Hybrid integrated electricity price prediction method and related device

Abstract

The invention belongs to the technical field of electricity markets, and discloses a hybrid integrated electricity price prediction method and a related device. The hybrid integrated electricity price prediction method comprises the following steps: based on a feature set of a to-be-predicted day, carrying out joint prediction by utilizing 24 trained statistical prediction models and one deep neural network model, and obtaining an electricity price prediction value of the to-be-predicted day; wherein in the 24 trained statistical prediction models, each statistical prediction model is used for correspondingly predicting the electricity price of each hour of the day to be predicted; in addition, in the process of joint prediction, a feature input type integrated path or a weighted fusion type integrated path is adopted. According to the technical scheme disclosed by the invention, the stability and interpretability advantages of the statistical model are organically combined with the nonlinear capture capability of the deep learning model, and the prediction precision, robustness and generalization capability are greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of electricity market technology, and specifically relates to a hybrid integrated electricity price forecasting method and related apparatus. Background Technology

[0002] In modern electricity markets, electricity prices are the core signal guiding the optimal allocation of electricity resources, exhibiting complex characteristics such as high volatility, high time-series correlation, and multiple seasonalities. Accurate electricity price forecasting is crucial for power generation companies to formulate pricing strategies, for electricity retailers to control electricity purchase costs, and for users to adjust their electricity consumption behavior. Therefore, researching high-precision electricity price forecasting models has significant theoretical and practical value.

[0003] Existing electricity price forecasting methods are mainly divided into two categories: statistical forecasting models and artificial intelligence forecasting models. However, single forecasting models are no longer sufficient to cope with the increasing complexity of the current electricity market. On the one hand, while statistical forecasting models are robust in linear relationships and periodic analysis, their predictive ability is limited when faced with the prevalent nonlinear and non-stationary characteristics of electricity prices, especially in accurately capturing price spikes caused by factors such as rapid changes in market supply and demand and the randomness of renewable energy output. On the other hand, while deep learning models possess strong nonlinear fitting capabilities, they also have inherent defects: their "black box" nature makes the model's decision-making process opaque, making it difficult to conduct effective error attribution and mechanism analysis, resulting in poor reliability in key decision-making applications; the models are highly dependent on massive amounts of high-quality data and complex hyperparameter tuning, and are prone to overfitting and decreased generalization ability when training data is insufficient or features are not significant; during the learning process, they may not be able to consistently capture the clear, physically meaningful autoregressive and periodic patterns in electricity prices.

[0004] In conclusion, designing a new forecasting framework to better address the increasing complexity of the electricity market has become an urgent technical challenge. Summary of the Invention

[0005] The purpose of this invention is to provide a hybrid integrated electricity price prediction method and related apparatus to solve one or more of the aforementioned technical problems. Specifically, the technical solution disclosed in this invention is a hybrid integrated electricity price prediction scheme based on a statistical prediction model and a deep neural network model. This scheme organically combines the stability and interpretability advantages of statistical models with the nonlinearity capture capability of deep learning models, resulting in significant improvements in prediction accuracy, robustness, and generalization ability.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a hybrid integrated electricity price forecasting method, comprising the following steps: Obtain the feature set of the day to be predicted; Based on the feature set of the day to be predicted, 24 pre-trained statistical prediction models and 1 deep neural network model are used to jointly predict the electricity price of the day to be predicted. Among the 24 pre-trained statistical prediction models, each statistical prediction model is used to predict the electricity price of each hour of the day to be predicted. In addition, in the process of joint prediction, a feature input integration path or a weighted fusion integration path is adopted.

[0007] A further improvement of the technical solution of the present invention is that, in the process of adopting the feature input integration path, firstly, based on the feature set of the day to be predicted, 24 statistical prediction models are used to make a preliminary prediction to obtain the 24-hour electricity price preliminary prediction vector of the day to be predicted; then, the feature set of the day to be predicted is concatenated with the 24-hour electricity price preliminary prediction vector of the day to be predicted to form an enhanced feature set; finally, the enhanced feature set is used as input, and a trained deep neural network model is used to make a final prediction to obtain the electricity price prediction value of the day to be predicted.

[0008] A further improvement of the technical solution of the present invention is that, in the process of adopting the weighted fusion integration path, the feature set of the day to be predicted is first input into 24 pre-trained statistical prediction models and 1 deep neural network model for parallel prediction, so as to obtain the first 24-hour prediction vector output by the 24 statistical prediction models and the second 24-hour prediction vector output by the deep neural network model; the first 24-hour prediction vector and the second 24-hour prediction vector are weighted to obtain the electricity price prediction value of the day to be predicted.

[0009] A further improvement of the technical solution of the present invention is that, in the step of weighting the first 24-hour prediction vector and the second 24-hour prediction vector, one or more of the following are employed: fixed weighting, dynamic weighting based on historical prediction errors, hourly independent weighting, adaptive weighting based on prediction confidence, conditional weighting based on market conditions, recursive update weighting, seasonal periodic weighting, segmented weighting based on temperature ranges, hierarchical weighting based on load levels, and differentiated weighting based on holiday types.

[0010] A further improvement to the technical solution of this invention is that the step of obtaining the feature set of the day to be predicted includes: Based on the selected date to be predicted and the time span, raw data is obtained; wherein, the raw data includes historical electricity price series, load data, renewable energy output data, fuel price data and meteorological data; Based on the selected date to be predicted and the acquired raw data, time features, historical price lag features, and exogenous variable lag features are constructed and concatenated in a predefined order to form a feature vector. The resulting feature vectors are processed to obtain the final feature set; the data processing includes detecting and handling missing values, and identifying and correcting outliers.

[0011] A further improvement to the technical solution of this invention lies in the step of constructing time characteristics, historical price lag characteristics, and exogenous variable lag characteristics. Based on the timestamp information of the day to be predicted, generate dummy variables for the week, month, holidays, and hours as time features; Extract the corresponding hourly electricity prices for the 1st, 2nd, 3rd, 7th, and 14th days prior to the predicted date to serve as historical price lag characteristics; For each exogenous variable, extract its corresponding hourly values ​​for the past 1, 2, and 7 days to construct lagged features of the exogenous variable.

[0012] A further improvement to the technical solution of this invention lies in that, among the 24 statistical prediction models, for hourly... The predicted electricity price from the statistical forecasting model is expressed as: ; In the formula, For hours h The predicted electricity price; For the intercept term; Input features eigencoefficients, For the first i Each input feature n For feature dimension, This is the error term; During the training of the statistical prediction model, L1 regularization is used for feature selection and coefficient estimation, and the objective function is... The expression is: ; In the formula, This is the scaling factor; The number of samples; For the first The actual target variable value for each observation; For the first The feature vector of each observation value For parameter vectors; For the first The predicted value of each observation is the dot product of the feature vector and the parameter vector; For regularization parameters; The L1 regularization term is a parameter vector. Each element The sum of the absolute values; n For parameter vectors Dimensions.

[0013] A second aspect of the present invention provides a hybrid integrated electricity price forecasting system, comprising: The data acquisition unit is used to acquire the feature set of the day to be predicted. The joint prediction unit is used to jointly predict the electricity price of the day to be predicted based on the feature set of the day to be predicted, using 24 pre-trained statistical prediction models and 1 deep neural network model. Among the 24 pre-trained statistical prediction models, each statistical prediction model is used to predict the electricity price for each hour of the day to be predicted. In addition, during the joint prediction process, a feature input integration path or a weighted fusion integration path is adopted.

[0014] In a third aspect, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the hybrid integrated electricity price prediction method as described in any one of the first aspects of the present invention.

[0015] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the hybrid integrated electricity price prediction method as described in any one of the first aspects of the present invention.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention innovatively proposes a hybrid integrated electricity price forecasting method. Its core technological innovation lies in its unique hybrid integrated forecasting architecture, which aims to achieve complementary advantages between statistical models and deep learning models. Specifically, this is reflected in the following aspects: Hybrid integrated architecture; In addressing the coexistence of linear trends and nonlinear fluctuations in electricity price sequences, the technical solution of this invention adopts a hybrid integrated architecture, combining statistical models (mainly capturing linear autoregressive relationships) with deep learning models (mainly capturing nonlinear mappings). This architecture features clear division of labor and complementary advantages, and can comprehensively cover the multiple characteristics of electricity prices when used to solve the above problems. Dual integration path; In response to the problem that different application scenarios have different requirements for integration methods, the technical solution of this invention adopts a dual integration path design, providing two optional solutions: "feature input integration" and "weighted fusion integration". This design has the characteristics of high flexibility and strong adaptability, and can meet the application requirements of different computing resources and accuracy requirements to solve the above problems. Error protection mechanism; In response to the high risk of prediction failure of a single model, the technical solution of this invention adopts an error protection mechanism. Through hybrid integration, the prediction deviation of one model can be compensated by another model. This mechanism has the characteristics of mutual backup and dynamic correction. It can significantly improve the robustness and reliability of the prediction system when used to solve the above problems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a hybrid integrated electricity price prediction method in an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall technical framework in an embodiment of the present invention; Figure 3 This is a schematic diagram of the feature input-based integration process in an embodiment of the present invention; Figure 4 This is a schematic diagram of the weighted fusion integration process in an embodiment of the present invention; Figure 5 This is a schematic diagram of the power load visualization in the Pennsylvania-New Jersey-Maryland Region (PJM) of the United States, as described in this embodiment of the invention. Figure 6 This is a schematic diagram comparing electricity price forecasts for a specific date in an embodiment of the present invention; Figure 7 This is a schematic diagram showing a comparison of electricity price forecasts over a consecutive week in an embodiment of the present invention; Figure 8 This is a schematic diagram of a hybrid integrated electricity price prediction system in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Based on the technical solutions disclosed in the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0021] Please see Figure 1 and Figure 2 The present invention provides a hybrid integrated electricity price forecasting method that closely integrates three stages: data preprocessing, parallel model construction, and integrated forecasting. The specific implementation process is as follows: Step 1: Data preprocessing and feature construction.

[0022] Specifically, the process begins by obtaining a raw dataset from the original data input, which includes historical electricity prices, multiple exogenous variables (such as load forecasts, wind / solar power output forecasts, etc.), and timestamps. Next, the following operations are performed on the original dataset in the data preprocessing module: Feature construction includes: generating time features (such as weekday dummy variables) based on timestamps, and constructing lagged terms of historical prices (such as hourly prices corresponding to the past 1, 2, and 7 days) and lagged terms of exogenous variables.

[0023] Data scaling includes processing numerical features to improve the stability and efficiency of model training. Specifically, for electricity price data with peaked and heavy-tailed distributions, arcsine-median transformation is used to stabilize the variance. For other variables, Z-score normalization is used to make them conform to a distribution with a mean of 0 and a standard deviation of 1, so as to avoid the unbalanced influence of features with different dimensions on the model.

[0024] In practical applications, the process of obtaining the feature set for a selected date to be predicted is as follows: Data acquisition steps: Extract raw data from the electricity market trading system, including historical electricity price sequences, load data, renewable energy output data, fuel price data, and meteorological data, ensuring that the data time span covers at least 30 days before the forecast date; Time feature construction steps: Based on the timestamp information of the day to be predicted, generate weekday dummy variables, month dummy variables, holiday identifiers and hour identifiers to capture the periodicity and seasonality of electricity prices; Steps for constructing historical price lag terms: Extract the corresponding hourly electricity prices of the 1st, 2nd, 3rd, 7th and 14th days before the date to be predicted as lag features, and calculate the moving average electricity prices of the past 24 hours and 168 hours to form an autoregressive feature vector; Steps for constructing lag terms of exogenous variables: For each exogenous variable (load, wind power output, photovoltaic power output, etc.), extract its corresponding hourly values ​​for the past 1 day, 2 days, and 7 days, and construct the time lag feature matrix of the exogenous variable; Data preprocessing steps: The electricity price data is processed using an arcsine-median transform to remove its peak and heavy-tailed characteristics, i.e.: ; in, The absolute deviation of the median; other numerical variables are standardized using Z-score: ; Feature concatenation steps: Concatenate time features, historical price lag features, and exogenous variable lag features in a predefined order to form a feature with dimensions of [dimensional value missing]. eigenvectors; where, Total feature dimension; Data quality control steps: Detect and handle missing values ​​(using linear interpolation or forward imputation), identify and correct outliers (based on...). (Principles) to ensure the integrity and reliability of the final feature set.

[0025] Step 2: Construct the LEAR model and DNN model in parallel. Feed the preprocessed data into two parallel model training modules, including: The LEAR model training process includes training an independent LASSO regression model for each hour of the future day (h=0, 1, ..., 23). This model utilizes input features to capture the main linear relationships and periodic patterns in electricity prices. During training, for the model at hour h, the target variable is the electricity price for that hour, and the input features are all the relevant features constructed in the first step. The LASSO algorithm used in training, through L1 regularization, can compress unimportant feature coefficients to zero while estimating regression coefficients, thus automatically achieving feature selection and effectively addressing the problems of collinearity and high dimensionality among features. Ultimately, 24 independent, optimized linear prediction models are obtained, which together constitute the LEAR prediction system, specifically designed to capture linear trends and periodic patterns in electricity prices.

[0026] DNN model training includes: constructing a multi-layer fully connected deep neural network designed to capture the complex nonlinear relationship between features and electricity prices; during training, using the backpropagation algorithm and optimizer to iteratively update the network weights with the goal of minimizing the loss function (mean squared error) between the predicted and true values.

[0027] In this embodiment of the invention, a typical deep neural network structure includes: Input layer: The number of neurons is equal to the total dimension of the input features.

[0028] Hidden layers: These consist of multiple fully connected layers, with the number of neurons in each layer decreasing in a progressively smaller manner. The activation function commonly used is ReLU (Rectified Linear Unit) to prevent gradient vanishing and accelerate training.

[0029] Regularization layer: Dropout layer (randomly deactivates some neurons to prevent overfitting) and batch normalization layer (stabilizes and accelerates the training process) are added between the hidden layers.

[0030] Output layer: Contains 24 neurons, corresponding to the electricity price forecast for the next 24 hours. It usually uses a linear activation function because the output is a continuous regression value.

[0031] In a specific exemplary technical solution, the steps for constructing the LEAR model are as follows: Independent LASSO regression models were established for each of the 24 hours of the day to be predicted, with each model corresponding to the electricity price prediction for a specific hour. For hours Model building: ;in, Forecast electricity price for hour h, For the intercept term, For characteristic coefficients, As input features, This is the error term; L1 regularization is used for feature selection and coefficient estimation. The objective function is: ;in, The value is a regularization parameter, and its optimal value is determined through cross-validation.

[0032] The complete training process for a statistical prediction model includes the following steps: The historical dataset was divided into a training set (70%), a validation set (15%), and a test set (15%). For each hour Train a LASSO regression model independently: solve the LASSO optimization problem using the coordinate descent algorithm; select the optimal regularization parameter on the validation set using 10-fold cross-validation. The search scope is Based on optimal The model is retrained on the complete training set to obtain the regression coefficients for each hour. ; Output 24 independently trained LASSO models to form a complete LEAR prediction system.

[0033] The deep neural network model uses a DNN model architecture, which specifically includes: Input layer: The number of neurons is equal to the feature dimension D, and it receives the preprocessed feature vector; Hidden layer structure: Contains 3 to 5 fully connected layers, with the number of neurons decreasing in sequence. A typical architecture is as follows. ; Activation function: The hidden layer uses the ReLU activation function. To prevent gradient vanishing; Regularization mechanisms: Dropout layer, which randomly deactivates 30%-50% of neurons during training to prevent overfitting; Batch normalization layer, which normalizes the output of each layer to accelerate convergence and improve stability. Output layer: Contains 24 neurons, corresponding to 24-hour electricity price prediction, using a linear activation function; Loss function: Mean squared error (MSE) loss is used. ; Optimization algorithm: The Adam optimizer is used, the learning rate is initially set to 0.001, and a learning rate decay strategy is adopted.

[0034] The complete training process of a deep neural network model includes the following steps: Data preparation: Input the preprocessed feature set into the network in batches (batch size=32 or 64); Forward propagation: The input layer receives a feature vector of dimension D; it undergoes linear transformation and ReLU activation through multiple hidden layers. Dropout and batch normalization are applied during the training phase; the output layer generates a 24-dimensional prediction vector. Backpropagation and parameter update: Calculating MSE loss: The network weights are updated using the Adam optimizer with an initial learning rate of 0.001. The learning rate is reduced to 0.9 times the original rate every 10 epochs. An early stopping mechanism is used, which stops training when the validation set loss does not decrease for 10 consecutive epochs to prevent overfitting. The optimal model parameters on the validation set are saved as the final model.

[0035] Step 3: Perform integrated prediction.

[0036] The integrated prediction module is the core of this invention. As an intermediate coordination layer, it receives the 24-hour linear prediction results output by the LEAR model and the 24-hour nonlinear prediction results output by the DNN model, and provides two optional integration paths to fuse these two types of predictions to generate the final prediction output.

[0037] Please see Figure 3 The first embodiment of this invention provides a feature-input ensemble path. This approach uses the output of the LEAR model as "prior knowledge" to guide the DNN model to perform more refined learning, specifically including: Preliminary prediction: Input the feature set of the day to be predicted into 24 pre-trained LEAR models to obtain a preliminary 24-hour electricity price prediction vector. .

[0038] Feature enhancement: This The vector is treated as an "expert feature" containing 24 new values, and it is concatenated with the original feature set to form an enhanced feature set with higher dimensions and richer information.

[0039] Final Prediction: This enhanced feature set is fed into a DNN model specifically trained for this approach. During training, this DNN model learns how to utilize the prediction baseline of LEAR and, based on this, captures and corrects nonlinear errors, thus outputting the final prediction result. This approach transforms the DNN's learning task from "predicting from zero" to "correcting based on expert opinions," reducing the learning difficulty and improving efficiency and accuracy.

[0040] Please see Figure 4 Scheme 2 of this invention provides a weighted fusion integration path. This scheme treats LEAR and DNN as two independent experts running in parallel, and aggregates their predictions through intelligent weighting, specifically including: Parallel prediction: The feature set of the day to be predicted is simultaneously input into the independently trained LEAR model and DNN model to obtain their respective 24-hour prediction vectors. and .

[0041] Weighted fusion: The predictions from the two models are weighted and averaged. For each hour h, the final predicted value is... The calculation formula is: ; Weight determination: The methods for determining the weight coefficient ω are flexible and varied, including: Fixed weights: It could be an optimal fixed value (such as 0.5) found by performing a grid search on the validation set. Dynamic weights: It can be dynamically adjusted. For example, it can be dynamically allocated based on the prediction errors (e.g., MAE) of the two models over a past period (e.g., the past 7 days), giving higher weight to the model that has performed better recently; it can even be adjusted every hour. Set independent weights To adapt to the different performance of different models during peak and off-peak periods.

[0042] Final output: Output the weighted and merged 24-hour electricity price forecast.

[0043] In existing technical solutions, autoregressive models based on LASSO (such as the LEAR model) belong to the category of statistical models. For example, advanced statistical forecasting methods construct multiple autoregressive models and use LASSO (Least Absolute Shrinkage and Selection Operator) regression for variable selection and parameter estimation. These models typically establish an independent LASSO regression model for each of the 24 hours of the next day. In terms of technical implementation, the input features mainly include historical electricity prices (such as the corresponding hourly electricity prices for the past 1, 2, 3, and 7 days), exogenous variables (such as load forecasts and renewable energy output forecasts), and dummy variables representing the weekday effect. This technical solution suffers from limited ability to capture nonlinear relationships and strong model assumptions. Its linear assumptions prevent it from effectively fitting the complex nonlinear mapping relationship between electricity prices and other influencing factors (such as fuel prices and weather changes). When market structures change or extreme events occur, statistical models based on linear extrapolation of historical data often produce large deviations and lack robustness. Specifically, LASSO is essentially a linear regression model. Although it can effectively handle high-dimensional features and collinearity, it is insufficient in capturing the complex nonlinear dynamics inherent in electricity price sequences (such as price spikes and the nonlinear impact of holiday effects). In addition, statistical models are usually based on certain mathematical assumptions (such as linear relationships and error distributions). When actual data does not meet these assumptions, the model's predictive performance will decline.

[0044] In existing technical solutions, prediction models based on deep neural networks (such as DNN models) belong to the category of artificial intelligence models. Deep Neural Networks (DNNs) can automatically learn and extract complex nonlinear features from large amounts of data by constructing multi-layered nonlinear neural networks. In electricity price forecasting, the input to DNN models typically includes historical electricity prices, exogenous variables, and time features (such as days of the week and hours). Technically, this is achieved by constructing a flexible network structure containing multiple fully connected layers, activation functions, and regularization techniques such as Dropout and batch normalization. The optimal model structure and feature combination are usually determined through hyperparameter optimization algorithms. This technical solution suffers from poor interpretability, sensitivity to data and hyperparameters, and the potential to overlook simple linear patterns. Specifically, DNNs are "black box" models, with their internal decision-making processes difficult to understand. This makes model tuning and building trust challenging, and they may not be able to clearly separate intuitive components such as linear trends and periodicity. Furthermore, DNN performance is highly dependent on large amounts of training data and meticulous hyperparameter tuning. In cases of sparse data or poor feature engineering, overfitting or underfitting can easily occur, leading to poor model generalization ability. Moreover, despite their powerful capabilities, DNNs may not be able to consistently capture fundamental, significant linear autoregressive relationships and periodicity during the learning process, unlike statistical models. In summary, each individual model has its advantages and disadvantages, making it difficult to simultaneously achieve multiple objectives such as linearity and nonlinearity, robustness, and high accuracy.

[0045] In view of the above, this invention presents a novel prediction framework that leverages the strengths of statistical models while mitigating their weaknesses, organically combining the stability and interpretability of statistical models with the nonlinearity capture capabilities of deep learning models. The hybrid integrated electricity price prediction method based on LEAR and DNN provided by this invention constructs a collaborative system. The LEAR model provides a reliable prediction baseline, while the DNN model learns and compensates for the nonlinear residuals. Alternatively, the results of both models are intelligently weighted and fused, ultimately achieving an electricity price prediction solution that surpasses any single model in prediction accuracy, robustness, and generalization ability.

[0046] In this embodiment of the invention, the statistical model is further explained as follows: In addition to the LEAR model, other advanced statistical models can also be used, such as the seasonal autoregressive moving average model with exogenous variables (SARIMAX), quantile regression, etc.

[0047] In this embodiment of the invention, the following further explanation of the deep learning model is provided: In addition to DNN, recurrent neural networks (RNN), long short-term memory networks (LSTM), or gated recurrent units (GRU), which are more suitable for processing time series, can also be used. These models are better able to capture long-range dependencies in time.

[0048] In this embodiment of the invention, the variations of the integration strategy are further explained as follows: Stacking can be more complex. It uses LEAR and DNN as base learners in the first layer, then trains a meta learner (which could be a simple linear regression model or even another small neural network) in the second layer. The meta learner learns how to optimally combine the outputs of the first layer model, rather than simply weighting them.

[0049] Hourly dynamic weights: In weighted fusion integration, the weights can be optimized for each hour, resulting in a weight vector containing 24 different values, because different models may have different predictive capabilities during peak and off-peak periods.

[0050] This invention effectively overcomes the limitations of single models in electricity price forecasting by integrating a LASSO-based autoregressive model (LEAR) with a deep neural network (DNN), achieving a significant improvement in forecasting performance. Addressing the issue of varying data characteristics across different electricity markets, this invention employs two specific integration paths (feature-input and weighted fusion). These two paths fuse data at the feature level and the result level respectively, exhibiting complementarity and versatility, thus adapting to the forecasting needs of various electricity markets. To address the need for the technical solution to adapt to different model types, the scope of this invention is not limited to the specific implementations of LEAR and DNN, but covers replacements with all functionally equivalent statistical models (such as ARIMA and SARIMAX) and deep learning models (such as LSTM and Transformer). This generalized design is highly scalable, ensuring the long-term effectiveness and wide applicability of the technical solution. All these alternative solutions adhere to the core idea of ​​this invention—integrating different types of models to achieve complementary advantages—and should be considered extensions of the scope of this invention.

[0051] Please see Figure 5To verify the effectiveness of the technical solution in this invention, six years of test data were selected for backtesting analysis, using the PJM electricity market as an example. The hybrid ensemble model was compared with single LEAR and DNN models. This invention uses two widely used evaluation metrics in electricity market forecasting: Mean Absolute Error (MAE) and Symmetric Mean Absolute Percentage Error (sMAPE) to measure the model's prediction accuracy. MAE can intuitively reflect the magnitude of the prediction error, while sMAPE can better assess the relative error under conditions of large price fluctuations.

[0052] In this embodiment of the invention, the single LEAR model, the single DNN model, and the hybrid model using the "weighted fusion integration" strategy (with weight ω=0.5) were evaluated on the entire test set, and the specific performance comparison results are shown in Table 1.

[0053] Table 1. Prediction errors of different models on NP and PJM test sets

[0054] As can be clearly seen from the data in Table 1, the prediction performance of the hybrid ensemble model of this invention far surpasses that of any single model. This result strongly demonstrates the "1+1>2" ensemble effect of this invention. While each single model has its limitations, the hybrid method of this invention effectively offsets their independent prediction errors by weightedly fusing the prediction results of the two models, thereby achieving a significant improvement in overall prediction accuracy.

[0055] Please see Figure 6 and Figure 7 In addition to lower overall error, the hybrid model also shows stronger robustness in dealing with sharp fluctuations in electricity prices and capturing price spikes. Figure 6 and Figure 7 The results show a comparison between the predicted electricity prices for a selected day and a consecutive week in the test set, respectively.

[0056] from Figure 6 The daily comparison shows that during periods of electricity price peaks and troughs, the predictions of the single model (blue and orange lines) often deviate significantly from the actual values ​​(black line); while the prediction curve of the hybrid integrated model (green) can more closely match the actual values, especially at price turning points, showing stronger dynamic tracking capabilities. Figure 7The weekly view further confirms this. In the continuous forecast over a week, the hybrid model not only generally aligns with the trend of actual electricity prices, but also exhibits significantly smaller prediction biases than the single model in several areas of sharp price fluctuations. This demonstrates that the method of this invention reduces reliance on the stability of a single model through model complementarity. When changes in market conditions cause one model to underperform, the other model can play a corrective and stabilizing role, thereby making the entire forecasting system more robust, reliable, and with stronger generalization capabilities. In summary, the hybrid integrated electricity price forecasting method based on LEAR and DNN proposed in this invention achieves a dual improvement in prediction accuracy and robustness by organically combining the stability of statistical models with the nonlinearity capture capabilities of deep learning models, providing more reliable decision support for electricity market participants.

[0057] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0058] Please see Figure 8 In this embodiment of the invention, a hybrid integrated electricity price forecasting system is provided, comprising: The data acquisition unit is used to acquire the feature set of the day to be predicted. The joint prediction unit is used to jointly predict the electricity price of the day to be predicted based on the feature set of the day to be predicted, using 24 pre-trained statistical prediction models and 1 deep neural network model. Among the 24 pre-trained statistical prediction models, each statistical prediction model is used to predict the electricity price for each hour of the day to be predicted. In addition, during the joint prediction process, a feature input integration path or a weighted fusion integration path is adopted.

[0059] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to execute the operation of a hybrid integrated electricity price prediction method.

[0060] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the hybrid integrated electricity price prediction method in the above embodiments.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0065] 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. A hybrid integrated electricity price forecasting method, characterized in that, Includes the following steps: Obtain the feature set of the day to be predicted; Based on the feature set of the day to be predicted, 24 pre-trained statistical prediction models and 1 deep neural network model are used to jointly predict the electricity price of the day to be predicted. Among the 24 pre-trained statistical prediction models, each statistical prediction model is used to predict the electricity price of each hour of the day to be predicted. In addition, in the process of joint prediction, a feature input integration path or a weighted fusion integration path is adopted.

2. The hybrid integrated electricity price prediction method according to claim 1, characterized in that, In the process of adopting the feature input integration path, firstly, based on the feature set of the day to be predicted, 24 statistical prediction models are used to make a preliminary prediction to obtain the 24-hour electricity price prediction vector of the day to be predicted; then, the feature set of the day to be predicted is concatenated with the 24-hour electricity price prediction vector of the day to be predicted to form an enhanced feature set; finally, the enhanced feature set is used as input, and a trained deep neural network model is used to make a final prediction to obtain the predicted electricity price value of the day to be predicted.

3. The hybrid integrated electricity price prediction method according to claim 1, characterized in that, In the process of adopting the weighted fusion integration path, the feature set of the day to be predicted is first input into 24 pre-trained statistical prediction models and 1 deep neural network model for parallel prediction, so as to obtain the first 24-hour prediction vector output by the 24 statistical prediction models and the second 24-hour prediction vector output by the deep neural network model; the first 24-hour prediction vector and the second 24-hour prediction vector are weighted to obtain the electricity price prediction value of the day to be predicted.

4. The hybrid integrated electricity price prediction method according to claim 3, characterized in that, In the step of weighting the first 24-hour forecast vector and the second 24-hour forecast vector, one or more of the following methods are used: fixed weighting, dynamic weighting based on historical forecast errors, hourly independent weighting, adaptive weighting based on forecast confidence, conditional weighting based on market conditions, recursive update weighting, seasonal periodic weighting, segmented weighting based on temperature ranges, hierarchical weighting based on load levels, and differentiated weighting based on holiday types.

5. The hybrid integrated electricity price prediction method according to claim 1, characterized in that, The steps to obtain the feature set of the day to be predicted include: Based on the selected date to be predicted and the time span, raw data is obtained; wherein, the raw data includes historical electricity price series, load data, renewable energy output data, fuel price data and meteorological data; Based on the selected date to be predicted and the acquired raw data, time features, historical price lag features, and exogenous variable lag features are constructed and concatenated in a predefined order to form a feature vector. The resulting feature vectors are processed to obtain the final feature set; the data processing includes detecting and handling missing values, and identifying and correcting outliers.

6. The hybrid integrated electricity price prediction method according to claim 5, characterized in that, In the steps of constructing time features, historical price lag features, and exogenous variable lag features, Based on the timestamp information of the day to be predicted, generate dummy variables for the week, month, holidays, and hours as time features; Extract the corresponding hourly electricity prices for the 1st, 2nd, 3rd, 7th, and 14th days prior to the predicted date to serve as historical price lag characteristics; For each exogenous variable, extract its corresponding hourly values ​​for the past 1, 2, and 7 days to construct and obtain the lagged features of the exogenous variable.

7. The hybrid integrated electricity price prediction method according to claim 1, characterized in that, Of the 24 statistical forecasting models, for hours The predicted electricity price from the statistical forecasting model is expressed as: ； In the formula, For hours h The predicted electricity price; For the intercept term; Input features eigencoefficients, For the first i Each input feature n For feature dimension, This is the error term; During the training of the statistical prediction model, L1 regularization is used for feature selection and coefficient estimation, and the objective function is... The expression is: ； In the formula, This is the scaling factor; The number of samples; For the first The actual target variable value for each observation; For the first The feature vector of each observation value For parameter vectors; For the first The predicted value of each observation is the dot product of the feature vector and the parameter vector; For regularization parameters; The L1 regularization term is a parameter vector. Each element The sum of the absolute values; n parameter vector Dimensions.

8. A hybrid integrated electricity price forecasting system, characterized in that, include: The data acquisition unit is used to acquire the feature set of the day to be predicted. The joint prediction unit is used to jointly predict the electricity price of the day to be predicted based on the feature set of the day to be predicted, using 24 pre-trained statistical prediction models and 1 deep neural network model. Among the 24 pre-trained statistical prediction models, each statistical prediction model is used to predict the electricity price for each hour of the day to be predicted. In addition, during the joint prediction process, a feature input integration path or a weighted fusion integration path is adopted.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the hybrid integrated electricity price forecasting method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the hybrid integrated electricity price prediction method as described in any one of claims 1 to 7.

Images