Self-adaptive adjustment method and system based on energy management system

Abstract

The invention discloses a self-adaptive adjustment method and system based on an energy management system, and relates to the technical field of energy management, and the method comprises the following steps: collecting multi-dimensional heterogeneous data including weather conditions, economic situations, social activities, user behavior patterns and energy price fluctuations; according to the method, the multi-dimensional heterogeneous data is fused, the space-time attention mechanism and the graph neural network are utilized to construct the high-precision prediction model, ultra-short-term to medium-and-long-term energy demand dynamic prediction is realized, the problem of insufficient multi-factor fusion is solved, the hierarchical response strategy is triggered based on the prediction result, and the prediction efficiency is improved. Distributed energy storage, a flexible load and a production end are linked to perform dynamic power adjustment, strategy autonomous optimization is realized through reinforcement learning, the adaptive adjustment ability to a complex scene is improved, an adaptive adjustment closed loop is formed through real-time error calculation and model parameter correction, and adjustment parameters and model weights are corrected in real time, so that the real-time performance of the system is improved. And the stability and accuracy of long-term operation of the system are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, in particular to an adaptive adjustment method and system based on an energy management system. BACKGROUND

[0002] An energy management system is an intelligent system that integrates sensors, data processing algorithms and control modules to monitor and optimize the production, transmission, storage and consumption of energy in real time. Its core goal is to improve energy utilization efficiency, reduce energy consumption costs and ensure the stability of energy supply and demand. With the popularization of distributed energy, the energy management system needs to cope with more complex supply and demand scenarios. For example, when user-side distributed energy is connected, traditional systems have difficulty in dynamically balancing the load of the power grid. In the event of a disaster or policy power cut, there is a lack of flexible emergency adjustment mechanism. At the same time, the peak and valley electricity prices in the electricity market differ significantly, and traditional systems cannot effectively utilize the "valley charging and peak discharging" mode to reduce electricity costs. Therefore, it is necessary to adaptively adjust the energy management system.

[0003] According to the invention patent with application number 202311741943.5, a platform-based adaptive energy management method and system are disclosed. The method includes setting the charging period and discharging period of the energy storage battery, obtaining the total load of the transformer, and distributing the charging power of the charging period and the discharging power of the discharging period according to the total load of the transformer, the charging period and the discharging period of the energy storage battery. The charging information and discharging information of the energy storage battery are obtained, and the maximum charging limit and discharging limit that the current energy storage battery can bear are calculated in real time according to the algorithm, and the charging and discharging strategy is set.

[0004] Although the above-mentioned scheme can realize basic charging and discharging periods and power management, it still has certain limitations. It only relies on a small number of parameters such as transformer load and energy storage SOC, lacks fusion analysis of multi-dimensional data such as weather, economic situation and user behavior, resulting in insufficient demand prediction accuracy. The adjustment strategy is based on a fixed threshold and cannot adapt to changes in factor weights in the time and space dimensions, and has limited ability to cope with complex load fluctuations. Therefore, it is of great significance to develop an adaptive adjustment method and system based on an energy management system. SUMMARY

[0005] The purpose of the present application is to overcome the shortcomings of the prior art and provide an adaptive adjustment method and system based on an energy management system. It can fuse multi-dimensional heterogeneous data, use a spatio-temporal attention mechanism and a graph neural network to construct a high-precision prediction model, realize dynamic prediction of energy demand from ultra-short term to medium and long term, trigger a hierarchical response strategy based on the prediction results, and dynamically adjust the power of distributed energy storage, flexible load and production end. Through reinforcement learning, the strategy is optimized automatically, and the adaptive adjustment capability for complex scenarios is improved.

[0006] The application provides the following technical solutions to solve the above technical problems: an adaptive adjustment method based on an energy management system, which comprises the following steps: Collecting multi-dimensional heterogeneous data, which includes weather conditions, economic situations, social activities, user behavior patterns, and energy price fluctuations, and performing cleaning, preprocessing, and standardization on the data; Using a space-time attention mechanism to construct an influence weight matrix in the time-space dimension, mining the correlation between factors through a graph neural network, generating an energy demand prediction model that fuses multi-dimensional data, training the model based on historical data, and realizing ultra-short-term, short-term, and medium-and-long-term energy demand prediction; Based on the prediction results, triggering a hierarchical response strategy, predicting demand growth, calling distributed energy storage discharge, adjusting industrial flexible load, and linking energy production end to increase power, predicting demand decline, reducing unnecessary equipment energy consumption, storing excess power, and participating in market transactions; Real-time collection of actual energy supply and demand data, calculation of prediction error, adjustment of prediction model parameters based on error correction strategy, and formation of a closed-loop optimization.

[0007] Further, the specific implementation of the space-time attention mechanism comprises: establishing a time dimension sliding window and a space grid division, and calculating the dynamic influence weight of a factor at time and space region through the following formula: wherein, is an attention temperature parameter determined through Bayesian optimization, , is a learnable weight matrix trained based on historical data through a back propagation algorithm, , are the time sequence feature and the space feature vector of the factor , respectively, and the weight matrix is automatically adjusted according to the season and regional characteristics.

[0008] Further, the construction of the graph neural network comprises: Mapping multi-dimensional data to graph nodes, taking the correlation coefficient between factors calculated dynamically based on the Pearson correlation coefficient as the edge weight, and updating the node features through the following formula: wherein, is the graph convolution layer number, is the neighborhood node set of the node , , is a learnable weight matrix optimized through a multi-head attention mechanism, is an activation function; ​The time sequence features are processed by a gated recurrent unit (GRU), and a multi-period demand prediction value is output by the following formula : , is the output feature of the last layer of the graph convolution network, Concat is a feature splicing operation, and Linear is a linear transformation layer.

[0009] Further, the closed-loop optimization process comprises: calculating a prediction error by the following formula : , is an actual energy demand value, is a predicted value, is a sampling period, is a preset error threshold determined by historical error distribution statistics, , a trigger factor importance reevaluation process is triggered, the spatio-temporal attention mechanism weight is updated by SHAP value analysis, and the graph neural network is incrementally learned.

[0010] Further, the weather condition data comprises temperature, humidity, and sunshine duration, the economic situation data comprises GDP growth rate and industrial structure adjustment data, the social activity data comprises holiday information and large-scale event holding time, and the user behavior pattern data comprises office overtime frequency and household electricity consumption time period distribution, and the energy price fluctuation data comprises real-time electricity price and futures contract price.

[0011] Further, the adjustment of the industrial flexible load comprises: based on the adjustable load period and power threshold value reported in advance by the user and stored in the system database, sending a load adjustment instruction containing a priority identifier through an intelligent terminal, the priority being determined by a fuzzy logic algorithm according to the emergency degree of the supply-demand gap, and a high-priority instruction corresponding to a scenario in which the real-time supply-demand difference exceeds a preset threshold.

[0012] Further, the energy production end linkage comprises: through a real-time communication interface between an edge computing node and a power generation equipment control system, generating a power generation power adjustment instruction containing an adjustment amplitude, a response time window, and a standby capacity reservation requirement according to a predicted demand curve and power generation equipment operating characteristic parameters, the power generation equipment operating characteristic parameters being obtained through an API interface provided by the equipment manufacturer.

[0013] An adaptive adjustment system based on an energy management system, applicable to the adaptive adjustment method based on the energy management system, the system comprising: a data acquisition and processing module for acquiring multi-dimensional heterogeneous data and performing cleaning, preprocessing, and standardization; The energy demand forecasting module includes a spatiotemporal attention mechanism unit and a graph neural network unit, which are used to generate energy demand forecasting models and output multi-period forecasting results. The adaptive adjustment module is used to execute a graded response strategy based on the prediction results, including an energy storage control unit, a load adjustment unit, and a production linkage unit; The feedback optimization module is used to collect actual supply and demand data, calculate prediction errors, and correct adjustment strategies and model parameters.

[0014] Furthermore, the energy storage control unit includes a distributed energy storage cluster control algorithm module, which generates an energy storage scheduling plan based on the predicted demand curve and real-time electricity price using a mixed integer programming algorithm. The load adjustment unit includes a user load profile database, which stores the flexible load parameters declared by users, historical adjustment response time, and error data.

[0015] Compared with existing technologies, this adaptive adjustment method and system based on an energy management system has the following advantages: This invention integrates multidimensional heterogeneous data and utilizes a spatiotemporal attention mechanism and graph neural network to construct a high-precision prediction model, enabling dynamic prediction of energy demand from the ultra-short to medium-to-long term. It solves the problem of insufficient integration of multiple factors. Based on the prediction results, it triggers a graded response strategy, linking distributed energy storage, flexible loads, and the production end for dynamic power adjustment. Through reinforcement learning, it achieves autonomous strategy optimization, improving the adaptive adjustment capability for complex scenarios. Through real-time error calculation and model parameter correction, it forms an adaptive adjustment closed loop, real-time correction of adjustment parameters and model weights, ensuring the stability and accuracy of the system's long-term operation.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

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

[0018] Figure 1 This is a flowchart illustrating an adaptive adjustment method based on an energy management system. Figure 2 This is a schematic diagram of an adaptive regulation system based on an energy management system. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Example 1 This embodiment focuses on the energy management scenario of an industrial park, detailing the adaptive adjustment method based on the energy management system and its specific application. By collecting multidimensional heterogeneous data within the park, an energy demand prediction model is constructed using a spatiotemporal attention mechanism and graph neural network. Based on the prediction results, a tiered response strategy is triggered, and real-time error correction forms a closed-loop optimization. The various modules of the system work collaboratively to achieve a dynamic balance between energy supply and demand in the park, significantly improving energy utilization efficiency and the system's adaptive adjustment capability. For specific processes and system architecture, please refer to [link / reference]. Figure 1 and Figure 2 .

[0021] Within the industrial park, the data acquisition and processing module first collects multi-dimensional heterogeneous data, covering multiple dimensions such as weather conditions, economic situation, social activities, user behavior patterns, and energy price fluctuations. After collecting this data, the data acquisition and processing module cleans, preprocesses, and standardizes it.

[0022] Regarding weather data, meteorological monitoring equipment deployed within the park collects real-time data on temperature, humidity, and sunshine duration. This data directly impacts the park's energy demand. Economic data primarily includes the GDP growth rate of the region where the park is located and data on industrial restructuring. The GDP growth rate reflects the regional economic development trend. Social activity data includes holiday information and the timing of major events. During holidays, some businesses within the park may suspend operations or reduce production, leading to significant fluctuations in energy demand. When major events are held, traffic and commercial activities around the park may increase, indirectly affecting the park's energy demand.

[0023] User behavior data primarily collects information such as office overtime frequency and household electricity consumption patterns. The frequency of overtime work in the office spaces of companies within the park directly affects the electricity consumption of the office area. The distribution of household electricity consumption patterns is related to the residential electricity needs of employees within the park. Understanding this data helps to more accurately predict energy demand. Energy price fluctuation data includes real-time electricity prices and futures contract prices. Changes in real-time electricity prices affect the electricity costs of enterprises, and enterprises may adjust their production plans and electricity consumption behaviors based on electricity prices. Futures contract prices provide a reference for enterprises to conduct energy procurement and storage, helping them optimize energy cost management.

[0024] When constructing the influence weight matrix of the time-space dimension, the spatiotemporal attention mechanism unit is used. First, a time-dimensional sliding window and spatial grid division are established. The time-dimensional sliding window is set to capture the changes in the influence of various factors on energy demand in different time periods. For example, the energy demand patterns may differ in different seasons, different weekdays and weekends. The spatial grid division divides the industrial park into multiple spatial regions in order to analyze the influence of factors on energy demand in different regions.

[0025] Factors are calculated using the following formula. At any moment spatial regions Dynamic influence weight : ,in, The attention temperature parameter, determined through Bayesian optimization, can be automatically determined through Bayesian optimization as well. The optimal value allows the attention mechanism to better capture the differences in importance among various factors. , This is a learnable weight matrix trained using the backpropagation algorithm based on historical data. (The process involves training on historical data.) and It can automatically learn the mapping relationship between the temporal and spatial characteristics of factors, thereby better reflecting the influence of factors in the temporal and spatial dimensions. , Factors The temporal and spatial feature vectors, and the weight matrix will be automatically adjusted according to seasonal and regional features.

[0026] Multidimensional data is mapped to graph nodes, and the correlation coefficient between factors is dynamically calculated based on the Pearson correlation coefficient as the edge weight. The Pearson correlation coefficient is a statistic used to measure the degree of linear correlation between two variables. By calculating the Pearson correlation coefficient between each factor, the strength of the association between factors can be determined, thereby providing edge weights for the graph neural network.

[0027] Update node features using the following formula : ,in, This represents the graph convolutional layer number. The function of the graph convolutional layer is to aggregate and transform the features of graph nodes, thereby extracting the implicit information in the graph structure. For nodes The set of neighboring nodes is determined based on the edge connections of the graph, i.e., the connections between nodes. Connected nodes form their neighborhood node set. , This is a learnable weight matrix optimized through a multi-head attention mechanism. Multi-head attention allows the model to capture different information in different representation subspaces, thereby improving the model's expressive power. The activation function is commonly used, such as the ReLU function, to introduce nonlinear factors, enabling the model to learn more complex relationships.

[0028] By combining gated recurrent units (GRUs) with time-series features, GRUs, a variant of recurrent neural networks (RNNs), can effectively handle long-term dependencies in time-series data. By processing the features output by graph convolutional networks using GRUs, the changing patterns of energy demand over time can be further captured.

[0029] The following formula outputs multi-period demand forecasts. : ,in, The output features of the last layer of the graph convolutional network are represented by Concat, which is a feature concatenation operation that concatenates the output features of the graph convolutional network with the features processed by GRU to fuse spatial and temporal information. Linear is a linear transformation layer that maps the concatenated features to the space of energy demand prediction values.

[0030] The generated energy demand forecasting model is trained based on historical data. By continuously adjusting the model parameters, the model can accurately predict energy demand in the ultra-short term, short term, and medium to long term.

[0031] Based on the forecast results output by the energy demand forecasting module, the adaptive adjustment module triggers a tiered response strategy.

[0032] When demand growth is predicted, the following operations are performed: Distributed energy storage is invoked to discharge; the distributed energy storage cluster control algorithm module adjusts the industrial flexible load based on the predicted demand curve and real-time electricity price; the load adjustment unit sends a load adjustment command containing a priority identifier through the smart terminal based on the adjustable load periods and power thresholds declared by users in advance and stored in the system database, thereby linking the energy production end to increase power.

[0033] When demand is predicted to decline, the following actions are taken: reduce energy consumption of non-essential equipment, send instructions to enterprises in the park through smart terminals to shut down or adjust the operating status of non-essential equipment, reduce energy consumption, store excess electricity, store excess electricity in a distributed energy storage system for use during peak demand, participate in market transactions, and sell excess electricity through the electricity market to improve energy utilization efficiency and economic benefits.

[0034] The feedback optimization module collects real-time energy supply and demand data and calculates the prediction error using the following formula. : ,in, This represents the actual energy demand value. For predicted values, The sampling period is used to evaluate the model's prediction accuracy by calculating the prediction error.

[0035] Preset error threshold Determined through historical error distribution statistics, when... At that time, the importance of triggering factors is reassessed. The weights of the spatiotemporal attention mechanism are updated through SHAP value analysis. SHAP value is a method used to interpret the prediction results of machine learning models. By calculating the contribution of each feature to the prediction results, the importance of factors is determined. Based on the SHAP value analysis results, the weight matrix in the spatiotemporal attention mechanism is adjusted to improve the model's attention to important factors. At the same time, incremental learning is performed on the graph neural network, and the model is trained using new error data to further optimize the model parameters and improve the model's prediction accuracy and adaptability.

[0036] In summary, in the industrial park application of this embodiment, the adaptive adjustment method and system based on the energy management system achieve accurate prediction and dynamic adjustment of energy demand through the coordinated operation of various modules. The fusion and processing of multi-dimensional heterogeneous data provides rich input information for the model. The combination of spatiotemporal attention mechanism and graph neural network effectively explores the spatiotemporal correlation between factors and constructs a high-precision energy demand prediction model. The flexible execution of the hierarchical response strategy realizes dynamic power adjustment of distributed energy storage, flexible loads and the production end. The closed-loop optimization process ensures the stability and accuracy of the system's long-term operation.

[0037] Example 2 This embodiment uses a large commercial complex as an application scenario to elaborate on the adaptive adjustment method based on an energy management system and the actual implementation process of the system. By integrating multi-dimensional data from both inside and outside the commercial complex, a spatiotemporal attention mechanism is used to dynamically allocate factor weights. A cross-period demand forecasting model is constructed by combining graph neural networks and gated cyclic units. Based on the forecast results, the energy storage system, adjustable load, and energy supplier are linked to implement hierarchical control. Through real-time error feedback, model parameters and adjustment strategies are optimized, achieving refined management of energy consumption and dynamic supply-demand balance in the commercial complex. This effectively improves energy utilization efficiency and system responsiveness. For specific processes and system architecture, please refer to [link to relevant documentation]. Figure 1 and Figure 2 .

[0038] In the context of commercial complexes, the data acquisition and processing module focuses on the collection of multi-dimensional data strongly related to commercial activities. It obtains real-time data such as temperature, humidity, and sunshine duration by accessing the city's meteorological API, collects data such as regional retail sales growth rate and service industry PMI index, captures holiday calendars, mall promotion schedules, and information on large-scale conferences, and collects data such as the spatiotemporal distribution of customer flow, the characteristics of store electricity consumption during the day, and the load curve of office buildings during overtime hours through Wi-Fi probes and smart meters. It also obtains real-time time-of-use electricity prices, peak-valley price difference data, and natural gas futures prices in the electricity market, and performs data preprocessing on the above-mentioned collected data.

[0039] In response to the spatiotemporal characteristics of commercial complexes, the spatiotemporal attention mechanism unit dynamically calculates the influence weights of various factors through a time-dimensional sliding window and spatial grid division.

[0040] Taking holiday promotional activities as an example, factors At any moment spatial regions Dynamic weights Calculated using the following formula: ,in, Automatic adjustment via Bayesian optimization is used to control the concentration of attention distribution. and By training with historical data, the system learns the mapping relationship between the temporal and spatial characteristics of promotional activities, and the weight matrix is ​​automatically adjusted with the business cycle.

[0041] The multidimensional data of the commercial complex is mapped to graph nodes, including: weather nodes, economic indicator nodes, passenger flow nodes, equipment load nodes, etc. The edge weights between nodes are dynamically calculated using the Pearson correlation coefficient. The graph neural network updates the node features through multiple layers of graph convolutional layers. Layer nodes The feature update formula is: Among them, the set of neighboring nodes This includes nodes related to the "air conditioning load" node, such as "temperature," "passenger flow," and "promotional activities." and By optimizing the multi-head attention mechanism, the model can capture the synergistic effects of different types of factors. Combined with gated recurrent units (GRUs) to process temporal features, the output features of the last layer of the graph convolutional network are then optimized. After being concatenated with the GRU output, multi-time-period predicted values ​​are generated through a linear transformation layer: The model can predict energy demand for the next 15 minutes (ultra-short term), 1 hour (short term), and 1 day (medium- to long term).

[0042] When predicting energy demand growth in commercial complexes, the adaptive adjustment module triggers a three-level response strategy: Level 1 response: When the real-time supply-demand difference exceeds a preset threshold, a high-priority instruction is generated using a fuzzy logic algorithm to call the distributed energy storage system for rapid discharge, while simultaneously sending load adjustment instructions to restaurants to temporarily reduce the power of non-essential equipment; Level 2 response: The lighting brightness in non-office areas of the office building is adjusted to energy-saving mode, the hotel's air conditioning energy-saving system is linked to raise the set temperature by 1°C, and a power increase request is sent to the energy supplier through edge computing nodes; Level 3 response: Before the start of promotional activities, the operating frequency of escalators in the shopping mall is pre-adjusted, and adjustable load reminders are sent to contracted merchants through the user load profile database; When demand decreases, the system automatically stores excess electricity into the energy storage system and participates in peak shaving through the electricity market trading platform, selling excess electricity during off-peak hours to generate revenue.

[0043] The feedback optimization module collects real-time supply and demand data from smart meters and energy storage systems, and calculates the prediction error. If the error exceeds the historical statistical threshold This triggers SHAP value analysis, reassessing the importance of each factor.

[0044] In summary, this embodiment constructs an energy management system with spatiotemporal dynamic adaptability by integrating multidimensional data with significant commercial characteristics in the context of large-scale commercial complexes. The combination of spatiotemporal attention mechanism and graph neural network effectively captures the complex correlation between factors such as customer traffic and promotional activities and energy demand. The hierarchical response strategy achieves precise coordination between energy storage, load, and production, and the closed-loop optimization mechanism ensures the system's predictive accuracy during changes in the business cycle.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive adjustment method based on an energy management system, characterized in that, The method includes the following steps: Collect multidimensional heterogeneous data, including weather conditions, economic situation, social activities, user behavior patterns, and energy price fluctuations, and clean, preprocess, and standardize the data. A temporal-spatial attention mechanism is used to construct an influence weight matrix in the time-space dimension. A graph neural network is used to mine the correlation between various factors to generate an energy demand prediction model that integrates multi-dimensional data. The model is trained based on historical data to achieve ultra-short-term, short-term, and medium-to-long-term energy demand prediction. Based on the forecast results, a tiered response strategy is triggered. If demand growth is predicted, distributed energy storage is called for discharge, industrial flexible loads are adjusted, and energy production power is increased in conjunction with energy production. If demand decline is predicted, energy consumption of non-essential equipment is reduced, excess electricity is stored, and market transactions are participated in. Real-time energy supply and demand data are collected, prediction errors are calculated, and closed-loop optimization is formed based on error correction and adjustment strategies and prediction model parameters.

2. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The specific implementation of the spatiotemporal attention mechanism includes: establishing a time-dimensional sliding window and spatial grid division, and calculating factors using the following formula. At any moment spatial regions Dynamic influence weight The formula is: ,in, The attention temperature parameters were determined through Bayesian optimization. , This is a learnable weight matrix trained using the backpropagation algorithm based on historical data. , Factors The temporal and spatial feature vectors are obtained, and the weight matrix is ​​automatically adjusted according to seasonal and regional features.

3. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The construction of the graph neural network includes: Multidimensional data is mapped to graph nodes, and the correlation coefficients between factors, dynamically calculated based on the Pearson correlation coefficient, are used as edge weights. The node features are then updated using the following formula. ,in, The convolutional layer number is shown in the image. For nodes The set of neighboring nodes, , For a learnable weight matrix optimized through a multi-head attention mechanism, For activation functions; Based on the timing characteristics of the gated recurrent unit (GRU), the demand forecast values ​​for multiple time periods are output using the following formula. : ,in, This represents the output features of the last layer of the graph convolutional network. Concat is the feature concatenation operation, and Linear is the linear transformation layer.

4. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The closed-loop optimization process includes: calculating the prediction error using the following formula. : ,in, This represents the actual energy demand value. For predicted values, The sampling period is The preset error threshold is determined through historical error distribution statistics. At the same time, the importance of triggering factors is re-evaluated, the weights of the spatiotemporal attention mechanism are updated through SHAP value analysis, and incremental learning is performed on the graph neural network.

5. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The weather data includes temperature, humidity, and sunshine duration; the economic data includes GDP growth rate and industrial restructuring data; the social activity data includes holiday information and the timing of major sporting events; the user behavior pattern data includes the frequency of overtime work in offices and the distribution of household electricity consumption during certain periods; and the energy price fluctuation data includes real-time electricity prices and futures contract prices.

6. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The adjustment of the industrial flexible load includes: based on the adjustable load time period and power threshold declared by the user in advance and stored in the system database, sending a load adjustment instruction containing a priority identifier through a smart terminal. The priority is determined by a fuzzy logic algorithm according to the urgency of the supply and demand gap, and the high priority instruction corresponds to the scenario where the real-time supply and demand difference exceeds a preset threshold.

7. The adaptive adjustment method based on an energy management system according to claim 1, characterized in that, The energy production end linkage includes: generating power generation adjustment instructions, including adjustment range, response time window and reserve capacity reservation requirements, based on the predicted demand curve and the power generation equipment operating characteristic parameters, through the real-time communication interface between the edge computing node and the power generation equipment control system.

8. An adaptive regulation system based on an energy management system, applicable to the adaptive regulation method based on an energy management system as described in claims 1-7, characterized in that, The system includes: The data acquisition and processing module is used to collect multidimensional heterogeneous data and perform cleaning, preprocessing and standardization. The energy demand forecasting module includes a spatiotemporal attention mechanism unit and a graph neural network unit, which are used to generate energy demand forecasting models and output multi-period forecasting results. The adaptive adjustment module is used to execute a graded response strategy based on the prediction results, including an energy storage control unit, a load adjustment unit, and a production linkage unit; The feedback optimization module is used to collect actual supply and demand data, calculate prediction errors, and correct adjustment strategies and model parameters.

9. The adaptive adjustment method and system based on an energy management system according to claim 8, characterized in that, The energy storage control unit includes a distributed energy storage cluster control algorithm module, which generates an energy storage scheduling plan based on the predicted demand curve and real-time electricity price using a mixed integer programming algorithm. The load adjustment unit includes a user load profile database, which stores the flexible load parameters declared by users, historical adjustment response time, and error data.

Images