Power equipment demand prediction method and system
By constructing a multi-dimensional phase space and delay coordinate mapping, and combining it with neural network learning of nonlinear characteristics, the uncertainty problem in power equipment market demand forecasting was solved, and more accurate and real-time equipment demand forecasting was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-06
- Publication Date
- 2026-04-07
AI Technical Summary
Existing technologies suffer from high uncertainty in forecasting demand for power equipment markets. Univariate time series reconstruction techniques cannot effectively capture the correlation characteristics of multiple factors, resulting in inaccurate forecasts, and there is a lack of real-time forecasting systems.
A multi-dimensional phase space is constructed based on the generalized embedding theorem. The topological conjugate relationship between the initial attractor and the delayed attractor is established through delayed coordinate mapping. The nonlinear features are learned by using neural networks to extract seasonal and long-term trend information from the time series. The embedding dimension is adjusted by combining spatiotemporal information to predict equipment demand.
It significantly improves the accuracy of power equipment demand forecasting, especially performing well in complex market systems, and provides real-time forecasting capabilities.
Smart Images
Figure CN121810334A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of power equipment, and particularly relates to a power equipment demand prediction method and system. BACKGROUND
[0002] In the power equipment market, the demand thereof is jointly affected by multiple factors and presents a highly nonlinear characteristic. Such a characteristic may cause a dramatic change in the prediction result even if there is a slight deviation in the observed data, thereby increasing the uncertainty of the future purchase quantity prediction of the power equipment. To cope with this challenge, it is necessary to establish a mathematical model of the market demand system, which is helpful to reveal the dynamic change characteristic of the power equipment purchase quantity and improve the prediction accuracy.
[0003] The power equipment purchase quantity is a complex time series. At present, although the single-variable time series reconstruction technology is relatively mature, it has limitations in capturing the multi-factor correlation characteristic of the complex power equipment market demand system and cannot effectively reveal the dynamic change characteristic of the market demand system. Moreover, there is no real-time prediction and practical system for the power equipment purchase quantity, which is not conducive to the real-time understanding of relevant personnel. SUMMARY
[0004] To solve the above problems, the application provides a power equipment demand prediction method and system. First, based on the generalized embedding theorem and according to the values of the power generation and power consumption at different time points, a multidimensional phase space is constructed. Then, based on the multidimensional phase space, the topological conjugate relationship between the initial attractor and the delay attractor is established through delay coordinate mapping to perform attractor reconstruction of the multivariate time series. Finally, the seasonal and long-term trend information in the time series is extracted by learning the nonlinear characteristics of the mapping using a neural network. After the spatiotemporal information is aggregated and the embedding dimension is adjusted, the delay attractor is determined, so that the equipment demand prediction quantity is obtained. The related data such as the power generation and power consumption are comprehensively considered, and the accuracy of the equipment demand prediction quantity is significantly improved.
[0005] To achieve the above purpose, the application is implemented by the following technical solutions: In a first aspect, the application provides a power equipment demand prediction method, comprising: acquiring the power generation and power consumption of the power industry; based on the generalized embedding theorem and according to the values of the power generation and power consumption at different time points, constructing a multidimensional phase space; based on the multidimensional phase space, establishing the topological conjugate relationship between the initial attractor and the delay attractor through delay coordinate mapping to perform attractor reconstruction of the multivariate time series; By utilizing the nonlinear characteristics of learning maps through neural networks, seasonal and long-term trend information in time series is extracted; then, by aggregating spatiotemporal information and adjusting the embedding dimension, the delayed attractor is determined, thereby obtaining the predicted equipment demand.
[0006] Furthermore, during attractor reconstruction: Let Indicates attractor, Indicates attractor Box dimension; time series Indicates time The state of the attractor at that time, where, Indicates the first Time of the first Each component; then delayed coordinate mapping for: ; in, The target variable is a positive number. satisfy Furthermore, based on the generalized embedding theorem, time series Reconstruction attractor With non-delayed attractors There is a topological conjugate relationship between them; For known 3D time series If there is If there are 10 sampling points, the following mapping relationship exists: ; Considering the reconstruction attractor Depend on Different moments Composition, then For a family of injective functions: ; The mapping process is represented as: .
[0007] Furthermore, the trend module in the neural network is used to mine the changing trend of the single variable in the initial attractor at different time scales. Through the trend decomposition module and the fully connected layer, the sliding window method is used to capture the dynamic trend of the time series, obtain the trend data of the single variable, and combine them into a matrix. Through the seasonal decomposition module and the fully connected layer, the seasonal data of the variable is extracted using the sliding window method and combined into a seasonal information matrix to capture the periodicity of the data. The information obtained from the trend module and the seasonal module is then aggregated.
[0008] Furthermore, the spatial variables in the initial attractor are: ; Long-term trend prediction information of all univariate attractors is extracted through trend mapping. Through seasonal mapping Extract seasonal forecast information ,Right now: ; Seasonal and trend information of univariate initial attractors is extracted using a matrix. With matrix Adding them together gives ; through trend mapping With seasonal forecasts Obtain intermediate variables ; intermediate variables Input spatiotemporal information aggregation module In this process, temporal and spatial information features are extracted and dimensions are adjusted to obtain delayed attractors.
[0009] Furthermore, the study explores the changing trends of single variables in the initial attractor at different time scales, captures the long-term trend changes of the initial attractor, and obtains trend prediction data. A sliding window method is used to capture the dynamic trend of the time series. The trend matrix is used as input, and trend information is extracted through a fully connected layer to generate a trend prediction matrix.
[0010] Furthermore, data on power grid construction investment, historical procurement volumes of different types of power equipment across the industry, GDP, and the consumer price index were also obtained.
[0011] Secondly, the present invention also provides a power equipment demand forecasting system, comprising: The data acquisition module is configured to acquire the power generation and power consumption data of the power industry. The data preprocessing module is configured to: construct a multi-dimensional phase space based on the generalized embedding theorem and the values of power generation and power consumption at different time points; The attractor reconstruction module is configured to: establish the topological conjugate relationship between the initial attractor and the delayed attractor based on the multi-dimensional phase space and through delayed coordinate mapping, and perform attractor reconstruction of multivariate time series. The prediction module is configured to: utilize the nonlinear features of the learning map through a neural network to extract seasonal and long-term trend information from the time series; then aggregate the spatiotemporal information and adjust the embedding dimension to determine the delayed attractor, thereby obtaining the predicted equipment demand.
[0012] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the power equipment demand forecasting method described in the first aspect.
[0013] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the power equipment demand forecasting method described in the first aspect.
[0014] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the power equipment demand forecasting method described in the first aspect.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention first constructs a multi-dimensional phase space based on the generalized embedding theorem and relevant data from the power industry, including GDP, consumer price index, and historical equipment procurement volumes at different time points. Then, based on this multi-dimensional phase space, a topological conjugate relationship between the initial and delayed attractors is established through delayed coordinate mapping, enabling attractor reconstruction for multivariate time series. Finally, the nonlinear characteristics of the learning mapping are utilized via neural networks to extract seasonal and long-term trend information from the time series. After spatiotemporal information aggregation and adjustment of the embedding dimension, the delayed attractor is determined, thus yielding the predicted equipment demand. This invention comprehensively considers relevant data such as power generation and consumption in the power industry, more effectively revealing the dynamic changes in the market demand system and significantly improving the accuracy of equipment demand forecasting, especially in complex market systems with multiple coupling effects.
[0016] 2. This invention comprises an interconnected data input unit, a data processing unit, and a display terminal. First, the data processing unit processes and analyzes the data to predict the demand for power equipment. Then, the display terminal displays the prediction results and other data in real time, which is beneficial for relevant personnel to understand the data in real time. Attached Figure Description
[0017] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.
[0018] Figure 1 This is a system schematic diagram of Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of two different variable association options in Embodiment 1 of the present invention; the yellow rectangle represents the area related to the purchase quantity to be predicted. Within this area, for any fixed data point, the red sliding rectangle centered on the location of the target variable contains the data points of each variable to be associated. Figure 3 The overall model architecture of Embodiment 1 of the present invention includes the layout of seasonality and trend mapping modules, and the process of obtaining delayed attractors through the time-space aggregation module; Figure 4 This is an overall architecture diagram of the single-step mapping for the initial attractor in Embodiment 1 of the present invention; First, the input data is processed through a sliding window to extract local trend features, and then processed through a trend decomposition module. Extracting trend information from time series The obtained trend features are then processed through multiple fully connected layers to capture long-term dependencies, ultimately generating a trend prediction matrix. ; Figure 5 This is the overall mapping structure diagram for the single-step mapping of a univariate initial attractor according to Embodiment 1 of the present invention; firstly, the input data is seasonally decomposed using a sliding window method to obtain a seasonal information matrix. Next, the seasonal matrix will be used. The data is fed into a fully connected network to further explore the periodic changes in seasonal features, thereby generating a seasonal prediction matrix. ; Figure 6 This is a complete flowchart of the initial model of Embodiment 1 of the present invention; Figure 7 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 8 This is a schematic diagram of the system in Embodiment 2 of the present invention. Detailed Implementation
[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0021] Spatio-Temporal Information Aggregation Module (STIAM): This module in the patent is used to aggregate information and predict the demand for power equipment. It is derived based on the generalized embedding theorem.
[0022] Example 1: This embodiment provides a method for forecasting power equipment demand, and correspondingly, a power equipment demand forecasting system. The forecasting system includes a data acquisition module, a data preprocessing module, a multivariate phase space construction module, an attractor reconstruction module, a neural network prediction module, a result output module, a data storage module, and a user interaction module. The forecasting method includes: S1. Data Acquisition: Users initialize the prediction system through the user interaction module, including the data acquisition time interval, sliding window size, relevant parameters of the neural network, and prediction period; the data storage module begins to prepare to receive and store data; the data acquisition module collects relevant data of the power industry, GDP, CPI, the industry's historical procurement volume of equipment, or other multivariate data at the set time intervals, and transmits the collected data to the data preprocessing module.
[0023] The data acquisition module is responsible for collecting relevant data from the power industry, GDP, CPI, historical procurement volume of equipment in the industry, or other multivariate data.
[0024] The relevant data for the power industry includes, but is not limited to, power generation, electricity consumption, and power grid construction investment; the historical procurement volume data can collect historical procurement data for different types of power equipment across the entire industry. Multiple data collection methods are employed to ensure the comprehensiveness and timeliness of the data.
[0025] S2, Data Preprocessing: The data preprocessing module processes the collected raw data, including removing noise, filling in missing values, correcting abnormal data, and standardizing data of different types and formats.
[0026] The data preprocessing module processes the collected raw data, removes noise, fills in missing values using mean or interpolation, and corrects outlier data by combining historical data and experience. At the same time, the data is standardized to convert data of different magnitudes and units into a unified standard format.
[0027] Optionally, missing values can be imputed using the mean or interpolation method, and outlier data can be imputed using historical data. Historical data and experience are used for correction; standardization converts data of different magnitudes and units into a unified standard format, which facilitates subsequent multivariate analysis and model training.
[0028] S3. Construction of multivariable phase space: The multivariable phase space construction module is based on the generalized embedding theorem to construct a multivariable phase space vector and analyze the intrinsic relationship between the variables.
[0029] Optionally, an embedding vector for the multivariate time series can be defined. This involves comprehensively considering the values of multiple variables, such as power industry data, economic indicators, and historical procurement volumes, at different points in time, to construct a multi-dimensional phase space. Specifically, based on the generalized embedding theorem, a multivariate phase space vector can be constructed to analyze the intrinsic relationships between multiple variables. The embedding vector for the multivariate time series is defined as follows: ; in, The number of variables; For time intervals.
[0030] S4, Attractor Reconstruction: The attractor reconstruction module constructs a spatiotemporal information transformation equation suitable for predicting power equipment procurement volume, realizing the transformation from the initial attractor to the delayed attractor. Specifically, the attractor reconstruction module constructs a spatiotemporal information transformation equation suitable for predicting power equipment procurement volume, considering the attractor observed by the system and its box dimension, and realizes the reconstruction of the initial attractor through delayed coordinate mapping; Optionally, when constructing the spatiotemporal information transformation equation, the attractors observed by the system and their box dimensions are considered. Through delayed coordinate mapping, a topological conjugate relationship between the initial attractor and the delayed attractor is established to realize the attractor reconstruction of multivariate time series; specifically, based on the generalized embedding theorem, its definition is as follows: S4.1, Order This represents the attractor observed by the system. Indicates attractor Box dimension; time series Indicates time Time system The state above, in which, Indicates the first Time of the first Each component; then delayed coordinate mapping Defined as: ; in, The target variable of interest in the system is a positive number. satisfy Furthermore, based on the generalized embedding theorem, time series... Reconstruction attractor With non-delayed attractors There is a topological conjugate relationship between them, that is, the two can be associated by constructing a smooth injective function.
[0031] S4.2, For the known 3D time series If there is If there are 10 sampling points, the following mapping relationship exists: ; Considering Depend on Different moments Composition, then It can be written as a family of injective functions: .
[0032] S4.3. The mapping process can be clearly represented by constructing the following matrix form: ; The matrix form above illustrates the variable mapping relationship at different time steps; by constructing this multidimensional mapping matrix, the nonlinear relationship between multiple time series can be captured more comprehensively.
[0033] S5, Neural Network Prediction: The neural network prediction module utilizes the nonlinear characteristics of neural network learning mapping, extracts seasonal and long-term trend information from the time series through the trend module and seasonal module, and then aggregates the information through the spatiotemporal information aggregation module STIAM and adjusts the embedding dimension to obtain the delayed attractor, that is, the future purchase quantity of equipment.
[0034] Optionally, the trend module in the neural network prediction module mines the changing trend of a single variable in the initial attractor at different time scales. Through the trend decomposition module and fully connected layers, it uses a sliding window method to capture the dynamic trend of the time series, obtains the trend data of the single variable, and combines it into a matrix. The seasonal module uses the seasonal decomposition module and fully connected layers to extract the seasonal data of the variable using a sliding window method, and combines it into a seasonal information matrix to capture the periodicity of the data. The spatiotemporal information aggregation module STIAM aggregates the information obtained from the trend module and the seasonal module, adjusts the embedding dimension, and solves the problems of insufficient model expressive power and unstable training caused by the small number of network layers in the original spatiotemporal fusion module.
[0035] From the perspective of market demand, the power equipment procurement system is a complex system influenced by multiple factors and changes continuously over time. For ease of calculation, a discretization method is used to describe the system. Sampling allows for efficient reconstruction of the power system. It is assumed that at any given moment... Below, there is A power equipment market demand system with variables, including variables (such as GDP, CPI, historical purchasing volume, etc.); in The initial attractor was obtained by observing the equipment purchase volume at different times. Define the initial time as The sampling interval is , then the first Each sampling time point for: ; It is important to note the sampling interval. The choice of time delay is crucial; an inappropriate time delay may cause the reconstruction attractor to fail to accurately characterize the dynamic properties of the system.
[0036] Under the above definition, Sampling points at time The purchase volume data is ,exist The relevant data for the entire market at this moment are expressed as follows: The discrete sampling method is as follows: in, ,but The initial attractor constructed by subsampling Represented as: ; For variables conduct Step prediction, to obtain To capture the spatiotemporal and nonlinear characteristics of high-dimensional dynamical systems, the delayed attractor is first represented as a matrix. : ; If the initial attractor The box dimension is Delayed attractor The box dimension is If satisfied Then there exists a smooth mapping. This makes the original attractor Topologically equivalent to the delayed embedding attractor The mapping It can reconstruct the nonlinear dynamic characteristics of the system; by selecting appropriate , the initial attractor Mapped to delayed attractor Thus, the spatiotemporal information transformation equation can be obtained: ; Based on the spatiotemporal information conversion equation, it is parallelized spatially, resulting in the following spatially parallel equation: ; Where D represents the total number of grid points contained within the region associated with the variable.
[0037] The mapping will be learned using neural networks. The nonlinear characteristics enable the initial attractor. To delayed attractor The transformation; if the initial attractor As input, delay attractor As output, the mapping relationship is represented as: ; By utilizing spatiotemporal information transformation, smooth mapping is achieved. It can be decomposed into a series of single-step prediction functions, i.e. Single-step prediction function Defined as: ; This step breaks down the complex nonlinear relationships in multidimensional time series into multiple single-step prediction tasks, thereby reducing the computational complexity of the model and improving prediction accuracy; the matrix form of the single-step prediction function is as follows: ; Next, a suitable neural network model is constructed to fit the nonlinear mapping. To capture the changing trends of power equipment procurement volume across different time scales, the model utilizes two modules to extract seasonal and long-term trend information from the time series. Based on this, the seasonal and long-term trend information is aggregated through the STIAM spatiotemporal information aggregation module, and the embedding dimension is adjusted to obtain the delayed attractor. ; The spatial variables in the initial attractor are: ; Long-term trend prediction information of all univariate attractors in the system is extracted through trend mapping. Through seasonal mapping Extract its seasonal forecast information ,Right now: ; Seasonal and trend information of univariate initial attractors is extracted using a matrix. With matrix Adding them together gives ; through trend mapping With seasonal forecasts Obtain intermediate variables Further intermediate variables Input spatiotemporal information aggregation module In the process, temporal and spatial information features are extracted and dimensions are adjusted to obtain delayed attractors: .
[0038] Furthermore, the trend module is used to mine single variables in the initial attractor. Trends in change at different time scales.
[0039] Trend function The goal is to capture the long-term trend changes of the initial attractor and obtain trend prediction data. Let the input be... Then the trend mapping function is expressed as: ; Trend mapping primarily reflects the trend information of the same variable at different times; to further improve the predictive ability of the model, Decomposed into a series of single-step mapping functions ,for The mapping is defined as: ; Trend function It consists of two parts: a trend decomposition module and a fully connected layer; the trend data is defined as... For each variable Define the trend variable corresponding to this variable as In this process, a sliding window method is used to capture the dynamic trend of the time series; the current window size is set to... Trend variables The formula is: ; Constructed based on trend mapping function To obtain a single variable Trend data If the trend data of all variables are combined into a matrix, then ,writing: ; Trend matrix As input, trend information is extracted through a fully connected layer, ultimately generating a trend prediction matrix. .
[0040] The seasonal module defines the seasonal mapping function as follows: The initial attractor and the seasonality prediction matrix of the time series are calculated using this mapping function. Relationship: ; If this mapping Decomposed into a family of injective functions Each of them This is considered a single-step mapping prediction of the univariate initial attractor, with the corresponding input being the univariate initial attractor. Define mapping for: ; Mapping functions include The system consists of two modules: a seasonal decomposition module and a fully connected layer. Similar to methods for extracting trend data, seasonal data is defined as... For each variable Define the seasonal variable corresponding to this variable as Here, a sliding window method is used, and the current window size is defined as... ,So The calculation formula is: ; A more general way to express this is: ; Using the above formula, the mapping function For each single variable Seasonal decomposition of the time series data yields the corresponding seasonal data. Combine the seasonal data of all variables at each time point to form a matrix. Its form is: ; Construct a seasonal information matrix Then, in order to better extract the periodic patterns in the data, the matrix was... The input is fed into a fully connected layer to obtain the predicted seasonality matrix. .
[0041] The spatiotemporal information aggregation module STIAM initially adjusted its model using a two-layer fully connected neural network. With delayed attractor Dimensional matching and information aggregation between them; that is: .
[0042] S6. Output of Results: The result output module will output the predicted future purchase volume of power equipment in an intuitive form, such as reports and charts.
[0043] S7, Data Storage: The data storage module is responsible for storing various types of data, including the collected raw data, preprocessed data, model parameters, and prediction results. The data storage module continuously stores all relevant data, including the collected raw data, preprocessed data, model parameters, and prediction results, ensuring data integrity and traceability, and facilitating subsequent model optimization and result verification.
[0044] S8, Data Visualization: The user interaction module provides a visual interface, making it convenient for users to view data, set parameters, and analyze prediction results.
[0045] Example 2: This embodiment provides a power equipment demand forecasting system, including a data input unit, a data processing unit, and a display terminal that are interconnected. The data input unit, the data processing unit, and the display terminal can be connected via wired and / or wireless means; the display terminal can be a computer, tablet, or mobile phone, etc., so that relevant personnel can view prediction results and other data in real time. The data input unit is used to acquire the power generation and power consumption of the power industry; the data processing unit is used to process and analyze the data and make power equipment demand forecasts; the display terminal is used to display forecast results and other data in real time, which is beneficial for relevant personnel to understand in real time.
[0046] The data processing unit operates in the same way as the power equipment demand forecasting method in Embodiment 1, and will not be described again here.
[0047] Example 3: This embodiment provides a power equipment demand forecasting system, including: The data acquisition module is configured to acquire the power generation and power consumption data of the power industry. The data preprocessing module is configured to: construct a multi-dimensional phase space based on the generalized embedding theorem and the values of power generation and power consumption at different time points; The attractor reconstruction module is configured to: establish the topological conjugate relationship between the initial attractor and the delayed attractor based on the multi-dimensional phase space and through delayed coordinate mapping, and perform attractor reconstruction of multivariate time series. The prediction module is configured to: utilize the nonlinear features of the learning map through a neural network to extract seasonal and long-term trend information from the time series; then aggregate the spatiotemporal information and adjust the embedding dimension to determine the delayed attractor, thereby obtaining the predicted equipment demand.
[0048] The operating method of the system is the same as that of the power equipment demand forecasting method in Example 1, and will not be repeated here.
[0049] Example 4: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power equipment demand forecasting method described in Embodiment 1.
[0050] Example 5: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the power equipment demand forecasting method described in Embodiment 1.
[0051] Example 6: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the power equipment demand forecasting method described in Embodiment 1.
[0052] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.
Claims
1. A method for forecasting power equipment demand, characterized in that, include: Obtain data on power generation and consumption in the power industry; Based on the generalized embedding theorem and according to the values of power generation and power consumption at different time points, a multi-dimensional phase space is constructed; Based on the multi-dimensional phase space, the topological conjugate relationship between the initial attractor and the delayed attractor is established through delayed coordinate mapping, and attractor reconstruction of multivariate time series is performed. By utilizing the nonlinear characteristics of learning maps through neural networks, seasonal and long-term trend information in time series is extracted; then, by aggregating spatiotemporal information and adjusting the embedding dimension, the delayed attractor is determined, thereby obtaining the predicted equipment demand.
2. The power equipment demand forecasting method as described in claim 1, characterized in that, During attractor reconstruction: Let Indicates attractor, Indicates attractor The box dimension; Time series Indicates time The state of the attractor at that time, where, Indicates the first Time of the first Each component; then delayed coordinate mapping for: ； in, The target variable is a positive number. satisfy Furthermore, based on the generalized embedding theorem, time series Reconstruction attractor With non-delayed attractors There is a topological conjugate relationship between them; For known 3D time series If there is If there are 10 sampling points, the following mapping relationship exists: ； Considering the reconstruction attractor Depend on Different moments Composition, then For a family of injective functions: ； The mapping process is represented as: 。 3. The power equipment demand forecasting method as described in claim 1, characterized in that, The trend module in the neural network is used to mine the changing trend of a single variable in the initial attractor at different time scales. Through the trend decomposition module and the fully connected layer, the sliding window method is used to capture the dynamic trend of the time series, obtain the trend data of the single variable and combine them into a matrix. The seasonal decomposition module and the fully connected layer are used to extract seasonal data of variables using the sliding window method and combine them into a seasonal information matrix to capture the periodic patterns of the data; the information obtained from the trend module and the seasonal module is aggregated.
4. The power equipment demand forecasting method as described in claim 3, characterized in that, The spatial variables in the initial attractor are: ； Long-term trend prediction information of all univariate attractors is extracted through trend mapping. Through seasonal mapping Extract seasonal forecast information ,Right now: ； Seasonal and trend information of univariate initial attractors is extracted using a matrix. With matrix Adding them together gives ; through trend mapping With seasonal forecasts Obtain intermediate variables ; intermediate variables Input spatiotemporal information aggregation module In this process, temporal and spatial information features are extracted and dimensions are adjusted to obtain delayed attractors.
5. The power equipment demand forecasting method as described in claim 4, characterized in that, By exploring the changing trends of a single variable in the initial attractor across different time scales, we can capture the long-term trend changes of the initial attractor and obtain trend prediction data. The sliding window method is used to capture the dynamic trends of time series. By taking the trend matrix as input, trend information is extracted through a fully connected layer to generate a trend prediction matrix.
6. The power equipment demand forecasting method as described in claim 1, characterized in that, It also obtained data on power grid construction investment, the industry's historical procurement volume of different types of power equipment, GDP, and the consumer price index.
7. A power equipment demand forecasting system, comprising a data input unit, a data processing unit, and a display terminal interconnected thereto; the data input unit is used to acquire power generation and power consumption data from the power industry; the data processing unit is used to process and analyze the data; the display terminal is used to display the forecast results in real time; the data processing unit is configured to: Based on the generalized embedding theorem and according to the values of power generation and power consumption at different time points, a multi-dimensional phase space is constructed; Based on the multi-dimensional phase space, the topological conjugate relationship between the initial attractor and the delayed attractor is established through delayed coordinate mapping, and attractor reconstruction of multivariate time series is performed. By utilizing the nonlinear characteristics of learning maps through neural networks, seasonal and long-term trend information in time series is extracted; then, by aggregating spatiotemporal information and adjusting the embedding dimension, the delayed attractor is determined, thereby obtaining the predicted equipment demand.
8. A power equipment demand forecasting system, characterized in that, include: The data acquisition module is configured to acquire the power generation and power consumption data of the power industry. The data preprocessing module is configured to: construct a multi-dimensional phase space based on the generalized embedding theorem and the values of power generation and power consumption at different time points; The attractor reconstruction module is configured to: establish the topological conjugate relationship between the initial attractor and the delayed attractor based on the multi-dimensional phase space and through delayed coordinate mapping, and perform attractor reconstruction of multivariate time series. The prediction module is configured to: utilize the nonlinear features of the learning map through a neural network to extract seasonal and long-term trend information from the time series; then aggregate the spatiotemporal information and adjust the embedding dimension to determine the delayed attractor, thereby obtaining the predicted equipment demand.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the power equipment demand forecasting method as described in any one of claims 1-5.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the power equipment demand forecasting method as described in any one of claims 1-5.