and enhanced features into the forecasting model. This approach allows the forecast results to better reflect the actual situation, especially under complex and volatile external environments, resulting in higher forecast accuracy.

[0015] This method combines clustering algorithms, mutual information analysis, and deep learning models, making it applicable to various industries and load types, and exhibiting strong versatility and adaptability. By combining kernel density estimation, standard mutual information, and an improved bidirectional LSTM, it can effectively improve the accuracy of monthly load forecasting for different industries, reduce forecasting errors, and provide more reliable data support for energy planning and scheduling.

[0016] As a preferred technical means, the clustering process in step 1) further includes:

[0017] 1.1) Determine the membership matrix: Based on the improved fuzzy C-means clustering algorithm, set the number of clusters, select the initial cluster centers based on the kernel density estimation method, and calculate the membership degree between the monthly load data and the cluster centers;

[0018] 1.2) Constructing the objective function: Define an objective function to minimize the Euclidean distance between the monthly load data and the cluster centers;

[0019] 1.3) Update cluster centers: Iteratively optimize the objective function and update the cluster center matrix until the objective function converges;

[0020] 1.4) Obtain the optimal clustering result: Obtain the clustering result based on the minimized objective function value;

[0021] 1.5) Determine the optimal number of clusters: Select the optimal number of clusters using the Xie & Ben index and determine the final clustering result.

[0022] The fuzzy C-means clustering algorithm allows data points to belong to multiple clusters simultaneously with different membership degrees, improving the flexibility of data processing and effectively identifying complex variation patterns and inherent structures in monthly load data. The algorithm employs the objective function of minimizing the Euclidean distance between data points and cluster centers, iteratively updating the cluster centers to ensure convergence to the global optimum. The Xie & Ben indices are used to dynamically determine the optimal number of clusters, avoiding reliance on preset parameters and improving the adaptability and intelligence of the clustering. The cluster centers obtained through multiple iterations accurately reflect the distribution characteristics of the load data, reducing analytical bias. The clustered monthly load data provides a high-quality foundation for correlation analysis and predictive modeling, enabling more accurate identification of potential patterns in the data. The iterative optimization process reduces the computational burden of full-space search, effectively reducing algorithm complexity and improving operational efficiency.

[0023] As a preferred technical approach: Step 1.1) Determine the number of clusters based on the number of months in the membership matrix. The kernel density estimation method is used to extract data from the electricity consumption matrix. Y Before being selected N l The data with the highest row density is used as the initial cluster center matrix. V =[ V 1,…, V l ,…, V Nl ] T Then calculate data for any month. Y m Belonging to the l The membership degree of each cluster center.

[0024] Step 1.2) Constructing the objective function, the objective function... G The sum of the squared weighted distances between the membership degrees of all historical months and all cluster centers;

[0025] Step 1.3) In updating the cluster centers, if the objective function... GIf the minimum value is not reached, then the cluster center matrix is ​​updated according to the given update strategy. V Update and recalculate the membership matrix until the objective function is achieved. G To reach the minimum;

[0026] Step 1.4) Obtaining the optimal clustering result involves analyzing the membership matrix and cluster center matrix. V The number of clusters obtained is N l The optimal clustering result corresponding to the minimum value of the objective function;

[0027] In step 1.5) determining the optimal number of clusters, the Xie & Ben index is used to reflect the intra-cluster compactness and inter-cluster separation of the clustering results. This is achieved by comparing the clustering quality indices corresponding to different numbers of clusters. I XB Select the optimal number of clusters And obtain the best clustering results.

[0028] By clustering historical monthly load data and external influencing factor data, the data is categorized before correlation analysis. This allows different categories of data to more clearly reveal their inherent patterns, improving the accuracy of subsequent correlation analysis and ensuring more reliable results. The membership matrix is ​​flexibly set and considers data uncertainty, allowing data points to belong to multiple cluster centers simultaneously. The fuzzy processing method better captures complex features in the data, reducing errors and increasing the algorithm's robustness. The introduction of the Xie & Ben index allows for dynamic evaluation of clustering effects under different numbers of clusters based on intra-cluster compactness and inter-cluster separation, thus selecting the optimal number of clusters, reducing human intervention and ensuring the scientific and rational selection of cluster numbers. The update of cluster centers is iteratively optimized based on the objective function of minimizing Euclidean distance. In each iteration, the cluster centers are quickly calculated and adjusted to ensure the objective function reaches its minimum value in a short time, thereby reducing computational complexity and improving the algorithm's efficiency. By optimizing the membership matrix and cluster center matrix, this method ensures that each data point is optimally assigned to its corresponding cluster. This not only improves the quality of the clustering results but also provides a more accurate foundation for subsequent data analysis based on the clustering results. This method can effectively handle load data and multidimensional external influencing factor data, fully utilizing the characteristics of the data for clustering and analysis, ensuring that the clustering results accurately reflect the inherent structure and similarity of the data.

[0029] As a preferred technical means: Step 2) further includes the following steps:

[0030] 2.1) Data fuzzing processing

[0031] Based on the clustering results obtained in step 1) by improving the fuzzy C-means (DFCM) clustering algorithm, the original external influencing factor values ​​x(m) and monthly load values ​​y(m) are replaced with the cluster center values ​​corresponding to each sampling point to obtain the fuzzy sequences U and V;

[0032] 2.2) Standard Mutual Information Calculation

[0033] 2.2.1) Mutual information calculation: The mutual information I(U;V) between the fuzzified data U and V is calculated to quantitatively describe the correlation between influencing factors and monthly load;

[0034] 2.2.2) Information entropy calculation: Calculate the information entropy H(U) and H(V) of sequences U and V;

[0035] 2.2.3) Standard mutual information calculation: The standard mutual information value J(U;V) between the influencing factors and the monthly load is calculated by the ratio of mutual information I(U;V) to information entropy H(U) and H(V);

[0036] 2.3) Quantitative assessment of correlation

[0037] 2.3.1) Results Analysis: Based on the standard mutual information value J(U;V), the correlation between external influencing factors and the monthly load of the industry is quantitatively assessed; the larger the value, the stronger the correlation.

[0038] 2.3.2) Parameter input: The correlation analysis results obtained based on standard mutual information are used as network hyperparameters and input into the monthly load forecasting network to optimize the model's forecasting ability.

[0039] Fuzzy clustering algorithms are used to preprocess the original data, mapping high-dimensional data to the feature space of cluster centers, achieving data noise reduction and feature enhancement, and building a stable data foundation for correlation analysis. A quantitative model of the dependency relationship between external factors and industry load is constructed based on standard mutual information theory, accurately characterizing this relationship. Compared to traditional correlation analysis, standard mutual information can deeply explore nonlinear correlation patterns and complex interactions between variables. The correlation analysis results are directly used as model hyperparameters, dynamically allocating weights according to the importance of each influencing factor, enhancing the model's responsiveness to core factors, and improving prediction accuracy and model stability. Sequence information entropy is calculated to measure data uncertainty, and a standard mutual information index is constructed in conjunction with mutual information to comprehensively evaluate the relative importance of influencing factors. This method considers both the direct correlation between data and information complexity, ensuring the scientific validity and accuracy of the analysis results. The standard mutual information value intuitively reflects the strength of the correlation between factors and load, helping to select key predictive variables and rationally allocate model weights. This data-driven parameter optimization strategy effectively improves model performance and prediction results. By accurately identifying and quantitatively analyzing the mechanisms by which external factors influence the model, it possesses stronger environmental adaptability and scenario generalization capabilities, making it widely applicable to various load forecasting tasks. Cluster-optimized data makes standard mutual information calculation more efficient and accurate, avoiding the complexity of processing high-dimensional raw data and improving overall analysis efficiency.

[0040] As a preferred technical means: In step 2.1), the data fuzzification process, for the monthly load of the industry to be analyzed and the external influencing factors, the data within a certain time period is known. N m Influencing factor values ​​of each sampling point x ( m ) and monthly load value y ( m ), m ∈{1,2,…, N m},all x ( m ) constitute a sequence X ,all y ( m ) constitute a sequence Y The original data was replaced with data processed by DFCM clustering, and standard mutual information was calculated. The sampled values ​​were recorded. x ( m The cluster center value of the cluster is u ( m ),all u ( m ) constitute a sequence U Sample value y ( m The cluster center value of the cluster isv ( m ),all v ( m ) constitute a sequence V ; u ( m Total F Possible values, F ≤ N m , recorded as , f ∈{1,2,…, F}, v ( m Total G Possible values, G ≤ N m , recorded as , ∈{1,2,…, G};

[0041] In step 2.2), the standard mutual information calculation involves the standard mutual information values ​​between the influencing factors and the monthly load. J ( U ; V The calculation method is as follows:

[0042]

[0043] In the formula,

[0044]

[0045]

[0046]

[0047] In the formula, I(U;V) for u(m) and v(m) Mutual information value; H(U) and H(V) They are respectively U and V Information entropy; To simultaneously satisfy u ( m )= and v ( m )= The proportion of sampling points to all sampling points; To meet u ( m )= The proportion of sampling points to all sampling points; To meet v(m) = The proportion of sampling times to all sampling times; J(U;V) The value range is [0,1]. The larger the value, the stronger the correlation between the influencing factor and the monthly load of the industry.

[0048] Standard mutual information can quantitatively assess the nonlinear relationship between monthly industry load and external influencing factors, avoiding complex correlations that may be overlooked by traditional linear correlation analysis methods, thus reflecting the actual connection between the two more comprehensively. By fuzzifying the original data and replacing the original values ​​with fuzzy C-means clustering results, errors caused by subtle differences in values ​​are reduced, allowing the analysis to focus more on the overall trend and main characteristics of the data, rather than being disturbed by minor changes, thereby improving the accuracy and robustness of correlation analysis. The correlation analysis results obtained based on standard mutual information can be used as network hyperparameter inputs for prediction models, helping to optimize the model structure and make it better adaptable to different external influencing factors and monthly load change scenarios, improving the accuracy and generalization ability of prediction. By calculating the mutual information value and information entropy, and further standardizing the obtained standard mutual information value, the correlation strength between influencing factors and monthly load can be quantitatively assessed. Since the standard mutual information value ranges from [0,1], this indicator is intuitive and has a clear physical meaning, facilitating the model to make more reasonable decisions in practical applications. By mapping complex raw data to a smaller number of cluster centers, computational complexity is reduced, lessening the burden on the model when processing large-scale, high-dimensional data and thus improving model efficiency. Employing fuzzy C-means clustering to process the data makes the analysis results more robust to noise and outliers, enhancing the stability and reliability of the load forecasting model in practical applications. Using standard mutual information analysis results as hyperparameters of the network allows the forecasting model to learn and adapt more specifically to key external influencing factors, thereby optimizing network performance and improving forecast accuracy.

[0049] As a preferred technical means, the monthly load forecasting in step 3) includes the following steps:

[0050] 3.1) Correlation weight calculation: Normalize the standard mutual information value between the industry's monthly load and each external influencing factor to determine the characteristic weight of the external influencing factor;

[0051] 3.2) Weighted Eigenvectors: Based on the calculated eigenweights, the eigenvectors of all external influencing factors are weighted and summed to obtain the enhanced comprehensive eigenvector;

[0052] 3.3) Bidirectional LSTM network modeling: A bidirectional long short-term memory network is used to capture the time series feature vector of historical monthly load as input to capture the time series change trend of industry monthly load;

[0053] 3.4) Transformer multi-head attention mechanism modeling: The Transformer multi-head attention mechanism is applied, taking multi-dimensional external feature vectors and comprehensive feature vectors as inputs, and capturing dynamic correlations across feature dimensions through parallel attention heads;

[0054] 3.5) Network Optimization and Output: The temporal feature vectors extracted by the bidirectional LSTM and the multi-head attention features encoded by the Transformer are used to construct a feature interaction layer through a convolutional neural network. Finally, the monthly load forecast value of the industry with environmental adaptability is output through a fully connected layer.

[0055] By calculating the correlation weights in step 3.1 and normalizing the standard mutual information values, the feature weights of external influencing factors are obtained. This allows the impact of each factor on monthly load forecasting to be reflected more accurately. The weighted feature vectors are further processed in step 3.2 to enhance the feature representation capability, thereby enabling the model to understand the data more deeply. Using the comprehensive feature vector of external influencing factors and the time series feature vector of historical monthly loads as input to the improved bidirectional LSTM network helps the model better capture the complex dynamics of monthly load changes. By weighted summing of the comprehensive feature vectors, the features input to the network become more representative and important, improving the model's prediction accuracy. The bidirectional LSTM network combines the advantages of forward and backward layers, enabling it to utilize both past and future information from time series data, thus capturing the regularity of monthly load changes more comprehensively. The bidirectional processing method can better utilize historical data and external influencing factors to make more accurate predictions. Through the modeling and optimization of the bidirectional LSTM network and Transformer multi-head attention mechanism in steps 3.3 and 3.4, the final output monthly load forecast results can fully reflect the complex time dependencies and the effects of external influencing factors. Bidirectional networks can more effectively process long-sequence data, reducing information loss caused by large time spans. Meanwhile, multi-head attention mechanisms, through parallel multi-group self-attention heads, dynamically capture global feature dependencies from different dimensions, enhancing the model's adaptability to multi-scale time patterns and compensating for the limitations of bidirectional LSTM models in terms of external feature input. By comprehensively considering the feature weights of various external influencing factors and bidirectional processing information, the model can more effectively utilize meaningful data features, reducing the risk of overfitting due to redundant information or noise, making the model more robust on new data and exhibiting good generalization ability. Combining the forward and backward layers of the bidirectional LSTM network allows for flexible adjustment of the network structure and optimization strategies according to actual conditions to achieve optimal prediction results, ensuring the model remains efficient and accurate when dealing with complex prediction tasks.

[0056] As a preferred technique, the bidirectional LSTM neural network model captures the long-term and short-term dependencies of the load sequence based on a bidirectional time-series feature extraction module. It controls information transmission through forget gates, input gates, and output gates. The formulas for calculating the forward and backward monthly load forecast outputs are as follows:

[0057]

[0058]

[0059]

[0060] In the formula, Represents the forward calculated value; Represents the backward calculated value; o t This represents the final output value calculated from the forward and backward values; oh 1 represents the weight matrix mapping from the input layer to the feedforward layer; oh 3 represents the weight matrix mapping from the input layer to the feed layer; oh 2 is the weight matrix that maps the output of the previous time step of the forward layer to the current time step; oh 5 is the weight matrix that maps the output of the backward layer at the previous time step to the current calculation time step; oh 4 is the weight matrix that maps the output of the feedforward layer to the output layer; oh 6 represents the weight matrix that maps the output of the backward layer to the weight matrix of the output layer;

[0061] The Transformer multi-head attention mechanism maps input features to different subspaces by computing multiple sets of attention weights in parallel. It captures the dynamic relationships between data in each subspace to effectively identify the heterogeneous impact of key external events, including abrupt temperature changes and holiday effects, on the load. Single-head attention is calculated using the following formula:

[0062]

[0063] In the formula, Q , K , V To query the key-value matrix; This is the scaling factor;

[0064] By dividing the model into multiple attention heads, forming multiple subspaces, the model can focus on different aspects of information. The specific expression is as follows:

[0065]

[0066]

[0067]

[0068] In the formula, Q i , K i , V i For the first i A query key-value matrix for each attention head; , , For the first i The weight matrix of each attention head; To output the projection matrix; h For the number of attention heads.

[0069] While bidirectional LSTM can process sequential time series information, it can only process data step-by-step, making it difficult to consider the relationships between global time points and unable to utilize future weather forecast information. The Transformer multi-head attention mechanism, by computing multiple sets of attention weights in parallel, maps input features to different subspaces, capturing the dynamic correlations between data in each subspace. This effectively identifies the heterogeneous impact of key external events such as sudden temperature changes and holiday effects on loads. This multi-head attention mechanism effectively compensates for the shortcomings of bidirectional LSTM.

[0070] The second technical solution of the present invention: provides an industry load forecasting system that integrates standard mutual information and improved bidirectional LSTM, the system being used to execute the aforementioned industry monthly load forecasting method, the system comprising:

[0071] Data receiving module: used to receive historical monthly industry load data and external influencing factor data;

[0072] Clustering module: The received data is clustered using a fuzzy C-means clustering algorithm based on kernel density estimation.

[0073] Correlation Analysis Module: Based on clustering results, standard mutual information is calculated to quantitatively analyze the correlation between external influencing factors and industry monthly load;

[0074] Monthly load forecasting module: It uses a bidirectional LSTM neural network and combines the results of correlation analysis to forecast the monthly load of the industry.

[0075] The data receiving module effectively integrates historical monthly industry load data and external influencing factor data, providing comprehensive data support for subsequent analysis and forecasting. Fuzzy C-means clustering is used to cluster the data, better identifying patterns and structures and improving data comprehension. Standard mutual information is used to quantify the relationship between external influencing factors and monthly industry load, revealing key factors affecting monthly load changes and improving forecast accuracy. Bidirectional LSTM (Bidirectional Long Short-Term Memory) captures time dependencies in the data and considers bidirectional information from historical data, thus improving the accuracy and reliability of monthly load forecasting. The system integrates clustering analysis, correlation analysis, and deep learning methods, fully utilizing data information to provide more comprehensive and accurate monthly load forecasts. The system can adjust according to changes in monthly industry load and external influencing factors, making the forecast results more aligned with reality and improving system adaptability.

[0076] As a preferred technical means: the clustering processing module includes a membership degree determination unit, an objective function construction unit, a cluster center update unit, an optimal clustering result acquisition unit, and an optimal number of clusters determination unit; the membership degree determination unit sets the number of clusters according to the improved fuzzy C-means clustering algorithm, selects the initial cluster centers based on the kernel density estimation method, and calculates the membership degree between the monthly load data and the cluster centers; the optimal number of clusters determination unit selects the optimal number of clusters through the Xie & Ben index and determines the final clustering result;

[0077] The correlation analysis module includes a data fuzzification processing unit, a standard mutual information calculation unit, and a correlation quantitative evaluation unit. The data fuzzification processing unit replaces the original external influencing factor values ​​and monthly load values ​​with cluster center values ​​to obtain a fuzzified sequence. The standard mutual information calculation unit calculates the standard mutual information values ​​between the fuzzified data to quantitatively describe the correlation between external influencing factors and monthly load.

[0078] The monthly load forecasting module includes a correlation weight calculation unit, a feature vector weighting unit, an improved bidirectional LSTM modeling unit, and a network optimization and output unit. The improved bidirectional LSTM modeling unit constructs a bidirectional LSTM neural network, uses the Transformer multi-head attention mechanism to analyze the influence of external features, and connects the Transformer and bidirectional LSTM models through a convolutional neural network.

[0079] The membership determination unit uses an improved fuzzy C-means clustering algorithm to effectively determine the number of clusters and calculate the membership degree between monthly load data and cluster centers, improving the understanding of data patterns and clustering effectiveness. The cluster center update unit iteratively optimizes the objective function to ensure that cluster centers fully reflect the data distribution, resulting in more accurate clustering results. The data fuzzification unit fuzzifies external influencing factors and monthly load data based on the clustering results, enhancing the ability to analyze data correlations. The standard mutual information calculation unit quantifies the correlation between external influencing factors and monthly load, providing more refined data relationship analysis. The correlation weight calculation unit determines the weights of external influencing factors by normalizing the standard mutual information value, helping to accurately identify key factors affecting monthly load changes. The improved bidirectional LSTM modeling unit utilizes a bidirectional long short-term memory network and a Transformer multi-head attention mechanism, combining comprehensive feature vectors and historical monthly load time-series feature vectors to effectively capture complex patterns and trends in time-series data, improving the accuracy and robustness of monthly load forecasting. The network optimization and output unit, combining a bidirectional LSTM network with a Transformer multi-head attention mechanism, generates robust and high-quality monthly load forecasts, meeting the needs of medium- and long-term planning and decision-making in the industry. The overall technical solution comprehensively utilizes clustering analysis, correlation analysis, and deep learning methods, possessing the ability to handle complex data and address real-world forecasting challenges, providing the industry with reliable forecasting tools and decision support.

[0080] An improved bidirectional LSTM modeling unit achieves bidirectional processing of sequence data through a combination of forward and backward LSTM networks. During each monthly load forecast, information transmission is controlled via forget gates, input gates, and output gates. The Transformer multi-head attention mechanism establishes direct correlations between any two time steps in the sequence through global recognition of enhanced integrated feature vectors and external feature vectors, effectively capturing long-distance dependent features and the impact of sudden events. A convolutional neural network layer spatially concatenates the output features of the bidirectional LSTM and Transformer, extracting local pattern features through multi-scale convolutional kernels to achieve deep fusion of heterogeneous features. A fully connected layer maps the fused high-dimensional feature vectors to the target output space, completing the final load forecast calculation. This hybrid architecture fully leverages the technical advantages of each component: the bidirectional LSTM ensures the continuity of time-series modeling, the Transformer enhances global correlation perception, the convolutional neural network achieves feature fusion, and the fully connected layer completes nonlinear mapping, forming an end-to-end high-precision monthly load forecasting solution.

[0081] The third technical solution of the present invention is to provide a computer device, the device including one or more processors and one or more memories, wherein at least one piece of program code is stored in the one or more memories, and when the program code is executed by the one or more processors, it implements the aforementioned industry load forecasting method that integrates standard mutual information and improved bidirectional LSTM.

[0082] The fourth technical solution of the present invention is to provide a storage medium storing at least one piece of program code, which, when executed by a processor, implements the steps of the aforementioned industry load forecasting method that integrates standard mutual information and improved bidirectional LSTM.

[0083] Beneficial Effects: This invention effectively integrates monthly load sequences of the industry with multiple external influencing factors such as consumption levels, temperature, and holidays to achieve accurate prediction of monthly industry load. Specifically, the first step uses an improved fuzzy C-means clustering algorithm that considers kernel density estimation to cluster load and external factor information, which can specifically capture hidden similar patterns and internal structures in the data, laying a data foundation for subsequent correlation analysis. The second step calculates standard mutual information based on the clustering results, which can quantitatively characterize the nonlinear correlation between external influencing factors and monthly industry load, breaking through the limitations of traditional linear correlation analysis methods and better reflecting the complex interaction mechanism between multiple factors and load in real-world scenarios. The third step assigns weights based on the correlation analysis results, captures the temporal variation pattern of load through a bidirectional LSTM neural network, and introduces a Transformer multi-head attention mechanism to analyze the influence of external features and achieves feature fusion through a convolutional neural network, effectively integrating temporal features and external factor features. This overcomes the problems of traditional methods that over-rely on the temporal continuity of historical load data and insufficient modeling of external dynamic influences, ultimately improving the accuracy of monthly industry load prediction and its adaptability to complex operating environments. Attached Figure Description

[0084] Figure 1 This is the basic model architecture of the present invention.

[0085] Figure 2 This is the LSTM network unit of the present invention.

[0086] Figure 3 This is the bidirectional LSTM network structure of the present invention.

[0087] Figure 4 These are the monthly load forecast results for the steel industry under different methods. Detailed Implementation

[0088] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0089] Example 1:

[0090] This invention proposes an industry monthly load forecasting method that integrates kernel density estimation, standard mutual information, and improved bidirectional LSTM. Its implementation process includes the following detailed steps:

[0091] Step 1: The improved fuzzy C-means clustering algorithm, which takes into account kernel density estimation, is used to cluster information including load sequence, consumption level, temperature, and vacation time.

[0092] Step 2: Based on the clustering results, standard mutual information is calculated to quantitatively analyze the correlation between information factors, including consumption level, temperature, and vacation time, and the monthly load of the industry.

[0093] Step 3: Based on the correlation analysis results, assign weights, construct a bidirectional LSTM neural network to capture the temporal variation pattern of industry load, use the Transformer multi-head attention mechanism to analyze the influence of external features, and connect them through a convolutional neural network to output the monthly load forecast results for the industry.

[0094] The basic model architecture of this method is as follows: Figure 1 As shown.

[0095] Step 1 uses an improved fuzzy C-means clustering algorithm to cluster the monthly load series with factors such as consumption level, temperature, and vacation information, as detailed below:

[0096] Before analyzing the correlation between external influencing factors and monthly industry load, the data is clustered to improve the accuracy of the correlation analysis. An improved fuzzy C-means clustering algorithm (Density-based Fuzzy C-Means, DFCM) and the Xie & Ben index are used to cluster historical monthly load data and external influencing factor data, respectively. The clustering process based on the DFCM algorithm is as follows:

[0097] Step 1.1: Determine the membership matrix: Set the number of clusters. , N m For the number of months. Based on the kernel density estimation method from the electricity consumption data matrix Y Before being selected N l The data with the highest row density is used as the initial cluster center matrix. V =[ V 1,…, V l ,…, V Nl ] T The density calculation method is as follows:

[0098]

[0099] In the formula, Representing History Month m The density near the monthly load data; Y m and Y j Each is a historical month m and j The corresponding monthly load data vector includes load values ​​and external influencing factors (such as temperature, consumption levels, etc.); Y m - Y j ||Representative Y m and Y j The Euclidean distance between them; d c The distance threshold parameter is specifically the bandwidth parameter for kernel density estimation, which is adaptively selected based on the data distribution or determined through cross-validation.

[0100] Y m Belonging to the l The membership degree of each cluster center is represented as:

[0101]

[0102] In the formula, Q m,l Representing History Month m The monthly load data belongs to the cluster center. l Membership degree, 0≤ Q m,l ≤1; q Membership factor Specifically, it is a fuzzification parameter that controls the degree of fuzziness in the membership degree, and is usually set to... q>1 To enhance the fuzziness of clustering results; V l Cluster center l Corresponding monthly load data; || Y m - V l ||Representative Y m and V l The Euclidean distance between them.

[0103] Step 1.2: Construct the objective function G :

[0104]

[0105] In the formula, The membership factor is q Historical month calculated under the following circumstances m Belonging to the cluster center l The degree of membership.

[0106] Step 1.3: Update the cluster center matrix V If the objective function obtained in step 2 G If the minimum value is not reached, then the cluster center matrix is ​​adjusted according to the following formula. V Update and recalculate the membership matrix until the objective function is achieved. J To reach the minimum.

[0107]

[0108] Step 1.4: Obtain the number of clusters N l The optimal clustering results are as follows: from the membership matrix and cluster center matrix. V The number of clusters obtained is N l The optimal clustering result corresponding to the minimum value of the objective function.

[0109] Step 1.5: Determine the optimal number of clusters N l *: The Xie & Ben index is selected to evaluate the quality of clustering results at each cluster size. This index can be expressed as:

[0110]

[0111] In the formula, the numerator reflects the intra-cluster compactness of the clustering results, while the denominator reflects the inter-cluster separation of the clustering results. I XB The smaller the value, the better the clustering effect. Therefore, by comparing the clustering quality index corresponding to different numbers of clusters... I XB This allows us to select the optimal number of clusters. N l * This yields the clustering results with the optimal number of clusters.

[0112] Furthermore, step 2 calculates standard mutual information based on the clustering results, and quantitatively analyzes the correlation between factors such as consumption level, temperature, and vacation information and the monthly load of the industry, as detailed below:

[0113] Standard mutual information (SMIS) is used to analyze the correlation between monthly industry load and external influencing factors to fully account for their nonlinear relationship. However, directly using SMIS from the raw data can lead to subtle numerical differences in the correlation analysis, making it difficult to grasp the main trends in variable changes. Therefore, based on SMIS, the clustering results obtained using the DFCM algorithm are employed to fuzzify these subtle numerical differences, thereby improving the accuracy of the results.

[0114] For the monthly load of the industry to be analyzed and the external influencing factors, it is known that within a certain time period... N m Influencing factor values ​​of each sampling point x ( m ) and monthly load value y ( m ), m ∈{1,2,…, N m},all x ( m ) constitute a sequence X ,all y ( m ) constitute a sequence Y The original data was replaced with data processed by FCM clustering, and standard mutual information was calculated. The sampled values ​​were recorded. x ( m The cluster center value of the cluster is u ( m ),all u ( m ) constitute a sequence U Sample value y ( m The cluster center value of the cluster is v ( m ),all v ( m ) constitute a sequence V . u ( m Total F Possible values ​​( F ≤ N m ), denoted as u 1 (f) , f ∈{1,2,…, F}, v ( m Total G Possible values ​​( G ≤ N m ), denoted as v 1( g ),g ∈{1,2,…, G The standard mutual information value between the influencing factors and the monthly load is then calculated. J ( U ; V The calculation method is as follows:

[0115]

[0116] In the formula,

[0117]

[0118]

[0119]

[0120] In the formula, I(U;V) for u ( m )and v ( m The mutual information value of ) H(U) and H(V) They are respectively U and V Information entropy; P(u 1 (f ) , v 1( g To simultaneously satisfy u ( m )= u 1 (f ) and v ( m )= v 1( g The proportion of sampling points to all sampling points; P(u 1 (f )) To meet u(m)=u 1 (f ) The proportion of sampling points to all sampling points; P ( v 1( g To satisfy v ( m )= v 1( g The proportion of sampling times to all sampling times; J(U;V) The value range is [0,1]. The larger the value, the stronger the correlation between the influencing factor and the monthly load.

[0121] It should be noted that the correlation analysis results obtained based on standard mutual information will be used as network hyperparameters input to the monthly load forecasting network.

[0122] Furthermore, step 3 inputs the external feature data and the comprehensive feature vector into the improved bidirectional LSTM neural network for industry monthly load forecasting, as detailed below:

[0123] Different external factors have varying degrees of impact on monthly load across industries. Therefore, when forecasting monthly load for each industry, the strength of the correlation between the industry's monthly load and each external influencing factor should be fully considered. To this end, the correlation results obtained from improved correlation analysis based on standard mutual information are used as hyperparameters of the monthly load forecasting network and integrated into its structure. The specific approach is as follows:

[0124] The industry's monthly load and the first r External influencing factors ( r ∈{1,2,.., R}, R The standard mutual information value (the number of external influencing factors) is denoted as J r The standard mutual information values ​​are normalized to obtain the industry monthly load and the first... r Normalized standard mutual information values ​​of external influencing factors J 0,r ,Right now:

[0125]

[0126] Next, with J 0,r For the first r The feature vector of an external influencing factor x r The weights of all external influencing factors are used to sum the eigenvectors of the factors, resulting in the enhanced comprehensive eigenvector. x o :

[0127]

[0128] Finally, the basic external feature vector, the enhanced integrated feature vector, and the historical monthly load feature vector are all used as inputs to the improved bidirectional LSTM monthly load prediction network.

[0129] The improved bidirectional LSTM monthly load forecasting network consists of a bidirectional LSTM module, a Transformer multi-head attention module, and a convolutional neural network module. The modules are described below:

[0130] Bidirectional LSTM is a combined forward and backward structure built on top of LSTM, consisting of a forward LSTM and a backward LSTM. It takes into account both past and future information, has the memory capacity of a long short-term memory network, and combines the characteristics of bidirectional recurrent neural networks and long short-term memory networks.

[0131] In a bidirectional LSTM network, a single LSTM unit transmits information through three basic "gate" structures: the forget gate, the input gate, and the output gate, such as... Figure 2 As shown.

[0132] 1) Gate of Oblivion

[0133] In LSTM, the forget gate determines the degree of forgetting of features in the state memory unit. The forget gate is derived from the monthly load forecast value from the previous time step. h t-1 and features of the current input x t Simultaneously, the sigmoid function is passed in, and the sigmoid function outputs... Forgetting weights are used as state memory units. f t It can be represented as:

[0134]

[0135] In the formula, W IF Output at the moment before the Gate of Forgetting h t-1 The weight matrix; W HF For the current input features x t The weight matrix; b IF and b HF Forget gate bias term; h t-1 This indicates the monthly load forecast output for the previous moment; x t This represents the current input monthly load forecast feature factors, including external features such as temperature.

[0136] 2) Input Gate

[0137] The input gate's function is to generate intermediate values ​​for the state memory unit. First, the input gate takes the predicted output value from the previous time step... h t-1 And new input x t Both are passed to the sigmoid function, and the output is... i t as input node state g t Update the weights. Input node state. g t Depend on h t-1 andx t A joint decision.

[0138] The sigmoid function output of the input gate i t The expression is as follows:

[0139]

[0140] In the formula, W II and W HI Passing the sigmoid function to the input gate h t-1 and input features x t The weight matrix of the product; b HI and b II This is the bias term for the input gate sigmoid function.

[0141] The output of the tanh function of the input gate g t The expression is as follows:

[0142]

[0143] In the formula, W IG and W HG Passing the tanh function to the input gate h t-1 and input features x t The weight matrix of the product; b IG and b HG This is the bias term for the input gate tanh function.

[0144] 3) Update the state memory unit

[0145] The third step in the LSTM workflow is to update the state memory cells, that is, to update the graph. C t First, the state of the previous time-to-time memory unit is compared with the output of the forget gate. f t Multiply the results to output the selective memory result. Then, add this value to the intermediate values ​​of the state memory units point by point to update the state memory units. C t , C t The expression is as follows:

[0146]

[0147] 4) Output gate

[0148] The output gate is used to determine the monthly load forecast value at the current moment. h t The output gate first outputs the monthly load forecast value from the previous moment. h t-1 and input features x t The weights are determined by passing the sigmoid function. o t Secondly, the output gate will update the state memory unit. C t Passing the weights to the tanh function will output the weights. o t Multiply by each other to obtain the monthly load forecast for the current time. h t Among them, the output weights o t The expression is as follows:

[0149]

[0150] In the formula, W IO and W HO for h t-1 and input features x t The weight matrix of the product; b IO and b HO This is the output gate bias term.

[0151] Current monthly load forecast h t The expression is as follows:

[0152]

[0153] In the formula, C t This is the updated state memory unit.

[0154] Bidirectional LSTM, based on LSTM, uses two independent hidden layers to process the sequence data in both forward and reverse directions. These two connection layers are linked to the same output layer, using both preceding and subsequent information as the current time basis for the time series data. Figure 3As shown. A bidirectional LSTM network is used for prediction, which has excellent expressive power for continuous time series, and the reuse of weight parameters makes it less demanding on data. The relevant calculation formulas are as follows:

[0155]

[0156]

[0157]

[0158] In the formula, Represents the forward calculated value; Represents the backward calculated value; o t This represents the final output value calculated from the forward and backward values; oh 1 represents the weight matrix mapping from the input layer to the feedforward layer; oh 3 represents the weight matrix mapping from the input layer to the feed layer; oh 2 is the weight matrix that maps the output of the previous time step of the forward layer to the current time step; oh 5 is the weight matrix that maps the output of the backward layer at the previous time step to the current calculation time step; oh 4 is the weight matrix that maps the output of the feedforward layer to the output layer; oh 6 represents the weight matrix that maps the output of the backward layer to the weight matrix of the output layer.

[0159] Since bidirectional LSTM models cannot utilize future weather forecast data and only have the capability for sequence trend analysis, a Transformer multi-head attention module is employed to capture the influence of multidimensional external features. By computing multiple sets of attention weights in parallel, the input features are mapped to different subspaces, capturing the dynamic correlations between data in each subspace. This effectively identifies the heterogeneous impact of key external events such as abrupt temperature changes and holiday effects on the load. This multi-head attention mechanism effectively compensates for the shortcomings of bidirectional LSTM. The single-head attention is calculated using the following formula:

[0160]

[0161] In the formula, Q , K , V To query the key-value matrix; This is the scaling factor.

[0162] Multi-head attention mechanisms divide the input feature space into multiple independent sub-representation spaces. Each attention head performs attention computation in parallel within its corresponding sub-space, thereby achieving parallel learning and representation of multi-dimensional features of the input sequence. This mechanism enables the model to simultaneously capture multiple potential association patterns and global information in the data. Its specific expression is as follows:

[0163]

[0164]

[0165]

[0166] In the formula, Q i , K i , V i For the first i A query key-value matrix for each attention head; , , For the first i The weight matrix of each attention head; To output the projection matrix; h For the number of attention heads.

[0167] Convolutional neural networks (CNNs) can serve as a bridge between bidirectional LSTM and Transformer models. They mainly consist of convolutional layers, pooling layers, and fully connected layers. Convolutional layers perform convolution operations on the input data using convolutional kernels (filters) to extract local features; pooling layers downsample the feature maps, reducing the number of parameters while retaining key information; and fully connected layers map the extracted features to the final output space.

[0168] Convolution is the core of convolutional neural networks, performing bit-by-bit multiplication and summation by sliding the convolution kernel across the input data. The weight-sharing mechanism ensures that the same convolution kernel shares the same weight parameters across the entire input region, significantly reducing the number of model parameters. The design of local receptive fields ensures that each neuron is connected only to a local region of the input, effectively capturing the spatial locality of the data. The specific formula for calculating the output features is as follows:

[0169]

[0170] In the formula, This is the output of the convolutional layer; For activation functions; These are the kernel weights; Input data; For bias.

[0171] To verify the effectiveness and accuracy of the proposed industry monthly load forecasting method that integrates kernel density estimation, standard mutual information, and bidirectional LSTM, an example is taken from the steel, shipbuilding, and textile industries in a certain city.

[0172] The monthly load data for the industry uses data from January 2016 to December 2020 for the city. Corresponding external influencing factors include resident consumption data, meteorological data, and holiday data. First, an improved clustering algorithm and standard mutual information are used to analyze the correlation between each external influencing factor and the monthly load of each industry. Then, the above data are divided into training and testing sets in a 5:1 ratio and input into the constructed monthly load prediction network to verify the effectiveness of the proposed algorithm. The training set is used to train the model, and the testing set is used to evaluate the model's prediction performance. The parameters of the bidirectional LSTM monthly load prediction model network constructed in this invention are selected as follows: time step of 1 month, maximum number of iterations of 1000, initial learning rate of 0.005, and 128 hidden layer units for both the forward and backward LSTM.

[0173] Based on standard mutual information, the correlation between the steel, shipbuilding, and textile industries and various external influencing factors was analyzed. These external influencing factors included residents' consumption levels, average monthly maximum temperature, average monthly minimum temperature, and number of monthly holidays. The correlation analysis results between each external influencing factor and the monthly load of each industry, obtained based on standard mutual information, are shown in Table 1.

[0174] Table 1. Results of correlation analysis based on standard mutual information.

[0175]

[0176] As shown in Table 1, among the various external factors, temperature has a significant impact on the monthly load of each industry, while the impact of consumption level and the number of monthly holidays is relatively small. This is mainly because temperature affects industrial heating and cooling air conditioning loads; when the monthly average maximum temperature is high, the cooling air conditioning load is high, while when the monthly average temperature is low, the heating air conditioning load is high. Nevertheless, the relative strength of the correlation between load and monthly average temperature varies among different industries, with the shipbuilding and textile industries significantly higher than the steel industry. This is mainly because the steel industry, due to its specific characteristics, has a significantly lower demand for air conditioning than the shipbuilding and textile industries. Furthermore, it is easy to see that the correlation between the number of monthly holidays and the textile industry is significantly higher than the other two industries. This is because the shipbuilding and steel industries belong to heavy industry, while the textile industry belongs to light industry, and heavy industry is generally not affected by holidays. Consumption level is national economic data, which is macroeconomic data, and therefore has a low correlation with monthly industry load data.

[0177] By comparing the correlation analysis results based on standard mutual information before and after clustering, it can be found that before clustering, the data in the table shows a weak correlation between consumption levels and the monthly load of the three industries (steel, shipbuilding, and textiles). After clustering, the correlation between consumption levels and monthly load further decreases, indicating that clustering effectively weakens the influence of non-critical factors. For temperature factors, which are highly correlated with industry monthly load, both the monthly average maximum and minimum temperatures show a significant increase in correlation after clustering. For example, the monthly average maximum temperature for the shipbuilding industry increased from 0.52 to 0.58, and for the textile industry from 0.49 to 0.59. This trend indicates that clustering effectively reduces random errors by minimizing the influence of outliers in the data, making important features more prominent in the data.

[0178] The correlation analysis results between various external influencing factors and the load of various industries were obtained. Based on this, the weighted external feature data was processed and then input into a bidirectional LSTM prediction network along with the load features to obtain the prediction results of the load of each industry. To verify the performance of the proposed bidirectional LSTM neural network prediction model that takes into account external influencing factors, its prediction results were compared with those obtained by load prediction models based on GRU, CNN, LSTM, bidirectional LSTM, exponential smoothing, and extreme gradient boosting decision tree (XGBoost). The prediction error comparison results for the shipbuilding industry load are shown in Table 2, and the prediction result curves are shown in... Figure 4 As shown.

[0179] Table 2. Prediction error results under different methods

[0180]

[0181] From Table 2 and Figure 4 It is evident that the LSTM prediction method based on traditional deep learning models and the exponential smoothing method based on traditional time series prediction models perform poorly. This is primarily because they only consider load time series data, neglecting the impact of external factors on industry load. GRU and XGBoost perform at a mid-range level, inferior to CNN and the method proposed in this invention. This is because the time span for monthly industry load prediction is long, GRU has many parameters, and it is prone to overfitting when the monthly data sample is relatively small. XGBoost, essentially a tree-based ensemble method, has limited ability to model the continuity and temporal dependencies of time series. While the time span for monthly industry load prediction is long, key features may be concentrated in the recent months. CNN's local receptive field is suitable for capturing such short- to medium-term correlations. However, monthly load also needs to be optimized against historically similar months. The method proposed in this invention possesses a global perspective that can effectively capture such long- to medium-term correlations, and combined with the temporal correlations that bidirectional LSTM excels at learning, it achieves the lowest prediction error.

[0182] Example 2:

[0183] This embodiment provides an industry monthly load forecasting system that integrates kernel density estimation, standard mutual information, and bidirectional LSTM. The system includes:

[0184] Data receiving module: used to receive historical monthly industry load data and external influencing factor data;

[0185] Clustering module: The received data is clustered using a fuzzy C-means clustering algorithm based on kernel density estimation.

[0186] Correlation Analysis Module: Based on clustering results, standard mutual information is calculated to quantitatively analyze the correlation between external influencing factors and industry monthly load;

[0187] Monthly load forecasting module: It uses a bidirectional LSTM neural network and combines the results of correlation analysis to forecast the monthly load of the industry.

[0188] The data receiving module effectively integrates historical monthly industry load data and external influencing factor data, providing comprehensive data support for subsequent analysis and forecasting. Fuzzy C-means clustering is used to cluster the data, better identifying patterns and structures and improving data comprehension. Standard mutual information is used to quantify the relationship between external influencing factors and monthly industry load, revealing key factors affecting monthly load changes and improving forecast accuracy. Bidirectional LSTM (Bidirectional Long Short-Term Memory) captures time dependencies in the data and considers bidirectional information from historical data, thus improving the accuracy and reliability of monthly load forecasting. The system integrates clustering analysis, correlation analysis, and deep learning methods, fully utilizing data information to provide more comprehensive and accurate monthly load forecasts. The system can adjust according to changes in monthly industry load and external influencing factors, making the forecast results more aligned with reality and improving system adaptability.

[0189] Membership Determination Unit: Used to set the number of clusters based on the improved fuzzy C-means clustering algorithm, select initial cluster centers based on the kernel density estimation method, and calculate the membership degree between monthly load data and cluster centers;

[0190] Objective function building block: Used to define the objective function to minimize the Euclidean distance between the monthly load data and the cluster centers;

[0191] Cluster center update unit: The cluster center matrix is ​​updated by iteratively optimizing the objective function until the objective function converges;

[0192] Optimal clustering result acquisition unit: used to obtain clustering results based on the minimized objective function value;

[0193] Optimal cluster number determination unit: Select the optimal cluster number using the Xie & Ben index and determine the final clustering result;

[0194] The correlation analysis module further includes:

[0195] Data fuzzification processing unit: Based on the clustering results of the clustering processing module, it replaces the original external influencing factor values ​​and monthly load values ​​with the cluster center values ​​corresponding to each sampling point to obtain the fuzzified sequence;

[0196] Standard mutual information calculation unit: used to calculate the standard mutual information value between the fuzzy data to quantitatively describe the correlation between external influencing factors and monthly load;

[0197] Quantitative Correlation Assessment Unit: Used to assess the correlation between external influencing factors and industry monthly load based on the standard mutual information value, and input the results into the monthly load forecasting module;

[0198] The monthly load forecasting module further includes:

[0199] Correlation weight calculation unit: used to normalize the standard mutual information value between the industry's monthly load and each external influencing factor in order to determine the characteristic weight of the external influencing factor;

[0200] Eigenvector weighting unit: Used to sum the eigenvectors of all external influencing factors based on the calculated eigenweights to obtain a comprehensive eigenvector;

[0201] Improved bidirectional LSTM modeling unit: Construct a bidirectional LSTM neural network to capture the temporal variation pattern of industry load, adopt the Transformer multi-head attention mechanism to analyze the influence of external features, and connect the Transformer and bidirectional LSTM models through a convolutional neural network;

[0202] Network optimization and output unit: Adjusts the shape of the final output data through a fully connected layer to form monthly load forecast results.

[0203] Through the foregoing detailed description of the industry monthly load forecasting method that integrates kernel density estimation, standard mutual information, and bidirectional LSTM, those skilled in the art can clearly understand that the industry monthly load forecasting system that integrates kernel density estimation, standard mutual information, and bidirectional LSTM in this embodiment, as corresponding to the method disclosed in Embodiment 1, has corresponding functional modules and beneficial effects. For relevant details, please refer to the method section.

[0204] Example 3

[0205] This embodiment provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the current limiting method for a grid-type flexible DC system based on power matching as described in any embodiment of the present invention.

[0206] Example 4

[0207] This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the current limiting method for a grid-type flexible DC system based on power matching, as described in any embodiment of the present invention.

[0208] Those skilled in the art will recognize that the units and algorithm steps described in the disclosed embodiments can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0209] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0210] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0211] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An industry load forecasting method integrating standard mutual information and improved bidirectional LSTM, characterized in that, Includes the following steps: 1) An improved fuzzy C-means clustering algorithm that takes kernel density estimation into account is used to cluster information including load sequences, consumption levels, temperature, and vacation time. 2) Based on the clustering results, standard mutual information is calculated to quantitatively analyze the correlation between information factors, including consumption level, temperature, and vacation time, and the monthly load of the industry; 3) Based on the correlation analysis results, weights are assigned, a bidirectional LSTM neural network is constructed to capture the temporal variation pattern of industry load, the Transformer multi-head attention mechanism is used to analyze the influence of external features, and the results are output through convolutional neural network connections to produce monthly industry load forecast results; Step 3) of the monthly load forecast includes the following steps: 3.1) Correlation weight calculation: Normalize the standard mutual information value between the industry's monthly load and each external influencing factor to determine the feature weights of information factors including consumption level, temperature, and vacation time; 3.2) Weighted Eigenvectors: Based on the calculated eigenweights, the eigenvectors of all external influencing factors are weighted and summed to obtain the enhanced comprehensive eigenvector; 3.3) Bidirectional LSTM network modeling: A bidirectional long short-term memory network is used to capture the time series feature vector of historical monthly load as input to capture the time series change trend of industry monthly load; 3.4) Transformer multi-head attention mechanism modeling: The Transformer multi-head attention mechanism is applied, taking multi-dimensional external feature vectors and comprehensive feature vectors as inputs, and capturing dynamic correlations across feature dimensions through parallel attention heads; 3.5) Network Optimization and Output: The temporal feature vectors extracted by the bidirectional LSTM and the multi-head attention features encoded by the Transformer are used to construct a feature interaction layer through a convolutional neural network. Finally, the monthly load forecast value of the industry with environmental adaptability is output through a fully connected layer.

2. The industry load forecasting method that integrates standard mutual information and improved bidirectional LSTM according to claim 1, characterized in that: The clustering process in step 1) further includes: 1.1) Determine the membership matrix: Based on the improved fuzzy C-means clustering algorithm, set the number of clusters, select the initial cluster centers based on the kernel density estimation method, and calculate the membership degree between the monthly load data and the cluster centers; 1.2) Constructing the objective function: Define an objective function to minimize the Euclidean distance between the monthly load data and the cluster centers; 1.3) Update cluster centers: Iteratively optimize the objective function and update the cluster center matrix until the objective function converges; 1.4) Obtain the optimal clustering result: Obtain the clustering result based on the minimized objective function value; 1.5) Determine the optimal number of clusters: Select the optimal number of clusters using the Xie & Ben index and determine the final clustering result.

3. The industry load forecasting method that integrates standard mutual information and improved bidirectional LSTM according to claim 2, characterized in that: Step 1.1) Determine the number of clusters based on the number of months in the membership matrix. The kernel density estimation method is used to extract data from the electricity consumption matrix. Y Before being selected N l The data with the highest row density is used as the initial cluster center matrix. V =[ V 1,…, V l ,…, V Nl ] T Then calculate data for any month. Y m Belonging to the l The membership degree of each cluster center; Step 1.2) Constructing the objective function, the objective function... G The sum of the squared weighted distances between the membership degrees of all historical months and all cluster centers; Step 1.3) In updating the cluster centers, if the objective function... G If the minimum value is not reached, then the cluster center matrix is ​​updated according to the given update strategy. V Update and recalculate the membership matrix until the objective function is achieved. G To reach the minimum; Step 1.4) Obtaining the optimal clustering result involves analyzing the membership matrix and cluster center matrix. V The number of clusters obtained is N l The optimal clustering result corresponding to the minimum value of the objective function; In step 1.5) determining the optimal number of clusters, the Xie & Ben index is used to reflect the intra-cluster compactness and inter-cluster separation of the clustering results. This is achieved by comparing the clustering quality indices corresponding to different numbers of clusters. I XB Select the optimal number of clusters And obtain the best clustering results.

4. The industry load forecasting method according to claim 3, which integrates standard mutual information and improved bidirectional LSTM, is characterized in that: Step 2) further includes the following steps: 2.1) Data fuzzing processing Based on the clustering results obtained in step 1) by improving the fuzzy C-means (DFCM) clustering algorithm, the original external influencing factor values ​​x(m) and monthly load values ​​y(m) are replaced with the cluster center values ​​corresponding to each sampling point to obtain the fuzzy sequences U and V; 2.2) Standard Mutual Information Calculation 2.2.1) Mutual information calculation: The mutual information I(U;V) between the fuzzified data U and V is calculated to quantitatively describe the correlation between influencing factors and monthly load; 2.2.2) Information entropy calculation: Calculate the information entropy H(U) and H(V) of sequences U and V; 2.2.3) Standard mutual information calculation: The standard mutual information value J(U;V) between the influencing factors and the monthly load is calculated by the ratio of mutual information I(U;V) to information entropy H(U) and H(V); 2.3) Quantitative assessment of correlation 2.3.1) Results Analysis: Based on the standard mutual information value J(U;V), the correlation between information factors including consumption level, temperature, and vacation time and the monthly load of the industry is quantitatively assessed; the larger the value, the stronger the correlation. 2.3.2) Parameter input: The correlation analysis results obtained based on standard mutual information are used as network hyperparameters and input into the monthly load forecasting network to optimize the model's forecasting ability.

5. The industry load forecasting method according to claim 4, which integrates standard mutual information and improved bidirectional LSTM, is characterized in that: In step 2.1), during the data fuzzification process, for the monthly load and external influencing factors of the industry to be analyzed, a certain time period is known. N m Influencing factor values ​​of each sampling point x ( m ) and monthly load value y ( m ), m ∈{1,2,…, N m },all x ( m ) constitute a sequence X ,all y ( m ) constitute a sequence Y The original data was replaced with data processed by DFCM clustering, and standard mutual information was calculated. The sampled values ​​were recorded. x(m) The cluster center value of the cluster is u(m) ,all u(m) constituting a sequence U Sample value y(m) The cluster center value of the cluster is v(m) ,all v(m) constituting a sequence V ; u(m) Total F Possible values, F ≤ N m , recorded as , f ∈{1,2,…, F }, v(m) Total G Possible values, G ≤ N m , recorded as , ∈{1,2,…, G }; In step 2.2), the standard mutual information calculation involves the standard mutual information values ​​between the influencing factors and the monthly load. J ( U ; V The calculation method is as follows: In the formula, In the formula, I(U;V) for u(m) and v(m) Mutual information value; H(U) and H(V) They are respectively U and V Information entropy; To simultaneously satisfy u ( m )= and v(m) = The proportion of sampling points to all sampling points; To meet u(m) = The proportion of sampling points to all sampling points; To meet v(m) = The proportion of sampling times to all sampling times; J(U;V) The value range is [0,1]. The larger the value, the stronger the correlation between the influencing factor and the monthly load of the industry.

6. The industry load forecasting method according to claim 5, which integrates standard mutual information and improved bidirectional LSTM, is characterized in that: The bidirectional LSTM neural network model captures the long-term and short-term dependencies of load sequences based on a bidirectional temporal feature extraction module. It controls the transmission of information through forget gates, input gates, and output gates. The formulas for calculating the forward and backward monthly load forecast outputs are as follows: In the formula, Represents the forward calculated value; Represents the backward calculated value; o t This represents the final output value calculated from the forward and backward values; ω 1 represents the weight matrix mapping from the input layer to the feedforward layer; ω 3 represents the weight matrix mapping from the input layer to the feed layer; ω 2 is the weight matrix that maps the output of the previous time step of the forward layer to the current time step; ω 5 is the weight matrix that maps the output of the backward layer at the previous time step to the current calculation time step; ω 4 is the weight matrix that maps the output of the feedforward layer to the output layer; ω 6 represents the weight matrix that maps the output of the backward layer to the weight matrix of the output layer; The Transformer multi-head attention mechanism maps input features to different subspaces by computing multiple sets of attention weights in parallel. It captures the dynamic relationships between data in each subspace to effectively identify the heterogeneous impact of key external events, including abrupt temperature changes and holiday effects, on the load. Single-head attention is calculated using the following formula: In the formula, Q , K , V To query the key-value matrix; This is the scaling factor; By dividing the model into multiple attention heads, forming multiple subspaces, the model can focus on different aspects of information. The specific expression is as follows: In the formula, Q i , K i , V i For the first i A query key-value matrix for each attention head; , , For the first i The weight matrix of each attention head; To output the projection matrix; h For the number of attention heads.

7. An industry load forecasting system integrating standard mutual information and improved bidirectional LSTM, characterized in that: The system is used to perform the method of any one of claims 1 to 6, the system comprising: Data receiving module: used to receive historical monthly industry load data and external influencing factor data; Clustering module: The received data is clustered using a fuzzy C-means clustering algorithm based on kernel density estimation. Correlation Analysis Module: Based on clustering results, standard mutual information is calculated to quantitatively analyze the correlation between external influencing factors and industry monthly load; Monthly load forecasting module: It uses a bidirectional LSTM neural network and combines the results of correlation analysis to forecast the monthly load of the industry.

8. An industry load forecasting system integrating standard mutual information and improved bidirectional LSTM according to claim 7, characterized in that: The clustering processing module includes a membership degree determination unit, an objective function construction unit, a cluster center update unit, an optimal clustering result acquisition unit, and an optimal number of clusters determination unit. The membership degree determination unit sets the number of clusters according to the improved fuzzy C-means clustering algorithm, selects the initial cluster centers based on the kernel density estimation method, and calculates the membership degree between the monthly load data and the cluster centers. The optimal number of clusters determination unit selects the optimal number of clusters using the Xie & Ben index and determines the final clustering result; The correlation analysis module includes a data fuzzification processing unit, a standard mutual information calculation unit, and a correlation quantitative evaluation unit. The data fuzzification processing unit replaces the original external influencing factor values ​​and monthly load values ​​with cluster center values ​​to obtain a fuzzified sequence. The standard mutual information calculation unit calculates the standard mutual information values ​​between the fuzzified data to quantitatively describe the correlation between external influencing factors and monthly load. The monthly load forecasting module includes a correlation weight calculation unit, a feature vector weighting unit, an improved bidirectional LSTM modeling unit, and a network optimization and output unit. The improved bidirectional LSTM modeling unit constructs a bidirectional LSTM neural network, uses the Transformer multi-head attention mechanism to analyze the influence of external features, and connects the Transformer and bidirectional LSTM models through a convolutional neural network.

9. A computer device, characterized in that: The device includes one or more processors and one or more memories, wherein at least one piece of program code is stored in the one or more memories, and when the program code is executed by the one or more processors, it implements an industry load forecasting method that integrates standard mutual information and improved bidirectional LSTM as described in any one of claims 1-6.

Images