A load regulation system for power grid distribution
By using intelligent load monitoring, electricity consumption behavior feature extraction, and dual-scenario load forecasting modules, combined with multi-source load collaborative regulation, a feedback mechanism is established to solve the feedback problem between load forecasting and regulation. This achieves a balance between the stability of load regulation and user comfort, thereby improving the economic efficiency of power grid operation and user participation.
Patent Information
- Application Number
- CN202611229996.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-13
- Publication Date
- 2026-11-17
AI Technical Summary
The existing load regulation system lacks feedback between forecasting and control, resulting in power tracking lag and supply-demand fluctuations. It is difficult to reduce the peak-valley difference of the distribution network while taking into account users' electricity demand and comfort. Furthermore, the existing methods do not fully integrate the electricity market pricing mechanism and renewable energy output forecasting, which affects the economic efficiency of grid operation and user participation.
The system employs an intelligent load monitoring module and high-frequency acquisition of multi-source load data. Combined with a user electricity behavior feature extraction module and a dual-scenario load prediction module, a feedback mechanism is established through a multi-source load collaborative control module. This mechanism calculates the dynamic load elasticity coefficient and prediction correction factor, generates the target load, and issues control commands. The system also utilizes the characteristics of distributed energy storage and electric heating equipment to smooth out secondary peaks during the load recovery phase.
It achieves closed-loop coordination of load forecasting and regulation, reduces power tracking lag and supply-demand fluctuations, improves the continuity and stability of distribution network load, balances user comfort and grid economy, and enhances users' willingness to participate in demand response.
Smart Images

Figure CN123121629A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a load regulation system for power grid distribution, belonging to the field of power grid distribution technology. Background Technology
[0002] Currently, the amount of distributed renewable energy connected to the power distribution network is constantly increasing, and the proportion of flexible loads such as electric vehicles, distributed energy storage, and electric heating is also rising. The load on the distribution side is gradually shifting from a traditional, relatively fixed electricity consumption pattern to a flexible electricity consumption pattern with greater scalability. To maintain the supply and demand balance of the distribution system, existing technologies typically collect load data through intelligent monitoring equipment, use time series models to predict short-term load change trends, and then issue demand response price signals or direct load control commands based on the prediction results to reduce the peak-valley difference in the distribution network. This is especially important under extreme weather conditions or when there are sudden changes in the output of new energy sources. Existing load regulation systems typically employ a unidirectional, serial approach of forecasting followed by regulation. The load forecasting stage primarily relies on historical electricity consumption data and meteorological factors to fit load curves, without fully considering the dynamic impact of regulation actions on user electricity consumption behavior and load response capabilities. The regulation decision-making stage uses the forecast results as a fixed input, issuing load reduction or load transfer commands according to preset thresholds. Due to the lack of feedback between forecasting and regulation, the system struggles to promptly reflect the recovery and rebound of loads with thermal inertia characteristics, such as electric heating, after regulation. This can easily lead to lag in real-time power tracking and cause repeated fluctuations in power supply and demand in the distribution network.
[0003] Simply increasing the sampling frequency of load data or adding layers to the neural network cannot solve the problem of the separation between load forecasting and regulation. Predictive models trained on historical data also struggle to anticipate changes in load response capabilities caused by regulatory actions. While tightening control thresholds or forcibly interrupting flexible loads can reduce peak loads in the short term, electric heating equipment may resume operation en masse after regulation is lifted, creating a secondary load peak. Furthermore, such regulation methods can negatively impact user experience and reduce user willingness to participate in demand response. For example, Chinese invention patent application CN121886738A discloses a multi-load... Collaborative optimization control methods, when dealing with nonlinear regulation of multi-variable loads, often rely on conventional optimization algorithms to tune controller parameters, which can easily get trapped in local optima when dealing with multi-dimensional strongly coupled variables. At the same time, under dynamic communication delays and non-ideal environmental disturbances, their real-time correction capability of control commands and dead zone compensation mechanism are not rigorous enough, making it difficult to effectively control power exchange deviations within a wide load fluctuation range, thus limiting the overall regulation efficiency and operational economy of the power grid. In addition, existing methods do not fully integrate electricity market pricing mechanisms, renewable energy output forecasting, and system physical constraints, making it difficult to improve the operational economy of the power grid and the renewable energy absorption rate while ensuring users' electricity demand.
[0004] Therefore, the technical problem to be solved by this invention is how to establish a closed-loop coordination mechanism between load forecasting and regulation, reduce power tracking lag and supply and demand fluctuations caused by regulation actions, and reduce the peak-valley difference of the distribution network while taking into account the heating comfort of different users. Summary of the Invention
[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A load regulation system for power grid distribution, comprising: The intelligent load monitoring module is deployed at the power distribution nodes of residential, commercial and industrial users to collect high-frequency load power consumption data and equipment operating status data from multiple sources. The user electricity consumption behavior feature extraction module is connected to the intelligent load monitoring module. It is used to receive multi-source load electricity consumption data and equipment operating status data, extract spatiotemporal correlated electricity consumption features, and generate feature index vectors that characterize user electricity consumption habits and response potential. The dual-scenario load forecasting module is connected to the user electricity behavior feature extraction module. It is used to analyze time series features based on feature index vectors and calculate the basic forecast load of each control time slot. The multi-source load coordinated control module is connected to the dual-scenario load forecasting module and the intelligent load monitoring module. It is used to obtain the load deviation calculated from the actual load and the forecast load of the previous control time slot, the dynamic load elasticity coefficient determined by the relative ratio of the load response change and the relative ratio of the electricity price change in the current control time slot, and the real-time supply and demand imbalance of the distribution network. It calculates the prediction correction factor that decays exponentially with the sampling time interval, feeds the prediction correction factor back to the dual-scenario load forecasting module, corrects the basic forecast load in the pre-control stage, generates the target load, and calculates and issues control instructions to the intelligent load monitoring module based on the target load and user comfort constraints.
[0006] Preferably, the intelligent load monitoring module has a built-in hardware encryption security unit and accesses the metering master station through a converged communication method of private network and public network, with a sampling period of 1 minute to 15 minutes; the equipment operation status includes the real-time temperature and operation mode of electric heating equipment with thermal inertia characteristics, the state of charge of distributed energy storage system, and the real-time power and connection status of electric vehicle charging piles.
[0007] Preferably, when the user electricity consumption behavior feature extraction module extracts spatiotemporally correlated electricity consumption features and generates feature index vectors, it constructs a time series array and a spatial topological adjacency relationship graph based on multi-source load electricity consumption data and equipment operating status quantities. It uses a convolutional neural network to extract local time series features reflecting daily electricity consumption cycle changes and peak-valley extremes along the time dimension, and uses a graph neural network to extract spatial global topological features between different distribution nodes along the spatial topological adjacency relationship graph. It then concatenates the local time series features and the spatial global topological features in parallel to generate a feature index vector containing qualitative and quantitative dimensions and response elasticity dimensions.
[0008] Preferably, in normal scenarios, the dual-scenario load forecasting module decomposes historical load time series into seasonal and trend components, extracts long-term dependencies and periodic patterns through recurrent neural networks to generate basic forecast load amounts; in abnormal and sudden scenarios, the forecast baseline is corrected based on meteorological change gradients or new energy output change gradients and historical response elasticity.
[0009] Preferably, when the multi-source load coordinated control module calculates the dynamic load elasticity coefficient, the dynamic load elasticity coefficient is determined by the ratio of the relative ratio of the load response change to the relative ratio of the electricity price or excitation signal change.
[0010] Preferably, when the multi-source load coordinated control module calculates control commands based on a multi-objective optimization model, the multi-objective optimization model takes minimizing the peak-valley difference cost of the distribution network and minimizing the user comfort loss cost as optimization objectives. The user comfort loss cost is determined by the deviation between the actual indoor temperature and the set comfort temperature.
[0011] Preferably, the control instructions include price-based demand response tariff signals for flexible loads, and direct load control instructions for specific interruptible loads in the event of a sudden emergency in the power grid.
[0012] Preferably, the controlled load equipment regulated by the intelligent load monitoring module includes electric heating equipment with thermal inertia characteristics, distributed energy storage systems, and electric vehicle charging piles. The multi-source load collaborative regulation module combines the equipment operating status data and utilizes the heat storage characteristics of electric heating equipment and the charging and discharging response of distributed energy storage systems to smooth out secondary load peaks generated during the load recovery phase.
[0013] Preferably, when the multi-source load collaborative control module switches control commands between adjacent control time slots, if the prediction correction factor exceeds the preset power safety threshold range, the prediction correction factor is limited to the preset power safety threshold range and sent to the intelligent load monitoring module in a step-by-step manner according to a preset step size.
[0014] Preferably, the system also includes a security isolation and authentication module located between the multi-source load collaborative control module and the intelligent load monitoring module. The security isolation and authentication module is used to authenticate and encrypt the issued control commands, and transmit the control commands to the intelligent load monitoring module after the verification is successful.
[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the load regulation of power grid distribution, by establishing a feedback relationship between load forecasting and control decisions, the multi-source load collaborative control module can promptly correct the basic forecast load based on the load deviation, dynamic load elasticity coefficient, and real-time supply and demand imbalance of the distribution network. This can reduce the deviation between the forecast results and the actual load, reduce power fluctuations caused by the lag in control actions, make the load tracking of the distribution network more continuous and stable, and reduce the regulation oscillations caused by supply and demand imbalance.
[0016] 2. The user electricity consumption behavior feature extraction module can reflect the electricity consumption habits and response potential of different users. Based on this, the multi-source load collaborative control module takes into account both the peak-valley difference cost of the distribution network and the user comfort loss cost, and forms control instructions adapted to the load characteristics of different users. Compared with the method of unified control with fixed thresholds, this invention can reduce the impact of forced load reduction on heating comfort. While smoothing the peak-valley difference of the distribution network, it also takes into account the actual electricity demand of different users and improves the degree of cooperation of users in demand response.
[0017] 3. The dual-scenario load forecasting module performs load forecasting for both normal and abnormal emergency scenarios. Under normal operating conditions, it can improve the stability of short-term forecasts based on long-term load changes and cyclical patterns. In cases of cold waves, sudden weather changes, or sudden changes in renewable energy output, it can correct forecast results by combining weather changes and historical response data. This reduces the forecasting bias caused by a single forecasting model under emergency conditions, improves the accuracy of load forecasting under complex operating conditions, and reduces the risk of supply and demand imbalance in the distribution network. Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the data interaction principle of the load regulation system of the present invention; Figure 2 This is a diagram illustrating the hierarchical deployment architecture of the load regulation system of this invention. Figure 3 This is a closed-loop structural diagram of the load regulation system of the present invention.
[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.
[0021] A load regulation system for power grid distribution includes: The intelligent load monitoring module is deployed at the power distribution nodes of residential, commercial and industrial users to collect high-frequency load power consumption data and equipment operating status data from multiple sources. The user electricity consumption behavior feature extraction module is connected to the intelligent load monitoring module. It is used to receive multi-source load electricity consumption data and equipment operating status data, extract spatiotemporal correlated electricity consumption features, and generate feature index vectors that characterize user electricity consumption habits and response potential. The dual-scenario load forecasting module is connected to the user electricity behavior feature extraction module. It is used to analyze time series features based on feature index vectors and calculate the basic forecast load of each control time slot. The multi-source load coordinated control module is connected to the dual-scenario load forecasting module and the intelligent load monitoring module. It is used to obtain the load deviation calculated from the actual load and the forecast load of the previous control time slot, the dynamic load elasticity coefficient determined by the relative ratio of the load response change and the relative ratio of the electricity price change in the current control time slot, and the real-time supply and demand imbalance of the distribution network. It calculates the prediction correction factor that decays exponentially with the sampling time interval, feeds the prediction correction factor back to the dual-scenario load forecasting module, corrects the basic forecast load in the pre-control stage, generates the target load, and calculates and issues control instructions to the intelligent load monitoring module based on the target load and user comfort constraints.
[0022] Preferably, the intelligent load monitoring module has a built-in hardware encryption security unit and accesses the metering master station through a converged communication method of private network and public network, with a sampling period of 1 minute to 15 minutes; the equipment operation status includes the real-time temperature and operation mode of electric heating equipment with thermal inertia characteristics, the state of charge of distributed energy storage system, and the real-time power and connection status of electric vehicle charging piles.
[0023] Preferably, when the user electricity consumption behavior feature extraction module extracts spatiotemporally correlated electricity consumption features and generates feature index vectors, it constructs a time series array and a spatial topological adjacency relationship graph based on multi-source load electricity consumption data and equipment operating status quantities. It uses a convolutional neural network to extract local time series features reflecting daily electricity consumption cycle changes and peak-valley extremes along the time dimension, and uses a graph neural network to extract spatial global topological features between different distribution nodes along the spatial topological adjacency relationship graph. It then concatenates the local time series features and the spatial global topological features in parallel to generate a feature index vector containing qualitative and quantitative dimensions and response elasticity dimensions.
[0024] Preferably, in normal scenarios, the dual-scenario load forecasting module decomposes historical load time series into seasonal and trend components, extracts long-term dependencies and periodic patterns through recurrent neural networks to generate basic forecast load amounts; in abnormal and sudden scenarios, the forecast baseline is corrected based on meteorological change gradients or new energy output change gradients and historical response elasticity.
[0025] Preferably, when the multi-source load coordinated control module calculates the dynamic load elasticity coefficient, the dynamic load elasticity coefficient is determined by the ratio of the relative ratio of the load response change to the relative ratio of the electricity price or excitation signal change.
[0026] Preferably, when the multi-source load coordinated control module calculates control commands based on a multi-objective optimization model, the multi-objective optimization model takes minimizing the peak-valley difference cost of the distribution network and minimizing the user comfort loss cost as optimization objectives. The user comfort loss cost is determined by the deviation between the actual indoor temperature and the set comfort temperature.
[0027] Preferably, the control instructions include price-based demand response tariff signals for flexible loads, and direct load control instructions for specific interruptible loads in the event of a sudden emergency in the power grid.
[0028] Preferably, the controlled load equipment regulated by the intelligent load monitoring module includes electric heating equipment with thermal inertia characteristics, distributed energy storage systems, and electric vehicle charging piles. The multi-source load collaborative regulation module combines the equipment operating status data and utilizes the heat storage characteristics of electric heating equipment and the charging and discharging response of distributed energy storage systems to smooth out secondary load peaks generated during the load recovery phase.
[0029] Preferably, when the multi-source load collaborative control module switches control commands between adjacent control time slots, if the prediction correction factor exceeds the preset power safety threshold range, the prediction correction factor is limited to the preset power safety threshold range and sent to the intelligent load monitoring module in a step-by-step manner according to a preset step size.
[0030] Preferably, the system also includes a security isolation and authentication module located between the multi-source load collaborative control module and the intelligent load monitoring module. The security isolation and authentication module is used to authenticate and encrypt the issued control commands, and transmit the control commands to the intelligent load monitoring module after the verification is successful.
[0031] Example 1: In a high-altitude cold region power distribution network demonstration area, after the peak period of electric heating begins, intelligent load monitoring modules are deployed at the metering cabinets and switchgear distribution nodes of residential, commercial, and industrial users. These modules collect multi-source load power consumption data, such as voltage, current, active power, reactive power, and ambient temperature, on a minute-by-minute basis. Simultaneously, they collect the operating status of heating equipment. Extreme cold waves cause sudden changes in the strong thermal inertia of heating loads, which are superimposed on the random fluctuations in renewable energy output. When load forecasting and regulation are performed using a unidirectional serial method, the load forecasting process cannot reflect the dynamic impact of regulation actions on user electricity consumption behavior and load response capabilities. The regulation link issues load reduction commands according to fixed thresholds. After the regulation is lifted, electric heating equipment is prone to concentrated resumption of operation, leading to secondary load peaks, power tracking phase lag, and power supply and demand oscillations.
[0032] The user electricity consumption behavior feature extraction module receives multi-source load electricity consumption data collected by the intelligent load monitoring module, extracts spatial global topological features and temporal local time series features from it, forms spatiotemporally correlated electricity consumption features, generates feature index vectors characterizing user electricity consumption habits and response potential, and performs user clustering based on these vectors. The dual-scenario load prediction module receives the feature index vectors and, in a normalized scenario, decomposes historical load time series into seasonal and trend components. It then extracts long-term dependencies and periodic patterns using a deep recurrent neural network to calculate the basic predicted load for the current control slot. The baseline load capacity of the regional distribution network is 120MW. It is 120MW.
[0033] The multi-source load coordinated control module sends price-based demand response electricity price signals to the residential electric heating user cluster. The change in this electricity price signal... The initial excitation baseline is 0.3 yuan / kWh. The price is 0.6 yuan / kWh; the intelligent load monitoring module provides feedback on the load adjustment amount of this user cluster. For -6MW, the multi-source load coordinated control module determines the dynamic load elasticity coefficient of the current control time slot based on the ratio of the relative ratio of load response change to the relative ratio of electricity price signal change. The calculation formula is as follows: ,in, The dynamic load elasticity coefficient. for Load adjustment amount at any time, Baseline load, For price-based demand response, the change in electricity price signals. For the initial excitation baseline, subscript This represents the current control time slot. Substituting the above values into the formula, we obtain... It is -0.1.
[0034] The multi-source load coordinated control module further obtains the actual load amount of the previous control time slot. The value is 128MW, and the predicted load from the previous control slot is also included. The calculated load deviation is 123MW. For 5MW, during the sampling time interval 15 min, response decay time constant 30 minutes, preset weighting coefficient 0.8, preset weighting coefficient The real-time supply and demand imbalance of the distribution network is 0.5. Calculate the prediction correction factor that decays exponentially with the sampling time interval, given a 10MW MW requirement. The calculation formula is as follows: ,in, To predict the correction factor, This is the load deviation amount of the previous control time slot. The sampling time interval, Let be the system's response decay time constant. The dynamic load elasticity coefficient. This refers to the real-time supply and demand imbalance in the distribution network. and For weighting coefficients, subscript This represents the previous control time slot. After substituting the above values, the decay exponent term... The predicted correction factor is 0.6065. It is 1.926MW.
[0035] The sign of the real-time supply-demand imbalance in the distribution network is determined according to the system dispatch direction. When the distribution network is in a state of supply shortage, the real-time supply-demand imbalance is negative. This negative value is multiplied by the negative dynamic load elasticity coefficient to form a positive power compensation demand increment, which is used to correct the target load upwards, ensuring that the prediction correction direction is consistent with the dispatch demand. The multi-source load coordinated control module will then use the prediction correction factor. Feedback is sent to the dual-scenario load forecasting module, where it is added to the base forecast load during the pre-control phase to generate the target load. The calculation formula is as follows: ,in, The corrected target load, Based on the predicted load, To predict the correction factor, after substituting the values, the target load is... It is 121.926MW.
[0036] The multi-source load coordination and control module determines the user comfort loss cost based on the corrected target load and, under user comfort constraints, comprehensively considers the deviation between the actual indoor temperature and the set comfort temperature. In addition to the peak-valley difference cost of the distribution network, a multi-objective optimization model is solved to generate control instructions, including price-based demand response electricity price signals and direct load control instructions; in the event of a sudden emergency in the power grid, direct load control instructions are issued to specific interruptible loads.
[0037] The control command is first authenticated and signed by the security isolation and authentication module located between the multi-source load collaborative control module and the intelligent load monitoring module. After the verification is successful, it is transmitted to the intelligent load monitoring module. The intelligent load monitoring module receives the control command through the built-in hardware encryption security unit and drives the distributed energy storage system, electric vehicle charging pile and thermal inertia-type electric heating equipment to perform load regulation.
[0038] After the control command is lifted, electric heating equipment may resume operation around the same time due to the drop in indoor temperature, causing its power to be superimposed on the distribution network side. The multi-source load collaborative control module utilizes the millisecond and second-level charging and discharging responses of distributed energy storage systems and electric vehicles, as well as the minute-level and hour-level thermal inertia time constant differences of thermal storage electric heating equipment, to stagger and disperse the heating load recovery demand, reduce the superposition of recovery power at the same time, and smooth out the secondary load peaks generated during the load recovery phase. This forms a closed-loop data flow and control flow that are interconnected by the prediction correction factor feedback, target load update, and control command issuance, so that the distribution network load power can be continuously tracked and the load can be adjusted according to the real-time supply and demand status.
[0039] Example 2: The closed-loop test platform of the 10kV distribution network demonstration area in the current high-altitude and cold region connects 10 actual physical nodes. Each node is equipped with a set of intelligent load monitoring modules to examine the control stability and power tracking accuracy under extreme cold waves and strong heat inertia heating load change conditions. The sampling period is set to 15 minutes. The input data includes minute-level power, voltage and temperature data of three typical user clusters: civil, commercial and industrial. To simulate electromagnetic interference in industrial sites, Gaussian white noise with a signal-to-noise ratio of 20dB and 50Hz power frequency harmonic interference are superimposed on the multi-source load power consumption data.
[0040] The experiment consisted of an experimental group and three control groups. The experimental group employed a load elasticity-control deviation closed-loop feedforward compensation mechanism. Based on the load deviation, dynamic load elasticity coefficient, and real-time supply-demand imbalance in the distribution network, a prediction correction factor was calculated, and then the baseline predicted load was corrected to generate the target load. Control group one eliminated closed-loop feedforward compensation and directly issued control commands based on the baseline predicted load. Control group two eliminated the dynamic load elasticity coefficient. The calculation fixed the coefficient at -0.05; control group three set the weighting coefficient to a value exceeding the safe convergence interval. and Each group used the same cold wave conditions and initial load baseline, and ran continuously for 8 control time slots, each control time slot being 15 minutes, for a total running time of 120 minutes.
[0041] After the electric heating load impact was introduced at 15 minutes, the load deviation of the experimental group and the three control groups was 5.0 MW. The experimental group used this load deviation, the dynamic load elasticity coefficient of -0.1, and the real-time supply and demand imbalance of the distribution network of 10 MW as inputs, and calculated the prediction correction factor according to the exponential decay relationship with the sampling time interval. The target load is 1.926MW; after adding this forecast correction factor to the base forecast load, the target load is... The power capacity is 121.926MW. The peak-valley difference of the distribution network in this control time slot is reduced by 12.1%, and the deviation between the actual indoor temperature and the set comfort temperature is 0.3℃.
[0042] When the closed-loop control operation reached the 30-minute mark, the load deviation of the test group decreased to 2.1MW, the prediction correction factor decreased accordingly to 0.824MW, and the target load was 120.824MW; the peak-valley difference reduction rate of the distribution network increased to 18.5%, and the temperature deviation was 0.5℃. At the 45-minute mark, the load deviation became -0.8MW, the prediction correction factor changed to -0.312MW, and the target load was adjusted to 119.688MW; the peak-valley difference reduction rate was 22.4%, and the temperature deviation was 0.6℃. At the 60-minute mark, the load deviation returned to 0.3MW and entered the fluctuation range of ±0.3MW, the prediction correction factor was 0.115MW, and the target load was 120.115MW; the peak-valley difference reduction rate reached 24.1%, and the temperature deviation reached the maximum value of 0.8℃ in this test.
[0043] At 75 minutes, the load deviation of the experimental group was -0.2 MW, the prediction correction factor was -0.078 MW, the target load was 119.922 MW, the peak-to-valley difference reduction rate was 24.8%, and the temperature deviation dropped to 0.7℃. At 90 minutes, the load deviation was 0.1 MW, the prediction correction factor was 0.038 MW, the target load was 120.038 MW, the peak-to-valley difference reduction rate was 25.2%, and the temperature deviation was 0.6℃. At 105 minutes, the load deviation was 0.0 MW, and the prediction correction factor was 0.01. With a target load of 120.012MW, the peak-to-valley difference reduction rate was 25.5%, and the temperature deviation was 0.5℃. At the 120-minute mark, the load deviation was 0.1MW, the prediction correction factor was 0.025MW, the target load was 120.025MW, the peak-to-valley difference reduction rate was 25.3%, and the temperature deviation dropped back to 0.4℃. From the 60-minute mark to the 120-minute mark, the load deviation of the test group remained within the aforementioned fluctuation range. The prediction correction factor changed its sign and amplitude according to the deviation direction, and the target load was continuously adjusted accordingly.
[0044] After the closed-loop feedback was removed in the control group, the load deviation increased from 8.3MW at 30 minutes to 12.6MW at 45 minutes, then to -15.4MW at 60 minutes, and to 18.2MW at 75 minutes. Subsequently, it was -11.5MW, 14.8MW, and -9.2MW at 90 minutes, 105 minutes, and 120 minutes, respectively. The positive and negative deviations alternated between adjacent control time slots, corresponding to the secondary load peak and power supply and demand phase oscillation during the load recovery phase.
[0045] After the control group adopted the static elastic coefficient, the load deviation from the 30th minute to the 120th minute was 4.2MW, 3.8MW, -2.9MW, 2.1MW, -1.8MW, 1.5MW and -1.2MW respectively; the absolute value of the residual deviation at the 120th minute was 1.2MW. The fixed elastic coefficient could not be updated with the change of user load response, and the convergence speed of the load deviation was lower than that of the test group.
[0046] After using the weighting coefficients that exceeded the safe convergence range, the load deviations from the 30th minute to the 120th minute in control group 3 were 9.6MW, -14.2MW, 18.8MW, -22.5MW, 25.1MW, -21.3MW and 19.7MW respectively. After the weighting coefficients were increased, the calculation results of the prediction correction factor were amplified, and the load deviation continued to oscillate significantly between adjacent control time slots.
[0047] Under the same cold wave conditions and initial load baseline, the prediction correction factor of the test group changes with the load deviation and changes its sign synchronously when the deviation direction changes, thereby continuously correcting the target load. The user temperature deviation gradually falls back after reaching its maximum value, and the load deviation does not show the divergent oscillation shown in control group 3. The load elasticity-regulation deviation closed-loop feedforward compensation mechanism can run continuously on this closed-loop test platform and simultaneously complete load power tracking, peak-valley difference adjustment and regulation under user comfort constraints.
[0048] Example 3: The current intelligent load monitoring module is deployed at each power distribution node. It continuously collects voltage, current, active power, reactive power, ambient temperature, and heating equipment operating status data at a sampling period of 15 minutes to form multi-source load power consumption data. The user power consumption behavior feature extraction module receives this data and, combined with the electrical connection relationship between different power distribution nodes and the load changes at each sampling time, extracts spatiotemporal correlation power consumption features, generates feature index vectors that characterize user power consumption habits and response potential, and classifies user clusters accordingly.
[0049] During feature extraction, the user electricity consumption behavior feature extraction module arranges voltage, current, active power, and ambient temperature into a time series array at 15-minute sampling intervals. Simultaneously, it constructs a spatial topological adjacency graph based on the electrical connections between nodes in the distribution network. A three-layer convolutional neural network performs convolution operations on the time series array along the time direction to extract local temporal features reflecting daily electricity consumption cycle changes and peak-valley extremes. The graph neural network aggregates information from adjacent nodes along the spatial topological adjacency graph to extract spatial global topological features between different nodes. The user electricity consumption behavior feature extraction module concatenates the two types of features in parallel along the feature channels to form a feature index vector containing 24 qualitative and quantitative dimensions and 8 response elasticity dimensions. This vector is used to characterize user electricity consumption habits, industry load baseline, and load response potential, and to complete user cluster classification.
[0050] The dual-scenario load forecasting module receives feature index vectors. Under normalized scenarios, it decomposes historical load time series into seasonal and trend components. It then extracts long-term dependencies and periodic patterns through deep recurrent neural networks. Finally, it superimposes the two types of forecast results to form short-term basic forecast load at the regional and user cluster levels.
[0051] In abnormal and sudden scenarios, the dual-scenario load forecasting module corrects the forecasting baseline based on the meteorological abrupt change gradient and historical response elasticity. The meteorological abrupt change gradient is determined based on the temperature change and wind speed change values at two adjacent collection times. The temperature change and wind speed change values are divided by the collection time interval, and the results are then weighted and summed. When the meteorological abrupt change gradient reaches a preset threshold of 5°C temperature drop per hour, the dual-scenario load forecasting module retrieves the actual power consumption rebound data of the user cluster in the same historical cooling event, and obtains the load growth ratio corresponding to the unit temperature drop as the historical response elasticity. The historical response elasticity is multiplied by the current meteorological abrupt change gradient to obtain the abrupt load increment, and this increment is superimposed on the forecasting baseline under normal scenarios to correct the basic forecasting load in abnormal and sudden scenarios.
[0052] The multi-source load coordinated control module is connected to the dual-scenario load forecasting module and the intelligent load monitoring module to form a closed-loop feedback channel. In each control time slot, the module determines the current load response potential and response sensitivity of each user cluster based on the characteristic index vectors of different user clusters, and obtains the load deviation of the previous control time slot, the dynamic load elasticity coefficient of the current control time slot, and the real-time supply and demand imbalance of the distribution network. Then, the load deviation is exponentially decayed according to the sampling time interval and the response decay time constant, and the prediction correction factor is calculated by combining the dynamic load elasticity coefficient, the real-time supply and demand imbalance of the distribution network, and the preset adaptive weighting coefficient.
[0053] The dual-scenario load forecasting module adds a forecast correction factor to the base forecast load in the current control time slot to generate a corrected target load. The multi-source load collaborative control module uses the target load as the control benchmark, taking into account the user comfort loss cost. Based on the peak-valley difference cost of the distribution network, a multi-objective optimization model is solved to generate price-based demand response electricity price signals or direct load control commands, which are then sent to each intelligent load monitoring module to drive distributed energy storage systems, electric vehicle charging piles, and electric heating equipment to perform peak shaving and valley filling regulation.
[0054] Example 4: This example combines Figures 1 to 3 This section describes a load regulation system for power grid distribution, such as... Figure 1As shown, the intelligent load monitoring module collects high-frequency multi-source load power consumption data and equipment operating status data, and transmits the multi-source load power consumption data to the user power consumption behavior feature extraction module; the user power consumption behavior feature extraction module extracts spatiotemporally correlated power consumption features and generates feature index vectors, which are then transmitted to the dual-scenario load prediction module; the dual-scenario load prediction module analyzes time series features and calculates the basic predicted load, which is then transmitted to the multi-source load collaborative control module; the multi-source load collaborative control module calculates the exponential decay prediction correction factor to generate the target load and control instructions, and simultaneously feeds back the prediction correction factor to the dual-scenario load prediction module, and transmits the target control instructions to the security isolation and authentication module; the security isolation and authentication module performs identity authentication and encrypted signature on the control instructions, and transmits the resulting encrypted signature and authentication instructions to the intelligent load monitoring module.
[0055] like Figure 2 As shown, the metering master station internally includes a user electricity consumption behavior feature extraction module, a dual-scenario load prediction module, and a multi-source load collaborative control module. The user electricity consumption behavior feature extraction module outputs signals to the dual-scenario load prediction module, and both the user electricity consumption behavior feature extraction module and the dual-scenario load prediction module output signals to the multi-source load collaborative control module. The output end of the metering master station is connected to a security isolation and authentication module, which connects to the intelligent load monitoring module in the metering cabinet and switchgear distribution nodes via a private network and public network converged communication method. The metering cabinet and switchgear distribution nodes are internally equipped with intelligent load monitoring modules, which integrate hardware encryption security units. The metering cabinet and switchgear distribution nodes are also connected downwards to electric heating equipment, distributed energy storage systems, and electric vehicle charging piles, respectively.
[0056] like Figure 3 As shown, the output of the intelligent load monitoring module is connected to the input of the user electricity consumption behavior feature extraction module, the output of the user electricity consumption behavior feature extraction module is connected to the input of the dual-scenario load prediction module, the output of the dual-scenario load prediction module is connected to the input of the multi-source load collaborative control module, the output of the multi-source load collaborative control module is connected to the input of the security isolation and authentication module, and the multi-source load collaborative control module has a feedback output connected to the feedback input of the dual-scenario load prediction module, and the output of the security isolation and authentication module is connected to the feedback input of the intelligent load monitoring module, forming a closed-loop signal transmission path between the modules of the system.
[0057] Example 5: When the system is deployed on-site, initial operating parameters are loaded, and the sampling time interval is set. Set to 15 minutes, response decay time constant Set to 30 minutes, adaptive weighting coefficients Set to 0.8, adaptive weighting coefficient The response decay time constant is set to 0.5, determined based on the thermal inertia delay of the electric heating equipment from temperature drop to the point where the user experiences a significant change in perceived temperature. When this parameter is below 15 minutes, the adjustment frequency of the control command is too high; when this parameter is above 60 minutes, the load deviation cannot be mitigated in a timely manner. (Adaptive weighting coefficient...) and Through simulation using offline historical data, it was determined that within the convergence domain constrained by the eigenvalues where the weighted sum of the two factors is less than 1.5, the prediction correction factor maintains monotonically convergent during the recursive calculation process, thus avoiding self-excited oscillations and uncontrolled divergence in the system.
[0058] During the initial system calibration process, maintenance personnel use the intelligent load monitoring module to read historical power data from each distribution node and calculate the baseline load of residential, commercial, and industrial user clusters under the current temperature gradient. and basic forecast load and the load deviation of the previous control time slot. And the real-time supply and demand imbalance of the distribution network during the current control time slot. Input the multi-source load coordinated control module to complete the initial state calibration and feedforward compensation loop joint debugging.
[0059] In the normalization of load regulation, the intelligent load monitoring module collects the actual changes in active power of residential, commercial, and industrial user clusters within 15 minutes before and after the issuance of control commands. The multi-source load coordinated control module multiplies the actual changes in active power of each user cluster by the corresponding response weight coefficient, where the response weight coefficient for residential user clusters is 0.5, for commercial user clusters it is 0.3, and for industrial user clusters it is 0.2. The weighted results of the three types of user clusters are summed to obtain the normalized aggregated single scalar load regulation, which is then used for the quantitative calculation of the dynamic load elasticity coefficient.
[0060] Example 6: Before the multi-source load coordinated control module is put into operation, the system completes parameter calibration and baseline model construction through an offline testing platform. The offline testing platform inputs a historical electricity load time series set, extracts seasonal and trend components using a time-series decomposition method, and performs sampling according to a preset sampling time interval. and response decay time constant For weighting coefficients and Optimization calibration was performed to determine the prediction correction factor under standard test conditions. The boundary threshold.
[0061] The preset power safety threshold range is set at -10% to 10% of the rated capacity of the distribution transformer, corresponding to -12MW to 12MW. When the multi-source load collaborative control module switches control commands between adjacent control time slots, it first determines whether the prediction correction factor exceeds this range. If it does not exceed this range, it performs subsequent processing according to the calculation results. If it exceeds this range, the prediction correction factor is limited to -12MW to 12MW, and then issued to the intelligent load monitoring module in a stepped manner with a preset step size of 1.5MW each time, so as to avoid local voltage fluctuations or equipment overload in the distribution network caused by excessive single power adjustment amplitude.
[0062] The system uses the calibrated parameter matrix as the operating benchmark. It receives minute-level active power, reactive power, and temperature data uploaded by each distribution node through the intelligent load monitoring module. The multi-source load collaborative control module calculates the peak-valley difference cost of the distribution network based on the target load of each control time slot, and calculates the user comfort loss cost based on the deviation between the actual indoor temperature and the set comfort temperature. Among them, the peak-valley difference cost of the distribution network is obtained by multiplying the difference between the maximum and minimum values of the target load within the forecast period by the unit price of operation and maintenance compensation for the peak-valley difference; the user comfort loss cost is obtained by multiplying the absolute difference between the actual indoor temperature and the set comfort temperature by the unit price of temperature deviation loss.
[0063] The multi-objective optimization model aims to minimize the sum of peak-valley difference cost and user comfort loss cost in the distribution network. It sets the upper limit of rated capacity of distribution transformers, the maximum charging and discharging power boundary of distributed energy storage system, and the maximum allowable temperature deviation range for users. In the solution process, particle swarm optimization algorithm or linear programming algorithm is used to search for the combination of price-based demand response electricity price signal and direct load control command within the feasible domain defined by each constraint condition, and generate control commands to be issued to each controlled load device.
[0064] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A load regulating system for power distribution of an electrical grid, characterized in that, The system includes: The intelligent load monitoring module is deployed at the power distribution nodes of residential, commercial and industrial users to collect high-frequency load power consumption data and equipment operating status data from multiple sources. The user electricity consumption behavior feature extraction module is connected to the intelligent load monitoring module. It is used to receive multi-source load electricity consumption data and equipment operating status data, extract spatiotemporal correlated electricity consumption features, and generate feature index vectors that characterize user electricity consumption habits and response potential. The dual-scenario load forecasting module is connected to the user electricity behavior feature extraction module. It is used to analyze time series features based on feature index vectors and calculate the basic forecast load of each control time slot. The multi-source load coordinated control module is connected to the dual-scenario load forecasting module and the intelligent load monitoring module. It is used to obtain the load deviation calculated from the actual load and the forecast load of the previous control time slot, the dynamic load elasticity coefficient determined by the relative ratio of the load response change and the relative ratio of the electricity price change in the current control time slot, and the real-time supply and demand imbalance of the distribution network. It calculates the prediction correction factor that decays exponentially with the sampling time interval, feeds the prediction correction factor back to the dual-scenario load forecasting module, corrects the basic forecast load in the pre-control stage, generates the target load, and calculates and issues control instructions to the intelligent load monitoring module based on the target load and user comfort constraints.
2. A load regulating system for power distribution in an electrical grid according to claim 1, characterized in that, The intelligent load monitoring module has a built-in hardware encryption security unit and connects to the metering master station through a converged communication method of private network and public network. The sampling period is 1 minute to 15 minutes. The equipment operation status includes the real-time temperature and operation mode of electric heating equipment with thermal inertia characteristics, the state of charge of distributed energy storage system, and the real-time power and connection status of electric vehicle charging piles.
3. A load regulating system for power distribution in an electrical grid according to claim 1, characterized in that, When the user electricity consumption behavior feature extraction module extracts spatiotemporally correlated electricity consumption features and generates feature index vectors, it constructs a time series array and a spatial topological adjacency graph based on multi-source load electricity consumption data and equipment operating status quantities. It uses a convolutional neural network to extract local time series features reflecting daily electricity consumption cycle changes and peak-valley extremes along the time dimension, and uses a graph neural network to extract spatial global topological features between different distribution nodes along the spatial topological adjacency graph. The local time series features and spatial global topological features are then concatenated in parallel to generate a feature index vector containing qualitative and quantitative dimensions and response elasticity dimensions.
4. The load regulating system for power distribution of claim 1, wherein, In normal scenarios, the dual-scenario load forecasting module decomposes historical load time series into seasonal and trend components, extracts long-term dependencies and periodic patterns through recurrent neural networks to generate basic forecast load amounts. In abnormal and sudden scenarios, the forecast baseline is corrected based on meteorological change gradients or new energy output change gradients and historical response elasticity.
5. The load regulating system for power distribution of claim 1, wherein, When the multi-source load coordinated control module calculates the dynamic load elasticity coefficient, the dynamic load elasticity coefficient is determined by the ratio of the relative ratio of the load response change to the relative ratio of the electricity price or excitation signal change.
6. The load regulating system for power distribution of claim 1, wherein, When the multi-source load coordinated control module calculates control commands based on the multi-objective optimization model, the multi-objective optimization model takes minimizing the peak-valley difference cost of the distribution network and minimizing the user comfort loss cost as optimization objectives. The user comfort loss cost is determined by the deviation between the actual indoor temperature and the set comfort temperature.
7. The load regulating system for power distribution of claim 1, wherein, The control instructions include price-based demand response tariff signals for flexible loads, and direct load control instructions for specific interruptible loads in the event of a sudden emergency in the power grid.
8. The load regulating system for power distribution of claim 1, wherein, The controlled load equipment regulated by the intelligent load monitoring module includes electric heating equipment with thermal inertia characteristics, distributed energy storage systems, and electric vehicle charging piles. The multi-source load collaborative regulation module combines the equipment operating status data and utilizes the heat storage characteristics of electric heating equipment and the charging and discharging response of distributed energy storage systems to smooth out secondary load peaks generated during the load recovery phase.
9. The load regulating system for power distribution of claim 1, wherein, When the multi-source load coordinated control module switches control commands between adjacent control time slots, if the prediction correction factor exceeds the preset power safety threshold range, the prediction correction factor will be limited to the preset power safety threshold range and sent to the intelligent load monitoring module in a step-by-step manner according to the preset step size.
10. The load regulating system for power distribution of claim 1, wherein, The system also includes a security isolation and authentication module located between the multi-source load collaborative control module and the intelligent load monitoring module. The security isolation and authentication module is used to authenticate and encrypt the control commands issued, and transmit the control commands to the intelligent load monitoring module after the verification is successful.
Citation Information
Patent Citations
Multi-element load collaborative optimization regulation and control method
CN121886738A
