A method and system for integrated energy optimization scheduling in industrial parks based on source-load-storage synergy
By using the improved DGCRN model to predict the actual response status of the source-load-storage operation data of the park's integrated energy system, a set of adjustable equipment is selected, and a coordinated adjustment quantity is generated. This solves the problems of equipment response delay and inconsistent adjustment direction in the existing scheduling methods, and achieves accurate and stable integrated energy optimization scheduling in the park.
Patent Information
- Application Number
- CN202611304247.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
AI Technical Summary
Existing integrated energy dispatching methods in industrial parks rely on the static access location and theoretical adjustable power of equipment, lacking the collection and utilization of the actual response status after small-scale trial adjustments. This leads to a mismatch between dispatching results and target feeder constraints, affecting the accuracy and stability of source-load-storage coordinated dispatching.
By collecting source-load-storage operation data and feeder operation data of the park's integrated energy system, the improved DGCRN model is used to predict the actual response status, generate formal dispatch response prediction data, screen out the set of adjustable equipment, and generate coordinated adjustment quantities to achieve accurate and stable optimized dispatch.
This improved the accuracy of equipment selection, reduced ineffective power regulation, enhanced the stability and reliability of the source-load-storage coordinated scheduling scheme in the park, and ensured that the target feeder load rate and renewable energy acceptance capacity were met.
Smart Images

Figure CN123121502A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of energy dispatching technology, and in particular to a method and system for comprehensive energy optimization dispatching in industrial parks based on source-load-storage coordination. Background Technology
[0002] As the number of photovoltaic, energy storage, charging piles, and air conditioning loads connected to the park's integrated energy system continues to increase, the park's operation and scheduling need to simultaneously consider target feeder load control, local consumption of new energy, and energy storage support. Existing scheduling methods typically calculate based on source-side output, load power, energy storage capacity, feeder load rate, and theoretical adjustable power of equipment, and allocate adjustment power according to equipment connection relationships or adjustable capacity to achieve unified scheduling of source, load, and storage resources within the park.
[0003] Existing methods primarily rely on the static access location and theoretical adjustable power of equipment when determining which devices participate in regulation. They lack the ability to collect and utilize the actual response status after small-scale trial adjustments, making it difficult to determine whether the equipment's regulation truly affects the target feeder. While some devices may possess theoretical regulation capabilities, they may exhibit response delays, response saturation, inconsistent regulation directions, or weak actual contributions. This can lead to ineffective regulation power during formal scheduling, causing a mismatch between scheduling results and target feeder constraints, thus affecting the accuracy and stability of the park's source-load-storage coordinated scheduling.
[0004] Therefore, how to provide a method and system for integrated energy optimization scheduling in industrial parks based on source-load-storage synergy is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose a method and system for integrated energy optimization scheduling in industrial parks based on source-load-storage coordination. This invention utilizes real-world trial commissioning and feeder contribution prediction to achieve source-load-storage coordinated optimization scheduling, which has the advantages of accuracy, stability, and reliability.
[0006] A method for integrated energy optimization scheduling in a park based on source-load-storage coordination according to an embodiment of the present invention includes the following steps:
[0007] Collect source-load-storage operation data, feeder operation data, and scheduling target data of the park's integrated energy system, and preprocess them to generate a comprehensive energy operation dataset for the park.
[0008] The target feeder for the target scheduling period is determined from the comprehensive energy operation data of the park, the candidate source-load-storage equipment is screened, and the target scheduling constraints are determined.
[0009] Send small-scale trial adjustment instructions to the source-load-storage candidate equipment and collect the actual response status after the instructions are executed;
[0010] The improved DGCRN model is input with the source-load-storage candidate equipment, actual response status, target feeder and target scheduling constraints to predict feeder contribution and response capability, and generate formal scheduling response prediction data.
[0011] Based on the formal scheduling response prediction data, determine the adjustment status and adjustable power of the source-load-storage candidate equipment, and group the equipment that meets the target scheduling constraints into the source-load-storage adjustable equipment set.
[0012] Based on the adjustable equipment set of source-load-storage, the operating status of the target feeder, and the target scheduling constraints, the coordinated adjustment quantities of source-side equipment, load equipment, and energy storage equipment are determined, and a comprehensive energy optimization scheduling scheme for the park is generated.
[0013] Optionally, the source-load-storage operation data includes the operating status of source-side equipment, the operating status of load equipment, the operating status of energy storage equipment, and the equipment access location; the feeder operation data includes feeder identifier, feeder power, feeder load rate, and feeder capacity boundary; the scheduling target data includes target scheduling period, scheduling target type, and target adjustment intensity; the preprocessing includes time base unification, equipment identifier unification, feeder affiliation association, operating status coding, and abnormal data removal.
[0014] Optionally, the screening of candidate source-load-storage devices and the determination of target scheduling constraints specifically include:
[0015] Based on the target scheduling period and scheduling target type in the scheduling target data, the feeder operation status within the target scheduling period is extracted from the park's comprehensive energy operation dataset to obtain target feeder determination data.
[0016] Based on the target feeder determination data and source-load-storage operation data, identify feeders with load reduction, new energy consumption or energy storage support needs, and determine the identified feeders as target feeders;
[0017] Based on the feeder identifier and equipment access location of the target feeder, equipment connected to the target feeder and having a regulation foundation is selected from the park's comprehensive energy operation dataset to generate source-load-storage candidate equipment. The source-load-storage candidate equipment includes candidate source-side equipment, candidate load equipment, and candidate energy storage equipment.
[0018] Configure equipment scheduling attributes for the selected source-load-storage candidate equipment, and determine the target scheduling constraints in combination with the target feeder's scheduling target type and target adjustment intensity. The target scheduling constraints include the target feeder identifier, target scheduling time period, scheduling target type, target adjustment intensity, feeder capacity boundary, and equipment operation boundary.
[0019] Optionally, the small-scale trial adjustment command includes source-side output adjustment command, load power adjustment command, and energy storage charge / discharge adjustment command; the actual response state includes source-side output change, load power change, energy storage charge / discharge change, response delay, response duration, and target feeder power change.
[0020] Optionally, the generation of the formal scheduling response prediction data specifically includes:
[0021] The improved DGCRN model is input with the candidate source-load-storage equipment, actual response status, target feeder and target scheduling constraints. The graph generator establishes the candidate source-load-storage equipment nodes, target feeder nodes and scheduling association edges between nodes according to the equipment access location and target feeder identifier, and generates the source-load-storage feeder scheduling graph.
[0022] The improved DGCRN model consists of a graph generator, an encoder, a feeder contribution convolution unit, and a response amplification decoding unit connected in sequence. Compared with the existing DGCRN model, the improved DGCRN model replaces the original dynamic graph convolution recurrent unit based on dynamic graph adjacency learning and temporal state propagation with a feeder contribution convolution unit oriented towards feeder adjustment contribution analysis of source-load-storage equipment. At the same time, it replaces the original decoding unit that focuses on historical response feature mapping with a response amplification decoding unit oriented towards extrapolating small-scale trial response to formal scheduling capability.
[0023] Input the source-load-storage feeder scheduling diagram into the encoder, and encode the node feature, response sequence and constraint conditions of the candidate equipment nodes of the source-load-storage system by encoding the equipment scheduling attributes, actual response status and target scheduling constraints of the candidate equipment nodes, and generate candidate equipment response coding data.
[0024] The candidate device response encoding data is input into the feeder contribution convolution unit. Adjacent contribution convolution matching, invalid response reduction and dynamic adjacent edge weight update are performed on the device power change of the source-load-storage candidate device and the target feeder power change to generate feeder contribution characterization data.
[0025] The feeder contribution characterization data, the actual response state, the target adjustment intensity, and the equipment operating boundary are input into the response amplification and decoding unit. The response growth process from the actual response state to the target adjustment intensity is amplified step by step, the boundary is verified, and saturation is determined to generate response capability decoding data.
[0026] Based on the response capability decoding data and feeder contribution characterization data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are collected to generate formal scheduling response prediction data.
[0027] Optionally, obtaining the source-load-storage adjustable device set specifically includes:
[0028] Based on the formal scheduling response prediction data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are extracted, and the adjustment state of the source-load-storage candidate equipment is determined in combination with the target scheduling constraints. The adjustment state includes an adjustable state and an unadjustable state.
[0029] For source-load-storage candidate devices in an adjustable state, the initial adjustable power is determined based on the theoretical adjustable power and the maximum executable response power, and power limiting is performed in combination with the operating boundary to generate the adjustable power of the source-load-storage candidate devices;
[0030] Based on the target feeder identifier, target scheduling constraints, and effective contribution power, effective regulating devices are selected from the source-load-storage candidate devices with adjustable power to generate a set of adjustable source-load-storage devices. The set of adjustable source-load-storage devices includes device identifier, device type, access feeder, adjustable power, and effective contribution power.
[0031] Optionally, the generation of the integrated energy optimization and scheduling scheme for the park specifically includes:
[0032] Based on the target feeder's operating status and target scheduling constraints, the power to be adjusted for the target feeder is determined, and the power to be adjusted is associated with the set of adjustable source-load-storage equipment to generate feeder adjustment demand data;
[0033] Using the power to be regulated in the feeder regulation demand data as the allocation object, the allocation order is determined based on the effective contribution power, and power is allocated to the participating regulation equipment to generate a coordinated regulation amount. The coordinated regulation amount includes the source-side output adjustment amount, the load power adjustment amount, and the energy storage charging and discharging adjustment amount.
[0034] The coordinated adjustment quantities are aggregated according to the participating adjustment equipment and the target scheduling period to generate a comprehensive energy optimization scheduling plan for the park.
[0035] According to an embodiment of the present invention, a comprehensive energy optimization and dispatching system for industrial parks based on source-load-storage coordination includes:
[0036] The data acquisition module is used to collect source-load-storage operation data, feeder operation data, and scheduling target data, and preprocess them to generate a comprehensive energy operation dataset for the park.
[0037] The constraint determination module is used to determine the target feeder from the comprehensive energy operation data set of the park, screen candidate source-load-storage equipment, and determine the target scheduling constraints;
[0038] The small-scale trial adjustment module is used to issue small-scale trial adjustment commands to the source-load-storage candidate equipment and collect the actual response status after the command is executed;
[0039] The response prediction module is used to input the source-load-storage candidate equipment, the actual response status, the target feeder, and the target scheduling constraints into the improved DGCRN model to generate formal scheduling response prediction data.
[0040] The equipment aggregation module is used to determine the adjustment status and adjustable power of candidate source-load-storage equipment based on the formal scheduling response prediction data, and to aggregate the equipment that meets the target scheduling constraints into a source-load-storage adjustable equipment set.
[0041] The scheme generation module is used to determine the coordinated adjustment amount based on the source-load-storage adjustable equipment set, the target feeder operating status, and the target scheduling constraints, and generate a comprehensive energy optimization scheduling scheme for the park.
[0042] The beneficial effects of this invention are:
[0043] This application collects source-load-storage operation data, feeder operation data, and scheduling target data, and processes them uniformly in conjunction with target scheduling periods, target feeders, candidate source-load-storage equipment, and target scheduling constraints. This allows the park's integrated energy scheduling to move beyond relying solely on the theoretical adjustable power of equipment for allocation. Instead, it first issues small-scale trial adjustment commands to candidate equipment and collects actual response data. This enables the understanding of the actual impact of source-side equipment, load equipment, and energy storage equipment on the target feeder before formal scheduling, avoiding the inclusion of equipment with satisfactory access locations but large response delays, insufficient response continuity, or inconsistent adjustment directions in formal scheduling, thus improving the accuracy of equipment selection for regulation.
[0044] This application inputs candidate source-load-storage equipment, actual response status, target feeder, and target scheduling constraints into an improved DGCRN model to predict feeder contribution and response capability. A feeder contribution convolutional unit identifies the correlation between equipment power changes and target feeder power changes, generating effective contribution power and dynamic adjacency weights. A response amplification and decoding unit extrapolates the actual response status under small-scale trial adjustments to the target adjustment intensity, obtaining the maximum executable response power and response saturation state. This allows for the differentiation between theoretical adjustable capability and actual executable adjustable capability, reducing ineffective adjustment power and mitigating scheduling deviations caused by equipment response saturation or boundary constraints.
[0045] This application further determines the adjustment status and adjustable power of candidate source-load-storage equipment based on formal dispatch response prediction data, and groups the equipment that meets the target dispatch constraints into a set of adjustable source-load-storage equipment. Then, combining the target feeder operating status and target dispatch constraints, it determines the source-side output adjustment, load power adjustment, and energy storage charging / discharging adjustment. This method enables the coordinated adjustment to act more accurately on the target feeder, making it easier for the target feeder load rate, renewable energy acceptance capacity, or power fluctuation status to meet dispatch requirements. Simultaneously, it avoids over-reliance on a single type of equipment, improving the stability, executability, and reliability of the source-load-storage coordinated dispatch scheme in the park. Attached Figure Description
[0046] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0047] Figure 1 This is a flowchart of a comprehensive energy optimization scheduling method for industrial parks based on source-load-storage coordination proposed in this invention;
[0048] Figure 2 This is a comparison diagram of the target feeder load rate adjustment process of the integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination proposed in this invention. Detailed Implementation
[0049] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0050] refer to Figure 1 A method for integrated energy optimization scheduling in industrial parks based on source-load-storage synergy includes the following steps:
[0051] Collect source-load-storage operation data, feeder operation data, and scheduling target data of the park's integrated energy system, and preprocess them to generate a comprehensive energy operation dataset for the park.
[0052] The target feeder for the target scheduling period is determined from the comprehensive energy operation data of the park, the candidate source-load-storage equipment is screened, and the target scheduling constraints are determined.
[0053] Send small-scale trial adjustment instructions to the source-load-storage candidate equipment and collect the actual response status after the instructions are executed;
[0054] The improved DGCRN model is input with the source-load-storage candidate equipment, actual response status, target feeder and target scheduling constraints to predict feeder contribution and response capability, and generate formal scheduling response prediction data.
[0055] Based on the formal scheduling response prediction data, determine the adjustment status and adjustable power of the source-load-storage candidate equipment, and group the equipment that meets the target scheduling constraints into the source-load-storage adjustable equipment set.
[0056] Based on the adjustable equipment set of source-load-storage, the operating status of the target feeder, and the target scheduling constraints, the coordinated adjustment quantities of source-side equipment, load equipment, and energy storage equipment are determined, and a comprehensive energy optimization scheduling scheme for the park is generated.
[0057] In this embodiment, the source-load-storage operation data includes the operating status of source-side equipment, the operating status of load equipment, the operating status of energy storage equipment, and the equipment connection location; the feeder operation data includes feeder identification, feeder power, feeder load rate, and feeder capacity boundary; the scheduling target data includes the target scheduling period, the scheduling target type, and the target regulation intensity, wherein the scheduling target type includes load reduction type, new energy consumption type, and energy storage support type; the target regulation intensity is the amount of power regulation that the target feeder needs to reduce, absorb, or compensate for during the target scheduling period;
[0058] Preprocessing includes unifying time base, unifying equipment identification, associating feeder affiliations, coding operating status, and removing abnormal data;
[0059] The time base is uniformly set as the start time of the target scheduling period as 0 seconds, and the collection time is uniformly converted into cumulative seconds; the data processing cycle is set to 60 seconds, data with a time deviation of no more than 30 seconds is included in the current processing cycle, data with a time deviation of more than 30 seconds but no more than 60 seconds is included in the adjacent processing cycle, and data with a time deviation of more than 60 seconds is marked as data to be reviewed.
[0060] The device identifier uses a 16-bit character encoding. The first 4 bits represent the access source, the middle 4 bits represent the device type, and the last 8 bits represent the device serial number.
[0061] Feeder affiliation is matched according to the device access location and feeder identifier; when more than two feeder identifiers are associated with the same device access location, the feeder with the same direction of power change of the device and a time difference of no more than 60 seconds is selected as the access feeder;
[0062] The operation status coding adopts a 4-level coding: 0 represents the stopped state, 1 represents the normal operation state, 2 represents the restricted operation state, and 3 represents the abnormal operation state.
[0063] Abnormal data removal includes removing records with missing device identifiers, acquisition times, device access locations, or feeder identifiers; when there are more than two records for the same device in the same processing cycle, the record with the latest acquisition time and whose operating status has been coded is retained.
[0064] In this embodiment, the screening of source-load-storage candidate equipment and the determination of target scheduling constraints specifically include:
[0065] Based on the target scheduling period and scheduling target type in the scheduling target data, the feeder operation status within the target scheduling period is extracted from the park's comprehensive energy operation dataset to obtain target feeder determination data.
[0066] The target feeder determination data includes feeder identifier, feeder power, feeder load rate, feeder capacity boundary, scheduling target type, and target scheduling period.
[0067] Based on the target feeder determination data and source-load-storage operation data, identify feeders with load reduction, new energy consumption or energy storage support needs, and determine the identified feeders as target feeders;
[0068] The load reduction requirement is determined based on the feeder load rate and feeder capacity boundary of the target feeder during the target scheduling period. The feeder capacity boundary corresponds to a load rate of 100%, and the allowable adjustment margin is 5%. When the feeder load rate is greater than or equal to 100%, the feeder is determined to be in an overload state. When the feeder load rate is greater than or equal to 95% and less than 100%, the feeder is determined to be in a near-overload state.
[0069] The demand for renewable energy consumption is determined based on the feeder power, feeder capacity boundary, and source-side equipment operating status of the target feeder during the target scheduling period. When the source-side equipment has limited output power, the target feeder still has power capacity to accept within the feeder capacity boundary, and the load equipment or energy storage equipment has an operating boundary to increase power consumption, it is determined that the feeder has renewable energy consumption demand. The limited output power of the source-side equipment is the power that the source-side equipment can generate during the target scheduling period that is not connected to the target feeder due to insufficient acceptance capacity of the target feeder.
[0070] The energy storage support requirement is determined based on the difference in feeder power between adjacent maintenance cycles during the target scheduling period and the operating status of the energy storage equipment. The allowable power fluctuation range of the target feeder is 10% of the feeder capacity boundary. The feeder power difference between adjacent maintenance cycles is the absolute value of the difference between the feeder power in the later maintenance cycle and the feeder power in the previous maintenance cycle. When the feeder power difference exceeds the allowable power fluctuation range of the target feeder, and the energy storage equipment has a charging power boundary or a discharging power boundary, it is determined that the feeder has an energy storage support requirement.
[0071] Based on the feeder identification and equipment access location of the target feeder, equipment connected to the target feeder and having a regulation foundation is selected from the park's integrated energy operation data to generate source-load-storage candidate equipment. The source-load-storage candidate equipment includes candidate source-side equipment, candidate load equipment, and candidate energy storage equipment.
[0072] The conditions for adjustment include that the equipment is in normal or restricted operation, the current power does not exceed the operating boundary, the theoretical adjustable power is not 0, and the equipment connected to the feeder is consistent with the target feeder.
[0073] Configure equipment scheduling attributes for the selected source-load-storage candidate equipment, and determine the target scheduling constraints in combination with the target feeder's scheduling target type and target adjustment intensity. The target scheduling constraints include the target feeder identifier, target scheduling period, scheduling target type, target adjustment intensity, feeder capacity boundary, and equipment operation boundary.
[0074] Equipment scheduling attributes include equipment identifier, equipment type, connected feeder, current power, theoretical adjustable power, and operating boundary; the equipment operating boundary consists of source-side output boundary, load regulation boundary, energy storage capacity boundary, energy storage charging power boundary, and energy storage discharging power boundary;
[0075] The determination of target scheduling constraints includes: when the scheduling target type is load reduction, the target regulation intensity is configured as the amount of feeder power reduction required to bring the target feeder load rate down to below 95%, and the feeder capacity boundary and load regulation boundary are used as constraint conditions; when the scheduling target type is renewable energy consumption, if the limited output power of the source-side equipment is less than or equal to the remaining power capacity of the target feeder, the target regulation intensity is configured as the limited output power of the source-side equipment; if the limited output power of the source-side equipment is greater than the remaining power capacity of the target feeder, the target regulation intensity is configured as the remaining power capacity of the target feeder, where the remaining power capacity of the target feeder is the difference between the feeder capacity boundary and the current feeder power of the target feeder, and the source-side output boundary, energy storage charging power boundary, and feeder capacity boundary are configured as constraint conditions in the target scheduling constraints; when the scheduling target type is energy storage support, the target regulation intensity is configured as the portion of the feeder power difference that exceeds the allowable power fluctuation range of the target feeder, and the energy storage capacity boundary, energy storage charging power boundary, energy storage discharging power boundary, and feeder capacity boundary are configured as constraint conditions in the target scheduling constraints.
[0076] In this embodiment, the small-scale trial adjustment command includes the source-side output adjustment command, the load power adjustment command, and the energy storage charging and discharging adjustment command. Before the small-scale trial adjustment command is issued, the operating data of the source-load-storage candidate equipment within 60 seconds before the command is issued is extracted as the baseline data. The small-scale trial adjustment command is issued at 5% of the theoretical adjustable power of the source-load-storage candidate equipment, and the single trial adjustment range is controlled between 1 kW and 10 kW. When there are multiple source-load-storage candidate equipment on the same target feeder, the small-scale trial adjustment command is issued sequentially according to the equipment identification. The execution time of the small-scale trial adjustment command is 180 seconds, and the equipment operating status and target feeder power change are recorded with a 10-second acquisition cycle. The data from the 30th to the 180th second after the command is executed is used as the stable response data. The small-scale trial adjustment command of the next source-load-storage candidate equipment is issued after the stable response data acquisition of the previous source-load-storage candidate equipment ends and after an interval of 120 seconds.
[0077] The actual response status includes changes in source-side output, load power, energy storage charging and discharging, response delay, response duration, and target feeder power. The response delay is recorded as the time when the equipment power change first reaches 30% of the command amplitude after the small-scale trial adjustment command is issued. The response duration is recorded as the continuous duration for which the equipment power change remains above 80% of the command amplitude. When the response delay is no more than 60 seconds, the response duration is no less than 120 seconds, and the direction of equipment power change is consistent with the direction of the small-scale trial adjustment command, the actual response status corresponding to this small-scale trial adjustment is retained as subsequent input data.
[0078] In this embodiment, the generation of formal scheduling response prediction data specifically includes:
[0079] The improved DGCRN model is input with the candidate source-load-storage equipment, actual response status, target feeder and target scheduling constraints. The graph generator establishes the candidate source-load-storage equipment nodes, target feeder nodes and scheduling association edges between nodes according to the equipment access location and target feeder identifier, and generates the source-load-storage feeder scheduling graph.
[0080] The improved DGCRN model consists of a graph generator, encoder, feeder contribution convolutional unit, and response amplification decoding unit connected sequentially. Compared with the existing DGCRN model, the improved model replaces the original dynamic graph convolutional recurrent unit based on dynamic graph adjacency learning and temporal state propagation with a feeder contribution convolutional unit oriented towards feeder regulation contribution analysis of source-load-storage equipment. This enables the model to distinguish the differences in the regulation effects of different source-load-storage equipment on the target feeder, improving the ability to identify effective regulation resources. Simultaneously, the original decoding unit, which focused on historical response feature mapping, is replaced with a response amplification decoding unit oriented towards extrapolating small-scale trial responses to formal scheduling capabilities. This allows the model to characterize the growth process of equipment response as scheduling intensity changes, improving the predictive ability for maximum executable response power and response saturation state. Through these improvements, the model's representation of the response capability of the source-load-storage coordinated scheduling process in the park is enhanced, improving the accuracy and reliability of formal scheduling scheme generation.
[0081] The improved DGCRN model's loss function is composed of a weighted average of the formal scheduling response power prediction loss, response saturation state discrimination loss, effective contribution power matching loss, and dynamic adjacency edge weight smoothing loss. Specifically, the formal scheduling response power prediction loss is calculated using the mean squared error method, the response saturation state discrimination loss using the binary cross-entropy method, the effective contribution power matching loss using the mean absolute error method, and the dynamic adjacency edge weight smoothing loss using the L2 regularization method of edge weight difference. The training samples consist of small-scale trial-tuning commands, actual response states, formal scheduling execution records, and target feeder power change records, and are trained using 32 groups. The samples constitute a batch-input improved DGCRN model, and the parameters of the graph generator, encoder, feeder contribution convolutional unit, and response amplification decoding unit are iteratively updated with an initial learning rate of 0.001. The weight decay coefficient is set to 0.0001, the hidden feature dimension is set to 64, the number of encoding layers is set to 2, the gradient clipping upper limit is set to 5, and the maximum number of training epochs is set to 200. When the validation loss decreases by less than 0.001 for 10 consecutive epochs, and the formal scheduling response power prediction error, response saturation state discrimination error, and effective contribution power matching error do not continue to decrease, the improved DGCRN model is considered to have converged.
[0082] The generation of the source-load-storage feeder scheduling diagram includes: nodalizing the candidate source-load-storage equipment by configuring the equipment identifier, equipment type, equipment access location, access feeder, current power, theoretical adjustable power, and operating boundary of each candidate equipment to its nodes, and configuring node indexes for these candidate equipment nodes in ascending order of equipment identifier; nodalizing the target feeder by configuring the target feeder identifier, target scheduling period, feeder power, feeder load rate, and feeder capacity boundary to its nodes; and establishing equipment access edges between the candidate source-load-storage equipment nodes and the target feeder nodes based on the equipment access location and access feeder, and configuring these equipment access edges... The initial edge weight is configured as 1; based on the changes in source-side output, load power, energy storage charging and discharging, and target feeder power in the actual response state, feeder regulation association edges are established between the candidate source-load-storage equipment nodes and the target feeder nodes, and the initial edge weight of the feeder regulation association edges is configured as 0.5; based on the scheduling target type and target regulation intensity, scheduling direction markers and regulation intensity markers are configured for the feeder regulation association edges. In the scheduling direction marker, 1 represents load reduction, 2 represents new energy consumption, and 3 represents energy storage support. The regulation intensity marker is generated according to the ratio of the target regulation intensity to the feeder capacity boundary and retains 3 decimal places to generate the source-load-storage feeder scheduling diagram.
[0083] Input the source-load-storage feeder scheduling diagram into the encoder, and encode the node feature, response sequence and constraint conditions of the candidate equipment nodes of the source-load-storage system by encoding the equipment scheduling attributes, actual response status and target scheduling constraints of the candidate equipment nodes, and generate candidate equipment response coding data.
[0084] Node feature encoding includes: converting device identifiers into node indices, converting device types into type codes, where 1 represents source-side devices, 2 represents load devices, and 3 represents energy storage devices; converting access feeders into feeder affiliation codes; and converting current power, theoretical adjustable power, and operating boundaries into power boundary codes. The current power, theoretical adjustable power, and operating boundaries are all normalized according to the ratio of their corresponding values to the feeder capacity boundary. The encoding results are retained to 3 decimal places, generating node encoding results.
[0085] The response sequence encoding includes: arranging the source-side output change, load power change, energy storage charging and discharging change, and target feeder power change into a response sequence according to the acquisition cycle; incorporating the response delay and response duration into the corresponding acquisition cycle; determining the response sequence length based on the stable response data from the 30th to the 180th second after the command is executed; forming 16 acquisition points when the acquisition cycle is 10 seconds; and generating the response sequence encoding result.
[0086] The constraint coding includes: converting the scheduling target type into a scheduling direction code, where 1 represents load reduction, 2 represents new energy consumption, and 3 represents energy storage support; converting the target regulation intensity into an intensity code; and converting the feeder capacity boundary and equipment operation boundary into boundary codes. The target regulation intensity, feeder capacity boundary, and equipment operation boundary are normalized and coded according to the ratio of their corresponding values to the feeder capacity boundary. The coding result is retained to 3 decimal places, and the constraint coding result is generated.
[0087] The node encoding results, response sequence encoding results, and constraint encoding results are associated according to the device access edge and the feeder adjustment association edge to generate candidate device response encoding data;
[0088] The candidate device response encoding data is input into the feeder contribution convolution unit. Adjacent contribution convolution matching, invalid response reduction and dynamic adjacency edge weight update are performed on the device power change of the source-load-storage candidate device and the target feeder power change to generate feeder contribution characterization data. The feeder contribution characterization data includes the effective contribution power and dynamic adjacency edge weight corresponding to the source-load-storage candidate device.
[0089] The adjacency contribution convolutional matching includes: determining the adjacency range between the source-load storage candidate device and the target feeder according to the device access edge and the feeder adjustment association edge; convolving and aggregating the device power change, the target feeder power change, the response delay and the response duration to generate the initial contribution response;
[0090] Invalid response reduction includes: determining the target feeder adjustment direction based on the scheduling target type, configuring initial contribution responses opposite to the target feeder adjustment direction as reverse contribution responses; reducing initial contribution responses with response delays greater than 60 seconds, response durations less than 120 seconds, or target feeder power changes that do not maintain a stable direction, while retaining valid contribution responses; target feeder power changes that do not maintain a stable direction refer to situations where the target feeder power change direction changes in the opposite direction more than twice within the acquisition period corresponding to the stable response data, or the target feeder power change direction is inconsistent with the target feeder adjustment direction corresponding to the scheduling target type;
[0091] Dynamic adjacency edge weight update includes: increasing the feeder adjustment associated edge weight based on the effective contribution power corresponding to the effective contribution response, decreasing the feeder adjustment associated edge weight based on the reverse contribution response or the reduced initial contribution response, and keeping the device access edge weight unchanged. The dynamic adjacency edge weight ranges from 0 to 1. The feeder adjustment associated edge weight corresponding to the effective contribution response increases by 0.2, the feeder adjustment associated edge weight corresponding to the reverse contribution response decreases by 0.3, and the feeder adjustment associated edge weight corresponding to the reduced initial contribution response decreases by 0.2. After the update, values exceeding the range are truncated to 0 or 1.
[0092] The effective contribution power is determined based on the changes in equipment power and target feeder power in the effective contribution response. When the scheduling target type is load reduction, the changes in equipment power that reduce the target feeder power are determined as the effective contribution power. When the scheduling target type is renewable energy consumption, the changes in equipment power that increase the target feeder power are determined as the effective contribution power. When the scheduling target type is energy storage support, the changes in energy storage charging and discharging power that reduce the target feeder power difference are determined as the effective contribution power.
[0093] The feeder contribution characterization data, the actual response state, the target adjustment intensity, and the equipment operating boundary are input into the response amplification and decoding unit. The response growth process from the actual response state to the target adjustment intensity is amplified step by step, the boundary is verified, and saturation is determined to generate response capability decoding data.
[0094] The stepwise amplification includes: taking the equipment power change in the actual response state as the initial response quantity, increasing the response simulation intensity stepwise from low to high according to the target adjustment intensity, and combining the effective contribution power and dynamic adjacent edge weight in the feeder contribution characterization data to determine the equipment power change and target feeder power change under each level of response simulation intensity;
[0095] The response simulation intensity is based on the trial adjustment range corresponding to the small trial adjustment command as the starting level, and the response simulation levels are formed in sequence according to 20%, 40%, 60%, 80% and 100% of the target adjustment intensity; when the target adjustment intensity is less than the trial adjustment range corresponding to the small trial adjustment command, the target adjustment intensity is used as the single-level response simulation intensity.
[0096] Boundary verification includes: comparing the equipment power changes under each level of response simulation intensity with the equipment operating boundary; comparing the target feeder power changes with the feeder capacity boundary; when the equipment power change exceeds the source-side output boundary, load regulation boundary, energy storage capacity boundary, energy storage charging power boundary, or energy storage discharging power boundary, the corresponding response simulation result is truncated, and the truncated equipment power change is taken as the maximum power change value allowed by the corresponding equipment operating boundary; when the target feeder power change exceeds the feeder capacity boundary, the corresponding response simulation result is truncated, and the truncated target feeder power change is taken as the maximum power change value allowed by the feeder capacity boundary.
[0097] The saturation determination includes: based on the response simulation results after boundary verification, determining whether the equipment power change of the source-load-storage candidate equipment continues to increase with the increase of the response simulation intensity; when the equipment power change remains unchanged under the two adjacent response simulation intensities, or when the equipment power change reaches the equipment operating boundary, the corresponding source-load-storage candidate equipment is marked as response saturated, wherein the difference in equipment power change under the two adjacent response simulation intensities is not greater than 0.5 kW, and the equipment power change is determined to remain unchanged;
[0098] Based on the step-by-step amplification results, boundary truncation markers, and response saturation states, the maximum device power change that is not truncated and has not reached response saturation is selected to determine the maximum executable response power and generate response capability decoding data.
[0099] Based on the response capability decoding data and feeder contribution characterization data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are collected to generate formal scheduling response prediction data.
[0100] In this embodiment, obtaining the adjustable source-load-storage device set specifically includes:
[0101] Based on the formal scheduling response prediction data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are extracted. Combined with the target scheduling constraints, the adjustment state of the source-load-storage candidate equipment is determined. The adjustment state includes adjustable and non-adjustable states. When the effective contribution power and maximum executable response power of the source-load-storage candidate equipment are greater than 0, and the response saturation state is unsaturated, the candidate equipment is determined to be in an adjustable state. When the effective contribution power and maximum executable response power of the source-load-storage candidate equipment are greater than 0, and the response saturation state is saturated, the candidate equipment is determined to be in an adjustable state. All other source-load-storage candidate equipment is determined to be in a non-adjustable state.
[0102] For source-load-storage candidate devices in an adjustable state, the initial adjustable power is determined based on the theoretical adjustable power and the maximum executable response power, and power limiting is performed in combination with the operating boundary to generate the adjustable power of the source-load-storage candidate devices;
[0103] The determination of the initial adjustable power includes: extracting the theoretical adjustable power and maximum executable response power of the source-load-storage candidate equipment in the adjustable state, determining the target feeder adjustment direction according to the scheduling target type, and determining the equipment adjustment direction according to the equipment type and theoretical adjustable power; when the equipment adjustment direction is consistent with the target feeder adjustment direction, the smaller power value between the theoretical adjustable power and the maximum executable response power is determined as the initial adjustable power; when the equipment adjustment direction is inconsistent with the target feeder adjustment direction, or when the maximum executable response power is 0, the initial adjustable power of the source-load-storage candidate equipment is configured to 0.
[0104] Power limiting includes: limiting the initial adjustable power based on the operating boundary of the source-load-storage candidate equipment, and determining the limited power as the adjustable power;
[0105] Based on the target feeder identifier, target scheduling constraints, and effective contribution power, effective regulating devices are selected from the source-load-storage candidate devices with adjustable power to generate a set of adjustable source-load-storage devices. The set of adjustable source-load-storage devices includes device identifier, device type, access feeder, adjustable power, and effective contribution power.
[0106] The selection of effective regulating equipment includes: source-side equipment, load equipment, and energy storage equipment whose access feeder and target feeder are identified in the same way, whose equipment regulation direction is consistent with the target feeder regulation direction, and whose adjustable power is greater than 0 and effective contribution power is greater than 0, are classified into the source-load-storage adjustable equipment set.
[0107] In this embodiment, the generation of the park's integrated energy optimization scheduling scheme specifically includes:
[0108] Based on the target feeder's operating status and target scheduling constraints, the power to be adjusted for the target feeder is determined, and the power to be adjusted is associated with the set of adjustable source-load-storage equipment to generate feeder adjustment demand data;
[0109] The power to be adjusted is determined based on the type of dispatch target and the intensity of target adjustment. When the dispatch target is load reduction, the power that the target feeder needs to reduce is determined as the power to be adjusted. When the dispatch target is renewable energy consumption, the power that the target feeder needs to increase is determined as the power to be adjusted. When the dispatch target is energy storage support, the power fluctuation that the target feeder needs to smooth is determined as the power to be adjusted.
[0110] The feeder regulation demand data includes the target feeder identifier, scheduling target type, power to be regulated, target scheduling period, and participating equipment; the participating equipment are source-side equipment, load equipment, and energy storage equipment whose source-load-storage adjustable equipment is centrally connected to the feeder and whose identifier is consistent with the target feeder identifier;
[0111] The allocation object is the power to be regulated in the feeder regulation demand data. The allocation order is determined according to the effective contribution power, and the power is allocated to the participating regulation equipment to generate the coordinated regulation quantity. The coordinated regulation quantity includes the source side output adjustment quantity, the load power adjustment quantity, and the energy storage charging and discharging adjustment quantity.
[0112] Power allocation involves selecting participating equipment in descending order of effective contribution power, with 1 kW as the minimum allocation unit for equipment adjustment power. The adjustment power of a single participating equipment shall not exceed its adjustable power. The equipment adjustment power shall be allocated within the equipment operation boundary and feeder capacity boundary limits in the target scheduling constraints until the allocated power meets the power to be adjusted or all participating equipment reaches its adjustable power limit. When the remaining power to be allocated is less than 1 kW and there are participating equipment that has not yet reached its adjustable power limit, the remaining power to be allocated shall be allocated to the participating equipment that has not yet reached its adjustable power limit and has the highest effective contribution power. When the difference between the allocated power and the power to be adjusted is not greater than 0.001 kW, the allocated power is determined to meet the power to be adjusted.
[0113] The coordinated adjustment quantities are aggregated according to the participating adjustment equipment and the target scheduling period to generate a comprehensive energy optimization scheduling plan for the park.
[0114] A comprehensive energy optimization and dispatching system for industrial parks based on source-load-storage coordination includes:
[0115] The data acquisition module is used to collect source-load-storage operation data, feeder operation data, and scheduling target data, and preprocess them to generate a comprehensive energy operation dataset for the park.
[0116] The constraint determination module is used to determine the target feeder from the comprehensive energy operation data set of the park, screen candidate source-load-storage equipment, and determine the target scheduling constraints;
[0117] The small-scale trial adjustment module is used to issue small-scale trial adjustment commands to the source-load-storage candidate equipment and collect the actual response status after the command is executed;
[0118] The response prediction module is used to input the source-load-storage candidate equipment, the actual response status, the target feeder, and the target scheduling constraints into the improved DGCRN model to generate formal scheduling response prediction data.
[0119] The equipment aggregation module is used to determine the adjustment status and adjustable power of candidate source-load-storage equipment based on the formal scheduling response prediction data, and to aggregate the equipment that meets the target scheduling constraints into a source-load-storage adjustable equipment set.
[0120] The scheme generation module is used to determine the coordinated adjustment amount based on the source-load-storage adjustable equipment set, the target feeder operating status, and the target scheduling constraints, and generate a comprehensive energy optimization scheduling scheme for the park.
[0121] Example 1: To verify the feasibility of this invention in practice, it was applied to an integrated energy system in an industrial park. The park is equipped with two main feeders, four rooftop photovoltaic inverters, six adjustable air conditioning loads, twelve charging pile loads, and two sets of energy storage batteries. The operation management platform performs a source-load-storage coordinated scheduling operation from 14:00 to 15:00. During this period, the capacity boundary of feeder F02 is 2000 kW, the real-time feeder power is 1952 kW, and the feeder load rate is 97.6%, which is close to overload. Meanwhile, some photovoltaic equipment still has output capacity, the energy storage battery state of charge is 62%, and the charging pile load has room for adjustment. The park's original scheduling method mainly allocated power based on theoretical adjustable capacity, failing to determine whether the equipment adjustments actually affected feeder F02, easily leading to the problem that the equipment executed the adjustment command but the target feeder power change was insufficient.
[0122] In this scenario, the system designates 14:00 to 15:00 as the target scheduling period and, based on the load rate and capacity boundary of feeder F02, determines the scheduling target type as load reduction and the target adjustment intensity as 58 kW. The system selects candidate source-load-storage devices from those connected to feeder F02, including 2 photovoltaic inverters, 3 sets of air conditioning loads, 5 charging pile loads, and 1 set of energy storage batteries. Each candidate device is configured with a device identifier, device type, connected feeder, current power, theoretical adjustable power, and operating boundary. Subsequently, the system issues small-scale trial adjustment commands to the candidate devices and collects the actual response status; the trial adjustment power for air conditioning loads is 3 kW, for charging pile loads it is 2 kW, for energy storage discharge it is 4 kW, and for photovoltaic output adjustment it is 2.5 kW.
[0123] The system inputs the aforementioned candidate devices, actual response status, F02 feeder, and target scheduling constraints into the improved DGCRN model to predict the feeder contribution and response capability of each candidate device, obtaining the maximum executable response power, response saturation state, and effective contribution power. Based on the prediction results, the system eliminates devices with zero effective contribution power to the F02 feeder or inconsistent adjustment directions, and groups the air conditioning load, charging pile load, and energy storage devices that still have adjustable power into a source-load-storage adjustable device set. Finally, the system uses 58 kW of power to be adjusted as the allocation object, determines the allocation order according to the effective contribution power from largest to smallest, and forms the source-side output adjustment amount, load power adjustment amount, and energy storage charging and discharging adjustment amount. The coordinated adjustment amount is then aggregated according to the participating adjustment devices and target scheduling time periods to generate a comprehensive energy optimization scheduling scheme for the park.
[0124] In the above-mentioned park application scenario, the scheduling records of 20 consecutive target scheduling periods in the same park were statistically analyzed, and the method of the present invention was compared with the traditional method of directly allocating according to theoretical adjustable power. The data shown in Table 1 is as follows:
[0125] Table 1. Performance Comparison of the Invention and Traditional Integrated Energy Optimization Scheduling Methods in Industrial Parks
[0126] Indicator Categories Traditional methods Method of the present invention Scheduling target satisfaction rate (%) 92.6 96.8 Effective device recognition accuracy (%) 87.9 93.4 Response power prediction error (kW) 7.8 5.6 Ineffective regulation power (kW) 11.4 6.8 Adjusted feeder load rate (%) 95.4 94.7 Feeder over-limit duration (min) 8.0 3.0 Energy storage SOC deviation (%) 4.8 3.6 Scheduling computation time (s) 6.8 7.4
[0127] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.
[0128] The scheduling target satisfaction rate of the method of this invention is 96.8%, which is higher than the 92.6% of the traditional method, representing an improvement of 4.2 percentage points. This indicates that under the same target feeder and the same target adjustment intensity, this invention can more effectively complete the tasks of load reduction, new energy consumption, or energy storage support. Combined with... Figure 2 It can be seen that when the initial load rate of feeder F02 is 97.6%, the traditional method adjusts the feeder load rate to 95.4%, failing to stably fall below the target load rate boundary. However, the method of this invention reduces the feeder load rate to 94.7%, meeting the scheduling constraints under load reduction scenarios. This result indicates that this invention does not directly allocate power according to the theoretically adjustable power, but first collects the actual response state through small-scale trial adjustments, and then combines this with the improved DGCRN model to predict the maximum executable response power, thereby improving the execution reliability of the formal scheduling scheme.
[0129] Regarding the accuracy of effective device identification, the method of this invention achieves 93.4%, while the traditional method achieves 87.9%, representing an improvement of 5.5 percentage points. This improvement stems from the correlation analysis of device power changes and target feeder power changes by the feeder contribution convolution unit. This analysis can identify whether the device regulation is truly effective on the target feeder, avoiding misclassification of devices with a connection relationship but weak actual contribution as effective regulating devices. Figure 2 The curve of the method of this invention drops more smoothly, which further illustrates that the equipment in the adjustable equipment set does not only meet the theoretical adjustable conditions, but also has an effective contribution power that can affect the load rate of the F02 feeder.
[0130] The response power prediction error was reduced from 7.8 kW to 5.6 kW using the traditional method, and the ineffective regulation power was reduced from 11.4 kW to 6.8 kW, indicating that the present invention's estimation of the equipment's response capability is closer to the actual execution state. This is because the response amplification and decoding unit extrapolates the true response state to the target regulation intensity and combines this with the equipment's operating boundaries and response saturation state for judgment, reducing the deviation caused by simply proportional amplification. Therefore, Figure 2 The method of this invention can continuously approach and exceed 95.0% of the target boundary during the scheduling period, rather than experiencing insufficient response at the end of the adjustment.
[0131] After adjustment, the feeder load rate decreased from 95.4% using the traditional method to 94.7% using the method of this invention, and the feeder over-limit time decreased from 8.0 minutes to 3.0 minutes, indicating that the adjustment of the target feeder by this invention is more stable. The energy storage SOC deviation decreased from 4.8% to 3.6% because the coordinated power allocation simultaneously considers the effective contribution power, adjustable power, and equipment operating boundaries, avoiding over-reliance on energy storage devices. The scheduling calculation time increased from 6.8 seconds to 7.4 seconds, mainly because the improved DGCRN model added the feeder contribution and response capability prediction process, but the increase was only 0.6 seconds, still meeting the requirements of hourly scheduling applications in the park.
[0132] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for integrated energy optimization scheduling in industrial parks based on source-load-storage synergy, characterized in that, Includes the following steps: Collect source-load-storage operation data, feeder operation data, and scheduling target data of the park's integrated energy system, and preprocess them to generate a comprehensive energy operation dataset for the park. The target feeder for the target scheduling period is determined from the comprehensive energy operation data of the park, the candidate source-load-storage equipment is screened, and the target scheduling constraints are determined. Send small-scale trial adjustment instructions to the source-load-storage candidate equipment and collect the actual response status after the instructions are executed; The improved DGCRN model is input with the source-load-storage candidate equipment, actual response status, target feeder and target scheduling constraints to predict feeder contribution and response capability, and generate formal scheduling response prediction data. Based on the formal scheduling response prediction data, determine the adjustment status and adjustable power of the source-load-storage candidate equipment, and group the equipment that meets the target scheduling constraints into the source-load-storage adjustable equipment set. Based on the adjustable equipment set of source-load-storage, the operating status of the target feeder, and the target scheduling constraints, the coordinated adjustment quantities of source-side equipment, load equipment, and energy storage equipment are determined, and a comprehensive energy optimization scheduling scheme for the park is generated.
2. The method for integrated energy optimization scheduling in industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The source-load-storage operation data includes the operating status of source-side equipment, the operating status of load equipment, the operating status of energy storage equipment, and the equipment access location; the feeder operation data includes feeder identification, feeder power, feeder load rate, and feeder capacity boundary; the scheduling target data includes target scheduling time period, scheduling target type, and target adjustment intensity; the preprocessing includes time base unification, equipment identification unification, feeder affiliation association, operating status coding, and abnormal data removal.
3. The integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The selection of candidate source-load-storage equipment and the determination of target scheduling constraints specifically include: Based on the target scheduling period and scheduling target type in the scheduling target data, the feeder operation status within the target scheduling period is extracted from the park's comprehensive energy operation dataset to obtain target feeder determination data. Based on the target feeder determination data and source-load-storage operation data, identify feeders with load reduction, new energy consumption or energy storage support needs, and determine the identified feeders as target feeders; Based on the feeder identifier and equipment access location of the target feeder, equipment connected to the target feeder and having a regulation foundation is selected from the park's comprehensive energy operation dataset to generate source-load-storage candidate equipment. The source-load-storage candidate equipment includes candidate source-side equipment, candidate load equipment, and candidate energy storage equipment. Configure equipment scheduling attributes for the selected source-load-storage candidate equipment, and determine the target scheduling constraints in combination with the target feeder's scheduling target type and target adjustment intensity. The target scheduling constraints include the target feeder identifier, target scheduling time period, scheduling target type, target adjustment intensity, feeder capacity boundary, and equipment operation boundary.
4. The integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The small-scale trial adjustment commands include source-side output adjustment commands, load power adjustment commands, and energy storage charge / discharge adjustment commands; the actual response status includes changes in source-side output, changes in load power, changes in energy storage charge / discharge, response delay, response duration, and changes in target feeder power.
5. The integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The generation of the formal scheduling response prediction data specifically includes: The improved DGCRN model is input with the candidate source-load-storage equipment, actual response status, target feeder and target scheduling constraints. The graph generator establishes the candidate source-load-storage equipment nodes, target feeder nodes and scheduling association edges between nodes according to the equipment access location and target feeder identifier, and generates the source-load-storage feeder scheduling graph. The improved DGCRN model consists of a graph generator, an encoder, a feeder contribution convolution unit, and a response amplification decoding unit connected in sequence. Compared with the existing DGCRN model, the improved DGCRN model replaces the original dynamic graph convolution recurrent unit based on dynamic graph adjacency learning and temporal state propagation with a feeder contribution convolution unit oriented towards feeder adjustment contribution analysis of source-load-storage equipment. At the same time, it replaces the original decoding unit that focuses on historical response feature mapping with a response amplification decoding unit oriented towards extrapolating small-scale trial response to formal scheduling capability. Input the source-load-storage feeder scheduling diagram into the encoder, and encode the node feature, response sequence and constraint conditions of the candidate equipment nodes of the source-load-storage system by encoding the equipment scheduling attributes, actual response status and target scheduling constraints of the candidate equipment nodes, and generate candidate equipment response coding data. The candidate device response encoding data is input into the feeder contribution convolution unit. Adjacent contribution convolution matching, invalid response reduction and dynamic adjacent edge weight update are performed on the device power change of the source-load-storage candidate device and the target feeder power change to generate feeder contribution characterization data. The feeder contribution characterization data, the actual response state, the target adjustment intensity, and the equipment operating boundary are input into the response amplification and decoding unit. The response growth process from the actual response state to the target adjustment intensity is amplified step by step, the boundary is verified, and saturation is determined to generate response capability decoding data. Based on the response capability decoding data and feeder contribution characterization data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are collected to generate formal scheduling response prediction data.
6. The integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The acquisition of the adjustable source-load storage device set specifically includes: Based on the formal scheduling response prediction data, the maximum executable response power, response saturation state, and effective contribution power of the source-load-storage candidate equipment are extracted, and the adjustment state of the source-load-storage candidate equipment is determined in combination with the target scheduling constraints. The adjustment state includes an adjustable state and an unadjustable state. For source-load-storage candidate devices in an adjustable state, the initial adjustable power is determined based on the theoretical adjustable power and the maximum executable response power, and power limiting is performed in combination with the operating boundary to generate the adjustable power of the source-load-storage candidate devices; Based on the target feeder identifier, target scheduling constraints, and effective contribution power, effective regulating devices are selected from the source-load-storage candidate devices with adjustable power to generate a set of adjustable source-load-storage devices. The set of adjustable source-load-storage devices includes device identifier, device type, access feeder, adjustable power, and effective contribution power.
7. The integrated energy optimization scheduling method for industrial parks based on source-load-storage coordination as described in claim 1, characterized in that, The generation of the comprehensive energy optimization and scheduling scheme for the park specifically includes: Based on the target feeder's operating status and target scheduling constraints, the power to be adjusted for the target feeder is determined, and the power to be adjusted is associated with the set of adjustable source-load-storage equipment to generate feeder adjustment demand data; Using the power to be regulated in the feeder regulation demand data as the allocation object, the allocation order is determined according to the effective contribution power, and power is allocated to the participating regulation equipment to generate a coordinated regulation amount. The coordinated regulation amount includes the source-side output adjustment amount, the load power adjustment amount, and the energy storage charging and discharging adjustment amount. The coordinated adjustment quantities are aggregated according to the participating adjustment equipment and the target scheduling period to generate a comprehensive energy optimization scheduling plan for the park.
8. A park integrated energy optimization scheduling system based on source-load-storage coordination, executing the park integrated energy optimization scheduling method based on source-load-storage coordination as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect source-load-storage operation data, feeder operation data, and scheduling target data, and preprocess them to generate a comprehensive energy operation dataset for the park. The constraint determination module is used to determine the target feeder from the comprehensive energy operation data set of the park, screen candidate source-load-storage equipment, and determine the target scheduling constraints; The small-scale trial adjustment module is used to issue small-scale trial adjustment commands to the source-load-storage candidate equipment and collect the actual response status after the command is executed; The response prediction module is used to input the source-load-storage candidate equipment, the actual response status, the target feeder, and the target scheduling constraints into the improved DGCRN model to generate formal scheduling response prediction data. The equipment aggregation module is used to determine the adjustment status and adjustable power of candidate source-load-storage equipment based on the formal scheduling response prediction data, and to aggregate the equipment that meets the target scheduling constraints into a source-load-storage adjustable equipment set. The scheme generation module is used to determine the coordinated adjustment amount based on the source-load-storage adjustable equipment set, the target feeder operating status, and the target scheduling constraints, and generate a comprehensive energy optimization scheduling scheme for the park.
