Electric quantity economy bidirectional prediction method and prediction system based on deep belief network
Through the two-way prediction method of electricity economy based on deep confidence network, and combining multiple information to build models and iteratively optimize, the shortcomings of traditional electricity prediction are solved, accurate prediction of electricity and economy are achieved, and power scheduling and the formulation of economic policies are supported.
Patent Information
- Application Number
- CN202510564024.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-15
AI Technical Summary
Traditional power forecasting methods are difficult to accurately capture the complexity and dynamics of power demand, affecting the stability of power supply and economic development.
A two-way prediction method of electricity economy based on deep confidence network is adopted, combining economic factors, meteorological conditions and historical power data, a deep confidence network model is built to conduct two-way predictions of electricity and economy, and to improve prediction accuracy through iterative optimization and correction.
It has achieved accurate prediction of future electricity demand, promoted the optimal allocation of power resources and the sustainable economic development, and provided strong support for policy formulation.
Smart Images
Figure CN120494170A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power analysis, and in particular to a method and system for bidirectional prediction of electricity economy based on a deep belief network. Background Art
[0002] With rapid economic development and growing energy demand, the stability and efficiency of power supply have become a major concern for society. Traditional power forecasting methods often struggle to accurately capture the complexity and dynamic nature of power demand. Consequently, a new technology, power-economy bidirectional forecasting technology based on deep belief networks (DBNs), has emerged. This technology aims to achieve accurate forecasts of power demand through a deep belief network algorithm, providing strong support for power dispatch, energy management, and economic policymaking.
[0003] There is also a close connection between electricity forecasting and economic forecasting. On the one hand, economic development conditions directly affect electricity demand. For example, electricity consumption in industrial production, commercial activities, and residents' lives will fluctuate with changes in the economic situation. On the other hand, the stability and cost of electricity supply also directly affect economic development. Therefore, two-way forecasting of electricity and economy will help to better understand the interaction between the two and provide strong support for policy formulation and decision-making. The specific logic is to first predict electricity development, then predict economic development, and then correct each other through the relationship between economy and electricity. Summary of the Invention
[0004] (1) Technical problems solved
[0005] To address the shortcomings of existing technologies, this paper aims to develop a method and system for bidirectional electricity and economic forecasting based on deep belief networks. This method and system can comprehensively consider various information, including economic factors, meteorological conditions, and historical electricity data, to accurately predict future electricity demand. Furthermore, based on the forecast results, this method and system can provide corresponding energy management recommendations to promote the optimal allocation of electricity resources and sustainable economic development.
[0006] (2) Technical solution
[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for bidirectional prediction of electricity economy based on a deep belief network, comprising the following steps:
[0008] Step 1: Data preprocessing
[0009] a. Data collection: Collect historical data on electricity demand and economic indicators from power system and economic statistics;
[0010] b. Data cleaning: pre-process the data, remove outliers and missing data, and ensure data quality;
[0011] c. Data standardization: Standardize power demand and economic indicators to eliminate dimensional differences and improve model training efficiency;
[0012] Step 2: Build a deep belief network model
[0013] d. Network structure: Based on the stacking of multiple layers of restricted Boltzmann machines (RBMs), a deep belief network model is constructed. Each RBM layer includes a visible layer, a hidden layer, and a fully connected structure.
[0014] e.RBM energy function: Define the energy function to represent the network state, including the activation states of the visible and hidden layers, as well as parameters such as weights and biases;
[0015] f. Model training: Model parameter optimization is performed in two stages: pre-training and fine-tuning. Pre-training uses a greedy unsupervised learning algorithm to train RBM layer by layer; fine-tuning uses a back-propagation algorithm to optimize supervised tasks.
[0016] Step 3: Electricity Economy Bidirectional Forecasting and Correction
[0017] g. Separate forecast: Use the established power demand forecast model and economic indicator forecast model to make preliminary forecasts respectively;
[0018] h. Two-way correction: correction of economic indicator forecasts based on power demand and correction of power demand forecasts based on economic indicators;
[0019] Step 4: Accuracy test and model verification
[0020] i. Error analysis: Evaluate the accuracy of the prediction results based on the mean square error (MSE) and accuracy indicators;
[0021] J. Multiple iterations: Improve prediction accuracy by iteratively optimizing model parameters and input features.
[0022] As a preferred solution, the data preprocessing in step 1 further includes:
[0023] Abnormal data processing, using statistical methods or machine learning algorithms (such as methods based on nonparametric kernel density estimation) to identify and remove outliers in power or economic data, which includes identifying outliers, evaluating outliers, and processing outliers;
[0024] Forecasting and correction: Use the constructed electricity forecasting model and economic forecasting model to make preliminary forecasts, obtain the forecast values of economy and electricity respectively, and then perform two-way correction.
[0025] As a preferred solution, the training model in step 2 specifically includes:
[0026] Pre-training phase
[0027] Layer-by-layer training: Starting from the bottom, RBM is trained layer by layer. Each layer of RBM uses the output of the previous layer as input to learn and capture features.
[0028] Greedy algorithm: Greedy unsupervised learning algorithm is used to initialize the parameters of the entire DBN model layer by layer;
[0029] Feature extraction: Through layer-by-layer training, the original features at the bottom of the deep architecture are combined into more compact high-level features, which will serve as input for subsequent classification or regression tasks;
[0030] Fine-tuning stage
[0031] Add classification layer: Add one or more classification layers (such as softmax layer) to the stacked RBMs to output classification recognition results;
[0032] Supervised training: A global learning algorithm (such as BP algorithm or wake-sleep algorithm) is used to supervise the entire network to optimize the performance of a specific task (such as classification or regression);
[0033] Backpropagation: During the fine-tuning phase, the weights of all layers are adjusted through the backpropagation algorithm to minimize the difference between the predicted and true labels.
[0034] As a preferred solution, the specific steps of step 4 precision test and model verification are as follows: first, select indicators such as mean square error and accuracy to evaluate the prediction and correction results; based on the evaluation results, adjust the model parameters, input features or correction methods to improve the accuracy of prediction and correction; compare the actual electricity and economic data with the predicted values; analyze the source of error; the mean square error (MSE) is used to evaluate the accuracy and error size of the prediction model, where the calculation formula of MSE is:
[0035]
[0036] in: It represents the true value; Represents the predicted value; m is the number of samples; Σ represents the summation operation. MSE measures the average error of the model prediction. The smaller the value, the more accurate the prediction result.
[0037] As a preferred solution, the accuracy test and model validation also include an accuracy rate, which is used to evaluate the performance of the classification model. It represents the proportion of correctly classified samples among all samples. The accuracy rate is calculated as follows:
[0038] Accuracy = Number of correctly classified samples / Total number of samples × 100% Accuracy = \frac{Number of correctly classified samples}{Total number of samples}\times 100% Accuracy = Number of correctly classified samples / Total number of samples × 100%.
[0039] A deep belief network-based two-way electricity and economic forecasting system accurately predicts future electricity demand based on economic factors, meteorological conditions, and historical electricity data. It includes an electricity subsystem and an economic subsystem. The electricity subsystem includes modules for total social electricity consumption, electricity consumption by industry, electricity consumption by key industries, new capacity, and business expansion registration.
[0040] The economic subsystem includes a per capita GDP module, an industry value-added module, and a total retail sales module of consumer goods.
[0041] As a preferred solution, it also includes:
[0042] Forward prediction port to predict future power demand and economic trends;
[0043] Reverse prediction port, which reversely estimates power demand based on economic indicators, or predicts economic changes based on power demand;
[0044] The prediction system is integrated with a real-time data processing unit for processing real-time data streams, a correction unit for correcting the power prediction model, and a visualization unit for displaying the prediction results through charts.
[0045] As a preferred solution, the prediction system is based on the RBM neural network, whose goal is to learn a probability distribution to generate data. The specific process is as follows:
[0046] A1. Determine the boundaries of the electricity economy system;
[0047] A2. Electricity and economic forecasting methods based on deep belief networks;
[0048] A3. Based on the predicted electricity and economic data, a two-way prediction and correction method is proposed;
[0049] A4. Accuracy test and model verification.
[0050] (3) Beneficial effects
[0051] Compared with the existing technology, the present invention provides a method and system for bidirectional prediction of electricity economy based on deep belief network, which has the following beneficial effects:
[0052] The research scope of this invention includes the construction and optimization of deep belief network models, the design and implementation of power forecasting algorithms, system testing and verification, etc. It is mainly used to analyze power problems and comprehensively consider future power demand under the influence of multiple factors. The power-economy bidirectional forecasting technology based on deep belief networks (DBNs) is applied, replacing traditional power forecasting methods and making power supply more accurate and efficient. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 Schematic diagram of the process steps of the present invention;
[0054] Figure 2 It is a schematic diagram of the system module of the present invention;
[0055] Figure 3 Schematic diagram of the deep belief network structure of the present invention. DETAILED DESCRIPTION
[0056] In order to better understand the purpose, structure and function of the present invention, the following will further illustrate the electricity economy bidirectional prediction method and prediction system based on deep belief network of the present invention in combination with the drawings and specific embodiments.
[0057] Example 1
[0058] This embodiment involves a method for bidirectional electricity and economic forecasting based on a deep belief network. The research scope includes the construction and optimization of the deep belief network model, the design and implementation of the electricity forecasting algorithm, and system testing and verification.
[0059] refer to Figure 1 The present invention provides a method for bidirectional prediction of electricity economy based on a deep belief network, comprising the following steps:
[0060] Step 1: Data preprocessing
[0061] a. Data collection: Collect historical data on electricity demand and economic indicators from power system and economic statistics;
[0062] b. Data cleaning: pre-process the data, remove outliers and missing data, and ensure data quality;
[0063] c. Data standardization: Standardize power demand and economic indicators to eliminate dimensional differences and improve model training efficiency;
[0064] Step 2: Build a deep belief network model
[0065] d. Network structure: Based on the stacking of multiple layers of restricted Boltzmann machines (RBMs), a deep belief network model is constructed. Each RBM layer includes a visible layer, a hidden layer, and a fully connected structure.
[0066] e.RBM energy function: Define the energy function to represent the network state, including the activation states of the visible and hidden layers, as well as parameters such as weights and biases;
[0067] f. Model training: Model parameter optimization is performed in two stages: pre-training and fine-tuning. Pre-training uses a greedy unsupervised learning algorithm to train RBM layer by layer; fine-tuning uses a back-propagation algorithm to optimize supervised tasks.
[0068] Step 3: Electricity Economy Bidirectional Forecasting and Correction
[0069] g. Separate forecast: Use the established power demand forecast model and economic indicator forecast model to make preliminary forecasts respectively;
[0070] h. Two-way correction: correction of economic indicator forecasts based on power demand and correction of power demand forecasts based on economic indicators;
[0071] Step 4: Accuracy test and model verification
[0072] i. Error analysis: Evaluate the accuracy of the prediction results based on the mean square error (MSE) and accuracy indicators;
[0073] J. Multiple iterations: Improve prediction accuracy by iteratively optimizing model parameters and input features.
[0074] This embodiment is based on the application of bidirectional electricity and economic forecasting technology using a deep belief network (DBN). It replaces traditional electricity forecasting methods and makes electricity supply more accurate and efficient. There is also a close connection between electricity forecasting and economic forecasting. On the one hand, economic development directly affects electricity demand. For example, electricity consumption in various aspects such as industrial production, commercial activities, and residents' lives will fluctuate with changes in the economic situation. On the other hand, the stability and cost of electricity supply also directly affect economic development. Therefore, bidirectional forecasting of electricity and the economy helps to better understand the interaction between the two and provide strong support for policy formulation and decision-making. The specific logic is to first predict electricity development, then predict economic development, and then correct each other through the relationship between economy and electricity. In its deep belief network (DBN), the DBN is composed of multiple stacked RBMs. Each RBM layer learns the distribution of the input data and attempts to capture the features in the data. Through stacking, each layer further abstracts the features captured by the previous layer, allowing the network to learn more complex data representations.
[0075] Specifically, refer to Figure 3As shown in the figure, the construction of the deep belief network model, its DBN is composed of multiple layers of restricted Boltzmann machines (RBMs). Each RBM is a two-layer neural network, including a visible layer and a hidden layer. At the top layer of the DBN, a classification layer (such as a softmax layer) is usually added to output the classification recognition results. The visible layer nodes correspond to the input data and can be any type of data point, such as pixel values, scores, or binary features. The hidden layer nodes are designed to capture the characteristics or patterns of the visible layer data, and their activation state can be understood as a high-level representation of the input data. The neurons within the layer are not connected to each other, and the neurons between layers are fully connected. The key feature of this structure is restriction (i.e., "restricted"), which reduces the complexity of the model and makes training feasible.
[0076] Furthermore, the next step is to train the deep belief network, which includes:
[0077] 1) Pre-training stage
[0078] Layer-by-layer training: Starting from the bottom, RBM is trained layer by layer. Each layer of RBM uses the output of the previous layer as input to learn to capture features.
[0079] Greedy Algorithm: A greedy unsupervised learning algorithm is used to initialize the parameters of the entire DBN model layer by layer. The purpose of this step is to effectively narrow the space for parameter optimization through unsupervised learning, laying the foundation for subsequent supervised training.
[0080] Feature extraction: Through layer-by-layer training, the original features at the bottom of the deep architecture are combined into more compact high-level features, which will serve as input for subsequent classification or regression tasks.
[0081] 2) Fine-tuning stage
[0082] Add classification layer: Add one or more classification layers (such as softmax layer) to the stacked RBMs to output classification recognition results.
[0083] Supervised training: A global learning algorithm (such as the BP algorithm or the wake-sleep algorithm) is used to perform supervised training on the entire network to optimize specific task performance (such as classification or regression).
[0084] Backpropagation: During the fine-tuning phase, the weights of all layers are adjusted through the backpropagation algorithm to minimize the difference between the predicted and true labels. The purpose of this step is to further optimize the network parameters and improve the prediction accuracy of the model.
[0085] More specifically, in this embodiment
[0086] (1) Energy function of RBM:
[0087] RBMs define the state of the system using an energy function, which is a function of weights, biases, and node states. The energy function is typically expressed as: E(v,h) = -aTv - bTh - vTWh, where v is the visible layer state, h is the hidden layer state, a and b are the biases for the visible and hidden layers, respectively, and W is the weight matrix between the visible and hidden layers. This energy function can be used to extract high-order features layer by layer, thereby achieving data dimensionality reduction and feature extraction.
[0088] (2) Probability distribution of RBM:
[0089] Using an energy function, RBMs can calculate the probability of the hidden layer state given the visible layer state, and vice versa. This calculation involves summing over all possible configurations, which is approximated in practice through sampling methods such as Gibbs sampling. This property enables RBMs to simulate the probability distribution of data, supporting the generative modeling properties of DBNs. In a DBN, each layer of RBMs can be viewed as a generative model that generates the inputs for the next layer of RBMs. This generative modeling property enables DBNs to capture the hierarchical structure of data, thereby achieving accurate predictions of complex data.
[0090] (3) Contrastive Divergence Algorithm:
[0091] RBM is trained using the Contrastive Divergence (CD) algorithm, which continuously adjusts weights to reduce the difference between the network's reconstructed input data and the original data, thereby learning the probability distribution of the data.
[0092] Example 2
[0093] This example aims to develop a bidirectional electricity and economic forecasting system based on a deep belief network. This system can comprehensively consider various information, including economic factors, meteorological conditions, and historical electricity data, to accurately predict future electricity demand. Furthermore, based on the forecast results, the system can provide corresponding energy management recommendations to promote the optimal allocation of electricity resources and sustainable economic development.
[0094] refer to Figure 2 The present invention provides a deep belief network-based two-way electricity and economic forecasting system. It accurately predicts future electricity demand based on economic factors, meteorological conditions, and historical electricity data. The system includes an electricity subsystem and an economic subsystem. The electricity subsystem includes a module for electricity consumption in the entire society, a module for electricity consumption in various industries, a module for electricity consumption in key industries, a module for newly added capacity, and a module for business expansion and installation.
[0095] The economic subsystem includes the per capita GDP module, the added value of various industries module and the total retail sales of consumer goods module.
[0096] Specifically, the system of this embodiment also includes:
[0097] Forward prediction port to predict future power demand and economic trends;
[0098] Reverse prediction port, which reversely estimates power demand based on economic indicators, or predicts economic changes based on power demand;
[0099] The system integrates a real-time data processing unit for processing real-time data streams, a correction unit for correcting the power forecast model, and a visualization unit for displaying the forecast results through charts.
[0100] The prediction system in this embodiment is based on the RBM neural network. Its goal is to learn a probability distribution to generate data. The specific process is as follows:
[0101] A1. Determine the boundaries of the electricity economy system;
[0102] A2. Electricity and economic forecasting methods based on deep belief networks;
[0103] A3. Based on the predicted electricity and economic data, a two-way prediction and correction method is proposed;
[0104] A4. Accuracy test and model verification.
[0105] Furthermore, based on the predicted electricity consumption and economic data, a two-way prediction and correction method is proposed, specifically including:
[0106] (1) Data preprocessing
[0107] Abnormal data processing uses statistical methods or machine learning algorithms to identify and eliminate outliers in power or economic data, which includes identifying outliers, evaluating outliers, and processing outliers. First, the first step: identifying outliers, finding "wrong" data, and using mathematical rules or intelligent algorithms to find data points that obviously deviate from the normal pattern. For example: using simple statistical rules: assuming that most data are concentrated in a certain range (such as temperature between 0-40°C), those outside the range may be marked as abnormal; using machine learning models: let the algorithm automatically learn the normal pattern of the data (such as the change pattern of electricity consumption over time), and the fluctuation points that cannot be explained by the model are considered abnormal; combined with professional knowledge: data that clearly violates physical laws or common sense, such as power in power data exceeding the equipment limit and GDP growth rate breaking through historical extremes in economic data, are directly judged as abnormal;
[0108] When evaluating outliers and determining whether they should be handled, not all outliers need to be deleted. It's necessary to verify whether they are truly errors or have research value. This includes: Data verification: checking whether the outlier conforms to the data distribution pattern (for example, a person aged 200 in the age data of 100 people is clearly an error); Time / space verification: if a region in the power data suddenly drops to zero load, retain it if it's a planned power outage; address it if it's a sensor failure; Business verification: if a surge in export volume in a certain month in economic data is found, confirm whether it's a statistical error or a real event (such as policy stimulus); Impact assessment: after deleting or retaining an outlier, calculate the impact on the overall analysis results (for example, whether deleting an outlier will reduce the average electricity price by 10%).
[0109] When dealing with outliers, choose "how to deal with" and adopt different strategies based on the evaluation results. Common methods include: direct deletion: when it is confirmed to be erroneous data (such as a damaged sensor causing a negative current); correction and replacement: replace with a reasonable value, such as using the average value of the two days before and after to fill in the missing electricity consumption; retain but mark: anomalies caused by special events (such as a sharp drop in economic data during the epidemic period), retain the data but analyze it separately; model adaptation: for anomalies whose causes cannot be determined, use an anti-interference algorithm (such as using the median instead of the average value for calculation).
[0110] Specifically, outliers in electricity or economic data can be identified and removed using statistical methods or machine learning algorithms (such as those based on nonparametric kernel density estimation). Missing electricity or economic data can be supplemented using interpolation, regression prediction, or filling in similar days based on historical data. Electricity and economic data can be standardized to eliminate dimensional differences and improve the stability and accuracy of model training.
[0111] (2) Prediction and correction
[0112] Use the constructed electricity forecasting model and economic forecasting model to make preliminary forecasts, obtain the forecast values of economy and electricity respectively, and then perform two-way correction.
[0113] Correction of economic data by electricity consumption:
[0114] Based on the electricity forecast results and the correlation between electricity consumption and economic development, the economic forecast model is corrected. If the electricity forecast shows a significant increase in electricity demand in a certain period of time in the future, it may mean that economic activity will also increase accordingly, so the relevant parameters or variables in the economic forecast can be adjusted.
[0115] Economic correction of electricity consumption:
[0116] The electricity forecast model can be adjusted based on the economic forecast results. If the economic forecast indicates that economic activity will increase significantly in a certain period in the future, then electricity demand may also increase accordingly. Therefore, the input features or parameters in the electricity forecast model can be adjusted.
[0117] Iterative optimization:
[0118] The corrected electricity and economic forecast values are used as new inputs, and the forecast and correction are performed again until the predetermined accuracy or number of iterations is reached.
[0119] System testing and verification includes accuracy testing and model validation. Specifically, the system selects indicators such as mean square error and accuracy to evaluate the prediction and correction results. Based on the evaluation results, the model parameters, input features, or correction methods are adjusted to improve the accuracy of the prediction and correction. The actual power and economic data are compared with the predicted values to analyze the sources of errors. The mean square error (MSE) is used to evaluate the accuracy and error size of the prediction model. The calculation formula of MSE is:
[0120]
[0121] in: It represents the true value; Represents the predicted value; m is the number of samples; Σ represents the summation operation. MSE measures the average error of the model prediction. The smaller the value, the more accurate the prediction result.
[0122] Its precision test and model validation also include accuracy to evaluate the performance of the classification model. It represents the proportion of correctly classified samples among all samples. The accuracy calculation formula is:
[0123] Accuracy = Number of correctly classified samples / Total number of samples × 100% Accuracy = \frac{Number of correctly classified samples}{Total number of samples}\times 100% Accuracy = Number of correctly classified samples / Total number of samples × 100%.
[0124] In information retrieval, the mathematical expression of accuracy can also be expressed as:
[0125] Accuracy = Number of documents retrieved correctly Number of documents retrieved Accuracy = \frac{Number of documents retrieved correctly}{Number of documents retrieved} Accuracy = Number of documents retrieved Number of documents retrieved correctly
[0126] The higher the accuracy, the better the classification performance of the model.
[0127] These two metrics play an important role in model evaluation and selection, but they focus on different aspects. MSE focuses more on measuring the error between the predicted value and the true value, while accuracy focuses more on measuring the classification model's ability to correctly classify. In practical applications, the appropriate evaluation metric should be selected based on the specific problem and data characteristics.
[0128] This invention utilizes a deep belief network-based bidirectional forecasting technology for electricity and economics. This technology incorporates a bidirectional forecasting method for electricity and economic forecasting, a nonlinear forecasting model for the deep belief network that incorporates dual indicators for electricity and economics, and a correction technique for mutual correction after the dual-system forecasting of electricity and economics. This bidirectional forecasting technology based on deep belief networks fully considers the acquisition of electricity and economic data, achieving accurate identification of system indicators. The computational framework is capable of accommodating dual systems, enabling better extraction of the inherent connections between the electricity and economic systems. The predicted results can be better verified, resolving issues related to empirical predictions based on single-system forecasts. This provides strong support for power dispatching, energy management, and economic policymaking.
[0129] It will be understood that the present invention is described by way of some embodiments, and it will be appreciated by those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are intended to be protected by the present invention.
Claims
1. A method for bidirectional prediction of electricity economy based on deep belief network, characterized by: The steps include: Step 1: Data preprocessing a. Data collection: Collect historical data on electricity demand and economic indicators from power system and economic statistics; b. Data cleaning: pre-process the data, remove outliers and missing data, and ensure data quality; c. Data standardization: Standardize power demand and economic indicators to eliminate dimensional differences and improve model training efficiency; Step 2: Build a deep belief network model d. Network structure: Based on the stacking of multiple layers of restricted Boltzmann machines (RBMs), a deep belief network model is constructed. Each RBM layer includes a visible layer, a hidden layer, and a fully connected structure. e.RBM energy function: Define the energy function to represent the network state, including the activation states of the visible and hidden layers, as well as parameters such as weights and biases; f. Model training: Model parameter optimization is performed in two stages: pre-training and fine-tuning. Pre-training uses a greedy unsupervised learning algorithm to train RBM layer by layer; fine-tuning uses a back-propagation algorithm to optimize supervised tasks. Step 3: Electricity Economy Bidirectional Forecasting and Correction g. Separate forecast: Use the established power demand forecast model and economic indicator forecast model to make preliminary forecasts respectively; h. Two-way correction: correction of economic indicator forecasts based on power demand and correction of power demand forecasts based on economic indicators; Step 4: Accuracy test and model verification i. Error analysis: Evaluate the accuracy of prediction results based on mean square error and accuracy metrics; J. Multiple iterations: Improve prediction accuracy by iteratively optimizing model parameters and input features.
2. The method for bidirectional prediction of electricity economy based on deep belief network according to claim 1 is characterized in that: The data preprocessing in step 1 further includes: Abnormal data processing, using statistical methods or machine learning algorithms to identify and remove outliers in power or economic data, including identifying outliers, evaluating outliers, and processing outliers; Forecasting and correction: Use the constructed electricity forecasting model and economic forecasting model to make preliminary forecasts, obtain the forecast values of economy and electricity respectively, and then perform two-way correction.
3. The method for bidirectional prediction of electricity economy based on deep belief network according to claim 1 is characterized in that: The training model in step 2 specifically includes: Pre-training phase Layer-by-layer training: Starting from the bottom, RBM is trained layer by layer. Each layer of RBM uses the output of the previous layer as input to learn and capture features. Greedy algorithm: Greedy unsupervised learning algorithm is used to initialize the parameters of the entire DBN model layer by layer; Feature extraction: Through layer-by-layer training, the original features at the bottom of the deep architecture are combined into more compact high-level features, which will serve as input for subsequent classification or regression tasks; Fine-tuning stage Add classification layer: Add one or more classification layers to the stacked RBMs to output classification recognition results; Supervised training: A global learning algorithm is used to supervise the training of the entire network to optimize the performance of a specific task; Backpropagation: During the fine-tuning phase, the weights of all layers are adjusted through the backpropagation algorithm to minimize the difference between the predicted and true labels.
4. The method for bidirectional prediction of electricity economy based on deep belief network according to claim 1 is characterized in that The specific steps of step 4, precision test and model verification, are as follows: first, select indicators such as mean square error and accuracy to evaluate the prediction and correction results. According to the evaluation results, adjust the model parameters, input features or correction methods to improve the accuracy of prediction and correction. Compare the actual power and economic data with the predicted values to analyze the error sources. The mean square error is used to evaluate the accuracy and error size of the prediction model. The calculation formula of MSE is: in: It represents the true value; Represents the predicted value; m is the number of samples; Σ represents the summation operation. MSE measures the average error of the model prediction. The smaller the value, the more accurate the prediction result.
5. The method for bidirectional prediction of electricity economy based on deep belief network according to claim 4 is characterized in that: The accuracy test and model validation also include accuracy, which is used to evaluate the performance of the classification model. It represents the proportion of correctly classified samples among all samples. The accuracy calculation formula is: Accuracy = Number of correctly classified samples / Total number of samples × 100% Accuracy = \frac{Number of correctly classified samples}{Total number of samples}\times 100% Accuracy = Number of correctly classified samples / Total number of samples × 100%.
6. A system for bidirectional prediction of electricity economy based on a deep belief network, used to implement the steps of the method for bidirectional prediction of electricity economy based on a deep belief network according to any one of claims 1 to 5, characterized in that: It also includes an electricity subsystem and an economic subsystem. The electricity subsystem includes a module for electricity consumption in the whole society, a module for electricity consumption in various industries, a module for electricity consumption in key industries, a module for newly added capacity, and a module for business expansion registration. The economic subsystem includes a per capita GDP module, an industry value-added module, and a total retail sales module of consumer goods.
7. The system for bidirectional prediction of electricity economy based on deep belief network according to claim 6 is characterized in that: Also included are: Forward prediction port to predict future power demand and economic trends; Reverse prediction port, which reversely estimates power demand based on economic indicators, or predicts economic changes based on power demand; The prediction system is integrated with a real-time data processing unit for processing real-time data streams, a correction unit for correcting the power prediction model, and a visualization unit for displaying the prediction results through charts.
8. The system for bidirectional prediction of electricity economy based on deep belief network according to claim 7 is characterized in that: The prediction system is based on the RBM neural network, whose goal is to learn a probability distribution to generate data. The specific process is as follows: A1. Determine the boundaries of the electricity economy system; A2. Electricity and economic forecasting methods based on deep belief networks; A3. Based on the predicted electricity and economic data, a two-way prediction and correction method is proposed; A4. Accuracy test and model verification.