Regional difference-oriented transformer substation bus-level adjustable load potential evaluation and hierarchical regulation and control method and system

Abstract

The invention relates to a transformer substation bus-level adjustable load potential evaluation and hierarchical regulation and control method and system for regional differences, and belongs to the technical field of intelligent power grid regulation and control. The method comprises the steps of obtaining transformer substation bus load data, and performing missing value processing on the transformer substation bus load data through a data missing processing model to obtain bus load reference data; performing curve clustering on the bus load reference data to obtain a load characteristic curve; performing potential evaluation based on the load characteristic curve and outputting a bus load potential evaluation result; and constructing a regional aggregation load characteristic model, and outputting an optimal load regulation and control strategy through hierarchical regulation and control according to the regional aggregation load characteristic model and a bus load potential evaluation result. And transformer substation bus-level adjustable load potential evaluation and hierarchical regulation and control oriented to regional differences are realized.

Description

Technical Field

[0001] This invention belongs to the field of smart grid control technology, specifically relating to a method and system for assessing and hierarchically controlling the adjustable load potential of substation busbars in response to regional differences. Background Technology

[0002] With the increasing demand for multi-source, load, and storage-based coordinated regulation in new power systems, traditional centralized load control strategies are gradually revealing limitations in response speed, regional adaptability, and resource identification accuracy. Especially in power distribution scenarios with significant regional differences, the load structure, pricing mechanisms, and electrical characteristics vary considerably across different geographical regions, making it difficult to accurately assess load response capabilities and regulation potential using a unified model. Furthermore, current regulation granularity is mostly concentrated at the regional or terminal level, failing to fully exploit intermediate adjustable resources at the substation bus level, resulting in long response chains, low efficiency, and high coupling in the regulation process. On the other hand, power systems are gradually evolving towards operating characteristics with high proportions of renewable energy integration and coexisting highly volatile loads, requiring load-side regulation to possess stronger hierarchical autonomy and precise local control capabilities. Therefore, there is an urgent need to establish a method for assessing and hierarchically regulating adjustable load potential, centered on substation buses and oriented towards regional differences. This method should fully consider regional load heterogeneity while enabling the decentralization of regulation granularity and optimization of response paths, thereby improving the overall flexibility, stability, and intelligence of the power grid system's regulation. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, this invention provides a method and system for assessing and hierarchically controlling the adjustable load potential at the substation bus level, taking into account regional differences. The objective of this invention can be achieved through the following technical solutions: A method for assessing and hierarchically controlling the adjustable load potential of substation busbars, oriented towards regional differences, includes: S1: Obtain substation bus load data, and use a data missing value processing model to process the missing values ​​of the substation bus load data to obtain bus load baseline data. S2: Perform curve clustering on the bus load baseline data to obtain load characteristic curves; S3: Based on the load characteristic curve, perform a potential assessment and output the bus load potential assessment result; S4: Construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

[0004] Preferably, the data missing information handling model in step S1 processes the following steps: S101: Obtain the substation bus load sequence by adding timestamps to the substation bus load data; S102: The substation bus load sequence is decomposed using a supplementary empirical mode decomposition algorithm to obtain the intrinsic mode function components and residual components; S103: Convert the residual components into shared feature components through a global residual gateway; S104: Construct a prediction model based on the intrinsic mode function components and the shared feature components; S105: Preset model training parameters, and train the prediction model based on the model training parameters to obtain the component prediction model; S106: The intrinsic mode function components and the shared feature components are predicted and output by the component prediction model to produce the predicted values ​​of the intrinsic mode function components and the residual components; S107: The bus load reference data is obtained by reconstructing the predicted values ​​of the intrinsic mode function components and the predicted values ​​of the residual components.

[0005] Preferably, the reconstruction method in step S107 is as follows: S107-1: Generate the weight matrix through a spatiotemporal weight generation network; S107-2: The bus load baseline data is obtained by performing a weighted fusion operation on the weight matrix, the predicted values ​​of the intrinsic mode function components, and the predicted values ​​of the residual components.

[0006] Preferably, the clustering process of curve clustering in step S2 is as follows: S201: Standardize the bus load reference data to obtain bus load standard data; S202: Determine the optimal number of clusters by calculating the profile coefficient of the bus load standard data; S203: Initialize cluster centers based on the optimal number of clusters, and calculate the distance metric value based on the cluster centers and the bus load standard data through time series morphological similarity measurement; S204: Construct the load characteristic curve based on the distance metric.

[0007] Preferably, the assessment strategy for potential evaluation in step S3 is as follows: S301: Extract the daily load curve of the substation busbar in the time dimension based on the load characteristic curve; S302: Extract the annual daily load curve in the spatial dimension based on the load characteristic curve; S303: Preset evaluation indicators, and analyze the daily load curve and the annual daily load curve of the substation bus to obtain the bus load potential assessment result based on the evaluation indicators.

[0008] Preferably, the evaluation indicators in step S303 include daily load factor, daily minimum load factor, daily peak-to-valley difference, and daily peak-to-valley difference rate.

[0009] Preferably, the modeling process of the regional aggregated load characteristic model in step S4 is as follows: Obtain the substation busbar belonging area unit, which includes the load characteristic curve and the busbar load potential assessment result; construct a load potential feature matrix based on the busbar load potential assessment result; extract the regional aggregation features of the load potential feature matrix through the mean and covariance; and construct the regional aggregation load characteristic model based on the regional aggregation features.

[0010] Preferably, the hierarchical control process in step S4 is as follows: S401: Divide the regional intensity level based on the aforementioned regional aggregated load characteristic model; S402: The hierarchical regulation is modeled as a Markov decision process, based on the regional intensity level, the regional aggregated load characteristic model, and the bus load potential assessment results, and presets the state space, action space, and reward function. S403: Calculate the action Q value using a deep Q-network based on the state space, action space, and reward function; S404: Based on the action Q-value, select the action with the highest Q-value as the optimal load control strategy; S405: The optimal load control strategy is tested for safety and feasibility. If the test fails, the strategy is adjusted using the strategy pruning function.

[0011] Preferably, the state space in step S402 includes: the regional aggregated load characteristic model, the bus load potential assessment result, and auxiliary state variables; the action space includes regulating the cluster center and regulating the regional intensity level; and the reward function is constructed based on load reduction, user dissatisfaction, response delay, and regulation cost.

[0012] A substation busbar-level adjustable load potential assessment and hierarchical control system oriented towards regional differences includes a data preprocessing module, a curve construction module, a load assessment module, and a strategy generation module, comprising: The data preprocessing module is used to acquire substation bus load data and process the missing values ​​of the substation bus load data through a data missing value processing model to obtain bus load baseline data. The curve construction module is used to perform curve clustering on the bus load reference data to obtain load characteristic curves; The load assessment module is used to perform potential assessment based on the load characteristic curve and output the bus load potential assessment result; The strategy generation module is used to construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

[0013] The beneficial effects of this invention are as follows: (1) By introducing multi-scale decomposition, modal prediction and spatiotemporal fusion mechanism, the problems of missing, fluctuating and interference in substation bus load data are effectively solved. By constructing a standardized clustering method with load characteristic curve as the core, the accuracy of identifying different bus load patterns is improved, and the regional adaptability and evaluation accuracy under complex load scenarios are significantly enhanced.

[0014] (2) By constructing a bus-level load potential characteristic model, the response capability of locally adjustable load resources is identified. The structural analysis and hierarchical organization of adjustable resources are realized without relying on strong user-side access, which improves the response accuracy and link efficiency of the control strategy and provides efficient support for multi-coupling regulation of source-load-storage.

[0015] (3) By constructing a regional load characteristic modeling method based on bus aggregation index, and combining multi-index aggregation and covariance analysis, the differences in electricity pricing mechanisms, load structures, and response habits among different geographical regions are fully characterized. By using reinforcement learning strategies to dynamically stratify and optimize regional control levels, local autonomous response and multi-regional collaborative control driven by differences are realized, significantly improving the system's intelligent scheduling capabilities and robustness.

[0016] (4) By modeling the regional regulation strategy as a Markov decision process and combining it with deep reinforcement learning algorithms, the regulation behavior can be continuously absorbed from environmental feedback and corrected in real time. The generated optimal regulation strategy has the ability to adaptively adjust, continuously learn and dynamically prune, and can cope with high uncertainty operating conditions such as new energy fluctuations and load changes, thereby improving the flexibility and stability of the entire power grid system. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart illustrating a method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, according to the present invention. Detailed Implementation

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

[0020] Please see Figure 1 A method for assessing and hierarchically controlling the adjustable load potential of substation busbars, oriented towards regional differences, includes: S1: Obtain substation bus load data, and use a data missing value processing model to process the missing values ​​of the substation bus load data to obtain bus load baseline data. S2: Perform curve clustering on the bus load baseline data to obtain load characteristic curves; S3: Based on the load characteristic curve, perform a potential assessment and output the bus load potential assessment result; S4: Construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

[0021] Specifically, the data missing information handling model in step S1 processes the following steps: S101: Obtain the substation bus load sequence by adding timestamps to the substation bus load data; S102: The substation bus load sequence is decomposed using a supplementary empirical mode decomposition algorithm to obtain the intrinsic mode function components and residual components; S103: Convert the residual components into shared feature components through a global residual gateway; S104: Construct a prediction model based on the intrinsic mode function components and the shared feature components; S105: Preset model training parameters, and train the prediction model based on the model training parameters to obtain the component prediction model; S106: The intrinsic mode function components and the shared feature components are predicted and output by the component prediction model to produce the predicted values ​​of the intrinsic mode function components and the residual components; S107: The bus load reference data is obtained by reconstructing the predicted values ​​of the intrinsic mode function components and the predicted values ​​of the residual components.

[0022] Specifically, the model training parameters in step S105 include the number of hidden layers, the learning rate, and the training coefficients.

[0023] Specifically, the reconstruction method in step S107 is as follows: S107-1: Generate the weight matrix through a spatiotemporal weight generation network; S107-2: The bus load baseline data is obtained by performing a weighted fusion operation on the weight matrix, the predicted values ​​of the intrinsic mode function components, and the predicted values ​​of the residual components.

[0024] Specifically, the spatiotemporal weight generation network includes a temporal feature extraction channel and a spatial feature extraction channel.

[0025] Specifically, the clustering process of curve clustering in step S2 is as follows: S201: Standardize the bus load reference data to obtain bus load standard data; S202: Determine the optimal number of clusters by calculating the profile coefficient of the bus load standard data; S203: Initialize cluster centers based on the optimal number of clusters, and calculate the distance metric value based on the cluster centers and the bus load standard data through time series morphological similarity measurement; S204: Construct the load characteristic curve based on the distance metric.

[0026] Specifically, the assessment strategy for potential evaluation in step S3 is as follows: S301: Extract the daily load curve of the substation busbar in the time dimension based on the load characteristic curve; S302: Extract the annual daily load curve in the spatial dimension based on the load characteristic curve; S303: Preset evaluation indicators, and analyze the daily load curve and the annual daily load curve of the substation bus to obtain the bus load potential assessment result based on the evaluation indicators.

[0027] Specifically, the evaluation indicators in step S303 include daily load factor, daily minimum load factor, daily peak-to-valley difference, and daily peak-to-valley difference rate.

[0028] Specifically, the modeling process of the regional aggregated load characteristic model in step S4 is as follows: Obtain the substation busbar belonging area unit, which includes the load characteristic curve and the busbar load potential assessment result; construct a load potential feature matrix based on the busbar load potential assessment result; extract the regional aggregation features of the load potential feature matrix through the mean and covariance; construct the regional aggregation load characteristic model based on the regional aggregation features; The mathematical expression for the regional aggregated load characteristic model is: , in, R k (t) This is the regional aggregated load characteristic model. f For the response capability estimation function, The load characteristic curve, This is the result of the bus load potential assessment.

[0029] Specifically, the hierarchical control process in step S4 is as follows: S401: Divide the regional intensity level based on the aforementioned regional aggregated load characteristic model; S402: The hierarchical regulation is modeled as a Markov decision process, based on the regional intensity level, the regional aggregated load characteristic model, and the bus load potential assessment results, and presets the state space, action space, and reward function. S403: Calculate the action Q value using a deep Q-network based on the state space, action space, and reward function; S404: Based on the action Q-value, select the action with the highest Q-value as the optimal load control strategy; S405: The optimal load control strategy is tested for safety and feasibility. If the test fails, the strategy is adjusted using the strategy pruning function.

[0030] Specifically, the state space in step S402 includes the regional aggregated load characteristic model, the bus load potential assessment result, and auxiliary state variables; the action space includes regulating the cluster center and regulating the regional intensity level; and the reward function is constructed based on load reduction, user dissatisfaction, response delay, and regulation cost.

[0031] In this embodiment, the safety and feasibility verification includes exceeding the maximum response capacity of the bus, violating user comfort constraints, and meeting the control time window.

[0032] In this embodiment, a substation bus-level adjustable load potential assessment and hierarchical control system oriented towards regional differences includes a data preprocessing module, a curve construction module, a load assessment module, and a strategy generation module, comprising: The data preprocessing module is used to acquire substation bus load data and process the missing values ​​of the substation bus load data through a data missing value processing model to obtain bus load baseline data. The curve construction module is used to perform curve clustering on the bus load reference data to obtain load characteristic curves; The load assessment module is used to perform potential assessment based on the load characteristic curve and output the bus load potential assessment result; The strategy generation module is used to construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

[0033] In this embodiment, the scenario is set as follows: Substation busbar: Substation A, responsible for supplying a mixed commercial and residential area; Data source: Bus load data from substation A for 7 consecutive days (168 hours), with a sampling interval of 1 hour. The load data simulates real-world electricity consumption patterns, including daytime peaks, nighttime off-peaks, and random fluctuations. Some data points are missing; this is used to test the data missing handling model.

[0034] Example of raw load data (unit: MW): Day1 00:00: 10.0, Day1 01:00: 12.0, Day1 02:00: 11.5, Day1 03:00: NaN (missing), ..., Day7 23:00: 9.0.

[0035] Missing value distribution: There are 5 missing values ​​in total, including Day 1 03:00, Day 2 10:00, and Day 4 15:00.

[0036] Objective: To process missing values ​​and obtain complete bus load baseline data.

[0037] S101: Add timestamp The load data is converted into a time series in the format "YYYY-MM-DD HH:MM" to form the substation bus load sequence.

[0038] S102: CEEMD decomposition The load sequence was decomposed using the Supplementary Empirical Mode Decomposition (CEEMD) algorithm, resulting in three intrinsic mode function (IMF) components and one residual component.

[0039] Example decomposition results (simplified): IMF1: [0.1, 0.2, ...] (High-frequency fluctuations) IMF2: [0.5, 0.6, ...] (Intermediate frequency variation) IMF3: [1.0, 1.1, ...] (Low-frequency trend) Residual: [8.0, 8.2, ...] (Long-term background) S103: Global Residual Gateway The residual components are transformed into shared feature components using a fully connected neural network to extract global features. The dimensions of the shared feature components are the same as those of the IMF components.

[0040] S104: Constructing a Predictive Model A prediction model is built using an LSTM network, with IMF components and shared feature components as inputs and predicted values ​​as outputs.

[0041] S105: Model Training Parameters Set the number of hidden layers to 2, learning rate to 0.001, and training coefficient to 0.8 (using 80% of the data for training and 20% for validation). Set the training epochs to 100 and the batch size to 32.

[0042] S106: Component Prediction After training, the model predicts the IMF components and residual component values ​​of the missing points. For example, predicting Day 1 03:00: IMF1 forecast: 0.15, IMF2 forecast: 0.55, IMF3 forecast: 1.05, residual forecast: 8.1.

[0043] S107: Reconstruct load data; S107-1: Spatiotemporal Weight Generation Network The network contains a temporal feature extraction channel (using one-dimensional convolution) and a spatial feature extraction channel (using graph convolution), and outputs a weight matrix W (the same size as the load sequence).

[0044] S107-2: Weighted Fusion The weight matrix W is then summed with the IMF predictions and residual predictions using a weighted average. Reconfiguration load = W * (IMF1_pred + IMF2_pred + IMF3_pred + Residual_pred) For example, the reconfigured load value at 03:00 on Day 1 is 10.2 MW.

[0045] Output: Complete bus load baseline data (168 points, no missing data).

[0046] S2: Curve clustering; Objective: To cluster load baseline data into typical load characteristic curves.

[0047] S201: Standardization Processing; The load baseline data is Z-score standardized so that the mean is 0 and the standard deviation is 1.

[0048] S202: Determine the optimal number of clusters; The silhouette coefficient is calculated for the number of clusters k from 2 to 5. The silhouette coefficient is the largest (0.65) when k=3, so the optimal number of clusters is 3.

[0049] S203: Cluster center initialization and distance metric; Three cluster centers were initialized using k-means++. The distance between each load curve and the cluster center was calculated based on the dynamic time warping (DTW) distance, and clusters were assigned accordingly.

[0050] Examples of distance metrics: Cluster1 average DTW=0.1, Cluster2 average DTW=0.2, Cluster3 average DTW=0.15.

[0051] S204: Construct load characteristic curves; The center curve of each cluster is the load characteristic curve: Cluster1 (smooth load): small load fluctuations, with an average daily load of 10-12 MW.

[0052] Cluster2 (Peak Type): High peak load, with an average daily load of 15-18 MW, and peak hours are from 10:00 to 18:00.

[0053] Cluster3 (Rough Season): Low load trough, with an average daily load of 8-10 MW, and the trough occurs at midnight.

[0054] S3: Potential Assessment; Objective: To assess the adjustable load potential of each load characteristic curve.

[0055] S301: Extraction of the time dimension; Extract typical daily load curves (24 hours) from the load characteristic curves. For example, the daily curve for Cluster2 shows a peak at 15:00 (18 MW).

[0056] S302: Spatial dimension extraction; The representative pattern of the annual daily load curve is extracted based on multi-day data (7 days), but this embodiment only uses 7 days to simulate spatial differences.

[0057] S303: Evaluation Index Analysis; Preset evaluation indicators: Daily load factor = Average load / Maximum load Daily minimum load factor = Minimum load / Maximum load Daily peak-to-valley difference = Maximum load - Minimum load Daily peak-to-valley difference rate = Daily peak-to-valley difference / Maximum load Calculate the indices for each cluster: Cluster1: Daily load factor = 0.85, Daily minimum load factor = 0.7, Daily peak-to-valley difference = 3 MW, Daily peak-to-valley difference rate = 0.25 → Moderate adjustable potential.

[0058] Cluster2: Daily load factor = 0.6, Daily minimum load factor = 0.4, Daily peak-to-valley difference = 8 MW, Daily peak-to-valley difference rate = 0.44 → High adjustability potential.

[0059] Cluster3: Daily load factor = 0.9, Daily minimum load factor = 0.8, Daily peak-to-valley difference = 2 MW, Daily peak-to-valley difference rate = 0.2 → Low adjustability potential.

[0060] Output: Bus load potential assessment results: Cluster2: High potential (can reduce peak load), Cluster1: Medium potential (can shift load), Cluster3: Low potential (basically unadjustable).

[0061] S4: Layered regulation: Objective: To generate the optimal load regulation strategy based on the potential assessment results.

[0062] S401: Regional Intensity Level Classification A regional aggregated load characteristic model Rk(t) is constructed. It is assumed that there are multiple similar buses in the area to which substation A belongs, but this embodiment only considers substation A itself.

[0063] Mathematical expression for the regional aggregated load characteristic model: , in, These are the load characteristic curves (Cluster1, 2, 3). As a result of the potential assessment, Features are extracted using the mean and covariance to classify regional intensity levels: High-intensity zone: Cluster2 dominant period (10:00-18:00); Moderate intensity zone: Cluster 1 dominant period (other daytime periods); Low-intensity zone: Cluster 3's dominant period (nighttime); S402: Markov decision process modeling; State space: includes Rk(t), θi, and auxiliary states (such as time and temperature).

[0064] Action space: This includes adjusting cluster centers (e.g., reducing Cluster2 load by 10%) or adjusting regional intensity levels (e.g., lowering the intensity level of high-intensity areas).

[0065] Reward function: Reward = w1 × load reduction - w2 × user dissatisfaction - w3 × response delay - w4 × control cost; Set the weights as w1=0.5, w2=0.3, w3=0.1, w4=0.1.

[0066] S403: Deep Q-Network Computation The policy network is trained using DQN, taking the state as input and outputting the Q-value of the action. For example, in the state "peak hours", the Q-value of the action "reduce the load of Cluster2 by 10%" is 0.8, and the Q-value of the action "reduce the load of Cluster1 by 5%" is 0.6.

[0067] S404: Select the optimal action Select the action with the highest Q value: Reduce Cluster2 load by 10% (approximately 1.8 MW) during the 15:00 peak period.

[0068] S405: Safety and Feasibility Verification: Verification conditions: Exceeding the maximum response capacity of the bus? Cluster2 can reduce it by up to 20%, with 10% within the acceptable range.

[0069] Violating user comfort constraints? Business users can accept a 10% reduction without comfort issues.

[0070] Does it meet the control time window? 15:00 is within the peak window.

[0071] The verification passed, and the optimal load control strategy was output: reduce the load of Cluster2 by 10% (1.8 MW) during the period from 15:00 to 16:00.

[0072] System module implementation: Data preprocessing module: Execute S1, input raw load data, and output bus load baseline data.

[0073] Curve construction module: Execute S2, input load baseline data, and output 3 load characteristic curves (Cluster1, 2, 3).

[0074] Load assessment module: Execute S3, input the load characteristic curve, and output the potential assessment results (high, medium, low).

[0075] Strategy generation module: Execute S4, input the potential assessment results and regional model, and output the optimal load control strategy.

[0076] If it fails, adjust using a strategy pruning function, for example, by reducing the reduction amount to 5%.

[0077] This embodiment demonstrates the complete process from handling missing data to hierarchical control. Through data reconstruction, clustering, potential assessment, and reinforcement learning-based control, refined management of substation bus loads is achieved. This method considers regional differences, ensuring the safety and feasibility of the control strategy. In practical applications, it can be extended to multiple buses and more complex scenarios.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for assessing and hierarchically controlling the adjustable load potential of substation busbars in response to regional differences, characterized in that, Includes the following steps: S1: Obtain substation bus load data, and use a data missing value processing model to process the missing values ​​of the substation bus load data to obtain bus load baseline data. S2: Perform curve clustering on the bus load baseline data to obtain load characteristic curves; S3: Based on the load characteristic curve, perform a potential assessment and output the bus load potential assessment result; S4: Construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

2. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 1, is characterized in that... The data missing handling model in step S1 processes the data missing information as follows: S101: Obtain the substation bus load sequence by adding timestamps to the substation bus load data; S102: The substation bus load sequence is decomposed using a supplementary empirical mode decomposition algorithm to obtain the intrinsic mode function components and residual components; S103: Convert the residual components into shared feature components through a global residual gateway; S104: Construct a prediction model based on the intrinsic mode function components and the shared feature components; S105: Preset model training parameters, and train the prediction model based on the model training parameters to obtain the component prediction model; S106: The intrinsic mode function components and the shared feature components are predicted and output by the component prediction model to produce the predicted values ​​of the intrinsic mode function components and the residual components; S107: The bus load reference data is obtained by reconstructing the predicted values ​​of the intrinsic mode function components and the predicted values ​​of the residual components.

3. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 2, is characterized in that... The reconstruction method in step S107 is as follows: S107-1: Generate the weight matrix through a spatiotemporal weight generation network; S107-2: The bus load baseline data is obtained by performing a weighted fusion operation on the weight matrix, the predicted values ​​of the intrinsic mode function components, and the predicted values ​​of the residual components.

4. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 1, is characterized in that... The clustering process of curve clustering in step S2 is as follows: S201: Standardize the bus load reference data to obtain bus load standard data; S202: Determine the optimal number of clusters by calculating the profile coefficient of the bus load standard data; S203: Initialize cluster centers based on the optimal number of clusters, and calculate the distance metric value based on the cluster centers and the bus load standard data through time series morphological similarity measurement; S204: Construct the load characteristic curve based on the distance metric.

5. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 1, is characterized in that... The assessment strategy for potential evaluation in step S3 is as follows: S301: Extract the daily load curve of the substation busbar in the time dimension based on the load characteristic curve; S302: Extract the annual daily load curve in the spatial dimension based on the load characteristic curve; S303: Preset evaluation indicators, and analyze the daily load curve and the annual daily load curve of the substation bus to obtain the bus load potential assessment result based on the evaluation indicators.

6. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 5, is characterized in that... The evaluation indicators in step S303 include daily load factor, daily minimum load factor, daily peak-to-valley difference, and daily peak-to-valley difference rate.

7. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 1, is characterized in that... The modeling process for the regional aggregated load characteristic model in step S4 is as follows: Obtain the substation busbar belonging area unit, which includes the load characteristic curve and the busbar load potential assessment result; construct a load potential feature matrix based on the busbar load potential assessment result; extract the regional aggregation features of the load potential feature matrix through the mean and covariance; and construct the regional aggregation load characteristic model based on the regional aggregation features.

8. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 1, is characterized in that... The hierarchical control process in step S4 is as follows: S401: Divide the regional intensity level based on the aforementioned regional aggregated load characteristic model; S402: The hierarchical regulation is modeled as a Markov decision process, based on the regional intensity level, the regional aggregated load characteristic model, and the bus load potential assessment results, and presets the state space, action space, and reward function. S403: Calculate the action Q value using a deep Q-network based on the state space, action space, and reward function; S404: Based on the action Q-value, select the action with the highest Q-value as the optimal load control strategy; S405: The optimal load control strategy is tested for safety and feasibility. If the test fails, the strategy is adjusted using the strategy pruning function.

9. The method for assessing and hierarchically controlling the adjustable load potential of substation busbars based on regional differences, as described in claim 8, is characterized in that... The state space in step S402 includes: the regional aggregated load characteristic model, the bus load potential assessment result, and auxiliary state variables; the action space includes regulating the cluster center and regulating the regional intensity level; and the reward function is constructed based on load reduction, user dissatisfaction, response delay, and regulation cost.

10. A substation bus-level adjustable load potential assessment and hierarchical control system oriented towards regional differences, comprising a data preprocessing module, a curve construction module, a load assessment module, and a strategy generation module, characterized in that, include: The data preprocessing module is used to acquire substation bus load data and process the missing values ​​of the substation bus load data through a data missing value processing model to obtain bus load baseline data. The curve construction module is used to perform curve clustering on the bus load reference data to obtain load characteristic curves; The load assessment module is used to perform potential assessment based on the load characteristic curve and output the bus load potential assessment result; The strategy generation module is used to construct a regional aggregated load characteristic model, and output the optimal load control strategy through hierarchical control based on the regional aggregated load characteristic model and the bus load potential assessment results.

Images