Power load prediction method, system and equipment based on multi-source data fusion and deep learning, and medium
By employing a power load forecasting method that combines multi-source data fusion and deep learning, and utilizing long short-term memory networks and optimized model parameters, this approach addresses the issues of insufficient data utilization and poor adaptability in existing technologies, achieving high-precision and robust load forecasting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-25
- Publication Date
- 2026-03-10
AI Technical Summary
Existing power load forecasting technologies suffer from insufficient data utilization, inadequate model expressive power, and lack of system adaptability, making it difficult to meet the requirements of high accuracy, robustness, and adaptability, especially in terms of insufficient forecast accuracy during extreme weather and load changes.
By fusing multi-source data to obtain historical power load, power generation, grid topology, and meteorological data, prediction is performed using long short-term memory networks and deep learning models. By combining forget gates, input gates, and output gates, the mean square error is used to optimize model parameters, enabling online learning and adaptive adjustment.
It significantly improves prediction accuracy, accurately captures complex nonlinear relationships and long-term dependency features, increases prediction accuracy by 10%-20%, enhances adaptability to complex scenarios, and rapidly responds to changes in load patterns.
Smart Images
Figure CN121642902A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power load forecasting technology, and in particular to a power load forecasting method, system, device and medium based on multi-source data fusion and deep learning. Background Technology
[0002] While current power load forecasting technology plays a fundamental role in power system operation, its core methodology still has significant limitations, making it difficult to meet the requirements of high-precision power spot trading in terms of forecast accuracy, robustness, and adaptability. The shortcomings of existing technologies are mainly reflected in three aspects: the one-sidedness of data utilization, the limitations of model architecture, and the sluggishness in response to changes in the external environment.
[0003] First, in terms of data, mainstream forecasting technologies primarily rely on the single time-series characteristics constructed from historical load data itself, failing to systematically and effectively integrate multi-source heterogeneous data such as meteorological, economic, and social activities into modelable feature inputs. A strong nonlinear coupling exists between meteorological parameters and electricity load. Especially during extreme weather events, sudden increases or decreases in air conditioning and heating loads can cause load curve spikes that traditional models struggle to capture. Existing technologies either treat these external data merely as simple auxiliary parameters or completely ignore their impact, resulting in forecasting techniques failing to reflect the complex dynamic mechanisms of the real world. This leads to a sharp increase in forecasting bias when structural changes occur in the load.
[0004] Secondly, at the model level, widely used prediction models include traditional time series methods such as autoregressive integral moving average and seasonal decomposition. These are essentially based on linear assumptions, and their expressive power cannot fully characterize the complex nonlinear and non-stationary interactions between electricity load and multi-dimensional influencing factors. Although some studies have introduced shallow machine learning models, their feature abstraction capabilities are limited, and their ability to fuse high-dimensional, multi-source data is insufficient. They struggle to automatically extract deep, highly indicative, and discriminative core feature combinations from massive datasets. This results in poor generalization performance and insufficient prediction accuracy when faced with long-term dependencies, abrupt changes, and complex cyclical patterns in load sequences.
[0005] Finally, the existing technological framework suffers from poor dynamic adaptability. With the increasing penetration of distributed energy in the construction of new power systems, the significant differences between the electricity load characteristics of emerging industries and traditional loads, and the continuous evolution of residential electricity consumption habits, the overall distribution of load time series may change over time. Traditional models typically rely on periodic manual retraining and parameter adjustments, lacking online learning and adaptive adjustment mechanisms. They cannot track and adapt to the evolution of load patterns in real time, leading to a decline in model performance over time and making it difficult to provide continuous and reliable predictive support in a rapidly changing electricity market environment.
[0006] In summary, existing power load forecasting technologies, due to insufficient data utilization, inadequate model expressive power, and lack of system adaptability, are no longer able to meet the higher requirements of modern power for ultra-short-term and short-term load forecasting in terms of accuracy, robustness, and intelligence. Summary of the Invention
[0007] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a power load forecasting method based on multi-source data fusion and deep learning to solve the above problems.
[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a power load forecasting method based on multi-source data fusion and deep learning, comprising: acquiring multi-source power data and performing data preprocessing; Based on the processed multi-source power data, a power load forecasting model is constructed to perform multivariate multi-step time series forecasting and output load forecasting results. Based on the power load prediction model, the model parameters are adjusted according to preset optimization indicators to train and optimize the power load prediction model.
[0009] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the multi-source power data includes historical power load data, power generation data, power grid topology, and meteorological data.
[0010] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the data preprocessing includes: Data cleaning and processing removes outliers and eliminates abnormal data from multiple sources; Data frequency alignment: Select parameters for the standard acquisition frequency and perform acquisition frequency synchronization processing; Data normalization processing involves standardizing the multi-source power data to obtain standardized data. The beneficial effect of this preferred solution is that it provides the model with richer, cleaner, and more discriminative input information through multi-source fusion and refined preprocessing.
[0011] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the construction of the power load forecasting model includes: A power load forecasting model is constructed using a long short-term memory network. The input is the feature sequence of the multi-source power data in the past T time steps, and the output is the power load value sequence at the next H time points. The beneficial effect of this preferred solution is that it solves the bottleneck of traditional models being unable to characterize complex nonlinear relationships and lacking long-term memory by leveraging the powerful expressive capabilities of the Long Short-Term Memory Network model, and solves the problem of poor model adaptability by utilizing its online learning capability.
[0012] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the power load forecasting model constructed using a long short-term memory network includes: The power load prediction model includes a forget gate, an input gate, and an output gate; The first memory superposition state is obtained through the forget gate, the first long-term memory state and the second memory superposition state are obtained through the input gate, and the third memory superposition state is output through the output gate.
[0013] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the forget gate is represented as follows: in, Indicates time The first memory superposition state; Represents the sigmoid function; Indicates weight; Indicates time Hidden memory state; Represents standardized feature data; This represents the bias vector of the forget gate.
[0014] As a preferred embodiment of the power load forecasting method based on multi-source data fusion and deep learning described in this invention, the training and optimization of the power load forecasting model according to preset optimization indicators includes: Using mean squared error as the optimization metric, minimizing the squared difference between the predicted and actual values is expressed as: Where K represents the total number of samples in the dataset; H represents the number of feature sequences used for prediction; This represents the true value of the i-th sample on the j-th feature. This represents the predicted value of the i-th sample on the j-th feature.
[0015] Secondly, the present invention provides a power load forecasting system based on multi-source data fusion and deep learning, comprising: The data acquisition and processing module is used to acquire multi-source power data and perform data preprocessing. The module is used to build a power load forecasting model based on the processed multi-source power data, so as to perform multivariate multi-step time series forecasting and output load forecasting results; The training and optimization module is used to train and optimize the power load prediction model by adjusting the model parameters according to preset optimization indicators.
[0016] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the power load forecasting method based on multi-source data fusion and deep learning.
[0017] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the power load forecasting method based on multi-source data fusion and deep learning.
[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention combines historical load data with meteorological and other data through multi-source data fusion, providing more comprehensive information for the model. At the same time, it adopts an advanced deep learning model that combines long short-term memory networks with attention mechanisms, which can effectively capture complex nonlinear relationships and long-term dependency features in load data, significantly improving prediction accuracy. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a schematic diagram of the overall process of a power load forecasting method based on multi-source data fusion and deep learning, as described in one embodiment of the present invention. Detailed Implementation
[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0022] Example 1, referring to Figure 1 As an embodiment of the present invention, a power load forecasting method based on multi-source data fusion and deep learning is provided, comprising: S101: Acquire multi-source power data and perform data preprocessing; S102, Based on the processed multi-source power data, construct a power load forecasting model to perform multivariate multi-step time series forecasting and output load forecasting results; S103, based on the power load forecasting model, adjusts the model parameters according to the preset optimization indicators to train and optimize the power load forecasting model.
[0023] In a preferred embodiment, the multi-source power data includes historical power load data, power generation data, power grid topology, and meteorological data.
[0024] Specifically, the system acquires multi-source data information from within the power system through data sharing interfaces. This includes historical load data, generation data, and grid topology data. Historical load data includes timestamps, load values, and energy-consuming user characteristics; generation data includes timestamps, generation volume, active power, and reactive power; and grid topology data includes plant and substation information, node information, transformer information, and switchgear information. Meteorological data, including temperature, air pressure, relative humidity, vapor pressure, wind, and precipitation, is also acquired through standard interfaces.
[0025] In a preferred embodiment, data preprocessing includes: Data cleaning and processing removes outliers and eliminates abnormal data from multiple sources; Data frequency alignment: Select parameters for the standard acquisition frequency and perform acquisition frequency synchronization processing; Data normalization processing involves standardizing multi-source power data to obtain standardized data.
[0026] Specifically, multi-source data is preprocessed and time-series aligned to obtain standardized data that conforms to prediction. This includes data cleaning to remove outliers. A sliding window is constructed based on the fluctuation amplitude of the multi-source data within a set time period, and this window is used to filter the multi-source data to eliminate outliers. For inconsistent data collection frequencies, linear interpolation is used to supplement the data to match the frequency of the highest frequency power load data collection value, as shown below: in, For multi-source data information, the first Missing values filled in at any time; The value collected at time t in the multi-source data; h represents the collected value at time t+k in the multi-source data; h represents the missing values. With collected values The step size between them; k is the number of collected values. With collected values The number of missing values between.
[0027] The multi-source data is normalized and represented as follows: in, For multi-source data The normalization characteristics, In a single category The minimum value; In a single category The maximum value is used to perform time-series alignment and data grouping on the normalized characteristics of multi-source data to obtain standardized data for load forecasting.
[0028] In one alternative implementation, data preprocessing also includes data augmentation, which involves learning the probability distribution of real load sequences through a generator network and learning to distinguish between real load sequences and fake sequences generated by the generator through a discriminator network. After training, the generator can create realistic and diverse load curves, including load peaks under various extreme weather conditions and special holiday patterns.
[0029] In another optional implementation, data preprocessing also includes time series feature extraction. The time series features can be constructed as periodic features, sliding statistical features, lag features, etc. Then, through algorithms such as mutual information and recursive feature elimination, the most effective feature subset for load forecasting is automatically selected from a large number of constructed features, thereby reducing noise and computational complexity.
[0030] In a preferred embodiment, constructing a power load forecasting model includes: A power load forecasting model is constructed using a long short-term memory network. The input is the feature sequence of power multi-source data within the past T time steps (review window), and the output is the power load value sequence for the next H time points (prediction horizon).
[0031] In a preferred embodiment, constructing a power load forecasting model using a long short-term memory network includes: The power load forecasting model includes a forget gate, an input gate, and an output gate; The first memory superposition state is obtained through the forget gate, the first long-term memory state and the second memory superposition state are obtained through the input gate, and the third memory superposition state is output through the output gate.
[0032] It should be noted that this embodiment is mainly based on multivariate multi-step time series prediction, and uses a Long Short-Term Memory (LSTM) network model to complete the power load prediction. LSTM is a variant of recurrent neural network in deep learning, used to solve the gradient vanishing and gradient explosion problems in long sequence data modeling.
[0033] Specifically, the forgetting gate allows users to choose to forget certain information from the past. The forget section of the control gate is represented as follows: in, Indicates time The first memory superposition state; Represents the sigmoid function; Indicates weight; Indicates time Hidden memory state; Represents standardized feature data; This represents the bias vector of the forget gate.
[0034] Furthermore, features are input into the input gate to obtain a first long-term memory state and a superimposed second memory state, through... Activate, then pass The control door is partially forgotten; the formula is as follows: in, This represents the second memory superposition state at time t; , All are weights; This is the bias vector for controlling the gate. This represents the first long-term memory state at time t; It is the hyperbolic tangent activation function.
[0035] Merging past and present memories can be represented as: in, This represents the second long-term memory state at time t. For a moment The second long-term memory state.
[0036] The output gate is represented as: in, This represents the third memory superposition state at time t; As weight; This is the unit output at time t; This is the bias vector for the output gate.
[0037] In an alternative implementation, the power load forecasting model can also be constructed using a temporal convolutional network, which, through the architecture of a convolutional neural network, processes time series using causal convolution (ensuring that future information is not leaked into the past) and dilated convolution (capturing long-term dependencies with an exponentially increased receptive field), capturing local patterns (such as rapid increases and decreases in load) and long-term cyclical patterns in the load series.
[0038] In another alternative implementation, the power load forecasting model can also be constructed using a Transformer model. Based on the self-attention mechanism, it can calculate the correlation weight between any two time points in the sequence, thereby directly capturing global dependencies and enabling the simultaneous modeling of ultra-long-term load forecasts for multiple periods (daily, weekly, and annual).
[0039] In a preferred embodiment, training and optimizing the power load forecasting model according to preset optimization indicators includes: Using mean squared error as the optimization metric, minimizing the squared difference between the predicted and actual values is expressed as: Where K represents the total number of samples in the dataset; H represents the number of feature sequences used for prediction; This represents the true value of the i-th sample on the j-th feature. This represents the predicted value of the i-th sample on the j-th feature.
[0040] In this embodiment, the collected data is input, and the same preprocessing and feature engineering process as the training data is performed. The processed features are constructed into a (1, T, d) dimensional input tensor required by the model. The model is forward propagated to obtain the predicted output tensor (1, H). The output is inversely standardized to convert the predicted value back to the original scale to obtain the final load prediction value. Based on the load prediction results, strategy suggestions are provided to electricity market participants.
[0041] It should be noted that this invention combines historical load data with meteorological and other data through multi-source data fusion, providing the model with more comprehensive information. Simultaneously, it employs an advanced deep learning model combining a long short-term memory network with an attention mechanism, effectively capturing complex nonlinear relationships and long-term dependencies in the load data, significantly improving prediction accuracy. Compared to traditional methods that rely on single historical load data and have simple models, the prediction accuracy can be improved by 10%-20%. For example, during the high-temperature period in summer, accurately considering meteorological factors such as temperature and humidity, it can precisely predict the load increase caused by the surge in air conditioning usage, avoiding power distribution imbalances due to prediction errors. It enhances adaptability to complex scenarios, enabling rapid adaptation to load pattern changes brought about by industrial restructuring and changes in residents' lifestyles. When emerging industries such as big data centers and 5G base stations are put into operation, their unique electricity consumption patterns can be quickly learned by the model. It can also respond promptly to the dispersed and complex changes in residents' electricity consumption habits caused by the popularization of smart homes. Traditional methods often have poor adaptability to such changes, while the technology of this invention can flexibly adjust the prediction strategy to ensure the accuracy of predictions in complex scenarios. The above is an illustrative scheme of a power load forecasting method based on multi-source data fusion and deep learning according to this embodiment. It should be noted that the technical solution of this power load forecasting system based on multi-source data fusion and deep learning belongs to the same concept as the technical solution of the power load forecasting method based on multi-source data fusion and deep learning described above. Details not described in detail in the technical solution of the power load forecasting system based on multi-source data fusion and deep learning in this embodiment can be found in the description of the technical solution of the power load forecasting method based on multi-source data fusion and deep learning described above.
[0042] Example 2: This example provides a power load forecasting system based on multi-source data fusion and deep learning, including: The data acquisition and processing module is used to acquire multi-source power data and perform data preprocessing. The module is used to build a power load forecasting model based on the processed multi-source power data, so as to perform multivariate multi-step time series forecasting and output load forecasting results; The training and optimization module is used to train and optimize the power load forecasting model by adjusting the model parameters according to preset optimization indicators.
[0043] This embodiment also provides a computer device suitable for power load forecasting based on multi-source data fusion and deep learning, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a power load forecasting method based on multi-source data fusion and deep learning, as proposed in the above embodiments.
[0044] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power load forecasting method based on multi-source data fusion and deep learning as proposed in the above embodiments.
[0045] The storage medium proposed in this embodiment and the method for power load forecasting based on multi-source data fusion and deep learning proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0046] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0047] Example 3 is an embodiment of the present invention, which provides a power load forecasting method based on multi-source data fusion and deep learning. In order to verify its beneficial effects, it is scientifically demonstrated through economic benefit calculation and simulation experiments.
[0048] Taking Kunming in summer as an example, load data was collected every 15 minutes every day for the past three summers, along with meteorological data such as temperature and humidity for the corresponding time period, as well as data on the city's industrial electricity consumption.
[0049] After data preprocessing according to the present invention, the data is input into the power load prediction model for training. After multiple rounds of training and parameter adjustment, the mean square error of the model on the test set has reached a low level. The accuracy of predicting the load for the next 24 hours is 15% higher than that of traditional methods. In practical applications, based on the prediction results, power companies can make preparations for power dispatch in advance and effectively cope with peak loads.
[0050] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A power load prediction method based on multi-source data fusion and deep learning, characterized in that, The method comprises the following steps: acquiring power multi-source data and performing data preprocessing; constructing a power load prediction model based on the processed power multi-source data to perform multi-variable multi-step time series prediction and output a load prediction result; training and optimizing the power load prediction model based on the power load prediction model and a preset optimization index to adjust model parameters.
2. The power load forecasting method based on multi-source data fusion and deep learning according to claim 1, characterized in that, The power multi-source data includes power historical load data, power generation data, power grid topology structure, and meteorological data.
3. The power load forecasting method based on multi-source data fusion and deep learning according to claim 2, characterized in that, The data preprocessing includes: data cleaning to remove data outliers and eliminate abnormal multi-source data; data frequency alignment to select a standard parameter of collection frequency and perform collection frequency synchronization processing; data normalization to standardize the power multi-source data and obtain standardized data.
4. The power load forecasting method based on multi-source data fusion and deep learning according to claim 1, characterized in that, The construction of the power load prediction model includes: using a long short-term memory network to construct the power load prediction model, inputting a feature sequence of the power multi-source data in the past T time steps, and outputting a power load value sequence of future H time points.
5. The power load forecasting method based on multi-source data fusion and deep learning according to claim 4, characterized in that, The construction of the power load prediction model using a long short-term memory network includes: the power load prediction model includes a forget gate, an input gate, and an output gate; a first memory superposition state is obtained through the forget gate, a first long-term memory state and a second memory superposition state are obtained through the input gate, and a third memory superposition state is output through the output gate.
6. The power load forecasting method based on multi-source data fusion and deep learning according to claim 5, characterized in that, The forget gate is expressed as: wherein, denotes a time instant of a first memory state; denotes a sigmoid function; denotes a weight; denotes a hidden memory state at a time instant of a first memory state; denotes a normalized feature data; denotes a bias vector for the forget gate.
7. The power load forecasting method based on multi-source data fusion and deep learning according to claim 1, characterized in that, training and optimizing the power load prediction model based on the preset optimization index includes: using mean square error as an optimization index to minimize the square difference between the predicted value and the true value, which is expressed as: Wherein K represents the total number of samples in the data set; H represents the number of feature sequences used for prediction; represents the true value of the i-th sample on the j-th feature, represents the predicted value of the i-th sample on the j-th feature.
8. A power load forecasting system based on multi-source data fusion and deep learning, applying a power load forecasting method based on multi-source data fusion and deep learning according to any one of claims 1 to 7, characterized in that, The method comprises the following steps: a collection and processing module for acquiring power multi-source data and performing data preprocessing; a construction module for constructing a power load prediction model based on the processed power multi-source data to perform multi-variable multi-step time series prediction and output a load prediction result; a training and optimization module for training and optimizing the power load prediction model based on the power load prediction model and a preset optimization index to adjust model parameters.
9. A computer device, comprising: The method comprises the following steps: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which realize the steps of the power load prediction method based on multi-source data fusion and deep learning according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The memory stores computer executable instructions, which realize the steps of the power load prediction method based on multi-source data fusion and deep learning according to any one of claims 1 to 7 when executed by the processor.