Smart community load prediction method and system based on ARIMA-BPNN
By leveraging the synergistic fusion mechanism of ARIMA and BPNN, we have achieved decomposition modeling and accurate prediction of the linear trend and nonlinear fluctuations of smart community loads, solving the problem of inaccurate prediction in existing technologies and improving prediction accuracy and robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-10
AI Technical Summary
In smart community load forecasting, existing technologies struggle to accurately fit both linear trends and complex nonlinear fluctuations simultaneously using a single ARIMA or BPNN model, resulting in inaccurate predictions and insufficient robustness.
A collaborative fusion mechanism of ARIMA and BPNN is adopted. The linear prediction component and residual sequence are extracted by the ARIMA model, and the nonlinear prediction component is learned by the BPNN model. The components are then weighted and fused to generate a comprehensive prediction value.
It significantly improves prediction accuracy and model robustness, enabling more accurate prediction of load changes in smart communities and providing reliable energy management decision support.
Smart Images
Figure CN121836015A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of load forecasting technology, and in particular to a smart community load forecasting method and system based on ARIMA-BPNN. Background Technology
[0002] With the deepening of new urbanization and smart city construction, smart communities, as important units of urban energy consumption, exhibit characteristics such as rapid growth in electricity load, complex composition, strong volatility, and significant influence from multiple coupled factors. Achieving accurate short-term load forecasting is a key prerequisite for optimizing the operation of community microgrids, demand-side response, efficient consumption of distributed energy resources, and reducing energy costs. It is also of great significance for improving grid security and stability and the overall efficiency of energy utilization.
[0003] Currently, load forecasting methods are mainly divided into two categories: classical statistical methods and artificial intelligence methods. Classical time series methods, represented by the Autoregressive Integral Moving Average (ARIMA) model, can effectively capture the linear trend and seasonality of load data, but their fitting ability for complex nonlinear characteristics and sudden fluctuations is limited, and the prediction accuracy is difficult to meet the requirements in smart community scenarios with significant nonlinearity. On the other hand, artificial intelligence methods, represented by the Backpropagation Neural Network (BPNN), have powerful nonlinear mapping and learning capabilities, but their training process is prone to getting trapped in local optima, and they do not fully explore the linear patterns in historical data, resulting in poor model stability. Existing technologies often use a single ARIMA or BPNN model for forecasting, failing to effectively leverage the advantages of both. This leads to problems such as inaccurate fitting of the linear part or incomplete capture of nonlinear residuals when facing the combined linear trend and complex nonlinear fluctuation characteristics of smart community loads. Ultimately, the accuracy and robustness of the prediction results need to be improved. Summary of the Invention
[0004] The purpose of this invention is to provide a smart community load forecasting method and system based on ARIMA-BPNN, which aims to solve the problem that traditional community load forecasting methods have problems such as inaccurate linear fitting or incomplete capture of nonlinear residuals when facing the linear trend and complex nonlinear fluctuation characteristics of smart community loads.
[0005] In a first aspect, the present invention provides a smart community load forecasting method based on ARIMA-BPNN, the method comprising: Acquire historical load time series data of the smart community, and preprocess the historical load time series data; Preliminary predictions are made on the pre-trained ARIMA model based on the pre-processed historical load time series data to obtain linear prediction components and residual sequences. A BPNN neural network model is constructed, and the residual sequence is predicted based on the BPNN neural network model to obtain the nonlinear prediction component; The linear prediction component and the nonlinear prediction component are fused to obtain the final prediction value.
[0006] In some embodiments, the step of acquiring historical load time series data of the smart community and preprocessing the historical load time series data includes: Outliers in the historical load time series are detected, and the detected outliers are processed using the mean comparison method. The historical load time series is normalized to map it to the interval [0, 1]. Missing points in the historical load time series are identified, and data is filled in using linear interpolation.
[0007] In some embodiments, the step of performing preliminary predictions on preprocessed historical load time series data based on a pre-trained ARIMA model to obtain linear prediction components and residual sequences includes: The preprocessed load sequence is input into the ARIMA model to perform forward prediction calculations and output the linear prediction components corresponding to future time periods. The residual sequence is obtained by subtracting the preprocessed historical load time series data from the linear prediction component.
[0008] In some embodiments, the step of constructing a BPNN neural network model and predicting the residual sequence based on the BPNN neural network model to obtain nonlinear prediction components includes: The residual sequence is divided into training and test sets. The training set is used as input, and the Adam optimizer is used to minimize the MSE loss function. The weight matrix and bias vector are iteratively updated to obtain the pre-trained BPNN neural network model. The test set is input into a pre-trained BPNN neural network model to obtain nonlinear prediction components.
[0009] In some embodiments, the step of constructing the pre-trained ARIMA model includes: The ARIMA model is fitted using the maximum likelihood estimation method, and the model coefficients are output. Input the preprocessed historical load time series data and model order parameters, and perform stationarity detection on the historical load time series data. If it is not stationary, perform differencing until the stationarity requirement is met. The stationary sequence is predicted for future time periods based on the trained ARIMA model, and the linear prediction component and residual sequence are output.
[0010] In some embodiments, the step of fusing the linear prediction component and the nonlinear prediction component to obtain the final prediction value includes: The fusion is performed according to the following formula: in, This is the final predicted value. To integrate weights, Let be the linear prediction component at time t. Let be the nonlinear prediction component at time t.
[0011] In some embodiments, the step of identifying missing points in the historical load time series and performing data imputation using linear interpolation includes: Using timestamps as indices, data segments with more than 3 consecutive missing timestamps are marked as invalid. The remaining missing timestamps are filled by linear interpolation of adjacent valid data, and the filled sequence is output. The steps of detecting outliers in the historical load time series and processing the detected outliers using the mean comparison method include: Based on the 3σ principle, outliers in the filled sequence that deviate from the mean by more than 3 times the standard deviation are identified, and replaced with the historical mean of the same period, thus outputting the corrected sequence.
[0012] Secondly, the present invention provides a smart community load forecasting system based on ARIMA-BPNN, the system comprising: The preprocessing module is used to acquire historical load time series data of the smart community and preprocess the historical load time series data. The first prediction module is used to make preliminary predictions on the pre-processed historical load time series data based on the pre-trained ARIMA model, and obtain linear prediction components and residual sequences. The second prediction module is used to construct a BPNN neural network model and predict the residual sequence based on the BPNN neural network model to obtain nonlinear prediction components. The fusion module is used to fuse the linear prediction component and the nonlinear prediction component to obtain the final prediction value.
[0013] Thirdly, the present invention provides a storage medium that stores one or more programs that, when executed by a processor, implement the above-described smart community load forecasting method based on ARIMA-BPNN.
[0014] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the above-mentioned smart community load forecasting method based on ARIMA-BPNN.
[0015] Compared with the prior art, the present invention has the following advantages: This invention utilizes a collaborative fusion mechanism of ARIMA and BPNN neural networks to achieve decomposition modeling and accurate prediction of linear trends and nonlinear residuals in load sequences of smart communities. The ARIMA model effectively extracts the deterministic linear patterns of the load, while the BPNN neural network specifically learns the complex nonlinear fluctuations that the linear model fails to explain. Finally, the two prediction results are fused through adaptive weighting. This method significantly improves prediction accuracy and model robustness, effectively overcoming the limitations of a single model in handling mixed characteristic loads, and providing more reliable decision support for energy management in smart communities. Attached Figure Description
[0016] Figure 1 This is a flowchart of a smart community load forecasting method based on ARIMA-BPNN proposed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a smart community load forecasting system based on ARIMA-BPNN proposed in an embodiment of the present invention.
[0017] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.
[0019] like Figure 1 As shown, an embodiment of the present invention proposes a smart community load forecasting method based on ARIMA-BPNN, which includes steps S101 to S104, wherein: Step S101: Obtain historical load time series data of the smart community and preprocess the historical load time series data; It should be noted that the preprocessing process includes detecting outliers in the historical load time series and processing the detected outliers using the mean comparison method; normalizing the historical load time series to map it to the [0,1] interval; identifying missing points in the historical load time series and imputing the data using linear interpolation. By preprocessing the historical load time series, high-quality and standardized input data can be provided for subsequent analysis.
[0020] Specifically, in some embodiments, data segments with more than three consecutive missing time points are first marked as invalid using timestamps as indexes. The remaining missing points are filled using linear interpolation of adjacent valid data, resulting in a filled sequence. Then, based on the 3σ principle, outliers in the filled sequence that deviate from the mean by more than three standard deviations are identified and replaced with the historical mean for the same period, resulting in a corrected sequence. Finally, the corrected sequence is normalized. This provides high-quality, standardized input for subsequent models. Outlier handling eliminates noise interference introduced by equipment failure or acquisition errors; normalization eliminates dimensional differences between different features, accelerating model convergence and improving numerical stability; and missing point filling ensures the continuity and integrity of the time series, preventing model training failure or prediction bias due to data loss, which is a prerequisite for ensuring reliable model operation and effective data utilization.
[0021] Step S102: Perform preliminary predictions on the pre-processed historical load time series data based on the pre-trained ARIMA model to obtain linear prediction components and residual sequences; It should be noted that in this step, the preprocessed load sequence is input into the ARIMA model, forward prediction calculation is performed, and the linear prediction component corresponding to the future time period is output; the preprocessed historical load time series data is subtracted from the linear prediction component to obtain the residual sequence.
[0022] The linear forecast component representing the basic load trend is obtained through forward forecasting, and then the residual sequence containing complex fluctuation information is extracted through interpolation. This step achieves the initial decomposition of load information, providing a clear learning objective for the subsequent BPNN neural network that focuses on nonlinear patterns.
[0023] In constructing a pre-trained ARIMA model, the ARIMA model needs to be fitted using the maximum likelihood estimation method to output the model coefficients. Then, the pre-processed historical load time series data and model order parameters are input, and the stationarity of the historical load time series data is tested. If it is not stationary, it is differentially processed until the stationarity requirement is met. Finally, the stationary sequence is predicted for future time periods based on the trained ARIMA model, and the linear prediction component and residual sequence are output.
[0024] It should be noted that parameter fitting through maximum likelihood estimation ensures the model's statistical optimality; stationarity detection and differencing are prerequisites for the correct application of the ARIMA model, as only stationary sequences can satisfy its modeling assumptions; training and prediction based on these assumptions can output reliable linear trend components and effective residual sequences, providing a credible linear analysis basis for the entire mixed prediction process.
[0025] Step S103: Construct a BPNN neural network model and predict the residual sequence based on the BPNN neural network model to obtain the nonlinear prediction component; It should be noted that by dividing the residual sequence into training and test sets, and using the training set as input, the Adam optimizer is used to minimize the MSE loss function, and the weight matrix and bias vector are iteratively updated to obtain the pre-trained BPNN neural network model; finally, the test set is input into the pre-trained BPNN neural network model to obtain the nonlinear prediction component.
[0026] In summary, dividing the training and test sets ensures the evaluation of the model's generalization ability; employing the Adam optimizer and MSE loss function enables efficient and stable network parameter learning, accelerating convergence and avoiding getting trapped in undesirable local optima; finally, the trained network is used to predict the test set, obtaining a nonlinear prediction component specifically for linear residuals, thus completing the accurate modeling of the complex fluctuations in the load sequence.
[0027] Step S104: Fuse the linear prediction component and the nonlinear prediction component to obtain the final prediction value.
[0028] It should be noted that the fusion is performed according to the following formula: in, This is the final predicted value. To integrate weights, Let be the linear prediction component at time t. Let be the nonlinear prediction component at time t.
[0029] In summary, based on the above, the linear trend prediction component output by the ARIMA model and the nonlinear fluctuation prediction component output by the BPNN neural network are weighted and fused to generate a comprehensive load forecast value that includes both macro trends and reflects micro fluctuations, significantly improving the completeness and accuracy of the prediction results.
[0030] In summary, this invention achieves decomposition modeling and accurate prediction of linear trends and nonlinear residuals in smart community load sequences through a collaborative fusion mechanism of ARIMA and BPNN neural networks. The ARIMA model effectively extracts the deterministic linear patterns of the load, while the BPNN neural network specifically learns the complex nonlinear fluctuations that the linear model fails to explain. Finally, the two prediction results are fused through adaptive weighting. This method significantly improves prediction accuracy and model robustness, effectively overcoming the limitations of a single model in handling mixed characteristic loads, and providing more reliable decision support for smart community energy management.
[0031] like Figure 2 As shown, one embodiment of the present invention proposes a smart community load forecasting system based on ARIMA-BPNN, the system comprising: The preprocessing module 10 is used to acquire historical load time series data of the smart community and preprocess the historical load time series data. The first prediction module 20 is used to make preliminary predictions on the preprocessed historical load time series data based on the pre-trained ARIMA model, and obtain linear prediction components and residual sequences. The second prediction module 30 is used to construct a BPNN neural network model and predict the residual sequence based on the BPNN neural network model to obtain nonlinear prediction components. The fusion module 40 is used to fuse the linear prediction component and the nonlinear prediction component to obtain the final prediction value.
[0032] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described ARIMA-BPNN-based smart community load forecasting method.
[0033] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the above-mentioned smart community load forecasting method based on ARIMA-BPNN.
[0034] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0035] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0036] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0037] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.
Claims
1. A smart community load forecasting method based on ARIMA-BPNN, characterized in that, The method includes: Acquire historical load time series data of the smart community, and preprocess the historical load time series data; Preliminary predictions are made on the pre-trained ARIMA model based on the pre-processed historical load time series data to obtain linear prediction components and residual sequences. A BPNN neural network model is constructed, and the residual sequence is predicted based on the BPNN neural network model to obtain the nonlinear prediction component; The linear prediction component and the nonlinear prediction component are fused to obtain the final prediction value.
2. The smart community load forecasting method based on ARIMA-BPNN according to claim 1, characterized in that, The steps of acquiring historical load time series data of the smart community and preprocessing the historical load time series data include: Outliers in the historical load time series are detected, and the detected outliers are processed using the mean comparison method. The historical load time series is normalized to map it to the [0,1] interval; Missing points in the historical load time series are identified, and data is filled in using linear interpolation.
3. The smart community load forecasting method based on ARIMA-BPNN according to claim 2, characterized in that, The step of performing preliminary predictions on preprocessed historical load time series data based on a pre-trained ARIMA model to obtain linear prediction components and residual sequences includes: The preprocessed load sequence is input into the ARIMA model to perform forward prediction calculations and output the linear prediction components corresponding to future time periods. The residual sequence is obtained by subtracting the preprocessed historical load time series data from the linear prediction component.
4. The smart community load forecasting method based on ARIMA-BPNN according to claim 1, characterized in that, The steps of constructing a BPNN neural network model and predicting the residual sequence based on the BPNN neural network model to obtain the nonlinear prediction component include: The residual sequence is divided into training and test sets. The training set is used as input, and the Adam optimizer is used to minimize the MSE loss function. The weight matrix and bias vector are iteratively updated to obtain the pre-trained BPNN neural network model. The test set is input into a pre-trained BPNN neural network model to obtain nonlinear prediction components.
5. The smart community load forecasting method based on ARIMA-BPNN according to claim 1, characterized in that, The steps for constructing the pre-trained ARIMA model include: The ARIMA model is fitted using the maximum likelihood estimation method, and the model coefficients are output. Input the preprocessed historical load time series data and model order parameters, and perform stationarity detection on the historical load time series data. If it is not stationary, perform differencing until the stationarity requirement is met to obtain a stationary series. The stationary sequence is predicted for future time periods based on the trained ARIMA model, and the linear prediction component and residual sequence are output.
6. The smart community load forecasting method based on ARIMA-BPNN according to claim 1, characterized in that, The step of fusing the linear prediction component and the nonlinear prediction component to obtain the final prediction value includes: The fusion is performed according to the following formula: in, This is the final predicted value. To integrate weights, Let be the linear prediction component at time t. Let be the nonlinear prediction component at time t.
7. The smart community load forecasting method based on ARIMA-BPNN according to claim 2, characterized in that, The step of identifying missing points in the historical load time series and performing data imputation using linear interpolation includes: Using timestamps as indices, data segments with more than 3 consecutive missing timestamps are marked as invalid. The remaining missing timestamps are filled by linear interpolation of adjacent valid data, and the filled sequence is output. The steps of detecting outliers in the historical load time series and processing the detected outliers using the mean comparison method include: Based on the 3σ principle, outliers in the filled sequence that deviate from the mean by more than 3 times the standard deviation are identified, and replaced with the historical mean of the same period, thus outputting the corrected sequence.
8. A smart community load forecasting system based on ARIMA-BPNN, characterized in that, The system includes: The preprocessing module is used to acquire historical load time series data of the smart community and preprocess the historical load time series data. The first prediction module is used to make preliminary predictions on the pre-processed historical load time series data based on the pre-trained ARIMA model, and obtain linear prediction components and residual sequences. The second prediction module is used to construct a BPNN neural network model and predict the residual sequence based on the BPNN neural network model to obtain nonlinear prediction components. The fusion module is used to fuse the linear prediction component and the nonlinear prediction component to obtain the final prediction value.
9. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the smart community load forecasting method based on ARIMA-BPNN as described in any one of claims 1-7.
10. An electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the smart community load forecasting method based on ARIMA-BPNN as described in any one of claims 1-7.