Power supply and distribution dynamic optimization method and system based on load space-time evolution

Abstract

The application discloses a power supply and distribution dynamic optimization method and system based on load space-time evolution, and belongs to the technical field of power system automation, which comprises the following steps: obtaining power supply and distribution capacity measurement data and power distribution network structure data, generating network basic data sets for analysis and fusion, generating a power supply and distribution network state image, then generating a space unit load matrix and performing decomposition and space correlation, generating load space-time evolution characteristics and performing calculation, generating an optimization demand vector, performing action influence assessment, generating an action influence matrix, executing action conflict resolution, generating an executable optimization strategy, issuing and collecting execution feedback, generating optimization closed-loop feedback data, and performing parameter updating. The application adopts the means of extracting load space-time evolution characteristics for prospective demand prediction, generating an optimization action sequence through quantitative assessment and collaborative arrangement, and performing closed-loop adaptive correction by using execution feedback, so that active and global regulation and control of the power distribution network operation state can be realized.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and in particular to a method and system for dynamic optimization of power supply and distribution based on load spatiotemporal evolution. Background Technology

[0002] Currently, the AC distribution network, as the final link in the power system, directly serves a large number of electricity users. Its safe, stable, and economical operation is crucial for ensuring social production and residents' lives. Dynamic optimization of power supply and distribution refers to adjusting controllable equipment in the network, such as transformer tap changes, reactive power compensation devices, and tie switches, to optimize the network's voltage distribution and power flow in real time, thereby reducing network losses, improving power quality, and ensuring grid security.

[0003] In related technologies, Chinese invention patent CN120511663A discloses an adaptive optimization method and system for power system load forecasting. This method includes collecting and fusing multi-source parameters and operation and maintenance event data, constructing a feature matrix with a unified time step, establishing a graph structure based on the power grid topology and extracting node features, and generating an initial population based on clustering. Individuals are iteratively optimized using a self-learning differential evolution algorithm, and control parameters are dynamically adjusted based on prediction error and fitness. A multi-model fusion strategy is then employed to fuse individual prediction results, ultimately achieving population adjustment and convergence determination based on error trend feedback, and outputting the optimal parameter combination and prediction results.

[0004] However, the above-mentioned technical solutions lack a mechanism to transform the prediction results into executable control strategies, and only stop at prediction; their optimization objectives are limited to prediction accuracy, without considering the actual operation objectives of the power grid such as loss reduction and quality improvement, and they cannot coordinate multiple control measures; and their feedback loop is only used to optimize the prediction model itself, and cannot learn and correct from the actual physical effects of the control actions. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a dynamic optimization method and system for power supply and distribution based on load spatiotemporal evolution. It employs methods such as extracting load spatiotemporal evolution characteristics for forward-looking demand forecasting, generating optimization action sequences through quantitative evaluation and collaborative orchestration, and utilizing execution feedback for closed-loop adaptive correction. This enables proactive and global control of the distribution network's operating status.

[0006] The above objectives can be achieved through the following approach: A method and system for dynamic optimization of power supply and distribution based on load spatiotemporal evolution includes: acquiring power supply and distribution measurement data and distribution network structure data and aligning them to generate a network basic dataset; parsing the network basic dataset to obtain node state variables and branch state variables, and performing state fusion to generate a power supply and distribution network state profile; extracting load sequences from the power supply and distribution network state profile and dividing them into spatial units to generate a spatial unit load matrix; performing temporal decomposition and spatial correlation calculation on the spatial unit load matrix to generate load spatiotemporal evolution features; and inputting the load spatiotemporal evolution features into... Constraint assessment is performed, and power flow consistency calculation is conducted in conjunction with the power supply and distribution network status profile to generate an optimization demand vector. Based on the optimization demand vector, a set of candidate optimization actions is generated from the adjustment resources, and the impact of these actions is assessed to generate an action impact matrix. Based on the action impact matrix, action conflict resolution is performed, and an action sequence arrangement is formed to generate an executable optimization strategy. The executable optimization strategy is issued, and execution feedback is collected to generate optimization closed-loop feedback data. The optimization closed-loop feedback data is used to update the load spatiotemporal evolution characteristics and the action impact matrix to generate update parameters for the next optimization cycle.

[0007] Optionally, the generation of the network basic dataset includes: acquiring voltage measurement parameters, current measurement parameters, and active and reactive power measurement parameters from terminal measurement devices to generate a measurement parameter set; acquiring switch position parameters, transformer tap changer status parameters, and reactive power compensation device status parameters from the operation management system to generate a device status parameter set; acquiring the topology connection relationship and line impedance parameters of the power distribution network and aligning them with the measurement parameter set and the device status parameter set using timestamps to generate the network basic dataset.

[0008] Optionally, generating the power supply and distribution network status profile includes: identifying abnormal measurements in the network basic dataset and performing confidence calibration to generate a set of reliable measurements; fusing the set of reliable measurements with the topological connectivity and performing state estimation to generate node state variables; calculating branch power flow and branch current from the node state variables and the line impedance parameters to generate branch state variables; and merging the node state variables and the branch state variables to form a searchable index to generate a power supply and distribution network status profile.

[0009] Optionally, generating the spatial unit load matrix includes: acquiring node load parameters and transformer area affiliation information from the power supply and distribution network status profile and performing spatial mapping to generate a spatially mapped load sequence; aggregating the spatially mapped load sequence by time slices to form a time-aligned sequence, thereby generating a time-slice load sequence; assembling the time-slice load sequence by spatial units while maintaining the spatial unit identifier to generate a spatial unit load matrix.

[0010] Optionally, the generation of load spatiotemporal evolution features includes: separating trend terms and extracting fluctuation terms from the spatial unit load matrix to generate load component features; performing spatial correlation calculation and migration direction identification on the load component features to generate spatial migration features; and jointly encoding the load component features and the spatial migration features to generate load spatiotemporal evolution features.

[0011] Optionally, generating the optimization demand vector includes: acquiring power quality constraint parameters, equipment operation constraint parameters, and grid safety constraint parameters and performing unified dimension processing to generate a constraint parameter set; performing constraint deviation quantization calculation based on the load spatiotemporal evolution characteristics to generate a constraint deviation vector; performing importance ranking on the constraint deviation vector and extracting constraint trigger links to generate an optimization demand vector.

[0012] Optionally, generating the action impact matrix includes: acquiring the status of adjustable resources and classifying them according to adjustment type to generate an adjustable resource catalog; mapping the optimization demand vector to the adjustable resource catalog to generate a candidate optimization action set; injecting the candidate optimization action set into the power supply and distribution network status profile one by one and performing power flow simulation to generate an action result set; extracting the impact of the action result set on voltage deviation, line loss change and operation cost and performing unified measurement to generate an action impact matrix.

[0013] Optionally, generating an executable optimization strategy includes: identifying mutually reinforcing and mutually canceling action pairs based on the action influence matrix, and generating an action coupling graph; performing conflict determination and constraint feasibility screening on the candidate optimization action set according to the action coupling graph, and generating an action subset; arranging the action subset in a temporal sequence and introducing execution window constraints to generate an action sequence; verifying the executability of the action sequence and forming a distribution format to generate an executable optimization strategy.

[0014] Optionally, generating the update parameters for the next tuning cycle includes: sending the executable tuning strategy to the execution terminal and recording the delivery receipt to generate a strategy execution record; acquiring the measurement data and device status parameters after execution and associating them with the strategy execution record to generate tuning closed-loop feedback data; calculating the strategy deviation from the tuning closed-loop feedback data and extracting the source of the deviation to generate a strategy deviation feature; using the strategy deviation feature to correct the spatial migration feature in the load spatiotemporal evolution feature and updating the action influence quantity in the action influence matrix to generate the update parameters for the next tuning cycle.

[0015] Based on the same inventive concept, this invention also provides a power supply and distribution dynamic optimization system based on load spatiotemporal evolution. The system includes: a data generation module for acquiring power supply and distribution measurement data and distribution network structure data, aligning them, and generating a network basic dataset; a state profiling module for parsing the network basic dataset to obtain node state variables and branch state variables, and performing state fusion to generate a power supply and distribution network state profile; a load matrix module for extracting load sequences from the power supply and distribution network state profile and dividing them into spatial units to generate a spatial unit load matrix; a spatiotemporal evolution module for performing temporal decomposition and spatial correlation calculations on the spatial unit load matrix to generate load spatiotemporal evolution characteristics; and a demand calculation module. The system comprises the following modules: a calculation module, which inputs the load spatiotemporal evolution characteristics into constraint evaluation and combines them with the power supply and distribution network status profile to perform power flow consistency calculation and generate an optimization demand vector; an action evaluation module, which generates a set of candidate optimization actions from the adjustment resources based on the optimization demand vector and performs action impact evaluation to generate an action impact matrix; a strategy orchestration module, which performs action conflict resolution based on the action impact matrix and forms an action sequence orchestration to generate an executable optimization strategy; and a closed-loop update module, which issues the executable optimization strategy and collects execution feedback to generate optimization closed-loop feedback data, and uses the optimization closed-loop feedback data to update the load spatiotemporal evolution characteristics and the action impact matrix to generate update parameters for the next optimization cycle.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention, by constructing load spatiotemporal evolution characteristics, can deeply explore the inherent correlations and migration patterns of loads in time and space dimensions, improving the prediction accuracy of future load changes. This makes optimization decisions no longer based solely on historical data and the current state, but rather forward-looking, enabling preventative adjustments before potential power quality problems or equipment exceeding limits occur, thus achieving proactive avoidance of grid operation risks.

[0017] 2. This invention achieves global collaborative optimization of multiple regulation resources by generating an action influence matrix and resolving action conflicts and arranging their order. It evaluates the comprehensive impact of each optimization action on the entire network, reduces systemic chain reactions or mutual cancellations caused by local or single regulation measures, and ensures that the generated optimization strategy achieves the best balance between voltage quality, network loss and operating cost as a whole, thereby improving the global effectiveness of regulation.

[0018] 3. This invention establishes a complete closed-loop feedback and adaptive update mechanism. By collecting actual data after strategy execution, it can quantify the deviation between expected results and actual outcomes, trace the source of deviation, and continuously correct and optimize the internal load forecasting model and action impact assessment model. This enables the system to have self-learning and evolutionary capabilities, continuously adapting to changes in power grid topology and user electricity consumption behavior, and maintaining high accuracy in optimization decisions over the long term.

[0019] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the dynamic optimization method for power supply and distribution based on load spatiotemporal evolution according to an embodiment of the present invention.

[0022] Figure 2 This is a thermal diagram of the spatial unit load matrix according to an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the cross-correlation analysis of spatial load migration according to an embodiment of the present invention.

[0024] Figure 4 This is a heatmap of the action influence matrix according to an embodiment of the present invention.

[0025] Figure 5 This is a schematic diagram of the power supply and distribution dynamic optimization system based on load spatiotemporal evolution according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 One embodiment of the present invention proposes a dynamic optimization method for power supply and distribution based on load spatiotemporal evolution. It uses the extraction of load spatiotemporal evolution characteristics for forward-looking demand forecasting, generates optimization action sequences through quantitative evaluation and collaborative arrangement, and uses execution feedback for closed-loop adaptive correction, thereby enabling proactive and global control of the distribution network operation status.

[0028] The method described in this embodiment specifically includes: Optionally, the underlying dataset for generating the network includes: Acquire voltage measurement parameters, current measurement parameters, and active and reactive power measurement parameters from the terminal measurement device, and generate a set of measurement parameters; Obtain switch position parameters, transformer tap changer status parameters, and reactive power compensation device status parameters from the operation management system, and generate a set of equipment status parameters; The topological connections and line impedance parameters of the power distribution network are obtained and timestamped with the measurement parameter set and the equipment status parameter set to generate a network basic dataset.

[0029] Specifically, the process begins by actively requesting data from terminal measurement devices in the distribution network via standard communication protocols, such as Modbus. These devices typically include feeder terminal units installed at key feeder nodes, distribution transformer monitoring terminal units, and smart meters on the user side. The collected data constitutes a set of measurement parameters, primarily reflecting power quality (voltage parameters such as phase voltage amplitudes), current parameters reflecting line load levels (such as branch current amplitudes), and active and reactive power parameters reflecting power flow. The data collection frequency is typically set between 1 and 15 minutes, and each record must be accompanied by a timestamp accurate to the second. Simultaneously, the data generation module interacts with the power grid's operation management system, such as the distribution automation master station or distribution management system (DMS), to obtain a set of equipment status parameters describing the power grid's operating mode. This parameter set includes switch position parameters reflecting network topology changes (values ​​are 0 or 1), transformer tap status parameters affecting voltage regulation (usually integer values, such as -3 to +3), and reactive power compensation device status parameters regulating reactive power, such as the number of capacitor banks switched. Data fusion and alignment are crucial for generating the final network foundation dataset. Static distribution network topology connections and line impedance parameters are loaded from a network model database, typically integrated with a Geographic Information System (GIS) or asset management system. Topology connections define the connection methods between nodes and branches, while line impedance parameters provide the resistance and reactance per unit length of the line. The core step is timestamp alignment, setting a reference synchronization time point, such as a whole 15-minute interval. Then, a small time tolerance window, such as ±30 seconds, is set around this time point. Within this window, the measurement parameter set and equipment status parameter set are searched, and data records with timestamps closest to the reference time point are selected. If a key data point is missing within the time window, a preset strategy is used, such as using data from the previous moment or performing short-term interpolation, to ensure the integrity of the dataset. The final generated network base dataset ,have: ; Where T represents the network topology; R represents the set of resistance parameters for all branches; and Z represents the set of reactance parameters for all branches. The set of measurement parameters aligned at the reference time t includes the voltage, current, active power, and reactive power of all measurement points. This is the set of device state parameters aligned to the reference time t, containing the state information of all controllable devices. This dataset structurally stores the complete physical and operational information of the distribution network at a specific time, providing high-quality input for subsequent state estimation and spatiotemporal evolution analysis.

[0030] For example, to construct a basic network dataset containing the operational status of a local power distribution network in a city at a specific time, using 10:15:00 AM on June 1st, XX year as the base time, requests were initiated via the Modbus protocol to the FTU-105 feeder terminal unit and the DTU-21_B distribution transformer monitoring terminal unit located on the 10 kV feeder FDR-01. At 10:14:58, FTU-105 reported the voltage measurement parameters of its node as 10.25 kV and the current measurement parameters as 150 amperes; at 10:15:03, DTU-21_B reported the total active and reactive power measurement parameters of the low-voltage side of the distribution transformer it monitored as 85 kW and 20 kVvar. These dynamic data constituted a preliminary set of measurement parameters. Simultaneously, the equipment status parameters were queried from the distribution automation master station. The query results show that at 10:14:50, the switch position parameter of the tie switch SW-112 on the feeder FDR-01 branch is 1, indicating that it is in the closed state; at 10:15:10, the transformer tap position parameter of the No. 1 main transformer in the main substation is +2. The fixed distribution network topology connection relationship for this area was loaded from the geographic information system database, confirming that FTU-105 is located at node N12, DTU-21_B serves node N21, and nodes N12 and N21 are connected by line L-12-21. The line impedance parameters were also obtained, such as the resistance per unit length of L-12-21 being 0.16 ohms / km and the reactance being 0.12 ohms / km. During timestamp alignment, a tolerance window of ±30 seconds is set with 10:15:00 as the center, matching the measurement data at 10:14:58 and the status data at 10:15:10 to the snapshot time of 10:15. If the smart meter at node N30 fails to report data within the window at this time, its measurement value at 10:00 is used as a supplement. Finally, this aligned and fused information is stored in a structured manner to form the network's basic dataset.

[0031] Optionally, generating the power supply and distribution network status profile includes: Identify and validate the confidence level of outliers in the network's basic dataset to generate a set of reliable measurements. By fusing the trusted measurement set with the topological connectivity and performing state estimation, node state variables are generated. The branch power flow and branch current are calculated by combining the node state variables with the line impedance parameters, and branch state variables are generated. The node status variables and the branch status variables are merged to form a searchable index, generating a power supply and distribution network status profile.

[0032] Specifically, the built-in abnormal measurement identification algorithm screens the data. For example, it marks potentially bad data by setting thresholds such as voltage deviation exceeding 7% of the rated voltage or current fluctuation rate exceeding 30% / s. Simultaneously, it performs cross-validation between data, such as verifying whether the total injected power of a node is approximately equal to the sum of the power flowing out of all branches of that node. For identified abnormal measurements, instead of simply discarding them, a confidence rating is applied, assigning them a low weight value. For measurements that pass the threshold judgment, such as voltage deviation exceeding ±7% of the rated value or current fluctuation rate exceeding 30% / s, the weight value is set to 0.2. For measurements that pass the consistency check, such as the absolute value of the difference between the node's injected power and outflow power being greater than 5% of the node's total injected power, the weight value is set to 0.4. For measurements that trigger both abnormality criteria simultaneously, the weight value is set to 0.1. Normal measurements that do not trigger any abnormality criteria have a weight value set to 1.0. After this step, the original measurement data is transformed into a set of reliable measurements with confidence weights. Distribution network state estimation typically employs weighted least squares, aiming to find a set of network state variables that best fits all measurement values, namely the voltage magnitude and phase angle at each node. For this given node state variable... Below, the weighted sum of squared errors for all measurement points. ,have: ; in, Let be the vector of nodal state variables to be solved, containing the voltage magnitude and phase angle of all nodes, usually written as , Let be the voltage magnitude at node i. Let n be the voltage phase angle of node i, n be the total number of nodes in the power grid, m be the total number of measurement points, and k be the index number of the measurement point, from 1 to m. The weight of the k-th measurement value is determined by the confidence level calibration during the preprocessing stage; The actual measured value of the k-th measurement point comes from the preprocessed reliable measurement set, such as node voltage amplitude and node injected active power. This is the theoretical value corresponding to the kth measurement calculated based on the currently estimated node state variables. This is the theoretical electrical value (such as power or voltage) of the k-th measurement point calculated based on the currently estimated node voltage amplitude and phase angle, combined with network parameters. Specifically, when the k-th measurement is the node voltage amplitude, When it injects active power into the node, ,in, Let i be the phase angle difference between nodes i and j. Let i be the set of nodes directly connected to node i. Let be the conductance of the element in the i-th row and j-th column of the nodal admittance matrix. Let be the susceptance of the element in the i-th row and j-th column of the nodal admittance matrix; when it is the active power of the branch. , For the series conductance of branch i to j, Let i be the series susceptance of branch i→j.

[0033] For any branch connecting nodes i and j, its branch current and power flow can be calculated from the voltage states of the two endpoints. These calculations constitute the branch state variables, which include the current magnitude, active power, and reactive power of each line in the network. Specifically, this involves calculating the effective value of the current from branch i to j. ,have: ; in, Let be the complex impedance of the branch connecting node i and node j, and its magnitude be used to calculate the current using Ohm's law. For calculating the active power of branch i to j... With reactive power ,have: ; in, The charge susceptance to ground is typically calculated by multiplying the charge susceptance per unit length of the line by the line length. The calculated node state variables and branch state variables are combined to form a complete dataset describing the power grid's operating state. To facilitate efficient use by subsequent modules, a searchable index is created for this dataset. For example, relevant voltage, current, and power information can be quickly retrieved based on node number, line name, or geographical region. This integrated and indexed data constitutes the final generated state profile of the power supply and distribution network.

[0034] For example, to generate a state profile of the power distribution network at 10:15 AM, we first identify anomalies in the measurement parameters of the network's basic dataset. Assuming the voltage measurement parameter of node N15 is 9.1 kV, which is 7% lower than the rated 10 kV (9.3 kV), this measurement is marked as anomaly, and its weight is set to 0.2. Simultaneously, at node N21, based on the sum of the power flowing into and out of the node from each branch, the calculated injected power is 85 kW, while the sum of the outflowing power is 92 kW, with an absolute difference of 7 kW, exceeding the injected power by 5% (4.25 kW). Therefore, the power injection measurement of N21 is also marked as anomaly, and its weight is set to 0.4. The weights of the remaining normal measurement points are all 1.0, thus generating a reliable measurement set. Subsequently, this reliable measurement set is fused with the topology connectivity, and weighted least squares state estimation is performed. Assuming the network is simplified to three nodes, the node state vector to be solved is: Node 1 is the equilibrium node. The objective function is constructed to minimize the weighted sum of squared errors of all measurement points. This minimization problem is solved using iterative algorithms such as Newton-Raphson to obtain a set of optimal node state variables, for example... It is 10.18 kV. It is -1.2 degrees. It is 10.15 kV. The value is -1.5 degrees. Next, we will use this set of precise node state variables to calculate the branch state variables. Taking the branch connecting node 2 and node 3 as an example, its complex impedance is known to be 0.75 ohms. Calculate the effective value of the branch current, and substitute the values ​​to get... (A). Similarly, calculate the active and reactive power of branches 2 to 3. Finally, merge the node state variables, such as voltage magnitude and phase angle, with the branch state variables, such as current and power, and establish a searchable index based on node number N21 or line name L-12-21 to form the final power supply and distribution network state profile.

[0035] Optionally, the generated spatial element load matrix includes: Obtain the node load parameters and transformer area affiliation information from the power supply and distribution network status profile and perform spatial mapping to generate a spatially mapped load sequence; The spatially mapped load sequences are aggregated by time slices to form time-aligned sequences, thereby generating time-sliced ​​load sequences; The time-slice load sequences are assembled into spatial units, while preserving the spatial unit identifiers, to generate a spatial unit load matrix.

[0036] Specifically, the system proactively retrieves the active and reactive load parameters of all load nodes from the power supply and distribution network status profile. Simultaneously, it accesses a pre-set geographic information or asset management database to obtain the transformer substation affiliation information corresponding to each electrical node. This substation affiliation information is crucial for mapping abstract electrical nodes to specific physical service areas, such as a residential community or commercial center. By associating node load parameters with their substation affiliation information, the original load data is transformed from a purely electrical dimension to a geospatial dimension, forming a spatially identified load data stream—a spatially mapped load sequence. Next, the spatially mapped load sequence is normalized in the time dimension. Since the timestamps of the original measurement data may not be perfectly normalized, a standard time slice length is set, such as 15 minutes or 30 minutes. Using this time slice as a unit, the continuous load sequence is segmented and aggregated. For all load data points falling within the same time slice and belonging to the same spatial unit, algorithms such as averaging are used to calculate the representative load value of that spatial unit within that time slice. This process ensures that the load data of all spatial units are strictly aligned on the time axis, forming a time-aligned sequence, i.e., a time-slice load sequence. Finally, the time-slice load sequences are assembled into the final spatial unit load matrix. This matrix is ​​a two-dimensional structure, where one dimension represents consecutive time slices and the other dimension represents the individual spatial units. Each element in the matrix represents the aggregated load value of a specific spatial unit within a specific time slice. During construction, the unique identifier of each spatial unit is strictly maintained, ensuring that the columns of the matrix correspond one-to-one with the actual geographical regions. For the element value of the p-th time slice and the q-th spatial unit in this spatial unit load matrix... ,have: ; in, This represents the total number of all load measurement points in the p-th time slice and q-th spatial unit; The h-th load measurement value can be either active power or reactive power. For time-related conditions, only timestamps falling within the p-th time slice are selected. Measurement points within; For spatial conditions, only spatial unit identifiers are selected. Equal to the unique identifier of the q-th spatial unit Measurement points; This involves summing all load measurements that simultaneously satisfy a specified time slice and a specified spatial unit to obtain the total load within that spatiotemporal unit. Through this process, discrete load data is transformed into a regular mathematical object containing spatiotemporal information, laying the foundation for in-depth analysis of load evolution patterns. For example... Figure 2As shown, the vertical axis of this heatmap represents different spatial units, such as commercial center A and residential area C, while the horizontal axis represents continuous time slices throughout the day. The brightness of each color block in the graph represents the load level of the corresponding spatial unit within a specific time slice; the darker the color, the higher the load. As shown in the figure, the load peaks in commercial areas (A, B) occur during the day, while the load peaks in residential areas (C, D) occur at night.

[0037] For example, load data is first extracted and spatially mapped from the power supply and distribution network status profile generated at 10:15 AM. The load parameters for node N21 are found to be 85 kW active power and 20 kVar reactive power, while the load for node N35 is 120 kW active power and 35 kVar reactive power. By querying the pre-set asset management database, it is found that nodes N21 and N35 belong to the commercial center A distribution area, while nodes N40 and N42 belong to residential area B. Based on this association, the two data points N21:{85kW,20kVar} and N35:{120kW,35kVar} are marked as belonging to commercial center A, forming a spatially mapped load sequence with spatial identifiers. Next, the load sequence is time-aggregated using 15-minute time slices. Assume that during the time slice from 10:15 AM to 10:30 AM, commercial center A has no other load measurement points besides the two nodes mentioned above. These two data points will be aggregated to form a representative load for that spatial unit within that time slice, i.e., the time slice load sequence. If here... The load measurements were 85 kW and 120 kW respectively, both timestamps falling within the 10:15 to 10:30 time slot, and both spatial unit identifiers were unique identifiers for Commercial Center A. (kW). Similarly, the same aggregation operation is performed on residential area B and all other spatial units within each 15-minute time slice. Finally, these calculation results are assembled into a spatial unit load matrix. The rows of this matrix represent consecutive time slices, such as 10:00-10:15, 10:15-10:30, etc., and the columns represent different spatial units, such as commercial center A and residential area B. The element value in the matrix located in the 10:15-10:30 row and the commercial center A column is 102.5 kW. This well-organized two-dimensional data structure intuitively shows the temporal changes in the load of each functional area.

[0038] Optionally, the spatiotemporal evolution characteristics of the generated load include: The spatial unit load matrix is ​​subjected to trend term separation and fluctuation term extraction to generate load component features; Spatial correlation calculation and migration direction identification are performed on the load component characteristics to generate spatial migration characteristics; The load component features and the spatial migration features are jointly encoded to generate load spatiotemporal evolution features.

[0039] Specifically, for each column of the spatial unit load matrix, i.e., the load time series of each spatial unit, a time series decomposition algorithm is used, such as STL decomposition based on local weighted regression, to decompose the original series into a trend term and a fluctuation term. The trend term reflects the smooth, deterministic changes in load over a longer time scale, such as hours or days, for example, the difference in patterns between day and night or weekdays and weekends. The fluctuation term represents the short-term disturbances with greater randomness remaining after removing the trend term. This step transforms the single load series of each spatial unit into two more analytically valuable component series, constituting the load component characteristics. The load component characteristics are analyzed in the spatial dimension, focusing on the correlation between the fluctuation term series of each spatial unit. A cross-correlation function with a time delay is used to calculate the correlation coefficient between any two spatial units, for example, a commercial area and an adjacent residential area. The cross-correlation coefficient between series x and series y is calculated when the time delay is τ. ,have: ; in, Let be the fluctuation component value of spatial unit A, such as a commercial area, at time t; For spatial unit B, such as a residential area, the fluctuation component value at time t; For time delay; Let x be the mean of the sequence. Let y be the mean of the sequence y; This represents the total number of time points in the sequence. A time-based index is used. When the correlation coefficient exceeds a preset threshold (e.g., 0.7) and there is a significant time delay (e.g., 15 to 60 minutes), a load migration relationship is identified. For example, the peak decrease in load in commercial areas always leads the peak increase in load in residential areas by about 30 minutes, revealing the load migration pattern of commuters moving from their workplaces to their residences. By calculating all spatial units across the entire network, spatial migration characteristics describing the intensity and direction of load impact between regions are constructed. Figure 3As shown in the figure, the curve illustrates the change of the cross-correlation function of the load fluctuation term with time delay τ between two different spatial units, such as a commercial area and a residential area. A significant peak appears at a non-zero delay τ, as indicated by the dashed line in the figure, indicating that the load fluctuation of one spatial unit is highly correlated with the fluctuation of the other spatial unit τ time units prior. Finally, the extracted load component features and spatial migration features are jointly encoded to generate the final spatiotemporal evolution features of the load. This process aims to fuse the separate features into a unified, high-dimensional feature vector to comprehensively characterize the evolutionary state of each spatial unit at a specific moment. For each spatial unit and its corresponding time slice, its own load component characteristics, including the trend and fluctuation values ​​at the current moment, are used as basic state variables. At the same time, all migration influences pointing to this spatial unit in the spatial migration characteristics are aggregated as migration influences at the receiving end, including the source unit identifier, migration intensity coefficient, optimal time delay, and migration outputs from this spatial unit to other units, which are used as migration outputs at the source end, to form a fixed-dimensional feature vector. This feature vector is then combined with the local load dynamics and the load diffusion effect in the neighborhood through methods such as concatenation, and finally outputs a joint encoding result, namely the load spatiotemporal evolution characteristics.

[0040] For example, the load time series corresponding to commercial center A might be [...98.7,102.5,105.3,101.0,...] kW. After applying the STL decomposition algorithm, this series is decomposed into a trend term representing the slow increase in daytime workload, such as [...100.1,101.2,102.3,103.4,...], and a fluctuation term reflecting random factors such as customer entry and exit, equipment start-up and shutdown, such as [...-1.4,+1.3,+3.0,-2.4,...]. These two component series together constitute the load component characteristics of commercial center A. Next, we analyze the spatial correlation between commercial center A and the adjacent residential community B. Assume that the load fluctuation term series of residential community B is [...+0.5,+2.5,+4.0,+1.5,...] in the evening. To identify the load migration relationship between the two, we calculate the cross-correlation coefficient with time delay. Setting the time slice length to 15 minutes, we calculate the cross-correlation at a time delay τ=2, i.e., 30 minutes, to verify the load change caused by commuters returning to residential areas from commercial areas. We take a 24-6 hour sequence for calculation, determining the cross-correlation coefficient, where x(t) is the fluctuation value of commercial center A at time t, and y(t) is the fluctuation value of residential area B at time t. If the calculation yields... The value is 0.82, exceeding the threshold of 0.7, indicating that the load fluctuation of commercial center A significantly leads that of residential community B by approximately 30 minutes, demonstrating a clear load migration phenomenon. This finding constitutes a spatial migration feature. Finally, joint encoding is performed. At 6:30 PM, the spatiotemporal evolution features of the load in residential community B are generated. This feature vector will contain: the trend term value (e.g., 250 kW) and fluctuation term value (e.g., +4.0) of residential community B at 6:30 PM, as well as the spatial migration feature describing its impact, i.e., {Source unit: commercial center A, migration intensity: 0.82, time delay: 30 minutes}. This high-dimensional feature vector comprehensively characterizes the load state and dynamic evolution trend of residential community B at this moment.

[0041] Optionally, the generation of the tuning requirement vector includes: Obtain power quality constraint parameters, equipment operation constraint parameters, and grid safety constraint parameters, and perform unified dimension processing to generate a constraint parameter set; Based on the aforementioned spatiotemporal evolution characteristics of the load, constraint deviation quantization calculation is performed to generate a constraint deviation vector; The constraint deviation vector is sorted by importance and the constraint triggering links are extracted to generate the tuning requirement vector.

[0042] Specifically, the parameters obtained include: power supply quality constraint parameters, reflecting power supply standards such as the allowable range of voltage deviation; equipment operation constraint parameters, reflecting the limits of physical equipment such as the maximum allowable current carried by a line; and network structure safety constraint parameters, ensuring the stability of the overall network topology such as the upper limit of node short-circuit current. Since these parameters have different physical units, direct comparison would lead to dimensional conflicts. Therefore, a unified dimensionality processing method is used, dividing by their respective benchmark values ​​and then weighted summing to generate a dimensionless comprehensive evaluation index, i.e., a constraint parameter set. The constraint parameter set values ​​for calculating branch i are then used. ,have: ; in, These are power supply quality constraint parameters; This serves as a baseline value for power quality. These are the equipment operation constraint parameters; This serves as the baseline value for equipment operation. These are the safety constraint parameters for the space frame; This serves as the safety benchmark value for the space frame. , , The values ​​represent the corresponding dimensionless weighting coefficients. The benchmark values ​​are set based on the State Grid operation standard documents, and the weighting coefficients are set based on the risk ratio derived from the statistical frequency of historical over-limit accidents. The load spatiotemporal evolution characteristics extracted in the preceding steps are then input into the power flow consistency calculation. Power flow consistency calculation refers to superimposing the predicted dynamic load changes onto the current network state based on the law of conservation of energy and the network physical topology, to predict the voltage and power distribution of each node in the future. In this process, combined with the power supply and distribution network state profile, the predicted future operating state values ​​are compared with the safety boundaries set by the constraint parameter set to calculate the specific proportion exceeding the safety threshold, i.e., the constraint deviation. The dimensionless constraint deviation of branch i is calculated. ,have: ; in, These are dimensionless mapping coefficients; This is the baseline value for active power; This is the apparent power reference value; The spatiotemporal evolution characteristics of the load at node i; To create a profile of the power supply and distribution network status for node i, features are extracted and dimensionality reduced from real-time monitoring data of the power supply and distribution network. The constraint deviations of all nodes and branches are arranged in an array according to spatial topology, generating a constraint deviation vector. Finally, the constraint deviation vectors are ranked by importance. Importance ranking is an evaluation process that comprehensively considers the severity of limit violations and the node's pivotal position in the network. The priority score for calculating the optimization requirements of branch i is then determined. ,have: ; in, Dimensionless node topological importance, reflecting the criticality of node connectivity; and This represents a dimensionless weighting coefficient. The node topological importance is set based on the betweenness centrality algorithm in graph theory, while the weighting coefficients are set based on historical decision preferences in the scheduling expert experience base. After sorting nodes by score from high to low, constraint triggering links are extracted for nodes with risk of exceeding limits. A constraint triggering link refers to the complete topological connection path between the source load node causing the limit exceedance and the physical device exceeding the limit. The priority score, limit-exceeding node information, and constraint triggering links are structured and encapsulated to generate an optimization requirement vector to guide subsequent resource scheduling.

[0043] For example, suppose a 10 kV line has the following power supply quality constraints: voltage must not be lower than 9.4 kV; equipment operation constraints: current must not exceed 400 amperes; and grid safety constraints: the power flow through the line must be stable. After standardization of dimensions, the voltage benchmark is set to 10 kV, the current benchmark to 500 amperes, and the grid safety benchmark to 1. The weighting coefficients are set based on risk assessment. It is 0.5. It is 0.3. =0.2. If, under the current simulated condition, the voltage at the end of the line is 9.3 kV, the current is 410 amps, and the grid structure is in good safety condition, corresponding to an index of 0.9, then the value of the constraint parameter set for this line is... Next, constraint deviation quantification calculation is performed based on the load spatiotemporal evolution characteristics. If the load node i connected to the end of the line has a load spatiotemporal evolution characteristic of 0.75 and a power distribution network state profile characteristic of 0.8, and the mapping coefficient is set to 1.2, with a power reference of 1 MVA, then the constraint deviation of this node is... This means that under the load evolution trend, this node has a dimensionless risk of exceeding the limit of 0.868. Combining the deviations of all nodes and branches in the network yields the constraint deviation vector. Finally, this deviation is ranked by importance. Assuming node i has a topological importance of 0.85, and the weight coefficients are... It is 0.7. Its optimization requirement priority score is 0.3. Based on this high score, the node was prioritized for processing, and the constraint triggering link causing its low voltage was extracted. Finally, the priority score of 0.8626, the information of the node exceeding the limit (Nodei), and this triggering link were encapsulated together to form a specific tuning requirement vector, guiding subsequent tuning actions.

[0044] Optionally, generating the action influence matrix includes: Obtain the status of adjustable resources and categorize them according to adjustment type to generate an adjustable resource catalog; Map the optimization requirement vector to the adjustment resource catalog to generate a set of candidate optimization actions; The candidate optimization action set is injected one by one into the power supply and distribution network state profile and power flow simulation is performed to generate an action result set. The impact of the action result set on voltage deviation, line loss change and operation cost is extracted and uniformly measured to generate an action impact matrix.

[0045] Specifically, the process begins with a resource inventory. By querying the power supply and distribution network status profile and equipment ledger database, the real-time status of all adjustable resources across the network is obtained and categorized by adjustment type, generating an adjustment resource catalog. This catalog details the equipment available for optimization and its current status. For example, it lists the current tap position and adjustable range of adjustable on-load tap-changing transformers, the current switching status and total capacity of switchable capacitor banks, and the adjustable power range and response rate of distributed power sources such as photovoltaics and energy storage. Next, based on the optimization demand vector generated in the previous step, a set of candidate optimization actions is generated from this adjustment resource catalog. This process is a mapping and matching: the constraint triggering links specified in the optimization demand vector are compared with the location information of the equipment in the adjustment resource catalog. If the demand vector indicates a low voltage problem at the end of a feeder, all equipment with voltage regulation capabilities, such as OLTCs and capacitors, within that feeder and its upstream substations are automatically retrieved, and all possible operations, such as raising the tap position or switching a group of capacitors, are included as candidate optimization actions in the set. This set encompasses all theoretically possible operational methods to solve the identified problem. The core step is to perform simulation evaluation on each action in the candidate optimization action set to generate an action result set. The current power supply and distribution network status profile is used as the baseline simulation model. It sequentially extracts one candidate optimization action from the set, temporarily modifies the state parameters of the corresponding equipment in the model (e.g., incrementing the tap position of an OLTC by 1), and then performs a complete power flow calculation. The results of the power flow calculation reveal how the voltage, current, and power distribution of the entire network will change after the action is executed. This process is repeated for each candidate action, and the network-wide status results from each simulation are stored to form an action result set. For each simulation result in the action result set, it is compared with the baseline state before the action is executed, extracting the impact quantities in three key dimensions: the improvement effect on voltage deviation, i.e., how much the voltage compliance rate of the entire network or key nodes has improved; the impact on line loss changes, i.e., whether the total active power loss of the entire network increases or decreases; and the operational cost, which is usually a comprehensive indicator and may include equipment operation limits, economic costs of regulation, etc. The range normalization method is used to uniformly measure the influence of these three different dimensions, ensuring they all fall within a comparable range, such as 0 to 1. Finally, all candidate optimization actions and their corresponding three normalized influence quantities are organized into a two-dimensional action influence matrix. The normalized value of the u-th action on the v-th influence dimension is... ,have: ; in, The target value for the v-th dimension after performing the u-th action is the overall network voltage qualification rate, total active power loss, etc. This represents the index value of the v-th dimension under the baseline condition; This represents the maximum absolute value of the change in this dimension among all candidate actions. For example... Figure 4 As shown, the rows of the matrix represent various candidate tuning actions, such as adjusting transformer taps and switching capacitor banks, while the columns represent multiple key dimensions for evaluating these actions, such as the improvement effect on voltage deviation, the impact on network losses, and operating costs. The shade of the color blocks intuitively quantifies the magnitude of the impact of each action on each dimension, with darker colors indicating a greater impact.

[0046] For example, a query revealed that there is an on-load tap changer (OLTC) at the substation, currently tapped at +1, with an adjustable range of -5 to +5. A 600 kV capacitor bank is connected upstream of feeder Line 4 and is currently disconnected. Based on the low-voltage issue and constraint triggering link indicated by the aforementioned optimization demand vector, two candidate optimization actions were generated: action u=1 is to raise the OLTC tap to +2; action u=2 is to switch the capacitors. Next, power flow simulations were performed on these two actions. Under baseline conditions, the voltage at node i is 9.3 kV, and the total active power loss of the entire network is 250 kW. Simulation results: After injecting action 1, the voltage at node i rises to 9.5 kV, the network loss decreases to 245 kW, and the operational cost is counted as 5 units; after injecting action 2, the voltage at node i rises to 9.6 kV, the network loss decreases to 242 kW, and the operational cost is counted as 2 units. These constitute the action result set. Finally, the results are uniformly quantified to generate an action impact matrix. The impact dimensions are: v=1 for voltage improvement, v=2 for network loss reduction, and v=3 for operational costs. The baseline values ​​are 9.3 kV, 250 kW, and 0, respectively. The absolute values ​​|Δu,v| of the impact of each action on each dimension are as follows: Action 1: 1000 volts kilowatt, Unit, Action 2: 1000 volts kilowatt, The units are 0.3 kV, 8 kW, and 5 units, respectively, for all actions. Calculate the normalized value of action 1 in the voltage influence dimension. ; Calculate the normalized value of action 1 in the dimension of network loss impact. ; Calculate the normalized value of action 2 in the voltage influence dimension. By doing so, after completing all the calculations, we obtain the action influence matrix, which clearly quantifies the benefits and costs of each candidate action on different objectives.

[0047] Optionally, the generation of an executable tuning strategy includes: Based on the action influence matrix, mutually reinforcing and mutually canceling action pairs are identified, and an action coupling relationship graph is generated. Based on the action coupling relationship diagram, conflict determination and constraint feasibility screening are performed on the candidate optimization action set to generate a subset of feasible actions; The subset of actionable actions is time-series orchestrated and execution window constraints are introduced to generate an action sequence; The action sequence is checked for executability and formatted for distribution, generating an executable optimization strategy.

[0048] Specifically, the algorithm iterates through all action pairs in the action influence matrix, determining their coupling type by comparing their respective influence vectors. For example, if two actions significantly boost the voltage of the same node—for instance, both increasing the voltage amplitude of the same node by more than a preset threshold, such as 0.5% or 1% of the rated voltage—they are marked as mutually reinforcing. If one action boosts the voltage while the other decreases it, they are marked as mutually canceling. This constructs a network graph where nodes represent candidate optimization actions, and edges represent their coupling relationships, such as reinforcement or cancellation—a graph of action coupling relationships. This graph reveals the potential superposition or weakening effects of executing multiple actions simultaneously. Action pairs with strong canceling relationships are identified as conflicting actions, and one is discarded based on preset rules, such as selecting the action with greater impact or lower cost. Simultaneously, the feasibility of action combinations is verified by simulating the simultaneous execution of a set of mutually reinforcing actions and checking whether their combined effect leads to new constraint violations, such as multiple voltage boosting actions causing excessively high voltage at a node. Only actions that, individually or in combination, do not violate any grid operation constraints are retained, forming a subset of actionable actions. This subset is then time-sequentially orchestrated, and execution window constraints are introduced to generate an action sequence. The execution order of certain actions affects the final outcome; for example, the main transformer tap changer should be adjusted first to stabilize the area voltage before refined reactive power compensation is performed on downstream lines. The optimal execution order of actionable actions is determined based on an expert rule base, a set of rules containing historical decision preferences of dispatching experts. This rule base determines the optimal execution order of actionable actions, such as adjusting the main transformer tap changer first to stabilize the area voltage before refined reactive power compensation on downstream lines. Simultaneously, considering factors such as communication delays and equipment response times in actual operation, an execution window is set for each action—an allowed start time range, such as within the next 5 to 10 minutes. This ensures the timeliness and operability of the strategy, ultimately forming an action sequence with a clear execution order and time window. Finally, the generated action sequence undergoes executability verification and is converted to a standard format to generate an executable optimization strategy. The verification process includes checking whether the equipment is currently under control and whether the communication link is normal. After confirmation, the action sequence is translated into a control message format conforming to the communication protocol, such as the Distributed Network Protocol version 3 (DNP3). This formatted instruction set constitutes the final executable tuning strategy. It contains precise device identification, target operations such as setting the tap changer to a specified position, execution timestamps, and other information, and can be directly sent to a remote control terminal for execution. The generation of this strategy represents a complete closed loop from problem identification to solution formulation.

[0049] For example, firstly, the action impact matrix is ​​analyzed, revealing that both action 1 (adjusting the OLTC) and action 2 (adding capacitors) improve the voltage of the target node; that is, their impact vectors are positive in the voltage dimension. When both actions occur simultaneously, the voltage improvement effects are superimposed, so they are marked as mutually reinforcing in the action coupling relationship diagram. Next, a power flow calculation is performed on the scenario where actions 1 and 2 are executed simultaneously. It is found that although the voltage of the target node returns to normal, the voltage of another node N25 in the network rises to 107.5% of its rated value, exceeding the 107% upper limit, resulting in a new over-limit. Therefore, the combination of actions 1 and 2 is determined to be infeasible. Action 2 is retained, and action 1 is discarded, generating a subset of feasible actions containing only action 2. Then, this subset is time-series orchestrated. Since there is only one action in the subset, the order is not an issue. An execution window constraint is set for it, i.e., execution within the next 5 to 10 minutes, generating an action sequence. Finally, executability is checked. The remote signaling status of the capacitor is queried, confirming that it is "controllable" and the communication link is normal. Therefore, the action sequence is converted into a control message in the DNP3 protocol format, forming the final executable tuning strategy. The strategy content is clearly defined as: {Target Device ID: Capacitor_ID, Opcode: Close, Additional Parameters: null, Execution Timestamp: XX-06-01T10:22:30Z}, and this strategy can be directly issued to the remote control device for execution.

[0050] Optionally, the update parameters for generating the next tuning cycle include: Send the executable tuning strategy to the execution terminal and record the delivery receipt to generate a strategy execution record; The measurement data and device status parameters after execution are obtained and associated with the strategy execution record to generate optimization closed-loop feedback data; The strategy deviation is calculated from the optimization closed-loop feedback data, and the source of the deviation is extracted to generate strategy deviation features; The spatial migration feature in the load spatiotemporal evolution feature is corrected using the strategy deviation feature, and the action influence quantity in the action influence matrix is ​​updated to generate the update parameters for the next tuning cycle.

[0051] Specifically, the generated executable tuning strategy is first sent to the corresponding execution terminal via a standard communication protocol. Simultaneously, the sending time, target device, and execution receipt status of each instruction are recorded, generating a strategy execution record. In the next measurement cycle after strategy execution, measurement data and device status parameters are re-collected from each terminal and matched with the action time window associated in the strategy execution record. The actual load sequence and network status parameters under the same spatiotemporal unit after execution are extracted, forming closed-loop feedback data for tuning. Based on the feedback data and the power supply and distribution network status profile before execution, the deviation between the actual and expected effects after strategy execution is calculated. By tracing the source of the deviation, it is identified whether it is caused by inaccurate spatial migration estimation in the load spatiotemporal evolution characteristics or by estimation errors in the action influence matrix, thus obtaining the strategy deviation characteristic. This strategy deviation characteristic is used to correct the spatial migration characteristics in the load spatiotemporal evolution characteristics, while simultaneously updating the influence of the corresponding actions in the action influence matrix. The core parameter in the spatial migration characteristic is the cross-correlation coefficient with time delay. Therefore, for a calculation time delay of τ, the corrected cross-correlation coefficient between sequence x and sequence y is... ,have: ; in, This is the cross-correlation coefficient recalculated based on the actual load fluctuation data after execution feedback; The learning rate has a value between 0 and 1. This represents the cross-correlation coefficients before correction. Simultaneously, for the action influence matrix, this relates to calculating the influence of the updated u-th candidate optimization action on the v-th influence dimension. ,have: ; in, The effect amount before correction; To update the gain coefficient, the value range is 0-1; This refers to the actual feedback of the same dimension of influence obtained after performing the action. The corrected spatial migration features, i.e., the updated cross-correlation matrix, and the updated action influence matrix are used as the initial parameters for the next tuning cycle, thereby achieving closed-loop adaptive optimization of the system.

[0052] For example, in the next measurement cycle at 10:30 AM, data was re-collected, revealing that the actual voltage at node i was 9.55 kV and the total active power loss across the network was 243 kW. This new data, correlated with the strategy execution record, constituted the optimization closed-loop feedback data. By comparing the actual results with the model's expected results—an expected voltage of 9.6 kV and an expected network loss of 242 kW—the strategy deviation was calculated as voltage -0.05 kV and network loss +1 kW. Source analysis revealed that the deviation primarily stemmed from the actual load growth in residential area B being slightly lower than the spatial migration model's prediction. This indicates that the cross-correlation coefficient in the model needs correction. If the original cross-correlation coefficient was 0.82, the actual correlation coefficient recalculated based on the feedback data was 0.80. A learning rate was then set. If it is 0.2, then Meanwhile, the actual impact of action 2 (capacitor placement) differs from the simulation, necessitating an update to the action impact matrix. The original normalized impact of action 2 on voltage improvement was 1.0. The actual impact was calculated based on the feedback data. Set and update the gain coefficient. The impact value is 0.3, and the update impact is: The corrected cross-correlation coefficient of 0.816 is updated in the spatial migration feature model, and the updated action influence value of 0.9499 is written back into the action influence matrix. These learned and corrected parameters will become the more accurate update parameters for the next tuning cycle.

[0053] Based on the same inventive concept, such as Figure 5 As shown, the present invention also provides a power supply and distribution dynamic optimization system based on load spatiotemporal evolution, the system comprising: The data generation module is used to acquire power supply and distribution measurement data and power distribution network structure data, perform alignment processing, and generate a basic network dataset. The status profiling module is used to parse the network basic dataset to obtain node status variables and branch status variables, and perform status fusion to generate a power supply and distribution network status profile. The load matrix module is used to extract the load sequence from the power supply and distribution network status profile and divide it into spatial units to generate a spatial unit load matrix. The spatiotemporal evolution module is used to perform temporal decomposition and spatial correlation calculation on the spatial unit load matrix to generate load spatiotemporal evolution characteristics; The demand calculation module is used to input the load spatiotemporal evolution characteristics into the constraint evaluation and combine them with the power supply and distribution network status profile to perform power flow consistency calculation and generate an optimized demand vector. The action evaluation module is used to generate a set of candidate optimization actions in the adjustment resources based on the optimization demand vector and to evaluate the impact of the actions, thereby generating an action impact matrix. The strategy orchestration module is used to resolve action conflicts and form an action sequence orchestration based on the action influence matrix, thereby generating an executable optimization strategy. The closed-loop update module is used to issue the executable optimization strategy and collect execution feedback, generate optimization closed-loop feedback data, update the load spatiotemporal evolution characteristics and the action influence matrix using the optimization closed-loop feedback data, and generate update parameters for the next optimization cycle.

[0054] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0055] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A dynamic optimization method for power supply and distribution based on load spatiotemporal evolution, characterized in that, The method includes: Acquire power supply and distribution measurement data and power distribution network structure data, perform alignment processing, and generate a network basic dataset; Based on the network basic dataset, the node state variables and branch state variables are parsed, and the state is fused to generate a power supply and distribution network state profile. Extract the load sequence from the power supply and distribution network status profile and divide it into spatial units to generate a spatial unit load matrix; The spatial unit load matrix is ​​subjected to temporal decomposition and spatial correlation calculation to generate load spatiotemporal evolution characteristics; The load spatiotemporal evolution characteristics are used as input constraints for evaluation, and combined with the power supply and distribution network state profile, power flow consistency calculation is performed to generate an optimized demand vector. Based on the optimization demand vector, a set of candidate optimization actions is generated from the adjustment resources, and the impact of the actions is evaluated to generate an action impact matrix; Based on the action influence matrix, action conflict resolution is performed and an action sequence arrangement is formed to generate an executable optimization strategy. The executable optimization strategy is issued and execution feedback is collected to generate optimization closed-loop feedback data. The optimization closed-loop feedback data is used to update the load spatiotemporal evolution characteristics and the action influence matrix to generate update parameters for the next optimization cycle.

2. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The underlying dataset for the generated network includes: Acquire voltage measurement parameters, current measurement parameters, and active and reactive power measurement parameters from the terminal measurement device, and generate a set of measurement parameters; Obtain switch position parameters, transformer tap changer status parameters, and reactive power compensation device status parameters from the operation management system, and generate a set of equipment status parameters; The topological connections and line impedance parameters of the power distribution network are obtained and timestamped with the measurement parameter set and the equipment status parameter set to generate a network basic dataset.

3. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 2, characterized in that, The generated power supply and distribution network status profile includes: Identify and validate the confidence level of outliers in the network's basic dataset to generate a set of reliable measurements. By fusing the trusted measurement set with the topological connectivity and performing state estimation, node state variables are generated. The branch power flow and branch current are calculated by combining the node state variables with the line impedance parameters, and branch state variables are generated. The node status variables and the branch status variables are merged to form a searchable index, generating a power supply and distribution network status profile.

4. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The generated spatial element load matrix includes: Obtain the node load parameters and transformer area affiliation information from the power supply and distribution network status profile and perform spatial mapping to generate a spatially mapped load sequence; The spatially mapped load sequences are aggregated by time slices to form time-aligned sequences, thereby generating time-sliced ​​load sequences; The time-slice load sequences are assembled into spatial units, while preserving the spatial unit identifiers, to generate a spatial unit load matrix.

5. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The spatiotemporal evolution characteristics of the generated load include: The spatial unit load matrix is ​​subjected to trend term separation and fluctuation term extraction to generate load component features; Spatial correlation calculation and migration direction identification are performed on the load component characteristics to generate spatial migration characteristics; The load component features and the spatial migration features are jointly encoded to generate load spatiotemporal evolution features.

6. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The generated optimization requirement vector includes: Obtain power quality constraint parameters, equipment operation constraint parameters, and grid safety constraint parameters, and perform unified dimension processing to generate a constraint parameter set; Based on the aforementioned spatiotemporal evolution characteristics of the load, constraint deviation quantization calculation is performed to generate a constraint deviation vector; The constraint deviation vector is sorted by importance and the constraint triggering links are extracted to generate the tuning requirement vector.

7. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The generated action influence matrix includes: Obtain the status of adjustable resources and categorize them according to adjustment type to generate an adjustable resource catalog; Map the optimization requirement vector to the adjustment resource catalog to generate a set of candidate optimization actions; The candidate optimization action set is injected one by one into the power supply and distribution network state profile and power flow simulation is performed to generate an action result set. The impact of the action result set on voltage deviation, line loss change and operation cost is extracted and uniformly measured to generate an action impact matrix.

8. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The generated executable tuning strategy includes: Based on the action influence matrix, mutually reinforcing and mutually canceling action pairs are identified, and an action coupling relationship graph is generated. Based on the action coupling relationship diagram, conflict determination and constraint feasibility screening are performed on the candidate optimization action set to generate a subset of feasible actions; The subset of actionable actions is time-series orchestrated and execution window constraints are introduced to generate an action sequence; The action sequence is checked for executability and formatted for distribution, generating an executable optimization strategy.

9. The power supply and distribution dynamic optimization method based on load spatiotemporal evolution according to claim 1, characterized in that, The update parameters for generating the next tuning cycle include: Send the executable tuning strategy to the execution terminal and record the delivery receipt to generate a strategy execution record; The measurement data and device status parameters after execution are obtained and associated with the strategy execution record to generate optimization closed-loop feedback data; The strategy deviation is calculated from the optimization closed-loop feedback data, and the source of the deviation is extracted to generate strategy deviation features; The spatial migration feature in the load spatiotemporal evolution feature is corrected using the strategy deviation feature, and the action influence quantity in the action influence matrix is ​​updated to generate the update parameters for the next tuning cycle.

10. A power supply and distribution dynamic optimization system based on load spatiotemporal evolution, applied to the power supply and distribution dynamic optimization method based on load spatiotemporal evolution as described in any one of claims 1-9, characterized in that, The system includes: The data generation module is used to acquire power supply and distribution measurement data and power distribution network structure data, perform alignment processing, and generate a basic network dataset. The status profiling module is used to parse the network basic dataset to obtain node status variables and branch status variables, and perform status fusion to generate a power supply and distribution network status profile. The load matrix module is used to extract the load sequence from the power supply and distribution network status profile and divide it into spatial units to generate a spatial unit load matrix. The spatiotemporal evolution module is used to perform temporal decomposition and spatial correlation calculation on the spatial unit load matrix to generate load spatiotemporal evolution characteristics; The demand calculation module is used to input the load spatiotemporal evolution characteristics into the constraint evaluation and combine them with the power supply and distribution network status profile to perform power flow consistency calculation and generate an optimized demand vector. The action evaluation module is used to generate a set of candidate optimization actions in the adjustment resources based on the optimization demand vector and to evaluate the impact of the actions, thereby generating an action impact matrix. The strategy orchestration module is used to resolve action conflicts and form an action sequence orchestration based on the action influence matrix, thereby generating an executable optimization strategy. The closed-loop update module is used to issue the executable optimization strategy and collect execution feedback, generate optimization closed-loop feedback data, update the load spatiotemporal evolution characteristics and the action influence matrix using the optimization closed-loop feedback data, and generate update parameters for the next optimization cycle.

Images