A flexible prediction method for industrial enterprise microgrid energy load model

Abstract

The application discloses a flexible prediction method for an industrial enterprise micro-grid energy load model, and particularly relates to the technical field of micro-grid energy load model flexible prediction. The application effectively distinguishes different physical stages of load climbing, inertia release and stable operation by continuously sampling and stage analysis of the load change process before and after the working condition switching of high inertia equipment, avoids misjudging the inertia slow change process as prediction model mismatch, simultaneously, quantifies the time decoupling degree between the production plan triggering and the actual load response by using the load inertia dominant factor, so that the flexible prediction model no longer blindly follows short-term fluctuations in the weight adjustment process, and realizes the continuous collaborative update of the load inertia characteristics and the prediction model parameters through the online correction mechanism based on the prediction residual.

Description

Technical Field

[0001] This invention relates to the field of flexible prediction technology for microgrid energy load models, and more specifically, to a flexible prediction method for energy load models of industrial enterprise microgrids. Background Technology

[0002] In the operation of microgrids in process industries such as steel, chemicals, and cement, the energy load structure is highly dependent on the operating status of large, high-inertia process equipment such as rotary kilns, compressors, and rolling mills. When these devices switch process sections, adjust operating conditions, or experience abnormal interventions, their electrical load does not exhibit an ideal instantaneous step change. Instead, it is constrained by multiple physical factors, including mechanical rotational inertia, the cumulative effect of heat capacity, and medium transport lag, resulting in a dynamic evolution process where power slowly climbs and the duration is difficult to define accurately in advance. During this process, the slope and duration of load changes are not entirely determined by control commands or production plans, but are dominated by the physical inertia release process of the equipment itself, leading to significant time decoupling between load response and information triggering. Existing load forecasting methods for industrial microgrids, especially forecasting models that introduce flexible weight adjustment mechanisms, are typically based on the assumption that load conditions can quickly respond to operating condition changes, dynamically adjusting the forecasting model parameters using recent data or status labels. However, this assumption is difficult to hold true in scenarios involving sudden switching of high-inertia equipment. Predictive models often fail to accurately perceive the inertia-driven load evolution rhythm in the initial stages of switching, easily leading to systematic underestimation or overestimation of the load growth process, thus affecting the stability of microgrid energy dispatch and energy storage control decisions. Therefore, how to effectively characterize the physical constraints of load changes in high-inertia process equipment within a flexible prediction framework remains a key technical problem that urgently needs to be solved in the energy load modeling and prediction of industrial enterprise microgrids. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a flexible prediction method for energy load models of microgrids in industrial enterprises, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A flexible prediction method for energy load models of industrial enterprise microgrids includes the following steps: The system collects the operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in industrial enterprise microgrids, continuously samples the load change process before and after equipment operating condition switching, and constructs a set of equipment load inertia response features. Based on the load inertia response feature set, the load change process is divided into stages, the load ramp-up stage, the inertia release stage and the stabilization stage are extracted, and the load change slope and duration of each stage are calculated to form a load inertia evolution description vector. The load inertia evolution description vector is compared and analyzed with the production plan triggering information to calculate the response lag between the planned load change and the actual load inertia response, and to construct the load inertia dominant factor. Based on the load inertia dominant factor and the trend of prediction residual change, the weights of prediction sub-models at different time scales in the flexible prediction model are dynamically adjusted. The prediction residuals of the flexible prediction model after weight adjustment are obtained, and the load inertia dominant factor is corrected online based on the prediction residuals to achieve adaptive updating of the flexible prediction model.

[0005] In a preferred embodiment, the process of collecting operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in an industrial enterprise microgrid, continuously sampling the load change process before and after equipment operating condition switching, and constructing the equipment load inertia response feature set is as follows: Electrical parameter acquisition units and equipment operation status acquisition units are deployed at the power supply nodes and high-inertia process equipment of industrial enterprise microgrids to continuously collect electrical load data and equipment operation status data of high-inertia process equipment. The acquisition unit is clock-aligned with the microgrid time synchronization system. A uniform sampling period is set for the electrical load data and equipment operating status data acquired by the acquisition unit, and a time identifier is added to each set of sampled data. When a device condition switching trigger signal is detected, the corresponding switching start time is marked, and a continuous observation window is set with the start time as the center to sample the load change process before and after the device condition switching to obtain complete load time series data. The load time series data is synchronized and associated with the equipment operating status data at the corresponding time to construct a load operating status corresponding sample, so that the electrical load value at each time corresponds to a clear physical operating status. Based on the samples corresponding to the load operating status, the samples are merged according to the equipment operating condition switching events to obtain the equipment load inertial response feature set.

[0006] In a preferred embodiment, the process of dividing the load change process into stages based on the load inertia response feature set, extracting the load ramp-up stage, inertia release stage, and stabilization stage, and calculating the load change slope and duration of each stage to form a load inertia evolution description vector is as follows: Based on the load inertial response feature set, continuous analysis is performed on the load time series data corresponding to the equipment operating condition switching event. Taking the starting time of the operating condition switching as the reference time point, a sliding window scan is performed on the load change trend over time. In the load time series data, the electrical load values ​​at adjacent sampling times are differentially processed to calculate the load change rate sequence, and the load change rate is statistically analyzed within a sliding time window. When the load change rate is greater than the first change rate threshold for a continuous time interval of not less than the length of the first preset window, and the window mean of the load change rate is monotonically decreasing, the corresponding time interval is determined to be the load ramp-up stage. When the load change rate decreases from greater than the first change rate threshold to between the first change rate threshold and the second change rate threshold within a continuous time interval, and the sign of the load change rate remains consistent within the continuous time interval, the corresponding time interval is determined to be the inertial release phase. When the absolute value of the load change rate is less than the second change rate threshold for a continuous time interval of not less than the second preset window length, and the window variance of the load change rate is less than the preset stability threshold, the corresponding time interval is determined to be a stable stage. For the load ramp-up phase, inertia release phase, and stabilization phase, linear fitting is performed on the load time series data within the corresponding time intervals to calculate the load change slope for each phase. The duration of each load phase is calculated by performing difference processing on the start and end times of each load phase. By combining the load change slope and duration corresponding to the load ramp-up phase, inertia release phase, and stabilization phase, a load inertia evolution description vector is obtained.

[0007] In a preferred embodiment, the process of comparing and analyzing the load inertia evolution description vector with the production plan triggering information, calculating the response hysteresis between the planned load-driven load change and the actual load inertia response, and constructing the load inertia dominant factor is as follows: Obtain production plan triggering information corresponding to equipment operating condition switching, including process segment switching instructions, effective timestamps and target operating load levels, and map the production plan triggering information to a time axis consistent with the load time series data to obtain a reference time series of plan-driven load changes; Based on the load inertia evolution description vector, the start time of the load ramp-up phase, the start time of the inertia release phase, and the start time of the stabilization phase are extracted, and the start time of each load phase is taken as the key time node of the actual load response. Using the production plan trigger timestamp as the theoretical starting point for plan-driven load changes, the time difference between the plan trigger timestamp and the start time of the load ramp-up phase is calculated to obtain the initial load response lag. By comparing the start time of the stable phase with the target load achievement time set in the production plan, the full load response lag is obtained. The load inertia dominance factor is constructed by combining the initial load response lag and the full load response lag, and by taking into account the duration ratio of the load ramp-up phase and the inertia release phase.

[0008] In a preferred embodiment, the process of dynamically adjusting the weights of the prediction sub-models at different time scales in the flexible prediction model based on the load inertia dominant factor and the trend of prediction residual changes is as follows: During the operation of the flexible prediction model, the prediction output value of each prediction sub-model at the current prediction time is obtained in real time, and the difference between the prediction output value and the actual measured microgrid load is calculated to obtain the prediction residual sequence. Based on the predicted residual sequence, the trend of the predicted residual within the sliding time window is analyzed, and the mean and standard deviation features of the predicted residual are extracted. If the mean of the predicted residuals is greater than the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is greater than the preset threshold for the standard deviation of the predicted residuals, then the prediction error is determined to be a systematic bias. If the mean of the predicted residuals is less than or equal to the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is less than or equal to the preset threshold for the standard deviation of the predicted residuals, then the predicted residuals are determined to be random fluctuations. Obtain the load inertia dominance factor corresponding to the current load evolution process, and determine the relative dominance of physical inertia factors and planned driving factors during the current load change process based on the load inertia dominance factor: When the load inertia dominance factor is greater than the preset inertia dominance threshold, it is determined that physical inertia is dominant. When the load inertia dominance factor is less than or equal to the preset inertia dominance threshold, the production plan is determined to be dominant. If the prediction error is a systematic bias and physical inertia dominates, then reduce the weight of the short-timescale prediction sub-model in the flexible prediction model, and correspondingly increase the weight of the medium-timescale and long-timescale prediction sub-models. If the predicted residuals fluctuate randomly and production planning dominates, then increase the weight of the short-timescale prediction sub-model in the flexible prediction model, and decrease the weight of the medium-timescale and long-timescale prediction sub-models.

[0009] In a preferred embodiment, the process of obtaining the prediction residuals of the weight-adjusted flexible prediction model and then correcting the load inertia dominant factor online based on the prediction residuals is as follows: After adjusting the weights of the flexible prediction model, the actual load data of the microgrid is acquired in real time, and the difference between the predicted output value and the actual load data of the microgrid is calculated to re-acquire the prediction residual sequence. Based on the predicted residual sequence, the predicted residuals are classified according to the load inertia stage, and the predicted residuals are respectively assigned to the residual sets of the load ramp-up stage, the inertia release stage and the stabilization stage, and the mean and standard deviation of the predicted residuals for each load stage are calculated. The load inertia dominant factor is then corrected online based on the mean and standard deviation of the predicted residuals. ,in For the first Mean of predicted residuals for each load phase For the first Standard deviation of each load phase The dominant load inertia factor at time t. This is the corrected load inertia dominant factor at time t+1.

[0010] The technical effects and advantages of this invention are as follows: 1. This invention introduces the physical constraint mechanism of load evolution of high-inertia process equipment into the flexible load forecasting framework, transforming the traditional information-triggered load forecasting process into a collaborative modeling process that takes into account both planning-driven and inertia-driven approaches, thereby achieving a more realistic depiction of the dynamic behavior of industrial enterprise microgrid loads.

[0011] 2. This invention, through continuous sampling and staged analysis of load changes before and after the switching of operating conditions of high-inertia equipment, can effectively distinguish different physical stages such as load ramp-up, inertia release, and stable operation. This avoids misjudging the gradual inertia process as a mismatch in the prediction model, thereby significantly reducing the risk of systematic overestimation or underestimation in the early stage of operating condition switching. At the same time, by using the load inertia-dominant factor to quantify the degree of time decoupling between production plan triggering and actual load response, the flexible prediction model no longer blindly follows short-term fluctuations during weight adjustment. Instead, it adaptively adjusts the role ratio of prediction sub-models at different time scales according to the degree of inertia constraint, improving the stability and consistency of prediction results in high-inertia scenarios. Through an online correction mechanism based on prediction residuals, continuous collaborative updating of load inertia characteristics and prediction model parameters is achieved, enabling the model to maintain long-term effectiveness when equipment ages, processes are adjusted, or external disturbance conditions change. Overall, this enhances the adaptability of industrial microgrid energy load prediction to complex operating conditions, providing a more reliable decision-making basis for energy dispatch, energy storage control, and safe operation. Attached Figure Description

[0012] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Example: Figure 1 This invention presents a flexible prediction method for energy load models of industrial enterprise microgrids, comprising the following steps: The system collects the operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in industrial enterprise microgrids, continuously samples the load change process before and after equipment operating condition switching, and constructs a set of equipment load inertia response features. Based on the load inertia response feature set, the load change process is divided into stages, the load ramp-up stage, the inertia release stage and the stabilization stage are extracted, and the load change slope and duration of each stage are calculated to form a load inertia evolution description vector. The load inertia evolution description vector is compared and analyzed with the production plan triggering information. The response lag between the planned load change and the actual load inertia response is calculated, and the load inertia dominant factor is constructed to characterize the degree to which load changes are constrained by physical inertia. Based on the load inertia dominant factor and the trend of prediction residual change, the weights of prediction sub-models at different time scales in the flexible prediction model are dynamically adjusted so that the response characteristics of the prediction model match the load inertia evolution process. The prediction residuals of the flexible prediction model after weight adjustment are obtained, and the load inertia dominant factor is corrected online based on the prediction residuals to achieve adaptive updating of the flexible prediction model.

[0015] In this embodiment of the invention, the process of collecting the operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in an industrial enterprise microgrid, continuously sampling the load change process before and after equipment operating condition switching, and constructing the equipment load inertia response feature set is as follows: Electrical parameter acquisition units and equipment operation status acquisition units are deployed at the power supply nodes and high-inertia process equipment of industrial enterprise microgrids to continuously collect electrical load data and equipment operation status data of high-inertia process equipment during normal operation and operation mode switching. The electrical load data includes active power, reactive power, current, voltage, etc. The equipment operating status data includes equipment speed, load rate, process temperature, and medium flow rate, etc. The acquisition unit is clock-aligned with the microgrid time synchronization system to ensure consistency of different data sources in the time dimension; A uniform sampling period is set for the electrical load data and equipment operating status data acquired by the acquisition unit, and a time identifier is added to each set of sampled data so that the load change process forms continuous and traceable time-series data on the time axis. Preferably, the sampling period is set according to the minimum time resolution requirement of load change during the typical operating condition switching process of the equipment, so as to ensure that the continuous evolution process of load over time can be completely recorded. When a device condition switching trigger signal is detected, the corresponding switching start time is marked, and a continuous observation window is set before and after the start time. The load change process before and after the device condition switching is sampled throughout, so as to obtain complete load time series data. The load time series data is synchronized and associated with the equipment operating status data at the corresponding time to construct a load operating status corresponding sample, so that the electrical load value at each time corresponds to a clear physical operating status. Based on the samples corresponding to the load operating status, the samples are merged according to the equipment operating condition switching events to obtain the equipment load inertia response feature set used to describe the entire load evolution process of high inertia process equipment during operating condition switching, which serves as the input data basis for subsequent load inertia evolution analysis and flexible prediction modeling. It should be noted that the physical significance of the load inertia response feature set is to provide a basic support for distinguishing between information-driven load changes and load changes dominated by physical inertia, thereby improving the adaptability and prediction stability of the industrial enterprise microgrid energy load flexible prediction model under high inertia switching scenarios.

[0016] In this embodiment of the invention, the process of dividing the load change process into stages based on the load inertia response feature set, extracting the load ramp-up stage, inertia release stage, and stabilization stage, and calculating the load change slope and duration of each stage to form a load inertia evolution description vector is as follows: Based on the load inertial response feature set, continuous analysis is performed on the load time series data corresponding to the equipment operating condition switching event. Taking the starting time of the operating condition switching as the reference time point, a sliding window scan is performed on the load change trend over time to identify the time series characteristics of the load change direction and intensity. In the load time series data, the electrical load values ​​at adjacent sampling times are differentially processed to calculate the load change rate sequence, and the load change rate is statistically analyzed within a sliding time window. When the load change rate is greater than the first change rate threshold for a continuous time interval of not less than the length of the first preset window, and the window mean of the load change rate is monotonically decreasing, the corresponding time interval is determined to be the load ramp-up stage, which is used to characterize the controlled growth process of the load of the equipment in the early stage of the working condition switch, which is triggered by the drive input and constrained by physical inertia. When the load change rate decreases from greater than the first change rate threshold to between the first change rate threshold and the second change rate threshold within a continuous time interval, and the sign of the load change rate remains consistent within the continuous time interval, the corresponding time interval is determined as the inertia release stage, which is used to describe the slow load change process caused by the gradual release of mechanical inertia, thermal inertia or fluid hysteresis effect inside the equipment under the condition that the external drive input remains basically unchanged. When the absolute value of the load change rate is less than the second change rate threshold for a continuous time interval of not less than the length of the second preset window, and the window variance of the load change rate is less than the preset stability threshold, the corresponding time interval is determined to be a stable stage, which is used to describe the process of the equipment load entering a relatively steady state after completing the inertial release. For the load ramp-up phase, inertia release phase, and stabilization phase, linear fitting is performed on the load time series data within the corresponding time intervals to calculate the load change slope for each phase, and the load change slope is used as a phase parameter describing the characteristics of the load change rate. Based on the predicted residual sequence, the predicted residuals are classified according to the load inertia stage, and the predicted residuals are respectively assigned to the residual sets of the load ramp-up stage, the inertia release stage and the stabilization stage, and the mean and standard deviation of the predicted residuals for each load stage are calculated. For example, for the first The slope of the load change during each load phase can be expressed as: ,in, and They represent the first The start and end times of each load phase Indicates the first Equipment electrical load values ​​at the start of each load phase Indicates the first Equipment electrical load values ​​at the end of each load phase; The duration of each load stage is calculated by performing difference processing on the start and end times of each load stage, which is used to characterize the proportion of different load evolution stages in the time dimension. By combining the load change slope and duration corresponding to the load ramp-up phase, inertia release phase and stabilization phase, a load inertia evolution description vector is obtained, which is used to comprehensively characterize the physical characteristics of the load evolution over time during the operation switching process of high inertia process equipment. It should be noted that the load inertia evolution description vector is not a simple piecewise fitting result of the load curve, but rather a distinction between the driving input influence and physical inertia release characteristics during the load change process through stage division and parameterization. This provides a physically meaningful quantitative basis for the subsequent construction of the load inertia dominant factor and the weight adjustment mechanism of the flexible prediction model.

[0017] In this embodiment of the invention, the process of comparing and analyzing the load inertia evolution description vector with the production plan triggering information, calculating the response hysteresis between the planned load change and the actual load inertia response, and constructing the load inertia dominant factor to characterize the degree to which load changes are constrained by physical inertia is as follows: Obtain production plan triggering information corresponding to equipment operating condition switching, including process segment switching instructions, effective timestamps and target operating load levels, and map the production plan triggering information to a time axis consistent with the load time series data to obtain a reference time series of plan-driven load changes; Based on the load inertia evolution description vector, the start time of the load ramp-up phase, the start time of the inertia release phase, and the start time of the stabilization phase are extracted, and the start time of each load phase is used as the key time node of the actual load response to describe the real evolution rhythm of the load during the switching of operating conditions. Using the production plan trigger timestamp as the theoretical starting point for plan-driven load changes, the time difference between the plan trigger timestamp and the start time of the load ramp-up phase is calculated to obtain the initial load response lag, which is used to characterize the first response delay of the load to the production plan command. For example, the initial load response hysteresis It can be represented as: ,in, For the production plan trigger timestamp, This marks the start of the load ramp-up phase. By comparing the start time of the stable phase with the target load achievement time set in the production plan, the full load response lag is obtained, which describes the time deviation required for the load to complete inertial release and reach the target operating state. For example, the load full response hysteresis It can be represented as: ,in, This marks the beginning of the stable phase. This refers to the expected time when the target load in the production plan will be achieved. By combining the initial load response lag and the full load response lag, and taking into account the duration ratio of the load ramp-up phase and the inertial release phase, a load inertia dominant factor is constructed to characterize the degree to which load changes are constrained by physical inertia. This factor can reflect the dominant relationship between planned load changes and actual inertial responses. For example, the load inertia dominant factor It can be represented as: ,in, The duration of the load ramp-up phase. This represents the total evolution time of the load from the ramp-up phase to the steady-state phase. These are the weighting coefficients for the ratio of the initial load response lag to the duration of the load ramp-up phase, and the ratio of the full load response lag to the total evolution time from the ramp-up phase to the steady-state phase, respectively, and they satisfy the following conditions: ; It should be noted that, The settings should be tailored to the specific circumstances. For example, an expert weighting method can be used, which involves inviting experts in relevant fields to determine the weighting coefficients of each indicator through professional opinion surveys and comprehensive evaluations. The initial value can be 0.5, 0.5; It should also be noted that the load inertia-dominant factor quantifies and aligns the production plan triggering sequence with the actual load evolution process, enabling the prediction model to distinguish between load changes directly driven by planning instructions and load changes dominated by equipment physical inertia. This provides a physically meaningful basis for the adaptive adjustment of prediction weights in the subsequent flexible prediction model.

[0018] In this embodiment of the invention, the process of dynamically adjusting the weights of the prediction sub-models at different time scales in the flexible prediction model according to the load inertia dominant factor and the trend of prediction residual changes, so that the response characteristics of the prediction model match the load inertia evolution process, is as follows: A flexible prediction model for load forecasting of microgrids in industrial enterprises is obtained. The flexible prediction model includes multiple prediction sub-models for different time scales, including at least a short-time scale prediction sub-model, a medium-time scale prediction sub-model, and a long-time scale prediction sub-model. Each prediction sub-model is used to characterize the rapid fluctuation characteristics, inertial slow change characteristics, and long-term trend characteristics of the load, respectively. During the operation of the flexible prediction model, the prediction output value of each prediction sub-model at the current prediction time is obtained in real time, and the difference between the prediction output value and the actual measured microgrid load is calculated to obtain the prediction residual sequence, which is used to reflect the fitting deviation of the prediction model to the current load change. Based on the predicted residual sequence, the trend of the predicted residual within the sliding time window is analyzed, and the mean and standard deviation features of the predicted residual are extracted to determine whether the prediction error exhibits systematic bias or random fluctuation characteristics. If the mean of the predicted residuals is greater than the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is greater than the preset threshold for the standard deviation of the predicted residuals, then the prediction error is determined to be a systematic bias that does not fluctuate over time. If the mean of the predicted residuals is less than or equal to the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is less than or equal to the preset threshold for the standard deviation of the predicted residuals, then the predicted residuals are determined to be random fluctuations and the error is in a relatively stable state. Obtain the load inertia dominance factor corresponding to the current load evolution process, and determine the relative dominance of physical inertia factors and planned driving factors during the current load change process based on the load inertia dominance factor: When the load inertia dominance factor is greater than the preset inertia dominance threshold, it is determined that physical inertia dominates the load change and has a greater impact on load evolution. When the load inertia dominance factor is less than or equal to the preset inertia dominance threshold, it is determined that the production plan and external input factors dominate the load change and occupy the dominant position. If the prediction error is a systematic bias and physical inertia dominates, then the weight of the short-timescale prediction sub-model in the flexible prediction model should be reduced, and the weight of the medium-timescale and long-timescale prediction sub-models should be increased accordingly to suppress the prediction bias caused by the overly fast response. It should be noted that when physical inertia dominates, it means that load changes are primarily governed by physical inertia. In this case, the system's load response is relatively slow, typically exhibiting a gradual change. If the prediction residuals show systematic bias, it indicates that the prediction model cannot effectively capture this inertial change process, resulting in persistent errors and significant fluctuations. Therefore, it is necessary to reduce the weight of short-timescale prediction sub-models, as short-timescale models are usually sensitive to rapidly changing loads, but in this case, actual load changes are dominated by inertia, leading to significant prediction bias in short-timescale models. Simultaneously, the weight of medium- or long-timescale prediction sub-models should be increased, as these sub-models can better fit the slow change process driven by inertial factors, thereby reducing prediction errors and improving the model's stability and accuracy. If the prediction residuals fluctuate randomly and production planning dominates, then increase the weight of the short-timescale prediction sub-model in the flexible prediction model, and decrease the weight of the medium-timescale and long-timescale prediction sub-models. It should be noted that when production planning dominates, it means that load changes are more driven by external plans or short-term disturbances, rather than slow changes dominated by physical inertia. In this case, the forecast residuals exhibit random fluctuations, indicating that system load changes are relatively frequent and do not have obvious trend deviations, with a small error fluctuation range. Therefore, the weight of the short-timescale forecast sub-model should be increased, because the short-timescale model can better track and capture the characteristics of rapidly changing load fluctuations. In this scenario, the fast-response forecast model can more accurately reflect short-term load fluctuations. At the same time, the weight of the medium- and long-timescale forecast sub-models should be reduced, because these sub-models are usually used to capture more stable or slowly changing load trends. In this case, the main characteristic of load changes is short-term fluctuation, and the longer-timescale forecast model may not provide sufficient accuracy or adaptability, thus introducing unnecessary errors. For example, the weight update of the prediction sub-model at each time scale can be expressed as: ,in, To predict the weights of the sub-model at time t, To predict the weights of the sub-model at time t+1, To predict the prediction residual of the sub-model at time t, The dominant factor of load inertia at time t; It should be noted that, Add a formula for weights. The formula for reducing weights; It should be noted that through the above-mentioned dynamic adjustment process of weights, the flexible prediction model exhibits slow response characteristics when load changes are dominated by physical inertia, and has rapid response capabilities when load changes are driven by plans or disturbances. This enables adaptive matching between the prediction model's response characteristics and the load inertia evolution process, thereby improving the stability and accuracy of energy load prediction for industrial enterprise microgrids under high inertia switching scenarios.

[0019] In this embodiment of the invention, the process of obtaining the prediction residuals of the flexible prediction model after weight adjustment, and then correcting the load inertia dominant factor online based on the prediction residuals to achieve adaptive updating of the flexible prediction model is as follows: After adjusting the weights of the flexible prediction model, the actual load data of the microgrid is acquired in real time, and the difference between the predicted output value and the actual load data of the microgrid is calculated to re-acquire the prediction residual sequence. Based on the predicted residual sequence, the predicted residuals are classified according to the load inertia stage, and the predicted residuals are respectively assigned to the residual sets of the load ramp-up stage, the inertia release stage and the stabilization stage, and the mean and standard deviation of the predicted residuals for each load stage are calculated. The load inertia dominant factor is then corrected online based on the mean and standard deviation of the predicted residuals. ,in, For the first Mean of predicted residuals for each load phase For the first Standard deviation of each load phase The dominant load inertia factor at time t. This is the corrected load inertia dominant factor at time t+1.

[0020] It should be noted that by continuously correcting the load inertia dominant factor, the prediction model can gradually adapt to the changes in inertia characteristics in the actual operating environment and output more accurate load prediction results. In particular, during the switching of operating conditions of high-inertia process equipment, it can effectively improve prediction accuracy and reduce the accumulation and amplification of prediction errors.

[0021] This invention introduces the physical constraint mechanism of load evolution of high-inertia process equipment into the flexible load forecasting framework, transforming the traditional information-triggered load forecasting process into a collaborative modeling process that takes into account both planning-driven and inertia-driven approaches, thereby achieving a more realistic portrayal of the dynamic load behavior of industrial enterprise microgrids.

[0022] This invention, through continuous sampling and phased analysis of load changes before and after switching operating conditions of high-inertia equipment, can effectively distinguish different physical stages such as load ramp-up, inertia release, and stable operation. This avoids misjudging the gradual inertia process as a mismatch in the prediction model, thus significantly reducing the risk of systematic overestimation or underestimation in the early stages of operating condition switching. Simultaneously, by utilizing the load inertia-dominant factor to quantify the time decoupling between production plan triggering and actual load response, the flexible prediction model no longer blindly follows short-term fluctuations during weight adjustment. Instead, it adaptively adjusts the proportion of prediction sub-models at different time scales based on the degree of inertia constraint, improving the stability and consistency of prediction results in high-inertia scenarios. Through an online correction mechanism based on prediction residuals, continuous collaborative updating of load inertia characteristics and prediction model parameters is achieved, enabling the model to maintain long-term effectiveness even with equipment aging, process adjustments, or changes in external disturbance conditions. Overall, this enhances the adaptability of industrial microgrid energy load prediction to complex operating conditions, providing a more reliable decision-making basis for energy dispatch, energy storage control, and safe operation.

[0023] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0024] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0025] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A flexible prediction method for energy load models of industrial enterprise microgrids, characterized in that: The steps include the following: The system collects the operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in industrial enterprise microgrids, continuously samples the load change process before and after equipment operating condition switching, and constructs a set of equipment load inertia response features. Based on the load inertia response feature set, the load change process is divided into stages, the load ramp-up stage, the inertia release stage and the stabilization stage are extracted, and the load change slope and duration of each stage are calculated to form a load inertia evolution description vector. The load inertia evolution description vector is compared and analyzed with the production plan triggering information to calculate the response lag between the planned load change and the actual load inertia response, and to construct the load inertia dominant factor. Based on the load inertia dominant factor and the trend of prediction residual change, the weights of prediction sub-models at different time scales in the flexible prediction model are dynamically adjusted. Obtain the prediction residuals of the flexible prediction model after weight adjustment, and correct the load inertia dominant factor online based on the prediction residuals; The process of comparing and analyzing the load inertia evolution description vector with the production plan triggering information, calculating the response hysteresis between the planned load-driven change and the actual load inertia response, and constructing the load inertia dominant factor is as follows: Obtain production plan triggering information corresponding to equipment operating condition switching, including process segment switching instructions, effective timestamps and target operating load levels, and map the production plan triggering information to a time axis consistent with the load time series data to obtain a reference time series of plan-driven load changes; Based on the load inertia evolution description vector, the start time of the load ramp-up phase, the start time of the inertia release phase, and the start time of the stabilization phase are extracted, and the start time of each load phase is taken as the key time node of the actual load response. Using the production plan trigger timestamp as the theoretical starting point for plan-driven load changes, the time difference between the plan trigger timestamp and the start time of the load ramp-up phase is calculated to obtain the initial load response lag. By comparing the start time of the stable phase with the target load achievement time set in the production plan, the full load response lag is obtained. The initial load response hysteresis and the full load response hysteresis are combined, and the duration ratio of the load ramp-up phase to the inertial release phase is combined to construct the load inertia dominant factor. Based on the load inertia dominant factor and the trend of prediction residual changes, the process of dynamically adjusting the weights of prediction sub-models at different time scales in the flexible prediction model is as follows: During the operation of the flexible prediction model, the prediction output value of each prediction sub-model at the current prediction time is obtained in real time, and the difference between the prediction output value and the actual measured microgrid load is calculated to obtain the prediction residual sequence. Based on the predicted residual sequence, the trend of the predicted residual within the sliding time window is analyzed, and the mean and standard deviation features of the predicted residual are extracted. If the mean of the predicted residuals is greater than the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is greater than the preset threshold for the standard deviation of the predicted residuals, then the prediction error is determined to be a systematic bias. If the mean of the predicted residuals is less than or equal to the preset threshold for the mean of the predicted residuals, and the standard deviation of the predicted residuals is less than or equal to the preset threshold for the standard deviation of the predicted residuals, then the predicted residuals are determined to be random fluctuations. Obtain the load inertia dominance factor corresponding to the current load evolution process, and determine the relative dominance of physical inertia factors and planned driving factors during the current load change process based on the load inertia dominance factor: When the load inertia dominance factor is greater than the preset inertia dominance threshold, it is determined that physical inertia is dominant. When the load inertia dominance factor is less than or equal to the preset inertia dominance threshold, the production plan is determined to be dominant. If the prediction error is a systematic bias and physical inertia dominates, then reduce the weight of the short-timescale prediction sub-model in the flexible prediction model, and correspondingly increase the weight of the medium-timescale and long-timescale prediction sub-models. If the predicted residuals fluctuate randomly and production planning dominates, then increase the weight of the short-timescale prediction sub-model in the flexible prediction model, and decrease the weight of the medium-timescale and long-timescale prediction sub-models.

2. The flexible prediction method for energy load models of industrial enterprise microgrids according to claim 1, characterized in that: The process of collecting operating status parameters and corresponding electrical load time-series data of high-inertia process equipment in industrial enterprise microgrids, continuously sampling the load change process before and after equipment operating condition switching, and constructing the equipment load inertia response feature set is as follows: Electrical parameter acquisition units and equipment operation status acquisition units are deployed at the power supply nodes and high-inertia process equipment of industrial enterprise microgrids to continuously collect electrical load data and equipment operation status data of high-inertia process equipment. The acquisition unit is clock-aligned with the microgrid time synchronization system. A uniform sampling period is set for the electrical load data and equipment operating status data acquired by the acquisition unit, and a time identifier is added to each set of sampled data. When a device condition switching trigger signal is detected, the corresponding switching start time is marked, and a continuous observation window is set with the start time as the center to sample the load change process before and after the device condition switching to obtain complete load time series data. The load time series data is synchronized and associated with the equipment operating status data at the corresponding time to construct a load operating status corresponding sample, so that the electrical load value at each time corresponds to a clear physical operating status. Based on the samples corresponding to the load operating status, the samples are merged according to the equipment operating condition switching events to obtain the equipment load inertial response feature set.

3. The flexible prediction method for energy load models of industrial enterprise microgrids according to claim 2, characterized in that: Based on the load inertia response feature set, the load change process is divided into stages, extracting the load ramp-up stage, inertia release stage, and stabilization stage. The load change slope and duration of each stage are calculated to form the load inertia evolution description vector. Based on the load inertial response feature set, continuous analysis is performed on the load time series data corresponding to the equipment operating condition switching event. Taking the starting time of the operating condition switching as the reference time point, a sliding window scan is performed on the load change trend over time. In the load time series data, the electrical load values ​​at adjacent sampling times are differentially processed to calculate the load change rate sequence, and the load change rate is statistically analyzed within a sliding time window. When the load change rate is greater than the first change rate threshold for a continuous time interval of not less than the length of the first preset window, and the window mean of the load change rate is monotonically decreasing, the corresponding time interval is determined to be the load ramp-up stage. When the load change rate decreases from greater than the first change rate threshold to between the first change rate threshold and the second change rate threshold within a continuous time interval, and the sign of the load change rate remains consistent within the continuous time interval, the corresponding time interval is determined to be the inertial release phase. When the absolute value of the load change rate is less than the second change rate threshold for a continuous time interval of not less than the second preset window length, and the window variance of the load change rate is less than the preset stability threshold, the corresponding time interval is determined to be a stable stage. For the load ramp-up phase, inertia release phase, and stabilization phase, linear fitting is performed on the load time series data within the corresponding time intervals to calculate the load change slope for each phase. The duration of each load phase is calculated by performing difference processing on the start and end times of each load phase. By combining the load change slope and duration corresponding to the load ramp-up phase, inertia release phase, and stabilization phase, a load inertia evolution description vector is obtained.

4. The flexible prediction method for energy load models of industrial enterprise microgrids according to claim 1, characterized in that: The process of obtaining the prediction residuals of the weight-adjusted flexible prediction model and then correcting the load inertia dominant factor online based on the prediction residuals is as follows: After adjusting the weights of the flexible prediction model, the actual load data of the microgrid is acquired in real time, and the difference between the predicted output value and the actual load data of the microgrid is calculated to re-acquire the prediction residual sequence. Based on the predicted residual sequence, the predicted residuals are classified according to the load inertia stage, and the predicted residuals are respectively assigned to the residual sets of the load ramp-up stage, the inertia release stage and the stabilization stage, and the mean and standard deviation of the predicted residuals for each load stage are calculated. The load inertia dominant factor is then corrected online based on the mean and standard deviation of the predicted residuals. ,in, For the first Mean of predicted residuals for each load phase For the first Standard deviation of each load phase The dominant load inertia factor at time t. This is the corrected load inertia dominant factor at time t+1.

Images