Power distribution system flexible regulation and control system based on load side behavior recognition and method thereof

Abstract

The invention discloses a power distribution system flexible regulation and control system based on load side behavior recognition and a method thereof, and relates to the technical field of power distribution of power systems. The method comprises the following steps: acquiring power utilization power, equipment state, environment parameters and power utilization preference information of a user side in real time; dynamically generating and updating behavior inertia factor data, and reporting high-frequency change data when the behavior inertia factor data exceeds a preset threshold; receiving regional load feature abstract data, and uploading the abstract data when the change of the abstract data exceeds a threshold value; historical period abstract data are fused, and a prediction matrix containing the partition load trend and the peak probability in the T time window is generated through a behavior inertia dynamic evolution model; and when the load rate of the system exceeds a safety threshold value, calculating an individual adjustment amplitude, and generating a regulation and control instruction containing a time-phased target and a flexible adjustment amplitude. Accurate prediction and flexible regulation and control of the load of the power distribution system are realized, and the operation stability and the power supply quality of the system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system distribution technology, and in particular to a flexible control system for a power distribution system based on load-side behavior identification and a method thereof. Background Art

[0002] In existing distribution system control technologies, traditional methods for sensing user electricity usage behavior remain at the static sampling stage, unable to dynamically capture transient load changes caused by equipment switching and sudden environmental changes. User electricity preference information is used only as offline reference data and is not integrated into real-time decision-making systems. Furthermore, the lack of quantitative modeling of load inertia makes it difficult to predict correlated load fluctuations such as air conditioning cluster startup and production line switching, resulting in significant lags in early warning of regional load changes.

[0003] Current load forecasts rely on extrapolation of historical data, without establishing a dynamic correlation model between environmental parameters and electricity consumption. Forecasts fail to reflect load evolution during holidays, extreme weather events, and other scenarios. Furthermore, forecast errors accumulate and lack a self-healing mechanism. When system load exceeds a limit, control strategies often simply shed load without considering equipment operating status constraints, potentially disrupting industrial production or causing a sudden drop in residential comfort.

[0004] Existing methods generate control commands based solely on the overall system load level, disregarding the physical constraints imposed by user equipment operating conditions (such as minimum air conditioner operating power and minimum production line start / stop cycles), individualized power preferences (temperature sensitivity and production priorities), and variations in compliance with historical control responses. This rigid control strategy results in frequent starts and stops of refrigeration equipment, damaging compressors, forcing manufacturing lines to interrupt precision processing, and severely degrading residents' comfort when they actively unload their loads. Furthermore, the lack of a feedback correction mechanism traps the system in a vicious cycle of "control-oscillation."

[0005] In response to the above problems, this field urgently needs to establish a flexible control system that integrates equipment characteristics, user preferences and compliance portraits, to achieve efficient matching load regulation while ensuring system safety, and to completely solve the problems of equipment loss and user resistance caused by rigid control. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides a flexible control method for a power distribution system based on load-side behavior identification.

[0007] In order to achieve the above object, the technical solution of the present invention is as follows: In a first aspect, the present invention discloses a method for flexible control of a power distribution system based on load-side behavior identification, comprising the following steps: Collect power consumption data, equipment operation status data and environmental parameter data from the user side in real time, and obtain user power preference information; Generate behavioral inertia factor data based on the power consumption data and the environmental parameter data, and dynamically update it according to a preset period; When the behavior inertia factor data exceeds a pre-stored local preset threshold, high-frequency change data is generated, and the high-frequency change data is reported to the edge node through an event-driven mechanism; Receive regional load characteristic summary data of the edge node; wherein the regional load characteristic summary data is generated by collecting the high-frequency change data of multiple user sides through local cluster analysis; Determine whether the regional load feature summary data in the current cycle exceeds the edge preset threshold. If so, fuse multiple regional load feature summary data from past cycles and input them into the pre-trained behavioral inertia dynamic evolution model to generate a partition load prediction matrix; wherein the partition load prediction matrix includes the partition load trend and peak probability within the T time window; The system load rate is calculated based on the partition load prediction matrix. When the system load rate exceeds the safety threshold, the adjustable load range is determined by using the partition load trend and the equipment operation status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information, and a control instruction including the time period adjustment target and the individual flexible adjustment range is generated.

[0008] Furthermore, the specific steps of generating behavioral inertia factor data based on the power consumption data and the environmental parameter data, and dynamically updating the data according to a preset period include: Calculating the power change gradient per minute of the electric power data; When the power change gradient exceeds the fluctuation threshold, the load jump frequency count is refreshed; Calculate the cumulative duration of excessive power on the day based on the load jump frequency, update the continuous high load duration, and generate a current power curve; Calculating a Pearson correlation coefficient between the current power curve and a pre-stored historical reference curve, and storing the Pearson correlation coefficient as a curve similarity parameter for dynamically updating the behavioral inertia factor data; The behavioral inertia factor data includes continuous high-load operation time, load jump frequency and intra-day curve similarity.

[0009] Furthermore, when the behavior inertia factor data exceeds a pre-stored local preset threshold, high-frequency change data is generated, and the high-frequency change data is reported to the edge node through an event-driven mechanism. The event-driven mechanism specifically includes: The local preset thresholds include a change rate absolute value threshold and a power absolute value threshold; When any preset threshold of the change rate absolute value threshold and the power absolute value threshold is exceeded, the upload of the high-frequency change data is triggered, and the high-frequency change data includes the power value of the behavioral inertia factor data and high-frequency data based on the equipment operation status data mark.

[0010] Furthermore, the specific steps of generating the regional load characteristic summary data by collecting the high-frequency change data of a plurality of the user sides through local cluster analysis include: The user side divides users into groups based on electrical and physical locations, and uses the OPTICS algorithm to automatically identify the core area of ​​user group density; Extracting the average load value of each group in the user group density core area within an N-minute sliding window as a window benchmark; Executing the local cluster analysis to calculate the standard deviation within the sliding window to generate a load fluctuation amplitude index; Analyzing the phase synchronization of the user load curve within the user group density core area by the Pearson correlation coefficient, and outputting a user linkage index; The calculation result of the user linkage index is encapsulated as the regional load characteristic summary data.

[0011] Furthermore, it is determined whether the regional load characteristic summary data in the current period exceeds a preset edge threshold. If yes, the regional load characteristic summary data is uploaded through a differential update mechanism. The specific steps performed by the differential update mechanism include: Retrieve a pre-stored cached copy of the historical summary; The regional load characteristic summary data in the current period is compared with the historical summary cache copy. When any of the following conditions is met, the regional load characteristic summary data is uploaded and marked for update: (1) The fluctuation amplitude changes by more than the first preset ratio; (2) The change in the user linkage indicator exceeds the second preset ratio; (3) The average load change exceeds the third preset ratio; Otherwise, a no-update confirmation signal is sent and the system waits for the next cycle.

[0012] Furthermore, the fusion of the plurality of regional load feature summary data of past cycles is input into a pre-trained behavior inertia dynamic evolution model to generate a partition load prediction matrix. The pre-trained behavior inertia evolution model specifically includes: encoding the user inertia state of the user side into a dynamic evolution vector based on the user power preference information; The dynamic evolution vector and the regional load characteristic summary data are weightedly integrated with the environmental disturbance factor, the time weighting factor and the holiday correction factor through the inertial state transfer equation; The partition load trend in a future period is calculated and the peak probability is output, where the peak probability includes a prediction matrix of the partition peak time and amplitude.

[0013] Furthermore, the system load rate is calculated based on the partition load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partition load trend and the equipment operation status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information. A control instruction including a time-based adjustment target and an individual flexible adjustment range is generated. The time-based adjustment target generation includes: identifying peak periods in the partitioned load forecast matrix; Prioritize the allocation of the golden control window with a preset time radius during the peak period; The enforcement priority label of the golden control window is marked in the control instruction.

[0014] Furthermore, the system load rate is calculated based on the partitioned load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partitioned load trend and the equipment operating status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information. A control instruction including a time-based adjustment target and an individual flexible adjustment range is generated. The individual flexible adjustment range specifically includes: Dividing the control priority zones according to the zone load forecast matrix; Generate a user's historical behavior profile based on the user's electricity usage preference information and set a personalized adjustment range; The individual flexibility adjustment amplitude calculation satisfies: ; in, is the maximum adjustable range, is the preference compromise coefficient, which is determined by the user's historical regulation compliance. is the current power, It is a historical peak; A control instruction set including a target load reduction value and a target time window is generated.

[0015] Furthermore, the method for flexible control of a power distribution system based on load-side behavior identification further includes: Sending the control instruction to the user side, and receiving actual control execution data generated by the user side executing the control instruction; The actual control execution data is collected and fed back to the partition load forecast matrix for real-time correction. The specific steps of the real-time correction include: Using the actual control execution data as a correction factor in the next forecast period; Inputting the correction factor into the partition load forecast matrix as a boundary constraint; Dynamically compress the floating interval of the subsequent partition load prediction matrix to a preset range.

[0016] In a second aspect, the present invention discloses an intelligent chassis control system with adaptive posture leveling, which uses the above-mentioned flexible control method of the power distribution system based on load-side behavior recognition, including: The data collection module is used to collect power consumption data, equipment operation status data and environmental parameter data on the user side in real time, and obtain user power preference information; an inertia processing module, configured to generate behavioral inertia factor data based on the power consumption data and the environmental parameter data, and dynamically update the data according to a preset period; An edge analysis module is configured to generate high-frequency change data when the behavior inertia factor data exceeds a pre-stored local preset threshold, and report the high-frequency change data to an edge node through an event-driven mechanism; A feature receiving module is configured to receive regional load feature summary data of the edge node; wherein the regional load feature summary data is generated by collecting the high-frequency change data of multiple user sides through local cluster analysis; A judgment fusion module is configured to, if the regional load feature summary data in the current cycle exceeds a preset edge threshold, fuse multiple regional load feature summary data from past cycles and input them into a pre-trained behavioral inertia dynamic evolution model to generate a partitioned load prediction matrix; wherein the partitioned load prediction matrix includes the partitioned load trend and peak probability within the T time window; A decision module is used to calculate the system load rate based on the partition load forecast matrix, determine the adjustable load range based on the partition load trend and the equipment operation status data when the safety threshold is exceeded, calculate the individual maximum adjustment range based on the user's electricity preference information, and generate a control instruction including the time period adjustment target and the individual flexible adjustment range.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention establishes a dynamic update mechanism for behavioral inertia factors, correlating power consumption with environmental parameter changes in real time, achieving a transition from passive monitoring to active behavior prediction. Specifically, under this event-driven mechanism, edge nodes can directly capture micro-behavioral characteristics such as air conditioning group control and production line switching, enabling the system to anticipate load fluctuations rather than respond lagging behind, thereby improving state perception accuracy from the source.

[0018] 2. This invention utilizes innovative closed-loop optimization of prediction and control, enabling the prediction model weights to be automatically adjusted based on measured deviations. Compared to traditional static models, this mechanism enables the model to continuously evolve, particularly adapting to load uncertainties caused by the integration of new energy sources. After multiple rounds of closed-loop iterations, the fit between the prediction curve and the actual load is significantly improved, providing a reliable basis for control decisions.

[0019] 3. This invention establishes a flexible command generation mechanism that integrates multiple factors. Based on load trend forecasts, it superimposes equipment operating status constraints and user preference profiles to create a dual-layer protection. Through flexible calculation of individual adjustment ranges, it preserves key user requirements while ensuring system security. The resulting cascaded decision-making system balances global optimization with local adaptability, enabling flexible and scientific regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0021] Figure 1 is a flow chart of the steps of the present invention;

[0022] Figure 2 A flow chart of the steps for generating and processing regional load characteristics of the present invention;

[0023] Figure 3 Flowchart of the generation process of the control instructions of the present invention;

[0024] Figure 4 This is a system module connection diagram of the present invention;

[0025] Figure 5 It is a flow chart of the system modules of the present invention. DETAILED DESCRIPTION

[0026] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0027] Application Overview:

[0028] Existing technologies often rely on rigid load control or static load forecasting models, making it difficult to balance system security with user electricity demand. Traditional methods lack the ability to dynamically detect the inertia of load changes when user electricity behavior changes suddenly, resulting in a significant lag in capturing regional load characteristics. Furthermore, existing systems are unable to effectively identify correlated load fluctuation patterns (such as when air conditioner clusters start up or production line switches). Especially during peak and off-peak periods, when loads fluctuate dramatically, single-point threshold control methods are prone to equipment shock or control failure, making it difficult to meet the flexible and interactive requirements of smart distribution networks.

[0029] To address the above issues, the inventors discovered a strong correlation between user load change behavior and environmental parameters and historical habits, and achieved precise control by establishing a dynamic mapping model between behavioral inertia factors and flexible adjustment ranges. During the research process, it was found that high-frequency electricity consumption behavior and regional load characteristics have spatiotemporal correlations, and edge node preprocessing can effectively extract key change characteristics; more importantly, the user's responsiveness to control instructions (preference compromise coefficient) and the proportion of real-time electricity consumption to historical peak values ​​are nonlinearly related. This proposed approach of dynamically calculating the individual flexible adjustment range based on electricity consumption status, verified through a feedback mechanism, and integrating equipment operation constraints and user preferences into the calculation of the adjustable load range to form a closed-loop optimization architecture.

[0030] Specifically, the control system first synchronously collects user-side power consumption data, equipment operating status data, and environmental parameter data to dynamically generate and update behavioral inertia factor data. When the behavioral inertia factor exceeds a locally preset threshold, it reports high-frequency changes to the edge node through an event-driven mechanism. The edge node generates regional load characteristic summary data through local cluster analysis. When the summary data changes beyond the edge's preset threshold, it is uploaded to the master control center using a differential update mechanism. The master control center integrates regional data from multiple periods and inputs it into a pre-trained behavioral inertia dynamic evolution model to generate a zoned load forecast matrix. The system periodically obtains measured load data to verify forecast deviations and automatically adjusts model weight parameters. When the system load rate exceeds a safety threshold, it determines the adjustable load range based on zoned load trends and equipment status. It then calculates the individual maximum adjustment range based on user electricity preferences. Finally, it generates and issues control instructions containing time-based targets and flexible adjustment ranges. Actual control data is fed back to the prediction model in real time, forming a dynamic calibration loop.

[0031] Compared with traditional technologies, existing solutions rely on static models and lack user behavior portraits, resulting in insufficient control accuracy when electricity consumption behavior suddenly changes. This solution innovatively integrates multi-source heterogeneous data, and realizes dynamic extraction and compressed transmission of key features through event-driven and differential update mechanisms; it originalizes the dynamic evolution model of behavioral inertia, and calculates the flexible adjustment amplitude based on the dual dimensions of device constraints and user preferences. Different from the existing rigid control mode, this solution can intelligently switch prediction-control strategies according to the real-time power consumption status, and continuously optimize the model parameters through a feedback closed loop, so that the system has the ability to evolve collaboratively. This application provides reliable flexible interactive technical support for smart distribution networks. On the premise of ensuring system safety, it avoids equipment damage such as frequent start and stop of air-conditioning compressors, and maintains the continuity of the production process, thus achieving the control goal of flexible control.

[0032] After introducing the basic concept of the present invention, embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0033] Example 1:

[0034] like Figure 1 As shown, the flexible control method of the distribution system based on load side behavior identification includes the following steps:

[0035] Collect power consumption data, equipment operation status data and environmental parameter data from the user side in real time, and obtain user power preference information;

[0036] Generate behavioral inertia factor data based on power consumption data and environmental parameter data, and dynamically update it according to the preset cycle;

[0037] When the behavioral inertia factor data exceeds the pre-stored local preset threshold, high-frequency change data is generated and reported to the edge node through the event-driven mechanism;

[0038] Receiving regional load characteristic summary data of the edge node; wherein the regional load characteristic summary data is generated by collecting high-frequency change data of multiple user sides through local cluster analysis;

[0039] Determine whether the regional load feature summary data in the current cycle exceeds the edge preset threshold. If so, fuse the regional load feature summary data of multiple past cycles and input the pre-trained behavioral inertia dynamic evolution model to generate a partition load prediction matrix; the partition load prediction matrix includes the partition load trend and peak probability within the T time window;

[0040] The system load rate is calculated based on the partition load forecast matrix. When the system load rate exceeds the safety threshold, the adjustable load range is determined by the partition load trend and equipment operation status data. The individual maximum adjustment range is calculated based on the user's electricity preference information, and a control instruction is generated that includes the time-sharing adjustment target and the individual flexible adjustment range.

[0041] Electricity power data refers to the series of active power values ​​of various user electrical devices collected in real time by smart meters. This data can be obtained using high-frequency sampling (≥1kHz) and RMS conversion techniques to reflect instantaneous load fluctuations. Equipment operating status data refers to the power-on / off status, operating mode, and operating parameters of electrical devices. Key information such as compressor start / stop times and motor speed can be obtained through IoT communication protocols (such as Modbus and MQTT) to assess the device's adjustable potential. Environmental parameter data refers to physical quantities such as temperature, humidity, and light intensity that are strongly correlated with electricity usage behavior. This data can be collected in real time through distributed sensor networks and used to correct load baselines for environmentally sensitive equipment such as air conditioners and lighting. User electricity preference information refers to strategic parameters pre-set by users, such as device usage time preferences and interruption tolerance. This data can be collected through the mobile app configuration interface to quantify individual acceptance thresholds for flexible adjustment. Behavioral inertia factor data refers to a dynamic indicator generated using a sliding window algorithm based on power data and environmental parameters. This data can be constructed using covariance analysis and a Markov transition probability model to characterize the temporal inertia of user electricity usage behavior. High-frequency change data refers to abnormal data packets triggered when the behavioral inertia factor exceeds a locally preset threshold (e.g., ±3σ). This data can be encapsulated using an event-driven transport protocol (e.g., CoAP) and used to report sudden changes in electricity consumption to edge nodes. Regional load characteristic summary data refers to the reduced-dimensional feature set extracted by edge nodes through DBSCAN clustering of multi-user high-frequency data. It can include statistics such as load change slope and peak-to-valley difference ratio, characterizing regional load fluctuation patterns. The partitioned load prediction matrix is ​​a multidimensional prediction result generated by integrating multi-period summary data through an LSTM model. It includes a 96-dimensional load trend curve and a peak probability matrix for the future T time window (e.g., 15 minutes), quantifying the load evolution patterns of the partitioned load. The adjustable load range refers to the interruptible load capacity determined based on the partitioned load trend forecast and equipment physical constraints. The upper limit of the total air conditioner group control capacity is calculated using electrical characteristic equations (e.g., the minimum compressor downtime τ min) to ensure safe equipment operation. Individual flexible adjustment range refers to the personalized adjustment amount calculated by combining user preference information and system requirements to generate a device-level adjustment instruction set to achieve humanized load control.

[0042] The core innovation of this application lies in constructing a closed-loop control architecture that uploads data to edge nodes for edge feature compression through local threshold perception and finally makes dynamic predictions at the center. It captures sudden changes in electricity consumption behavior in real time through behavioral inertia factors, and generates differentiated adjustment strategies based on the dual dimensions of equipment physical constraints and user preferences to solve the problems of response lag and user resistance caused by the rigid regulation of traditional distribution networks.

[0043] like Figure 2 and Figure 3 As shown, the working process and principle of this application are as follows: first, the power consumption data, equipment operation status data and environmental parameter data of the user side are collected in real time, and the user's power preference information is obtained; based on the power consumption data and environmental parameter data, behavioral inertia factor data is generated, and dynamically updated according to a preset period; when the behavioral inertia factor data exceeds the local preset threshold, high-frequency change data is generated and reported to the edge node through an event-driven mechanism; the edge node collects high-frequency change data from multiple user sides through local clustering analysis to generate regional load feature summary data; it is determined whether the regional load feature summary data of the current period exceeds the edge preset threshold. If so, the historical summary data is integrated and input into the behavioral inertia dynamic evolution model to generate a partition load prediction matrix; the system load rate is calculated based on the prediction matrix, and when it exceeds the safety threshold, the adjustable load range is determined in combination with the partition load trend and the equipment status; the individual maximum adjustment range is calculated in combination with the user's power preference information, and a control instruction including the time-sharing adjustment target and the flexible adjustment range is generated.

[0044] This application further proposes that the specific steps of generating behavioral inertia factor data based on power consumption data and environmental parameter data and dynamically updating it according to a preset period include:

[0045] Calculate the power change gradient of the electric power data every minute;

[0046] When the power change gradient exceeds the fluctuation threshold, the load jump frequency count is refreshed;

[0047] Calculate the cumulative duration of excessive power on the day based on the load jump frequency, update the continuous high load duration, and generate the current power curve;

[0048] Calculate the Pearson correlation coefficient between the current power curve and the pre-stored historical benchmark curve, and store the Pearson correlation coefficient as a curve similarity parameter for dynamically updating the behavioral inertia factor data;

[0049] Behavioral inertia factor data include continuous high-load operation duration, load jump frequency and intra-day curve similarity.

[0050] Specifically, the power change gradient is calculated to capture the user's minute-by-minute power load fluctuation intensity in real time. The formula is as follows:

[0051]

[0052] in is the power change gradient at time t, is the power consumption at time t, is the power consumption at time t-1, is a minute-level time step. Exceeding the preset fluctuation threshold When the load jump frequency counter is triggered, the load jump frequency The counter is refreshed incrementally. It accumulates the number of high-frequency fluctuation events through a sliding window mechanism (such as the time window of 0:00-24:00 on the same day) to reflect the instantaneous volatility of users' electricity consumption behavior.

[0053] based on The system will calculate the duration of the excessive power on that day. (i.e., the cumulative duration of power change gradient exceeding the threshold continuously), combined with the duration of continuous high load (The duration of the current load ≥ 80% of the rated load) to jointly construct the time dimension characteristics that reflect the continuity of the user's power intensity. The above time series data is further generated through the curve reconstruction algorithm to generate the current power curve The curve uses time as the horizontal axis and power as the vertical axis to intuitively present the user's daily electricity usage pattern.

[0054] To quantify the continuity between current electricity consumption behavior and historical habits, the system introduces the Pearson correlation coefficient r as the intraday curve similarity parameter, and its formula is as follows:

[0055]

[0056] in, is the historical baseline curve, and n is the length of the time series (e.g., if a 1-minute sampling period is used for a 1-hour period, then n=60). When r approaches 1, it indicates that the current electricity consumption curve is highly consistent with the historical pattern, indicating strong user behavior inertia; otherwise, it indicates a significant shift in the behavior pattern.

[0057] in, is the real-time power value of the day at the i-th time point, and the power curve of the day generated by the curve reconstruction algorithm ; The power curve for the day The average power value, , used to reflect the average power consumption intensity of the user on the day, is the historical benchmark power value at the i-th time point, and the pre-established historical benchmark curve (t), select the electricity consumption data of the same type of day in history (such as the same day last week), Historical benchmark curve The average power value of (t), , reflecting the user's typical electricity consumption level during the same period in history.

[0058] Historical benchmark curve The calculation formula of (t) is as follows:

[0059] represents the power value at time t on day d; Represents the time decay weight (new data has higher weight), calculated as: , D represents the total number of historical days participating in the weighted average, and the value is Based on the results of optimization of exponential smoothing coefficients in time series analysis.

[0060] The above continuous high-load operation time (Duration of power ≥ 80% of the rated power of the equipment), load jump frequency The number of times the power change gradient exceeds the threshold per unit time (for example, a power change exceeding 1.5kW within 1 minute is counted as one jump) and the Pearson correlation coefficient r (generated using the above formula) together constitute the behavioral inertia factor data. , and uses the exponential smoothing algorithm according to the preset period of 15 minutes , is the smoothing coefficient, with a value of [0,1]) and is updated dynamically, taking into account both historical inertia and real-time changes.

[0061] Inertia factor data When r approaches 1, it means that the current electricity consumption curve is highly consistent with the historical pattern, the user behavior inertia is strong, and the system can make accurate predictions based on historical data; when r approaches 0 or a negative value, the electricity consumption behavior deviates significantly from the historical pattern, and the prediction model parameters need to be dynamically adjusted to trigger the special event handling mechanism.

[0062] In the exponential smoothing algorithm middle, Represents the new predicted value, which is used to output the prediction result of the next cycle. Represents the historical forecast value, which is the forecast result of the previous period. Indicates the current actual value, which is the real-time collected power consumption monitoring data; when the power consumption behavior is stable ( ≈ ), the prediction results remain continuous; when the electricity consumption behavior changes suddenly ( Significant deviation ), and the forecast results quickly respond to new trends.

[0063] This application solves the problem of a single threshold being unable to capture the intensity of fluctuations through real-time linkage between gradient analysis and frequency statistics. By combining curve similarity measurement with time series reconstruction, it breaks through the limitations of traditional methods that only focus on extreme values ​​while ignoring pattern continuity. Through a periodic dynamic update mechanism, the behavioral inertia factor can adapt to seasonal adjustments in users' electricity usage habits (such as increased air conditioning load in summer) or temporary changes (such as short-term high loads caused by family gatherings). The various parameters interact progressively, with the fluctuation gradient triggering frequency counting, which drives duration calculation. Duration and historical curves jointly participate in similarity evaluation. The final output of the behavioral inertia factor provides key input for subsequent event-driven reporting, regional load clustering, and other links that reflect the essential characteristics of users' electricity usage behavior, effectively improving the distribution system's sensitivity to changes in load-side behavior and the adaptability of control strategies.

[0064] The present application further proposes that when the behavioral inertia factor data exceeds a pre-stored local preset threshold, high-frequency change data is generated and reported to the edge node through an event-driven mechanism. The event-driven mechanism specifically includes:

[0065] The local preset thresholds include the absolute value threshold of the rate of change and the absolute value threshold of the power;

[0066] When any preset threshold of the absolute value threshold of the change rate and the absolute value threshold of the power is exceeded, the upload of high-frequency change data is triggered. The high-frequency change data includes the power value of the behavioral inertia factor data and high-frequency data based on the device operation status data mark.

[0067] Specifically, the local preset threshold is composed of the absolute value threshold of the change rate and power absolute value threshold Among them, the absolute value threshold of the rate of change is used to capture the instantaneous fluctuation intensity of the load ( is the power change gradient at time t), when When , it indicates that the user's electricity consumption behavior has changed dramatically; the power absolute value threshold The absolute level of load is monitored directly (e.g. Indicates that the current load exceeds the safe operating benchmark value). The two thresholds characterize abnormal power consumption from the two dimensions of "change speed" and "operation intensity", forming complementary trigger conditions.

[0068] When any preset threshold is exceeded (i.e. or ), the system immediately generates high-frequency change data and reports it to the edge node through an event-driven mechanism. This high-frequency data is not a simple raw power value, but a composite information that combines behavioral inertia factors and device status: its core is the continuous high-load operation time in the behavioral inertia factor. , load jump frequency and intra-day curve similarity r (reflecting the stability of users' electricity consumption behavior), while superimposing tag information based on equipment operating status data (such as the real-time operating mode and rated power of adjustable equipment such as air conditioners and water heaters), so that edge nodes can quickly locate abnormal sources and potential control objects when receiving data.

[0069] Compared with the traditional timed reporting mechanism, this event-driven mechanism avoids the redundant transmission of meaningless data through dual threshold triggering. For example, when the user is in a stable power consumption state ( and ), the system retains only the local cache and does not upload, significantly reducing the load on the communication link. When abnormal fluctuations or high load are detected, high-frequency data containing behavioral characteristics and device information is immediately reported, ensuring that edge nodes can promptly detect sudden changes in regional load trends. This "silent-trigger" dynamic switching mode reduces data transmission volume compared to traditional solutions and shortens the latency of capturing critical events from minutes to seconds.

[0070] The behavioral inertia factor of this application reflects the user's electricity consumption pattern (such as Characterizes the risk of sustained high load, characterizes the risk of instantaneous fluctuations), and the dual threshold is dynamically adjusted based on historical operating data and system safety requirements (e.g. (This can be adjusted upward to accommodate increased air conditioning loads.) When a dimension of the behavioral inertia factor exceeds its threshold, high-frequency data reporting is immediately triggered, providing event-level response input for local cluster analysis at edge nodes, rather than traditional time-slice-level data. This effectively improves the accuracy and timeliness of regional load feature summaries. This avoids bandwidth waste caused by scheduled reporting while ensuring the timely delivery of critical anomalies. The inclusion of characteristic parameters reflecting the inertia of power consumption behavior and the association with the physical properties of device status provide multi-dimensional, high-value input support for subsequent load clustering, trend forecasting, and control decisions at the edge, significantly enhancing the distribution system's perception and response efficiency to dynamic changes on the load side.

[0071] This application further proposes that the specific steps for generating regional load characteristic summary data by collecting high-frequency change data from multiple user sides through local cluster analysis include:

[0072] The user side divides users into groups based on electrical and physical locations, and uses the OPTICS algorithm to automatically identify the core area of ​​user group density;

[0073] Extract the average load value of each group in the core area of ​​user group density within an N-minute sliding window as the window benchmark;

[0074] Perform local cluster analysis to calculate the standard deviation within the sliding window to generate the load fluctuation amplitude index;

[0075] The phase synchronization of user load curves within the user group density core area is analyzed by Pearson correlation coefficient, and the user linkage index is output;

[0076] The calculation results of user linkage indicators are encapsulated as regional load characteristic summary data.

[0077] Specifically, the system first divides users into several geographically adjacent groups based on their electrical and physical locations (such as distribution transformer coverage and feeder topology). Based on each group's high-frequency changing data (including behavioral inertia factors and device status information), the OPTICS (Ordering Points To Identify the Clustering Structure) algorithm automatically identifies the density core within the user group. This algorithm calculates the reachable distance and core distance between data points using the following formula:

[0078]

[0079] in The OPTICS algorithm can adaptively identify uneven user load distribution, avoiding the defect of the traditional K-means algorithm that is sensitive to the initial cluster center. It is especially suitable for areas with large load density differences, such as urban-rural fringe areas.

[0080] After identifying the density core area, the system extracts the average load value of each core area in the N-minute sliding window as the window benchmark (Calculation formula: , is the average power of users in the core area at time t). This benchmark is used to measure the degree of deviation between the current load and the historical average level. On this basis, the standard deviation within the sliding window is calculated through local cluster analysis. (formula: ), generates a load fluctuation amplitude index, reflecting the synchronous fluctuation intensity of user load in the region—— The larger it is, the higher the short-term uncertainty of the load in the area.

[0081] To further characterize the collaborative behavior between users, the Pearson correlation coefficient r is introduced to analyze the phase synchronization of the user load curve within the core area group. When the Pearson correlation coefficient r approaches 1, it indicates that the user load curve within the group shows a synchronous characteristic of "rising and falling together" (for example, the air conditioning load in the community starts synchronously with the temperature change). This strong linkage will amplify the regional load peak; on the contrary, it means that user behavior is relatively independent and the regional load fluctuation is more stable. This user linkage index is packaged together with the fluctuation amplitude index and the window benchmark to form the regional load characteristic summary data. .

[0082] Compared with traditional regional load statistics methods, this mechanism uses the OPTICS algorithm to address clustering distortion caused by uneven user grouping density (traditional DBSCAN requires a preset neighborhood radius, which can easily miss low-density core areas). By combining a sliding window benchmark with standard deviation, this mechanism overcomes the limitation of single extreme value statistics in reflecting sustained load fluctuations. By quantifying user connectivity through the Pearson correlation coefficient, this mechanism overcomes the shortcomings of traditional methods that focus solely on individual loads while ignoring group synergy. The OPTICS algorithm provides effective core area boundaries for subsequent statistics, the sliding window benchmark provides a reference anchor for fluctuation calculations, and the standard deviation and correlation coefficient complement regional load characteristics, ultimately generating summary data.

[0083] This application achieves intelligent condensation from user-level discrete data to regional-level features, retaining key information reflecting the essence of regional load (such as fluctuation intensity and user collaboration) while significantly reducing data dimensionality (the summary data only includes three core indicators), reducing the computational complexity of edge nodes compared to directly processing raw user data. Furthermore, the multidimensional characteristics of the summary data provide more accurate input for the subsequent behavioral inertia dynamic evolution model—the fluctuation amplitude indicator is used to predict the load uncertainty range, the user linkage indicator is used to correct the peak probability calculation, and the window benchmark is used to calibrate the load trend baseline. Together, these improvements improve the accuracy of the partitioned load forecast matrix and the targeted regulation strategy, providing solid regional feature support for the flexible regulation of the distribution system.

[0084] The present application further proposes to determine whether the regional load characteristic summary data in the current period exceeds a preset edge threshold. If so, the regional load characteristic summary data is uploaded through a differential update mechanism. The specific steps performed by the differential update mechanism include:

[0085] Retrieve a pre-stored cached copy of the historical summary;

[0086] The regional load characteristic summary data for the current period is compared with the historical summary cache copy. When any of the following conditions are met, the regional load characteristic summary data is uploaded and marked for update:

[0087] (1) The fluctuation amplitude changes by more than the first preset ratio;

[0088] (2) The change in the user linkage indicator exceeds the second preset ratio;

[0089] (3) The average load change exceeds the third preset ratio;

[0090] Otherwise, a no-update confirmation signal is sent and the system waits for the next cycle.

[0091] Specifically, the system first retrieves the pre-stored cached copy of the historical summary (including the window benchmark, fluctuation amplitude index and user linkage index of the previous period), and obtain the new summary data of the current period .in, and Represents the average load of the core area of ​​the old window (previous period) and the new window (current period) (through the window benchmark calculate); are the standard deviations of the old window (previous cycle) and the new window (current cycle), respectively, indicating the load fluctuation amplitude in the core area (measured by the standard deviation Calculation), used to quantify the fluctuation of load; r is the Pearson correlation coefficient, and Reflects the core area user linkage index of the old window (previous period) and the new window (current period) respectively (unitless, range [-1, 1]).

[0092] To quantify the difference between new and old data, the system calculates the change ratio in three dimensions:

[0093] Fluctuation amplitude change ratio: (Reflects the degree of change in regional load fluctuation intensity);

[0094] Change ratio of user linkage indicators: (Reflects the extent of change in user collaborative behavior);

[0095] Average load change ratio: (An offset that reflects the overall level of regional load).

[0096] When any change ratio exceeds the preset threshold ( or or ,in When the load characteristics of the current area change significantly (the first, second, and third preset proportions respectively), the system immediately uploads new summary data and updates the historical cache; if all change proportions do not exceed the threshold, a no-update confirmation signal is sent and the next cycle is waited for, ensuring that only valid change data that affects system decisions is transmitted.

[0097] Compared with the traditional scheduled full upload mechanism, this differential update mechanism effectively solves the problem of data redundancy. The traditional method uploads the complete summary at a fixed period regardless of whether the data has changed, resulting in the communication bandwidth being occupied by a large amount of data. However, this application uses change ratio verification to trigger upload only when the data has substantially changed, thereby reducing communication traffic. At the same time, the transmission delay of key changes is shortened from the traditional fixed period (such as 15 minutes) to the instant when the change occurs, significantly improving data timeliness. After the regional load characteristic summary is screened, only the summary data with significant changes is retained to input the behavior inertia dynamic evolution model, so that it focuses more on the trend changes of load characteristics (such as the sudden increase in user linkage indicates the potential risk of load peak superposition), thereby improving the accuracy of the partition load forecast matrix. For example, when the user linkage indicator change ratio When it exceeds β, the model can identify the increase in user collaborative behavior, and then increase the risk factor of the corresponding period when predicting the peak probability; conversely, if all change ratios are lower than the threshold, the model can use historical trends for prediction and reduce computing resource consumption.

[0098] The differential update mechanism of this application realizes the timely transmission of key changes in regional load characteristic data, and avoids the waste of redundant data on communication links and computing resources; it not only forms an effective connection with the output of the edge analysis module, but also provides high-quality input support for the prediction model, providing a solid technical guarantee for the precision and intelligence of load forecasting and control decisions.

[0099] This application further proposes to integrate the summary data of multiple regional load characteristics from past cycles and input them into a pre-trained behavioral inertia dynamic evolution model to generate a partitioned load forecast matrix. The behavioral inertia evolution model specifically includes:

[0100] Encoding the user inertia state on the user side into a dynamic evolution vector based on the user's power preference information;

[0101] The dynamic evolution vector and regional load characteristic summary data are weighted and fused with the environmental disturbance factor, time weighting factor and holiday correction factor through the inertial state transfer equation;

[0102] Calculate the partition load trend in the future period and output the peak probability, which contains the prediction matrix of the partition peak time and amplitude.

[0103] Specifically, the behavioral inertia evolution model first encodes the user-side inertia state into a dynamic evolution vector based on the user's electricity preference information (such as user historical regulation compliance, device type (air conditioner / water heater, etc.), and work and rest time rules). .in, Corresponding to inertial characteristics of different dimensions: for example The user's historical control compliance (value range is 0-1, 1 means full compliance with the control), is the equipment adjustability coefficient (e.g. 0.8 for air conditioner, 0.2 for refrigerator), This multidimensional vector encoding mechanism converts individual user behavior characteristics into computable numerical representations, providing structured input for subsequent fusion.

[0104] The model will then dynamically evolve the vector and regional load characteristics summary data (including window benchmark, fluctuation amplitude index and user linkage index) are weighted fused through the inertial state transfer equation. The state transfer equation is in the form of:

[0105]

[0106] in, is the fusion state vector at the next moment; is the environmental disturbance factor (such as the impact of temperature and humidity on air conditioning load, which is the deviation ratio between the current temperature and the historical average temperature); is the time weighting factor (e.g. 1.2 for peak hours and 0.8 for off-peak hours); is the holiday correction factor (1.1 for weekends / holidays and 1.0 for weekdays); is the weight coefficient of each factor (determined through historical data training). This multi-factor fusion mechanism not only retains the inertial characteristics of individual user behavior (V), but also incorporates the group characteristics of regional load (S). It also corrects for the influence of external factors such as environment, time, and holidays, making the fusion state more closely aligned with actual load fluctuations.

[0107] Based on the fusion state vector The model calculates the partition load trend in the future T time window through the pre-trained LSTM neural network (long short-term memory network) , and combined with the Gaussian mixture model to output the peak probability matrix The peak probability matrix contains the probability p(t) that the load exceeds the rated capacity at each time point t and the corresponding peak amplitude A(t), which is expressed as follows:

[0108]

[0109] Compared with traditional load forecasting models, this solution solves the problem of traditional models focusing only on load values ​​and ignoring user behavioral inertia through dynamic evolution vectors (for example, users with high compliance are more likely to cooperate with regulation and have greater load reduction potential); it overcomes the limitation of traditional models' insensitivity to external environmental changes through multi-factor weighted fusion (for example, when the environmental disturbance factor E is increased during high temperatures in summer, the model automatically increases the growth forecast of air-conditioning load); and it replaces traditional single-point predictions with peak probability matrices, providing an uncertainty range for load changes (for example, the peak probability in a certain period is 80%, and the amplitude is 120% of the rated capacity), reserving flexibility for regulation strategies.

[0110] This application integrates regional load characteristic summary data (reflecting group load characteristics) with user dynamic evolution vectors (reflecting individual behavioral inertia) through state transition equations. External environmental, time, and holiday factors are used as correction terms in the calculation. The final output of the partitioned load forecast matrix provides a time-dimensional load panorama for subsequent system load rate calculation and control instruction generation. For example, when the peak probability matrix shows that the peak probability p(t) in a certain period t is greater than 90% and the amplitude A(t) is greater than 110% of the rated capacity, the decision module can identify high-risk periods in advance, calculate the adjustable load range based on user electricity preferences, and generate targeted control instructions.

[0111] The behavioral inertia dynamic evolution model proposed in this application captures the inertial characteristics of individual user behaviors while reflecting the group synergy of regional loads. It considers the deterministic trends of load changes while quantifying the uncertain risks. This multi-dimensional, multi-factor prediction mechanism improves the load forecasting accuracy of the distribution system compared to traditional models, provides reliable time and probability support for the precise formulation of flexible control strategies, and significantly enhances the system's resilience to load fluctuations.

[0112] This application further proposes to calculate the system load rate based on the partition load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partition load trend and equipment operating status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information. The control instructions containing the time-based adjustment target and the individual flexible adjustment range are generated. The time-based adjustment target generation includes:

[0113] Identify peak periods in the partitioned load forecast matrix;

[0114] Prioritize the allocation of golden control windows with preset time radius during peak periods;

[0115] Mark the enforcement priority label of the golden control window in the control instructions.

[0116] Specifically, the system first uses the partition load forecast matrix (Including load trend in T time window With peak probability , identify the peak period by double threshold judgment: when the load trend of a certain period t More than 90% of the system rated capacity and peak probability When , the period is determined to be the peak period The dual-dimensional identification avoids misjudgments caused by traditional methods that rely solely on load values ​​(such as short-term load spikes with extremely low probability), ensuring that the identification of peak periods focuses on both load intensity and risk probability.

[0117] Determining the peak period Afterwards, the system As the center, expand forward and backward (The preset time radius is usually 15-30 minutes, which is determined by the standard deviation of historical load fluctuations. Dynamic tuning, formula: , k is the empirical coefficient), forming a golden control window The core logic of this window is: before and after the peak period Within this range, the load is most sensitive to the control command (for example, reducing the load in advance can effectively suppress the peak value, while delayed control will weaken the effect). The golden window can adapt to the load fluctuation characteristics of different areas (such as industrial areas Larger, expand accordingly).

[0118] To ensure that control resources are prioritized in critical periods, the control instructions are marked with the mandatory execution priority label L (values ​​range from 1 to 3, with 1 being the highest priority). The label assignment logic combines the peak probability and device adjustability: the higher the peak probability (e.g. The greater the proportion of adjustable devices within the window (e.g., air conditioning load > 60%), the smaller the L value (higher priority). This tagging mechanism enables execution terminals (such as user-side intelligent controllers) to quickly identify control tasks that require priority response, avoiding response delays caused by concurrent multiple commands.

[0119] Compared with traditional control strategies, this application utilizes dual-threshold identification during peak periods to address the traditional approach's tendency to misjudge short-term peaks based solely on load exceeding rated capacity. The dynamic time radius design of the golden control window overcomes traditional limitations (such as a fixed time radius of 30 minutes before the peak, which cannot accommodate areas with large load fluctuations). The introduction of mandatory priority tags optimizes the allocation logic of control resources.

[0120] This application provides a time-dimensional load overview through a partitioned load forecast matrix ( ), equipment operating status data (such as equipment type and rated power) determines the adjustable potential of the golden window, and user electricity preferences (such as historical compliance) affect the assignment of priority labels. For example, when the peak probability of a peak period is 85% and the air conditioning load in the window accounts for 70% (high adjustability), its priority label will be set to 1, and control resources will be allocated first; if the peak probability of another period is only 60% and it is mainly composed of non-adjustable refrigerator loads, the label will be set to 3, and control will be postponed. It not only ensures the control effect during critical peak periods, but also avoids the waste of resources in unnecessary periods; it combines the trend of load forecasting and takes into account the actual characteristics of equipment and users. This precise control in the time dimension shortens the time that the system load rate exceeds the safety threshold compared to traditional strategies, significantly improves the distribution system's resilience to load peaks, and provides key time decision support for the implementation of flexible control.

[0121] This application further proposes to calculate the system load rate based on the partitioned load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partitioned load trend and equipment operating status data. The individual maximum adjustment range is calculated based on the user's electricity preference information. The control instructions containing the time-based adjustment target and the individual flexible adjustment range are generated. The individual flexible adjustment range specifically includes:

[0122] Divide the control priority zones according to the zone load forecast matrix;

[0123] Generate user historical behavior profiles based on user electricity preference information and set personalized adjustment ranges;

[0124] The calculation of individual flexible adjustment amplitude satisfies:

[0125]

[0126] in, is the maximum adjustable range, is the preference compromise coefficient, which is determined by the user's historical regulation compliance. is the current power, It is a historical peak;

[0127] A control instruction set including a target load reduction value and a target time window is generated.

[0128] First, the control priority zones are divided according to the zone load forecast matrix. The zone load forecast matrix contains key information such as the zone load trend and peak probability within the T time window. By analyzing and processing this information, the entire control area can be reasonably divided into different priority zones. This allows subsequent control operations to be carried out in a focused and sequential manner, ensuring that when the system load rate exceeds the safety threshold, the key or high-load trend zones are given priority for effective regulation.

[0129] Next, a user historical behavior profile will be generated based on the user's electricity usage preference information, and a personalized adjustment range will be set. User electricity usage preference information plays an important role in the entire regulation process. It reflects the user's habits and characteristics in electricity usage. Through in-depth analysis and processing of this information, an accurate user historical behavior profile can be constructed. The profile covers many aspects such as the user's electricity usage habits at different times and preferences for using different devices. Based on this profile, combined with relevant calculation rules and system requirements, a personalized adjustment range that meets the user's own characteristics is set for each user. This can not only meet the overall regulation needs of the system, but also take into account the user's personalized electricity needs to a certain extent, avoiding excessively unified regulation and causing greater inconvenience to users.

[0130] When calculating the specific individual flexibility adjustment range, The maximum adjustable range is an important parameter determined based on the overall control capability of the system and the current load. It represents the maximum range that a user or device can adjust under ideal conditions. is the preference compromise coefficient, which is determined by the user's historical regulation compliance. It reflects the degree of compromise on the user's electricity preferences during the regulation process. If the user has a high degree of compliance in the past regulation process, then The value of may be relatively small, which means that relatively small changes can be made to their electricity usage habits in this regulation; otherwise, it may be large; The current power reflects the actual power consumption of the user or device at the current moment; This is the historical peak value, the highest historical power consumption of a user or device, calculated based on past power consumption data. By comprehensively considering various factors, the individual flexible adjustment range is determined to meet system control requirements while also taking into account user power usage habits to a certain extent.

[0131] Finally, based on the above calculation results, a control instruction set containing the target load reduction value and target time window is generated. These control instruction sets will be sent to the user side to guide the user-side equipment to perform corresponding adjustment operations, thereby achieving effective control of the system load. At the same time, the actual control execution data generated after the user side executes the control instructions will be collected and fed back to the partition load forecast matrix for real-time correction, forming a complete closed-loop control system. This allows the flexible control of the entire distribution system to be continuously optimized and improved, continuously improving the accuracy and effectiveness of the control, and ensuring that the distribution system can operate stably and efficiently under various load conditions.

[0132] This application further proposes that the flexible control method of the distribution system based on load-side behavior identification also includes:

[0133] Sending control instructions to the user side and receiving the actual control execution data generated by the user side executing the control instructions;

[0134] The actual control execution data is collected and fed back to the partition load forecast matrix for real-time correction. The specific steps of real-time correction include:

[0135] In the next forecast period, the actual control execution data is used as a correction factor;

[0136] The correction factors are input into the partition load forecast matrix and used as boundary constraints;

[0137] Dynamically compress the floating interval of the subsequent partition load prediction matrix to a preset range.

[0138] Specifically, after control instructions (including time-based control targets and individual flexible control ranges) are issued through user-side smart terminals, user-side devices (such as smart air conditioners and energy storage devices) adjust their operating states accordingly, generating actual control execution data (such as the actual load reduction, response time, and changes in equipment operating parameters). This data is transmitted back to the main distribution station in real time via the IoT communication module, providing direct feedback on the control effect. For example, if a user actually reduces their load by 2kW (the target is 3kW) within the golden control window, the execution data will record this deviation and the specific response time, providing empirical evidence for subsequent forecast corrections.

[0139] The actual control execution data fed back to the master station is used as a correction factor to be input into the partitioned load forecast matrix in the next forecast cycle. The correction factor acts as a boundary constraint, limiting the output range of the forecast model. For example, if a user's historical control compliance is 0.8 (preferred compromise coefficient), but they actually reduce their target value by 85%, the model will adjust the maximum adjustable range for that user in the next cycle from "historical peak value × 0.8" to "historical peak value × 0.85," narrowing the forecast fluctuation range. Furthermore, the correction factor acts as a dynamic optimization parameter and participates in the state transition equation calculation of the behavioral inertia dynamic evolution model. For example, the load reduction response time in the actual execution data is encoded as a new dimension of user behavioral inertia (response agility), updating the dynamic evolution vector V and making the model more consistent with the user's actual behavior patterns.

[0140] After this correction, the floating interval of the subsequent partition load forecast matrix is ​​dynamically compressed to a preset range (e.g., from ±15% to ±8%). Dynamic compression makes adaptive adjustments based on the statistical characteristics of actual execution data (such as mean deviation and variance). If the deviation between actual execution data and the forecast value is less than 5% for three consecutive cycles, the floating interval is tightened to improve forecast accuracy. If a sudden deviation occurs (e.g., a user fails to respond to regulation due to a special event), the interval is temporarily relaxed to maintain forecast flexibility. This dynamic balanced compression avoids the error accumulation caused by the fixed interval in traditional forecast models and prevents forecast fluctuations caused by over-correction.

[0141] Through the above design, this application achieves self-updating load forecasting, preserving the inertial characteristics of historical data while incorporating real-time feedback from control execution. This ensures both forecast stability and adaptability to dynamic changes. This closed-loop correction capability enables the distribution system to maintain high control accuracy in complex scenarios such as sudden changes in user electricity usage habits (such as a surge in air conditioning load due to extreme summer temperatures) and equipment failures (such as the failure of a user's energy storage device), ensuring the continued and effective implementation of flexible control.

[0142] The following is a specific scenario implementation of the distribution system flexible control method based on load-side behavior identification - taking the summer peak power consumption control of the "Green Source Community" as an example:

[0143] As the high summer weather continues, the load rate of the power distribution system in a certain city's "Green Source Community" (including 200 households) exceeded the safety threshold of 80% for three consecutive days. The traditional power-rationing mode caused some users (such as elderly families and mother-and-child families) to have a poor electricity experience, and the load forecast error was as high as 15%, making it difficult to accurately control.

[0144] 1. Data Collection Phase (from 8:00 AM on July 15, 2024):

[0145] The community power distribution master station collects user-side data through the following methods, including:

[0146] Electricity usage data: Each household is equipped with a smart meter (sampling frequency: 1 minute / time) to collect real-time, time-based power data for major appliances such as air conditioners, water heaters, and refrigerators. For example, the meter for Ms. Wang's family in Room 302 (two working parents and one elementary school student) shows that during breakfast time from 7:00 to 8:00, the water heater (2kW) and microwave oven (1.5kW) were running simultaneously, totaling 3.5kW. During lunch break from 12:00 to 14:00, the air conditioner (1.8kW) ran continuously, maintaining a stable power consumption.

[0147] Device operating status data: The IoT module monitors the device's on / off status. For example, Ms. Wang's air conditioner frequently turns on between 6:00 PM and 10:00 PM (after get off work), and her water heater runs regularly between 6:30 AM and 7:30 AM, and again between 8:00 PM and 9:00 PM (during shower time).

[0148] Environmental parameter data: The community weather station collects real-time outdoor temperature (maximum 38°C on the day) and humidity (65%), and correlates them with user electricity usage behavior (for example, when the temperature is >35°C, the air conditioner activation rate increases to 90%).

[0149] User electricity preference information: User settings are collected through an APP questionnaire (e.g., Ms. Wang prefers an air conditioning temperature of 25°C and turns off the air conditioning in the living room after 23:00 at night), and historical control compliance is recorded (the response rate to "lowering the air conditioning temperature between 19:00-20:00" in the past three months was 85%).

[0150] 2. Behavioral inertia factor generation stage (8:00-9:00): The main station dynamically analyzes the electricity consumption data of Ms. Wang's home: Calculate the power change gradient: From 8:00 to 8:01, Ms. Wang turned off the water heater (power dropped from 3.5 kW to 1.5 kW), with a gradient of -2 kW / minute. This exceeded the fluctuation threshold (-1.5 kW / minute. The IEEE 1547 standard stipulates that the load mutation threshold range is ±1.0-2.0 kW / min. The actual process is -1.5 kW / min based on equipment tolerance testing), triggering the load jump frequency count (a total of 3 jumps that day).

[0151] Update continuous high load duration: The air conditioner in Ms. Wang's home ran continuously for 2 hours from 12:00 to 14:00, with the power stable at 1.8kW (≥1.5kW high load threshold), and recorded a continuous high load duration of 2 hours.

[0152] Calculate the intraday curve similarity: the power curve of Ms. Wang's home from 8:00 to 9:00 on the same day ( ) and the stored historical benchmark curve (same period last week ) was used to calculate the Pearson correlation coefficient and the similarity was 0.89 (threshold 0.8, verified based on the clustering effect of the load curve), which was judged as “high similarity”.

[0153] The final behavioral inertia factor for Ms. Wang's family on that day was: continuous high load duration of 2 hours, load jump frequency of 3 times, and intra-day curve similarity of 0.89.

[0154] 3. High-frequency data reporting and regional characteristics analysis (9:00-10:00):

[0155] Event-driven reporting: The continuous high load duration (2 hours) at Ms. Wang's home exceeded the local preset threshold (1.5 hours), triggering the generation of high-frequency change data (including a power value of 1.8kW and the device status "air conditioning running"), which was reported to the community edge node (deployed in the power distribution room of Building 1) via the narrowband Internet of Things (NB-IoT).

[0156] Local cluster analysis: The edge node collects high-frequency data from 30 users that trigger reports, divides them into four user groups based on their electrical and physical locations (by unit building), and uses the OPTICS algorithm to identify the core area of ​​group density (for example, units 1-3 in Building 2 are the high-load core area).

[0157] The average load value of core area users within a 15-minute sliding window was extracted (the average load value in the core area of ​​Building 2 was 4.2kW). The standard deviation within the window was calculated (1.1kW) to generate a load fluctuation amplitude index (1.1kW). The Pearson correlation coefficient was used to analyze the phase synchronization of the load curves of users within the group (the air conditioning start-up time synchronization rate of users in the core area of ​​Building 2 was 82%), and the user linkage index was output as 0.82.

[0158] The final packaged area load characteristic summary data: average load 4.2kW, fluctuation amplitude 1.1kW, user linkage 0.82.

[0159] 4. Load forecasting and control instruction generation (10:00-12:00):

[0160] The edge node retrieves the historical summary cache (the average load in the same time period of the previous day was 3.8kW, the fluctuation amplitude was 0.9kW, and the user linkage was 0.75), and finds that the current average load change (+10.5%) exceeds the third preset ratio (5%), triggering the upload of the regional load feature summary to the master station.

[0161] By integrating regional feature data from the past three days and inputting the pre-trained behavioral inertia dynamic evolution model (which has encoded the electricity consumption preference vectors of users such as Ms. Wang), combined with the environmental disturbance factor (the high temperature of the day was 38°C) and the time weighting factor (the weight of peak electricity consumption in summer is 0.9), the output is the partitioned load prediction matrix: 18:00-20:00 (the time when people return home from get off work) is the peak load period, with a peak probability of 90%, and a predicted load of 5.8kW (safety threshold 5.5kW).

[0162] The master station calculates the system load factor (predicted 5.8kW / capacity 6.0kW = 96.7% > 85% (safety threshold)). Based on the zone load trend (peak period 18:00-20:00) and equipment status (air conditioners and water heaters are adjustable loads), it determines the adjustable load range (1-2kW per household). Combined with user preferences (e.g., Ms. Wang prefers 25°C and a compliance rate of 85%), it calculates the individual flexible adjustment range:

[0163]

[0164] in, (maximum adjustable range), β=0.15 (compliance 85% corresponding to the compromise coefficient), (Current air conditioning power), (historical peak), calculated (It is recommended to adjust the air conditioning temperature from 25℃ to 27℃, reducing 1.2kW).

[0165] Finally, the control instructions are generated, including: 1.2kW load reduction will be implemented at Ms. Wang's home from 18:00 to 19:00 (the golden control window, 1 hour before the peak), with the target time window of 18:00-19:00, and a "priority response" label.

[0166] 5. Regulation Implementation and Feedback Correction (18:00-19:00):

[0167] Ms. Wang received the control instructions through the smart meter app. After agreeing, the air conditioner automatically raised the temperature to 27°C, and the actual load was reduced by 1.1kW (a deviation of 0.1kW from the target of 1.2kW). The execution data (1.1kW, response time 18:02) was transmitted back to the main station.

[0168] The main station uses the actual execution data of Ms. Wang's home as a correction factor, inputs it into the partition load forecast matrix for the next cycle (19:00-21:00), adjusts its dynamic evolution vector (response agility is increased from 0.85 to 0.88), and compresses the floating range (from the original ±15% to ±8%).

[0169] Through the closed-loop control of this embodiment, the actual load of the "Green Source Community" from 18:00 to 20:00 on the same day dropped from the predicted 5.8kW to 5.2kW (below the safety threshold of 5.5kW), and the peak load decreased by 10.3%. Because the control instructions met her electricity usage preferences (only adjusting the temperature for one hour), user Ms. Wang's satisfaction increased from 60% during traditional power restrictions to 90%. The load forecast error dropped from 15% to 5%, and system stability was significantly enhanced. This application effectively solves the problems of poor user experience, large forecast errors, and delayed response in traditional power distribution control, and provides technical support for the efficient and stable operation of the power distribution system.

[0170] Example 2:

[0171] like Figure 4 and Figure 5 As shown, an intelligent chassis control system with adaptive posture leveling uses the above-mentioned flexible control method of the power distribution system based on load-side behavior identification, including:

[0172] The data acquisition module is used to collect power consumption data, equipment operation status data, and environmental parameter data from the user side in real time, and obtain user power preference information; the inertia processing module is used to generate behavioral inertia factor data based on power consumption data and environmental parameter data, and dynamically update it according to a preset period; the edge analysis module is used to generate high-frequency change data when the behavioral inertia factor data exceeds a pre-stored local preset threshold, and report the high-frequency change data to the edge node through an event-driven mechanism; the receiving feature module is used to receive regional load feature summary data from the edge node; the regional load feature summary data is generated by collecting high-frequency change data from multiple users through local cluster analysis;

[0173] The judgment fusion module is used to fuse the regional load feature summary data of multiple past cycles if the regional load feature summary data in the current cycle exceeds the edge preset threshold, and input the pre-trained behavioral inertia dynamic evolution model to generate a partition load prediction matrix; wherein the partition load prediction matrix includes the partition load trend and peak probability within the T time window;

[0174] The decision-making module is used to calculate the system load rate based on the partition load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partition load trend and equipment operation status data. The individual maximum adjustment range is calculated based on the user's electricity preference information, and a control instruction is generated that includes the time-sharing adjustment target and the individual flexible adjustment range.

[0175] The data acquisition module, as the front-end data sensing component of the entire system, collects real-time user-side power consumption data, equipment operating status data, and environmental parameter data, and obtains information about user electricity preferences. For example, by monitoring power consumption data in real time, it is possible to understand the changes in user electricity demand at different times, while equipment operating status data can reflect the normal working status of related electrical equipment.

[0176] The inertia processing module generates behavioral inertia factor data based on the collected power consumption data and environmental parameter data, and dynamically updates it according to a preset cycle. The specific process includes calculating the power change gradient of the power consumption data every minute, refreshing the load jump frequency count when the power change gradient exceeds the fluctuation threshold, and then calculating the cumulative duration of excessive power on the day based on the load jump frequency and updating the continuous high load duration to generate the current power curve. The Pearson correlation coefficient between the current power curve and the pre-stored historical benchmark curve is then calculated, and the Pearson correlation coefficient is stored as a curve similarity parameter for dynamically updating the behavioral inertia factor data. The behavioral inertia factor data covers information such as the continuous high load operation duration, load jump frequency, and intra-day curve similarity. This series of processing enables the system to mine the inertial characteristics of users' electricity consumption behavior from the data, providing an important basis for subsequent judgment and decision-making.

[0177] When behavioral inertia factor data exceeds pre-set thresholds (including the absolute value threshold for the rate of change and the absolute value threshold for power), the edge analysis module generates high-frequency change data and reports it to the edge node through an event-driven mechanism. This high-frequency change data includes the power value of the behavioral inertia factor data and high-frequency data tagged with device operating status data. This event-driven reporting mechanism promptly transmits important power usage change information to the edge node for further analysis, avoiding unnecessary data transmission and improving system efficiency.

[0178] The receiving feature module receives the regional load feature summary data of the edge node. The regional load feature summary data is generated by collecting high-frequency change data from multiple user sides through local cluster analysis. Specifically, the user groups are first divided based on the electrical and physical location, and the OPTICS algorithm is used to automatically identify the core area of ​​user group density. The average load value of each group in the core area of ​​user group density within the N-minute sliding window is extracted as the window benchmark. Then, local cluster analysis is performed to calculate the standard deviation within the sliding window to generate the load fluctuation amplitude index. The phase synchronization of the user load curve within the group of the core area of ​​user group density is analyzed by the Pearson correlation coefficient, and the user linkage index is output. Finally, the calculation result of the user linkage index is encapsulated as regional load feature summary data. In this way, the system can understand the overall characteristics and change trends of the load at the regional level.

[0179] After receiving the regional load characteristic summary data, the judgment fusion module determines whether the regional load characteristic summary data for the current cycle exceeds the edge preset threshold. If so, the regional load characteristic summary data from multiple past cycles are fused and input into the pre-trained behavioral inertia dynamic evolution model to generate a partitioned load prediction matrix. In the behavioral inertia evolution model, the user's inertia state on the user side is encoded into a dynamic evolution vector based on the user's electricity preference information. The dynamic evolution vector and the regional load characteristic summary data are then weighted and fused using the inertia state transfer equation with the environmental disturbance factor, time weighting factor, and holiday correction factor. This calculates the partitioned load trend for the future period and outputs the peak probability (a prediction matrix containing the partitioned peak time and amplitude). This step enables the system to make relatively accurate predictions about future partitioned load conditions, providing strong support for subsequent control decisions.

[0180] The decision module calculates the system load factor based on the partitioned load forecast matrix. When the system load factor exceeds a safety threshold, it determines the adjustable load range based on partitioned load trends and equipment operating status data. It then calculates individual maximum adjustment ranges based on user electricity preferences, generating control instructions that include time-based adjustment targets and individual flexible adjustment ranges. When generating time-based adjustment targets, the module identifies peak periods within the partitioned load forecast matrix and prioritizes golden adjustment windows within a preset time radius during these peak periods. The control instructions are then labeled with a mandatory execution priority tag for the golden adjustment windows. When calculating individual flexible adjustment ranges, the module prioritizes control zones based on the partitioned load forecast matrix. User historical behavior profiles are generated based on user electricity preferences, and personalized adjustment ranges are set. Ultimately, a control instruction set is generated that includes target load reduction values ​​and target time windows. Through this decision-making process, the system can formulate targeted and flexible control instructions based on the characteristics of individual users and the overall load situation, ensuring safe and stable system operation while minimizing user electricity needs.

[0181] The generated control instructions are then sent to the user side. After the user side executes the control instructions, actual control execution data is generated. This data is collected and fed back into the partitioned load forecast matrix. In the next forecast cycle, the actual control execution data is input into the partitioned load forecast matrix as a correction factor and serves as a boundary constraint, dynamically compressing the floating range of the subsequent partitioned load forecast matrix to a preset range. Through this real-time correction mechanism, the system can continuously optimize and adjust predictions and decisions based on actual execution, improving the accuracy and adaptability of the entire control system.

[0182] This application forms a complete closed loop from data collection and analysis to control decision-making, execution and feedback correction, realizing flexible control of the distribution system based on load-side behavior identification, effectively improving the management capability and operation efficiency of the distribution system in the face of complex and changeable power consumption conditions, ensuring the safe and stable operation of the system, and to a certain extent meeting the personalized power consumption needs of users.

[0183] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. A flexible control method for a power distribution system based on load-side behavior identification, characterized by: The following steps are involved: Collect power consumption data, equipment operation status data and environmental parameter data from the user side in real time, and obtain user power preference information; Generate behavioral inertia factor data based on the power consumption data and the environmental parameter data, and dynamically update it according to a preset period; When the behavior inertia factor data exceeds a pre-stored local preset threshold, high-frequency change data is generated, and the high-frequency change data is reported to the edge node through an event-driven mechanism; Receive regional load characteristic summary data of the edge node; wherein the regional load characteristic summary data is generated by collecting the high-frequency change data of multiple user sides through local cluster analysis; Determine whether the regional load feature summary data in the current cycle exceeds the edge preset threshold. If so, fuse multiple regional load feature summary data from past cycles and input them into the pre-trained behavioral inertia dynamic evolution model to generate a partition load prediction matrix; wherein the partition load prediction matrix includes the partition load trend and peak probability within the T time window; The system load rate is calculated based on the partition load prediction matrix. When the system load rate exceeds the safety threshold, the adjustable load range is determined by using the partition load trend and the equipment operation status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information, and a control instruction including the time period adjustment target and the individual flexible adjustment range is generated.

2. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: The specific steps of generating behavioral inertia factor data based on the power consumption data and the environmental parameter data and dynamically updating the data according to a preset period include: Calculating the power change gradient per minute of the electric power data; When the power change gradient exceeds the fluctuation threshold, the load jump frequency count is refreshed; Calculate the cumulative duration of excessive power on the day based on the load jump frequency, update the continuous high load duration, and generate a current power curve; Calculating a Pearson correlation coefficient between the current power curve and a pre-stored historical reference curve, and storing the Pearson correlation coefficient as a curve similarity parameter for dynamically updating the behavioral inertia factor data; The behavioral inertia factor data includes continuous high-load operation time, load jump frequency and intra-day curve similarity.

3. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: When the behavior inertia factor data exceeds a pre-stored local preset threshold, high-frequency change data is generated, and the high-frequency change data is reported to the edge node through an event-driven mechanism. The event-driven mechanism specifically includes: The local preset thresholds include a change rate absolute value threshold and a power absolute value threshold; When any preset threshold of the change rate absolute value threshold and the power absolute value threshold is exceeded, the upload of the high-frequency change data is triggered, and the high-frequency change data includes the power value of the behavioral inertia factor data and high-frequency data based on the equipment operation status data mark.

4. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 2, characterized in that: The specific steps of generating the regional load characteristic summary data by collecting the high-frequency change data of multiple user sides through local cluster analysis include: The user side divides users into groups based on electrical and physical locations, and uses the OPTICS algorithm to automatically identify the core area of ​​user group density; Extracting the average load value of each group in the user group density core area within an N-minute sliding window as a window benchmark; Executing the local cluster analysis to calculate the standard deviation within the sliding window to generate a load fluctuation amplitude index; Analyzing the phase synchronization of the user load curve within the user group density core area by the Pearson correlation coefficient, and outputting a user linkage index; The calculation result of the user linkage index is encapsulated as the regional load characteristic summary data.

5. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: Determine whether the regional load characteristic summary data in the current cycle exceeds the edge preset threshold. If yes, upload the regional load characteristic summary data through the differential update mechanism. The specific steps performed by the differential update mechanism include: Retrieve a pre-stored cached copy of the historical summary; The regional load characteristic summary data in the current period is compared with the historical summary cache copy. When any of the following conditions is met, the regional load characteristic summary data is uploaded and marked for update: (1) The fluctuation amplitude changes by more than the first preset ratio; (2) The change in the user linkage indicator exceeds the second preset ratio; (3) The average load change exceeds the third preset ratio; Otherwise, a no-update confirmation signal is sent and the system waits for the next cycle.

6. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: The fusion of the multiple regional load feature summary data of the past cycles is input into the pre-trained behavior inertia dynamic evolution model to generate the partition load prediction matrix. The pre-trained behavior inertia evolution model specifically includes: encoding the user inertia state of the user side into a dynamic evolution vector based on the user power preference information; The dynamic evolution vector and the regional load characteristic summary data are weightedly integrated with the environmental disturbance factor, the time weighting factor and the holiday correction factor through the inertial state transfer equation; The partition load trend in a future period is calculated and the peak probability is output, where the peak probability includes a prediction matrix of the partition peak time and amplitude.

7. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: The system load rate is calculated based on the partition load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partition load trend and the equipment operation status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information. A control instruction including a time-based adjustment target and an individual flexible adjustment range is generated. The time-based adjustment target generation includes: identifying peak periods in the partitioned load forecast matrix; Prioritize the allocation of the golden control window with a preset time radius during the peak period; The enforcement priority label of the golden control window is marked in the control instruction.

8. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: The system load rate is calculated based on the partitioned load forecast matrix. When the safety threshold is exceeded, the adjustable load range is determined based on the partitioned load trend and the equipment operating status data. The individual maximum adjustment range is calculated in combination with the user's electricity preference information. A control instruction including a time-based adjustment target and an individual flexible adjustment range is generated. The individual flexible adjustment range specifically includes: Dividing the control priority zones according to the zone load forecast matrix; Generate a user's historical behavior profile based on the user's electricity usage preference information and set a personalized adjustment range; The individual flexibility adjustment amplitude calculation satisfies: ； in, is the maximum adjustable range, is the preference compromise coefficient, which is determined by the user's historical regulation compliance. is the current power, It is a historical peak; Generate a control instruction set including a target load reduction value and a target time window.

9. The method for flexible control of a power distribution system based on load-side behavior identification according to claim 1, characterized in that: The distribution system flexible control method based on load side behavior identification also includes: Sending the control instruction to the user side, and receiving actual control execution data generated by the user side executing the control instruction; The actual control execution data is collected and fed back to the partition load forecast matrix for real-time correction. The specific steps of the real-time correction include: Using the actual control execution data as a correction factor in the next forecast period; Inputting the correction factor into the partition load forecast matrix as a boundary constraint; Dynamically compress the floating interval of the subsequent partition load prediction matrix to a preset range.

10. A flexible control system for distribution systems based on load-side behavior identification, characterized by: The method for flexible control of a power distribution system based on load-side behavior identification according to any one of claims 1 to 9 is used, comprising: A data acquisition module is configured to collect user-side power consumption data, equipment operating status data, and environmental parameter data in real time, and obtain user power consumption preference information; an inertia processing module is configured to generate behavioral inertia factor data based on the power consumption data and the environmental parameter data, and dynamically update the data according to a preset period; an edge analysis module is configured to generate high-frequency change data when the behavioral inertia factor data exceeds a pre-stored local preset threshold, and report the high-frequency change data to the edge node through an event-driven mechanism; a feature receiving module is configured to receive regional load feature summary data from the edge node; wherein the regional load feature summary data is generated by collecting the high-frequency change data from multiple user sides through local cluster analysis; A judgment fusion module is configured to, if the regional load feature summary data in the current cycle exceeds a preset edge threshold, fuse multiple regional load feature summary data from past cycles and input them into a pre-trained behavioral inertia dynamic evolution model to generate a partitioned load prediction matrix; wherein the partitioned load prediction matrix includes the partitioned load trend and peak probability within the T time window; A decision module is used to calculate the system load rate based on the partition load forecast matrix, determine the adjustable load range based on the partition load trend and the equipment operation status data when the safety threshold is exceeded, calculate the individual maximum adjustment range based on the user's electricity preference information, and generate a control instruction including the time period adjustment target and the individual flexible adjustment range.

Images