A system and method for prosumer participation in auxiliary peak shaving based on neural algorithms

Abstract

The present application relates to the technical field of power system operation control and power transaction, and particularly relates to a system and method for prosumers participating in auxiliary peak shaving based on a neural algorithm. The method acquires prosumer types, target peak shaving time periods and production process state records, matches flexible resource parameters, and generates dynamic adjustment boundary records. The peak shaving demand and the load are bound as a peak shaving task context, which is input into a neural network model to generate candidate adjustment task records for verification and standby states. The production process state records are reacquired, the dynamic adjustment boundary is updated, the effective adjustment amount and the verification failure amount are determined, the standby candidate tasks are screened according to the remaining peak shaving gap, and an executable task set is formed. A peak shaving instruction package is generated, and an execution evaluation record and an adjustment capacity update record are generated according to the baseline load and the actual load. The present application matches the candidate adjustment tasks with the production process state before execution, forms a gap supplementing and result writing back relationship, and improves the executability of the peak shaving task.

Description

Technical Field

[0001] This invention relates to the fields of power system operation control and power trading technology, and in particular to a system and method for auxiliary peak shaving based on neural algorithms involving producer-consumer participation. Background Technology

[0002] In the fields of power system operation and control and power trading, existing schemes for producer-consumer (PPC) participation in auxiliary peak shaving typically collect real-time load, given load forecast data, peak shaving demand, market prices, and elastic resource parameters. These data are then used to formulate PPC output schemes or peak shaving control strategies through neural network models or peak shaving calculation models. Corresponding control commands are then issued to PPC terminals, and an execution evaluation is generated based on the actual load during the execution period. Such schemes can handle the selection of adjustable resources from multiple PPCs, the allocation of regulated power, and the issuance of commands.

[0003] Existing solutions often use adjustable power limits, adjustable duration, ramp-up constraints, or capacity status as resource constraints during model calculations, or form peak-shaving tasks based on pre-configured producer-consumer elastic resource parameters. For producer-consumers with continuous operation, downtime, restart preheating, and resumption restrictions, their production process status may change after candidate adjustment tasks are generated. The resource status used in model calculations is prone to inconsistency with the actual operating status before the peak-shaving command is issued, making it difficult for candidate adjustment power, candidate adjustment duration, or candidate execution period to correspond to the production process conditions at the time of execution.

[0004] When candidate regulation tasks cannot be executed at the original regulation power due to changes in production process status, task occupancy, or lack of terminal confirmation, the actual effective regulation provided by multiple producers and consumers may be less than the peak-shaving demand. Existing solutions typically recalculate the overall peak-shaving plan or delete and replace the original tasks. The regulation power gap caused by the failure of candidate tasks lacks a continuous record and calling relationship with subsequent resource selection, which can easily lead to discontinuities in the connection between task status, remaining peak-shaving demand, and subsequent regulation tasks.

[0005] For multi-producer-consumer peak shaving handling under conditions of changing production process status after candidate regulation tasks are generated, existing technologies still have shortcomings in processing coordination regarding pre-execution regulation capacity verification, failure regulation power recording, and subsequent task allocation for remaining peak shaving gaps. Therefore, it is necessary to address the issue of generating executable peak shaving tasks when candidate regulation tasks and execution conditions are inconsistent, by focusing on the correlation between producer-consumer production process status, candidate regulation tasks, and peak shaving needs. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for assisted peak shaving based on neural algorithms involving producer-consumer participation, comprising:

[0007] S100: Obtain the producer-consumer type, target peak-shaving period, and production process status records of multiple producer-consumers; match the elastic resource parameters of producer-consumers, calculate the adjustable power range, adjustable duration, and allowed execution status based on the current process stage, continuous running time, and restart preheating status, and generate dynamic adjustment boundary records;

[0008] S200. Based on the dynamic adjustment boundary record, the peak-shaving demand and real-time load are bound with the task identifier to generate a peak-shaving task context; the peak-shaving task context is input into the neural network model to generate candidate adjustment task records with a state to be verified and a standby state.

[0009] S300: Reacquire the production process status record and update the dynamic adjustment boundary record; compare the candidate adjustment task record of the state to be verified with the updated dynamic adjustment boundary record to determine the effective adjustment amount and the verification failure amount; generate the remaining peak shaving gap record according to the peak shaving demand and the effective adjustment amount; when the remaining peak shaving gap is greater than the allowable threshold, filter the candidate adjustment tasks of the standby state and compare them again to generate an executable task set.

[0010] S400: Generate a peak shaving instruction package based on the set of executable tasks and send it to the producer-consumer terminal; determine the actual adjustment amount based on the baseline load and actual load in the peak shaving instruction package, and generate an execution evaluation record and an adjustment capability update record.

[0011] Furthermore, the production process status record includes producer / consumer identifier, production equipment identifier, current process stage, current operating status, continuous running time, cumulative downtime, remaining restart preheating time, number of adjustments executed in the current scheduling cycle, and status acquisition time;

[0012] The status acquisition time of the production process status record re-acquired in S300 is later than the generation time of the candidate adjustment task record.

[0013] Further, in S100, the allowable execution state is matched according to the current process stage, and the adjustable power range, adjustable duration and resumption conditions are calculated according to the continuous running time, the cumulative downtime, the remaining restart preheating time and the existing peak shaving tasks;

[0014] In S300, the adjustable power range, the adjustable duration, the allowed execution state, and the boundary version are updated based on the reacquired current process stage, continuous running time, cumulative downtime, and remaining restart preheating time.

[0015] Furthermore, in S200, the following information is obtained: given load forecast data, peak-shaving price ceiling for ancillary services market, data cutoff time, boundary version, market rule version, and model version.

[0016] Match the peak shaving demand, real-time load, given load forecast data, and the peak shaving price ceiling of the ancillary service market according to the target peak shaving period, and bind the matching result, the data cutoff time, the boundary version, the market rule version, and the model version with the task identifier to generate the peak shaving task context.

[0017] Furthermore, the neural network model generates candidate producers and consumers, candidate adjustment power, candidate adjustment duration, candidate execution period, and candidate priority based on the context of the peak shaving task;

[0018] Candidate adjustment tasks with priority within the first priority range are marked as pending verification, and the remaining candidate adjustment tasks are marked as standby.

[0019] The candidate adjustment task record includes task identifier, producer-consumer identifier, production equipment identifier, candidate adjustment power, candidate adjustment duration, candidate execution period, candidate priority, candidate task status, model version, and boundary version.

[0020] Further, in S300, the candidate adjustment power is compared with the updated adjustable power range, the candidate adjustment duration is compared with the updated adjustable duration, the candidate execution period is matched with the updated allowed execution state and recovery operation conditions, and the boundary version of the candidate adjustment task record is verified to be the updated boundary version.

[0021] Based on the comparison, matching, and version verification results, candidate adjustment tasks are determined to be in a pass, fix, or fail state.

[0022] Furthermore, for candidate regulation tasks that are in the pass state, the candidate regulation power is included in the effective regulation amount;

[0023] For candidate regulation tasks in the correction state, the regulation power corrected by the adjustable power range is included in the effective regulation amount, and the difference between the candidate regulation power and the corrected regulation power is included in the verification failure amount.

[0024] For candidate regulation tasks that are in a failed state, the candidate regulation power is included in the verification failure amount;

[0025] The remaining peak-shaving gap record is generated based on the difference between the peak-shaving demand and the effective adjustment amount;

[0026] Candidate adjustment tasks in the standby state are selected according to candidate priority and the comparison, matching and version verification as described in claim 6 are performed until the remaining peak-shaving gap is not greater than the allowable threshold or the maximum replenishment round is reached.

[0027] Furthermore, the peak shaving instruction package includes a task identifier, a producer-consumer identifier, a production equipment identifier, a target regulation power, an execution start time, an execution end time, a baseline load, an allowable deviation, a boundary version, an instruction version, and a response period;

[0028] If no instruction response is received within the response period, the target regulation power is deducted from the effective regulation amount, the remaining peak shaving gap record is updated, and candidate regulation tasks in standby status are selected and executed in step S300.

[0029] Furthermore, the actual load from the start time to the end time of execution is obtained according to the task identifier;

[0030] The actual adjustment amount is determined based on the difference between the baseline load and the actual load, and the metering period.

[0031] The execution compliance rate is generated based on the actual adjustment amount and the target adjustment amount. The execution compliance rate, the peak-shaving price ceiling of the ancillary service market and the market rule version are bound to the task identifier to generate a subsidy settlement record.

[0032] The adjustment capability update record is generated by updating the continuous running time, cumulative downtime, remaining restart and warm-up time, number of adjustments performed, and remaining adjustable power based on the actual execution start time, actual execution end time, and actual adjustment amount.

[0033] Furthermore, a system for producer-consumer participation in assisted peak shaving based on neural algorithms includes: an information acquisition module, an adjustment boundary generation module, a peak shaving task generation module, an executability verification and remaining peak shaving gap processing module, and an instruction execution and feedback module; the system is used to implement the method described in any of the above embodiments.

[0034] The key innovations of this invention include:

[0035] (1) Combine the elastic resource parameters of the producer-consumer type with the production process status record, and form the adjustable power range, adjustable duration, allowed execution status and boundary version corresponding to the target peak shaving period according to the current process stage, continuous running time and restart preheating status, and generate a dynamic adjustment boundary record for candidate adjustment task generation and subsequent verification.

[0036] (2) Organize the output of the neural network model into candidate adjustment task records with a state to be verified and a standby state; after the candidate adjustment task records are generated, reacquire the production process state records and update the dynamic adjustment boundary records; compare the candidate adjustment task records with the state to be verified with the updated dynamic adjustment boundary records to form effective adjustment amount and verification failure amount; generate the remaining peak shaving gap record based on the effective adjustment amount, and trigger the screening and re-comparison of the standby state candidate adjustment tasks by the remaining peak shaving gap.

[0037] (3) Generate peak shaving instruction packages associated with task identifiers, execution periods and baseline loads from the set of executable tasks. Form actual adjustment amount and execution evaluation record based on the actual load corresponding to the same task identifier. Generate adjustment capacity update record based on actual execution period and actual adjustment amount. Return the target adjustment power corresponding to the instruction that has not been confirmed to the remaining peak shaving gap processing link.

[0038] The following are its main beneficial effects:

[0039] (1) To address the problem that static elastic resource parameters are difficult to reflect the current process stage and recovery conditions of production equipment, dynamic adjustment boundary records are generated by associating production process status records with target peak shaving periods, so that the adjustable power range, adjustable duration and allowed execution status correspond to the current operating status and target peak shaving periods of specific production equipment.

[0040] (2) In response to the problem that the production process status changes after the candidate adjustment task is generated, and the adjustment power gap caused by the failure of the task is not properly connected with the subsequent task selection, the dynamic adjustment boundary is updated before execution, the effective adjustment amount is quantified and the failure amount is verified, and the backup state candidate adjustment task is called according to the remaining peak shaving gap, so that a continuous power allocation relationship is formed between the candidate task verification result, the uncovered peak shaving demand and the supplementary task.

[0041] (3) To address the problem of discontinuous correspondence between peak shaving task generation, instruction issuance, execution measurement and subsequent adjustment capacity records, the peak shaving instruction package, baseline load, actual load and execution evaluation record are associated through task identifier, and the adjustment capacity update record is used for subsequent production process status record, so that the actual execution result enters the dynamic adjustment boundary generation process of the next target peak shaving period. Attached Figure Description

[0042] Figure 1 A flowchart illustrating a method for auxiliary peak shaving based on neural algorithms involving prosumers and consumers, provided in an embodiment of this application;

[0043] Figure 2 This is a structural block diagram of a system for auxiliary peak shaving based on a neural algorithm, provided in an embodiment of this application. Detailed Implementation

[0044] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for assisting peak shaving based on neural algorithms using prosumers, provided in an embodiment of the present invention. The process may include at least steps S100-S400:

[0045] S100: Obtain the producer-consumer type, target peak-shaving period, and production process status records of multiple producer-consumers; match the elastic resource parameters of producer-consumers, calculate the adjustable power range, adjustable duration, and allowed execution status based on the current process stage, continuous running time, and restart preheating status, and generate dynamic adjustment boundary records;

[0046] S200. Based on the dynamic adjustment boundary record, the peak-shaving demand and real-time load are bound with the task identifier to generate a peak-shaving task context; the peak-shaving task context is input into the neural network model to generate candidate adjustment task records with a state to be verified and a standby state.

[0047] S300: Reacquire the production process status record and update the dynamic adjustment boundary record; compare the candidate adjustment task record of the state to be verified with the updated dynamic adjustment boundary record to determine the effective adjustment amount and the verification failure amount; generate the remaining peak shaving gap record according to the peak shaving demand and the effective adjustment amount; when the remaining peak shaving gap is greater than the allowable threshold, filter the candidate adjustment tasks of the standby state and compare them again to generate an executable task set.

[0048] S400: Generate a peak shaving instruction package based on the set of executable tasks and send it to the producer-consumer terminal; determine the actual adjustment amount based on the baseline load and actual load in the peak shaving instruction package, and generate an execution evaluation record and an adjustment capability update record.

[0049] S100: Obtain the producer-consumer type, target peak-shaving period, and production process status records of multiple producer-consumers; match the elastic resource parameters of producer-consumers, calculate the adjustable power range, adjustable duration, and allowed execution status based on the current process stage, continuous running time, and restart preheating status, and generate dynamic adjustment boundary records.

[0050] In this embodiment, S100 is executed by an information acquisition module and a regulation boundary generation module. The information acquisition module is connected to a producer-consumer registration system, a production control system, an energy metering device, and a peak-shaving task database. The producer-consumer registration system provides producer-consumer identifiers, producer-consumer types, production equipment identifiers, and producer-consumer elastic resource parameters. The production control system provides the current process stage, current operating status, continuous operating time, cumulative downtime, and remaining restart preheating time for the production equipment. The energy metering device provides the real-time load of the producers-consumers. The peak-shaving task database provides peak-shaving tasks that have been generated, issued, or are being executed within the current scheduling cycle. The regulation boundary generation module receives the above data and generates dynamic regulation boundary records for the target peak-shaving period issued by the scheduling agency.

[0051] The target peak-shaving period is a time interval set by the dispatching agency for a single peak-shaving demand, including start and end times. In one implementation, the target peak-shaving period is consistent with the trading period used by the power trading center; in another implementation, a trading period is divided into multiple scheduling cycles, each of which forms a target peak-shaving period. The real-time load is the active power of the producer-consumer at the time of status acquisition; for producers-consumers participating in peak shaving by reducing power consumption, the real-time load is the active power absorbed from the grid at their grid connection point; for producers-consumers equipped with distributed power sources or energy storage systems, the real-time load can also form a net active load based on the active power input and output at the grid connection point.

[0052] The production process status record uses producer-consumer identifier and production equipment identifier as associated fields, including current process stage, current operating status, continuous running time, cumulative downtime, remaining restart preheating time, number of adjustments executed in the current scheduling cycle, and status acquisition time. The current process stage indicates the production stage the production equipment is in at the time of status acquisition; the current operating status includes operating status, power reduction status, shutdown status, preheating status, and resumed operating status; the continuous running time indicates the cumulative time the production equipment has maintained operation since its most recent entry into the operating state; the cumulative downtime indicates the cumulative time the production equipment has maintained the shutdown state since its most recent stop; the remaining restart preheating time indicates the preheating time required for the production equipment to recover from its current state to a permissible production state; and the number of adjustments executed indicates the number of peak-shaving tasks the production equipment has accepted and completed within the current scheduling cycle.

[0053] Upon receiving a new peak-shaving request, the information acquisition module reads the producer-consumer identifiers of multiple producers within the target peak-shaving period's affected area; it then reads the corresponding producer-consumer type and production equipment identifier based on the producer-consumer identifier; and finally, based on the production equipment identifier, it requests a production process status record from the production control system. Upon receiving the request, the production control system converts equipment operation signals, production stage signals, and status time records into the aforementioned production process status record. For implementations where the production control system actively uploads status data according to a fixed acquisition cycle, the information acquisition module selects the record from the current status records whose status acquisition time is closest to the current time and has not exceeded the data validity period.

[0054] The adjustment boundary generation module first verifies the producer-consumer identifier, production equipment identifier, and status acquisition time. If the producer-consumer identifier does not exist in the producer-consumer registration system, no dynamic adjustment boundary record is generated for that identifier. If the binding relationship between the production equipment identifier and the producer-consumer identifier does not exist, an equipment binding exception record is generated. If the time difference between the status acquisition time and the current time exceeds the data validity period, the corresponding production process status record is marked as data timeout, and the boundary validity status is set to invalid. For cases where multiple production process status records exist for the same production equipment, the record with the later acquisition time and complete fields is selected; earlier records are retained in the status record but do not participate in this round of dynamic adjustment boundary calculation.

[0055] The producer-consumer elastic resource parameters are stored in the data and model storage module according to producer-consumer type and production equipment type. These parameters include adjustable power lower limit, adjustable power upper limit, maximum adjustable duration, minimum continuous running time, maximum allowable downtime, restart warm-up time, recovery running time, and maximum number of adjustments. Different producer-consumer types correspond to different parameter records; when multiple production equipment types are configured for the same producer-consumer type, the equipment parameter records are matched according to the production equipment identifier; when multiple versions of the same equipment parameter exist, the parameter version that takes effect at the start time of the target peak-shaving period is matched.

[0056] The adjustment boundary generation module matches the allowed execution status based on the current process stage. Specifically, the correspondence between process stages and allowed execution statuses is set in the producer-consumer elastic resource parameters; when the current process stage corresponds to a state that allows power reduction, shutdown, or increased output, the allowed execution status of the production equipment is set to valid; when the current process stage corresponds to a stage that prohibits interruption, shutdown, or resumption of operation, the allowed execution status is set to invalid. For electric heating producers-consumers, the municipal heating network status can also be recorded in the production process status record; when the municipal heating network status meets the preset switching conditions, the allowed execution status corresponds to allowing a reduction in electric heating load; when the municipal heating network status does not meet the switching conditions, the allowed execution status corresponds to invalid.

[0057] The adjustment boundary generation module compares the continuous runtime with the minimum continuous runtime. When the continuous runtime is less than the minimum continuous runtime, the allowed execution status within the current target peak shaving period is set to invalid. When the continuous runtime is not less than the minimum continuous runtime, the adjustable duration is calculated based on the target peak shaving period, the maximum allowable downtime, and the recovery runtime. For production equipment that is allowed to stop, the adjustable duration is the smaller of the target peak shaving period length and the maximum allowable downtime. For production equipment that is only allowed to operate at reduced power, the adjustable duration is the smaller of the target peak shaving period length and the remaining time of the process stage. For production equipment in the preheating state, no allowed execution period is formed until the remaining preheating time ends.

[0058] When production equipment is shut down, the adjustment boundary generation module compares the cumulative downtime with the maximum allowable downtime. If the cumulative downtime reaches the maximum allowable downtime, the production equipment is no longer considered an adjustable resource for further load reduction; if the cumulative downtime does not reach the maximum allowable downtime, the adjustable duration is calculated based on the difference between the maximum allowable downtime and the cumulative downtime. When production equipment resumes operation, the resumption end time is calculated based on the resumption time; if the target peak-shaving period overlaps with the resumption time, the overlapping time interval is excluded from the allowable execution time.

[0059] The adjustment boundary generation module also reads existing peak-shaving tasks within the current scheduling cycle. These existing tasks include tasks awaiting execution, tasks already assigned, and tasks in execution. For existing peak-shaving tasks overlapping with the target peak-shaving period, the adjustment boundary generation module deducts the upper limit of adjustable power based on the target adjustment power and the remaining adjustable duration based on the execution duration. When the number of adjustments performed in the current scheduling cycle reaches the maximum number of adjustments, the allowed execution status is set to invalid. For existing peak-shaving tasks on the same production equipment with non-overlapping times, the next allowed execution time is calculated based on the end time and recovery time of the previous task. If the target peak-shaving period is earlier than the next allowed execution time, no valid boundary is formed for that target peak-shaving period.

[0060] The adjustable power range is calculated based on real-time load, producer-consumer elastic resource parameters, and existing peak-shaving tasks. For producer-consumers that reduce load, the upper limit of adjustable power is no greater than the difference between the real-time load and the reserved operating power of the production equipment; the lower limit of adjustable power is set according to the minimum adjustment power that the production equipment can stably execute. For producer-consumers that increase output, the upper limit of adjustable power is calculated based on the remaining output of distributed power sources or the remaining discharge power of energy storage systems. For producer-consumers that have both interruptible loads and distributed power sources, the load adjustment power and elastic resource output are calculated separately and summarized according to the target peak-shaving direction to form the adjustable power range of the producer-consumer.

[0061] The dynamic adjustment boundary record uses producer-consumer identifier, production equipment identifier, and target peak-shaving period as associated fields, including adjustable power lower limit, adjustable power upper limit, adjustable duration, allowed execution status, resumption conditions, boundary validity period, and boundary version. The boundary validity period starts from the status acquisition time and terminates when a new production process status record is generated, the target peak-shaving period ends, or the data validity period expires. The boundary version is generated based on the producer-consumer identifier, production equipment identifier, status acquisition time, and producer-consumer elastic resource parameter version; when the production process status of the same production equipment changes, a new boundary version is generated, and the original boundary version is retained and marked as a historical version.

[0062] In one implementation, the producer-consumer elasticity resource parameter record for silicon carbide smelting producers sets the maximum single downtime to 4 hours and the minimum continuous operating time after resumption to 6.5 hours. When the production equipment has been running continuously for 8 hours and the current process stage allows for downtime, the adjustment boundary generation module calculates the adjustable duration based on the target peak shaving period length and the 4-hour downtime limit; when the production equipment has only been running continuously for 3 hours after the previous peak shaving task, the allowed execution status is set to invalid. This parameter only applies to production equipment that matches its equipment type and parameter version.

[0063] In another implementation, the adjustable power range of the electric calcined coal production equipment is set to 500–650 kW, and the restart preheating time is set to 1 hour. When the target peak-shaving period overlaps with the preheating state, the allowed execution state in the dynamic adjustment boundary record is invalid; after preheating, the adjustable power range is recalculated based on the real-time load and existing tasks. For calcium carbide production equipment, the adjustable power range is recorded according to the corresponding parameter version; the restart preheating time is matched based on the cumulative duration of this shutdown, and the adjustment boundary generation module writes the matched restart preheating time into the recovery operation conditions.

[0064] The dynamic adjustment boundary record generated in S100 is transmitted to the peak shaving task context generation module and the executability verification module. The peak shaving task context generation module uses the producer-consumer identifier, production equipment identifier, target peak shaving period, adjustable power range, adjustable duration, allowed execution status, and boundary version to form the candidate producer-consumer input for S200. The executability verification module retains the dynamic adjustment boundary record for comparison by S300 on the candidate adjustment task record. Boundary invalid states and equipment binding anomaly records are synchronously written to the status record, and S100 is re-executed after subsequent status re-collection.

[0065] S200. Based on the dynamic adjustment boundary record, the peak-shaving demand and real-time load are bound to the task identifier to generate a peak-shaving task context; the peak-shaving task context is input into the neural network model to generate candidate adjustment task records with a state to be verified and a standby state.

[0066] S200 is executed by the peak shaving task context generation module and the prediction module. The peak shaving task context generation module receives the dynamic adjustment boundary record generated in S100, and receives peak shaving demand from the dispatching agency, real-time load from the electricity metering device, given load prediction data from the dispatching prediction system, the peak shaving price ceiling and market rule version from the peak shaving ancillary service market, and reads the neural network model and model version from the data and model storage module. S200 starts after the dynamic adjustment boundary record is generated; when the dispatching agency updates the peak shaving demand, the market rule version changes, or the target peak shaving period changes, the peak shaving task context is regenerated.

[0067] The peak-shaving demand includes the peak-shaving area, target peak-shaving period, peak-shaving direction, and target regulating power. Peak-shaving direction includes reducing power consumption, increasing power consumption, increasing producer output, and reducing producer output; this implementation method uniformly converts the peak-shaving direction into the regulating power that producers need to provide during the target peak-shaving period. The given load forecast data includes the forecasted load of each producer during the target peak-shaving period and the forecasted total load of the peak-shaving area; the ancillary service market peak-shaving price ceiling includes the applicable trading period, peak-shaving direction, price ceiling, and market rule version; the data cutoff time indicates the deadline for generating and receiving input data for this round of peak-shaving tasks.

[0068] The peak-shaving task context generation module first matches peak-shaving demand, real-time load, given load forecast data, and the peak-shaving price ceiling of the ancillary service market according to the target peak-shaving period. If the real-time load collection time is later than the data cutoff time, the real-time load is not written into the current peak-shaving task context; if the real-time load collection time is earlier than the data cutoff time and has not exceeded the data validity period, it is used as the current load input. When the given load forecast data contains multiple forecast time intervals, the forecast value that overlaps with the target peak-shaving period is selected; if the forecast time granularity is smaller than the target peak-shaving period, the forecast load within the target peak-shaving period is summarized in chronological order; if the forecast time granularity is larger than the target peak-shaving period, the corresponding forecast load is converted according to the duration occupied by the target peak-shaving period in the forecast time interval.

[0069] The peak-shaving task context generation module filters records with valid boundary states from the dynamic adjustment boundary records to form a candidate producer-consumer set. If a single producer-consumer includes multiple production devices, each production device is designated as a candidate. In one implementation, the adjustable power ranges of multiple production devices can be aggregated according to the producer-consumer identifier and then allocated to the production devices after the candidate adjustment task records are generated. Records with invalid boundary states, boundary validity periods earlier than the start time of the target peak-shaving period, or adjustable power upper limits not exceeding adjustable power lower limits are not included in the candidate producer-consumer set.

[0070] The task identifier consists of a peak-shaving area, a target peak-shaving period, a peak-shaving direction, and a task sequence number. In one implementation, the task identifier is also associated with a system scheduling instruction identifier issued by the scheduling agency. The peak-shaving task context generation module binds the task identifier, peak-shaving demand, real-time load, given load forecast data, ancillary service market peak-shaving price ceiling, data cutoff time, dynamic adjustment boundary version, market rule version, and model version to generate a peak-shaving task context. Each candidate prosumer in the peak-shaving task context is also associated with a prosumer identifier, production equipment identifier, adjustable power range, adjustable duration, and allowed execution status.

[0071] The neural network model employs a multilayer perceptron architecture. Model inputs include peak-shaving demand, target peak-shaving period, real-time producer and consumer load, given load forecast data, adjustable power lower limit, adjustable power upper limit, adjustable duration, number of adjustments already executed in the current scheduling cycle, ancillary service market peak-shaving price upper limit, and historical execution deviation. Model outputs include candidate adjustment power, candidate adjustment duration, candidate execution period, and candidate priority. The neural network model does not directly generate peak-shaving instruction packets; instead, the model output first forms candidate adjustment task records, which are then used by S300 to call the updated dynamic adjustment boundary records for executability verification.

[0072] In one implementation, the training samples for the neural network model consist of historical peak-shaving demands, historical loads, historical dynamic adjustment boundary records, and historical execution evaluation records. Each training sample is organized according to the target peak-shaving period and associated with the actual adjustment power and execution period. After training, the model parameters, model input field versions, and model output field versions are written to the model version record. When the prediction module performs inference, it first verifies the model version in the peak-shaving task context against the model input field version; if the versions are inconsistent, the model is not invoked, and a model version anomaly record is generated.

[0073] The prediction module generates input data from the candidate prosumer set according to the model input order; it converts the power, duration, and price fields according to the dimensional conversion parameters recorded in the data and model storage modules; candidate objects lacking real-time load, dynamic adjustment boundary records, or production equipment identifiers are not included in the model input. After the model completes the calculation, the prediction module associates the output candidate adjustment power with the prosumer identifier, production equipment identifier, and task identifier; it converts the candidate adjustment duration into candidate execution periods; and it forms a candidate order based on candidate priority.

[0074] The candidate adjustment task record includes task identifier, producer / consumer identifier, production equipment identifier, candidate adjustment power, candidate adjustment duration, candidate execution period, candidate priority, candidate task status, model version, and boundary version. Candidate adjustment power represents the adjustment power generated by the model for that production equipment; candidate adjustment duration represents the duration for which the adjustment power is maintained; candidate execution period consists of candidate start time and candidate end time; candidate priority indicates the order in which multiple candidate adjustment tasks enter executability verification or backup screening.

[0075] The prediction module sets a first priority range based on candidate priorities. In one implementation, the first priority range is determined by a preset number of tasks; candidate adjustment tasks with priority within the aforementioned task number range are marked as pending verification, and the remaining candidate adjustment tasks are marked as standby. In another implementation, the prediction module aggregates candidate adjustment power sequentially according to candidate priority. When the accumulated candidate adjustment power reaches the power corresponding to the peak-shaving requirement, the aggregated candidate adjustment tasks are marked as pending verification, and the unaggregated candidate adjustment tasks are marked as standby. The pending verification and standby states only indicate the processing order of candidate adjustment tasks in S300, and do not indicate that they have met the updated dynamic adjustment boundary record.

[0076] When the candidate adjustment power output by the model is null, the candidate adjustment duration is null, or the candidate execution period does not overlap with the target peak-shaving period, no corresponding candidate adjustment task record is generated, and it is written to the model output anomaly record. When the candidate adjustment power is negative and the peak-shaving direction requires an increase in positive adjustment power, the output record is written to the model output anomaly record. When the candidate adjustment power output by the model exceeds the numerical range configured by the model, a candidate adjustment task record in a pending verification state can still be formed, which is compared by S300 according to the updated dynamic adjustment boundary record; the original output value of the model is retained in the candidate adjustment task record.

[0077] In one implementation, the target peak-shaving period is from 18:00 to 19:00, and the peak-shaving demand is 100MW. The effective dynamic adjustment boundary record generated in S100 corresponds to 20 production devices. The prediction module inputs the real-time load, adjustable power range, adjustable duration, and historical execution deviation of the 20 production devices into the neural network model to obtain candidate adjustment task records arranged by candidate priority. The prediction module summarizes the candidate adjustment power according to the candidate priority. When the cumulative value reaches 105MW, the candidate adjustment tasks participating in the accumulation are marked as pending verification, and the remaining candidate adjustment tasks are marked as standby. The 105MW is only the cumulative value of candidate adjustment power output by the model, and whether it is included in the effective adjustment amount is determined by the executability verification result of S300.

[0078] The candidate adjustment task records, peak shaving task contexts, and model output anomaly records generated by S200 are passed to the executability verification module. The task identifier, producer / consumer identifier, production equipment identifier, candidate adjustment power, candidate adjustment duration, candidate execution period, candidate priority, candidate task status, model version, and boundary version in the candidate adjustment task records serve as input to S300; the peak shaving demand in the peak shaving task context serves as input to S300 for calculating the remaining peak shaving gap. The dynamic adjustment boundary records generated by S100 are stored in the data and model storage module and updated after S300 re-acquires the production process status records.

[0079] S300: Reacquire the production process status record and update the dynamic adjustment boundary record; compare the candidate adjustment task record of the state to be verified with the updated dynamic adjustment boundary record to determine the effective adjustment amount and the verification failure amount; generate the remaining peak shaving gap record according to the peak shaving demand and the effective adjustment amount; when the remaining peak shaving gap is greater than the allowable threshold, filter the candidate adjustment tasks of the standby state and compare them again to generate an executable task set.

[0080] S300 is jointly executed by an information acquisition module, a regulation boundary generation module, an executability verification module, and a remaining peak-shaving gap processing module. S300 receives the candidate regulation task record and peak-shaving task context generated by S200, and then retrieves the production process status record of the production equipment corresponding to the candidate regulation task to be verified from the production control system. The status acquisition time of the re-acquired production process status record is later than the generation time of the candidate regulation task record; the re-acquisition action is initiated after the candidate regulation task record is generated, or it can be initiated at a preset time point before the planned issuance of the peak-shaving instruction package.

[0081] When re-acquiring the production process status record, the information acquisition module sends a status request to the production control system based on the producer / consumer identifier and production equipment identifier in the candidate adjustment task record. The production control system returns the current process stage, current operating status, continuous running time, cumulative downtime, remaining restart / preheating time, number of adjustments executed, and status acquisition time. For production equipment that does not return a status request, the information acquisition module resends the request according to the status request interval. If no response is received after reaching the configured number of requests, the corresponding candidate adjustment task is marked as a status acquisition failure and handed over to the executability verification module to determine it as an invalid state.

[0082] The adjustment boundary generation module updates the dynamic adjustment boundary record using the newly acquired production process status record, following the parameter matching and calculation method in S100. The updated content includes the adjustable power range, adjustable duration, allowed execution status, resumption conditions, boundary validity period, and boundary version. After the candidate adjustment task record is generated, when the production equipment enters a new process stage, begins shutdown, begins preheating, completes a previous peak-shaving task, or accepts a new peak-shaving task, the updated dynamic adjustment boundary record will differ from the boundary version bound in S200.

[0083] The executability verification module first verifies the task identifier, producer identifier, and production equipment identifier in the candidate adjustment task record; then it verifies the boundary version in the candidate adjustment task record against the updated boundary version. If the boundary versions match, the updated boundary fields are used for comparison; if the boundary versions do not match, the candidate adjustment task is not directly determined to be in a failed state. Instead, the comparison is re-performed based on the updated adjustable power range, adjustable duration, allowed execution status, and recovery conditions, and the version change status is written into the constraint verification record.

[0084] Specifically, the executability verification module compares the candidate adjustable power with the updated adjustable power range. If the candidate adjustable power is between the lower and upper limits of the adjustable power, the power verification status is passed; if the candidate adjustable power is greater than the upper limit, the power verification status is pending correction, and the upper limit is used as the corrected adjustable power; if the candidate adjustable power is less than the lower limit, the power verification status is failed. For production equipment with an updated invalid execution status, power correction is not performed, and the corresponding candidate adjustment task enters an invalid state.

[0085] The executability verification module compares the candidate adjustment duration with the updated adjustable duration. If the candidate adjustment duration is not greater than the updated adjustable duration, the duration verification status is passed; if the candidate adjustment duration is greater than the updated adjustable duration and the updated adjustable duration is greater than 0, the candidate adjustment duration is corrected to the updated adjustable duration; if the updated adjustable duration is 0, the corresponding candidate adjustment task enters a failed state. For implementations that calculate the effective adjustment amount according to power metrics, the corrected adjustment duration is still written into the executable task set for S400 to generate the execution start time and execution end time; for implementations that calculate the effective adjustment amount according to power metrics, the corrected adjustment duration participates in the effective adjustment amount calculation.

[0086] The executability verification module also matches candidate execution periods with the updated allowed execution status and recovery conditions. When all candidate execution periods are within the allowed execution time interval, and the time from the end of a candidate execution period to the start of the next existing task is not less than the recovery time, the execution period verification status is passed. When a candidate execution period partially overlaps with the recovery time, the candidate execution start time is adjusted to the recovery end time, and the candidate adjustment duration is recalculated. When the adjusted candidate adjustment duration is less than the lower limit of the execution duration corresponding to the adjustable power, the corresponding candidate adjustment task enters the failure state.

[0087] For the same production equipment where tasks have been issued or are in progress, the executability verification module compares the execution time periods and target adjustment power of the two tasks. If the execution time periods overlap and the sum of the target adjustment powers of the two tasks exceeds the updated adjustable power limit, the later-generated candidate adjustment task is determined to be in a failed state. If the execution time periods partially overlap but the non-overlapping time still meets the adjustable duration, the candidate execution time period is adjusted to the non-overlapping time interval. If the number of adjustments performed in the current scheduling cycle of the same production equipment has reached the maximum number of adjustments, the candidate adjustment task enters a failed state.

[0088] The constraint verification record uses task identifier and production equipment identifier as associated fields, including verification time, candidate adjustable power, updated adjustable power range, candidate adjustment duration, updated adjustable duration, candidate execution period, allowed execution status, resumption conditions, candidate task boundary version, updated boundary version, power verification status, duration verification status, execution period verification status, version change status, value before correction, value after correction, and task verification status. Task verification status includes pass status, correction status, and failure status.

[0089] When the candidate adjustment power, candidate adjustment duration, and candidate execution period all meet the updated dynamic adjustment boundary record, the candidate adjustment task is determined to be in the passed state; when the candidate adjustment power, candidate adjustment duration, or candidate execution period meets the updated dynamic adjustment boundary record after correction, the candidate adjustment task is determined to be in the corrected state; when the execution status is invalid, the updated adjustable duration is 0, the candidate adjustment power is lower than the lower limit of adjustable power, the conditions for resuming operation are not met, the status acquisition fails, or the task conflict cannot be eliminated, the candidate adjustment task is determined to be in the failed state.

[0090] Effective regulation and verification failure are calculated using a unified standard. In the implementation using power as the standard, for candidate regulation tasks in a pass state, the candidate regulation power is included in the effective regulation; for candidate regulation tasks in a corrected state, the corrected regulation power is included in the effective regulation, and the difference between the candidate regulation power and the corrected regulation power is included in the verification failure; for candidate regulation tasks in a failed state, the candidate regulation power is included in the verification failure. In the implementation using electrical quantity as the standard, the product of the corresponding regulation power and the regulation duration is included in either the effective regulation or verification failure.

[0091] The technical characteristic of the verification failure quantity is represented by the regulation power or regulation amount that is not included in the effective regulation quantity because the candidate regulation task fails to pass the verification of the updated dynamic regulation boundary record. The verification failure quantity and the model prediction error are recorded separately; the model prediction error is formed by the difference between the model output and the historical execution results, while the verification failure quantity is formed by changes in the production process status, version changes, task conflicts, or boundary corrections after the candidate regulation task is generated. The failure reason corresponding to the verification failure quantity is written in the constraint verification record.

[0092] The remaining peak-shaving gap processing module summarizes the effective regulation amounts corresponding to candidate regulation tasks based on their current and corrected states; it then writes the difference between peak-shaving demand and effective regulation amounts into the remaining peak-shaving gap record. When using power metrics, the remaining peak-shaving gap represents the difference between the target peak-shaving power and the effective regulation power; when using electricity metrics, the remaining peak-shaving gap represents the difference between the target peak-shaving electricity and the effective regulation electricity. Peak-shaving demand and effective regulation amounts use the same peak-shaving direction and the same metering caliber.

[0093] The remaining peak-shaving gap record includes task identifier, target peak-shaving period, peak-shaving demand, effective adjustment amount, verification failure amount, remaining peak-shaving gap, allowable threshold, replenishment round, and gap status. When the remaining peak-shaving gap is not greater than the allowable threshold, the gap status is written as completed; when the remaining peak-shaving gap is greater than the allowable threshold, the gap status is written as pending replenishment, and the candidate adjustment task selection for the backup status is triggered.

[0094] The remaining peak-shaving gap processing module selects supplementary tasks from the candidate adjustment task records in the standby state according to candidate priority. In one implementation, one standby state candidate adjustment task is selected in each round; in another implementation, multiple standby state candidate adjustment tasks are selected sequentially according to candidate priority until the sum of the selected candidate adjustment powers is not less than the current remaining peak-shaving gap. The selected standby state candidate adjustment task is changed to a pending verification state, and the production process status record of its corresponding production equipment is re-acquired.

[0095] The supplementary task undergoes executability verification again following the aforementioned boundary update, version verification, power comparison, duration comparison, execution time period matching, and task conflict verification process. When a supplementary task reaches a pass or corrected state, its effective regulation amount is added to the original effective regulation amount, and the remaining peak-shaving gap record is updated; when a supplementary task reaches a fail state, its candidate regulation power is included in the failure verification amount, and the next backup candidate regulation task is selected. Each round of backup task screening and verification increases the supplementary round by 1.

[0096] The replenishment process terminates when the remaining peak-shaving gap is no greater than the allowable threshold, the standby candidate adjustment tasks are empty, or the maximum replenishment rounds are reached. When the maximum replenishment rounds are reached and the remaining peak-shaving gap is still greater than the allowable threshold, an insufficient adjustment status is written into the remaining peak-shaving gap record, and the remaining peak-shaving gap is sent to the scheduling agency. When the scheduling agency reissues the peak-shaving demand, S100 is started with a new task identifier; if the peak-shaving demand is not reissued, the insufficient adjustment portion is not written into the executable task set.

[0097] In one implementation, the peak-shaving demand is 100MW, and the candidate adjustment power corresponding to the candidate adjustment task in the verification state generated in S200 is 105MW. After S300 re-acquires the production process status record, a 20MW candidate adjustment task enters a failed state due to a change in the current process stage; 5MW of the remaining candidate adjustment tasks is corrected to 3MW due to a change in the adjustable power limit. The effective adjustment amount in the first round is 83MW, the verification failure amount is 22MW, and the remaining peak-shaving gap is 17MW. The remaining peak-shaving gap processing module selects two supplementary tasks with candidate adjustment powers of 12MW and 8MW from the standby candidate adjustment tasks; after re-verification, the 12MW task passes, the 8MW task is corrected to 5MW, the updated effective adjustment amount is 100MW, and the remaining peak-shaving gap is 0. The supplementary tasks, the first-round passed tasks, and the corrected tasks together form an executable task set.

[0098] The executable task set consists of candidate regulation tasks in the pass and correction states, including task identifier, producer / consumer identifier, production equipment identifier, target regulation power, execution start time, execution end time, boundary version, and task verification status. The target regulation power and execution period of the candidate regulation tasks in the correction state use corrected values. The executable task set is passed to the instruction generation and distribution module as input for the S400 to generate peak shaving instruction packages.

[0099] S300 also receives an unacknowledged instruction status returned by S400. If the consumer terminal fails to return an instruction acknowledgement within the acknowledgement period, S400 deducts the corresponding target regulation power from the effective regulation amount and updates the remaining peak-shaving gap record; the remaining peak-shaving gap processing module selects a new supplementary task according to the backup task screening and re-comparison process in this step. Thus, the constraint verification record, the remaining peak-shaving gap record, and the instruction acknowledgement record jointly participate in the update of the executable task set.

[0100] S400: Generate a peak shaving instruction package based on the set of executable tasks and send it to the producer-consumer terminal; determine the actual adjustment amount based on the baseline load and actual load in the peak shaving instruction package, and generate an execution evaluation record and an adjustment capability update record.

[0101] S400 is executed by an instruction generation and distribution module, a producer-consumer terminal, an energy metering device, and an execution and feedback module. The instruction generation and distribution module receives the set of executable tasks generated by S300, reads the baseline load of the corresponding producer-consumer during the execution period from the baseline load record, reads the task identifier and market rule version from the peak-shaving task context, and reads the ancillary service market peak-shaving price ceiling from the market parameter record. For each production device in the set of executable tasks, a peak-shaving instruction package is generated separately.

[0102] The baseline load represents the load reference value for the corresponding metering period when the current peak shaving command is not executed. The baseline load record includes task identifier, producer / consumer identifier, production equipment identifier, metering period, baseline power value, baseline generation time, and baseline version. In one embodiment, the baseline load is formed based on the average load of multiple similar dates within the same metering period prior to the execution date; in another embodiment, the baseline load is formed based on the actual load of multiple consecutive metering periods prior to the start of peak shaving execution; in yet another embodiment, the baseline load uses given load forecast data matched with the production plan in the scheduling and forecasting system. The same baseline version is used within a single peak shaving task.

[0103] The peak-shaving instruction package includes a task identifier, producer / consumer identifier, production equipment identifier, target regulation power, execution start time, execution end time, baseline load, allowable deviation, boundary version, instruction version, and response period. The allowable deviation represents the permissible power difference between the actual regulation power and the target regulation power. The instruction version is generated based on the task identifier and instruction generation time; a new instruction version is generated when the target regulation power or execution period changes for the same task. The peak-shaving instruction package may also include peak-shaving direction and market rule version; these fields, along with the task identifier, are written into the instruction record.

[0104] The instruction generation and distribution module sends peak-shaving instruction packets according to the communication address of the prosumer terminal. After receiving the peak-shaving instruction packet, the prosumer terminal verifies the task identifier, prosumer identifier, production equipment identifier, and instruction version. If the verification is successful, it returns an instruction receipt indicating acceptance. If the production equipment is in a locally prohibited control state, the production equipment identifier does not match, or the instruction version is earlier than the received version, it returns an instruction receipt indicating rejection. The instruction receipt record includes the task identifier, instruction version, receipt status, and receipt time.

[0105] If no instruction acknowledgment is received within the acknowledgment period, the instruction generation and issuance module marks the peak-shaving instruction package as unacknowledged; deducts the corresponding target regulation power from the effective regulation amount generated in S300; updates the remaining peak-shaving gap record according to the deducted effective regulation amount, and returns the updated remaining peak-shaving gap record to the remaining peak-shaving gap processing module. When the remaining peak-shaving gap is greater than the allowable threshold, the candidate regulation tasks for the standby state are screened according to the process recorded in S300 and compared again; after the new supplementary task passes the verification, a new peak-shaving instruction package is generated.

[0106] For producer-consumer terminals receiving peak-shaving command packets, control commands are sent to the producer-consumer power equipment according to the execution start time. Control commands for load-reducing producer-consumers include reducing operating power, stopping interruptible loads, or switching power supply modes; control commands for output-increasing producer-consumers include increasing the output of distributed power sources or increasing the discharge power of energy storage. The producer-consumer terminal records the actual execution start time, the actual execution end time, and the equipment execution status, and returns the equipment execution status to the execution and feedback module.

[0107] The electricity metering device collects the actual load within the execution period according to the metering cycle. The actual load record includes task identifier, producer / consumer identifier, metering time, actual power value, and metering data quality status. The metering data quality status is determined based on whether the metering value is missing, whether the metering time is continuous, and the status of the metering equipment. When metering data is missing, the execution and feedback module initiates a re-collection request. Metering periods that are still missing after re-collection are marked as abnormal metering data and are not included in the effective execution time.

[0108] For load-reducing producers and consumers, the execution and feedback module subtracts the actual load from the baseline load at the same metering time to obtain the actual regulated power. For load-increasing producers and consumers, the actual load is subtracted from the baseline load to obtain the actual regulated power. For tasks that increase producer and consumer output, the increase in actual output power relative to the baseline output power is used as the actual regulated power. When the direction of the actual regulated power is inconsistent with the peak-shaving direction in the peak-shaving instruction packet, the actual regulated power for the corresponding metering period is recorded as 0, and the inconsistency status is written in the execution evaluation record.

[0109] The actual regulation amount is calculated based on the actual regulation power and the metering period. The execution and feedback module multiplies the actual regulation power of each metering period by the length of that metering period to obtain the actual regulation power for that metering period; then it summarizes the actual regulation power with normal metering data quality from the start time to the end time of execution to form the actual regulation amount. For implementation methods that use power metrics for execution evaluation, the execution and feedback module calculates the average actual regulation power within the execution period and writes it into the execution evaluation record.

[0110] The execution evaluation record uses task identifier and production equipment identifier as associated fields, including target adjustment power, target adjustment amount, actual adjustment power, actual adjustment amount, effective execution duration, execution compliance rate, execution status, measurement data quality status, actual execution start time, and actual execution end time. The execution compliance rate is generated as the ratio of the actual adjustment amount to the target adjustment amount; when the actual adjustment amount is greater than the target adjustment amount, the execution compliance rate is written according to the upper limit value recorded in the market rule version. In another implementation, the execution compliance rate is generated as the ratio of the effective execution duration during which the actual adjustment power is within the allowable deviation range of the target adjustment power to the target execution duration.

[0111] The execution and feedback module binds the execution compliance rate, actual adjustment volume, peak-shaving price ceiling for ancillary services, and market rule version to the task identifier to generate subsidy settlement records. These records include the task identifier, producer / consumer identifier, production equipment identifier, actual adjustment volume, effective execution duration, execution compliance rate, peak-shaving price ceiling for ancillary services, market rule version, subsidy calculation parameters, and settlement status. The subsidy calculation parameters are read from the corresponding peak-shaving direction and execution compliance rate range in the market rule version. The subsidy settlement records are sent to the external transaction settlement system, which registers the settlement data according to the aforementioned market rule version.

[0112] When the actual load record does not match the task identifier of the peak-shaving instruction package, no corresponding subsidy settlement record is generated, and the task matching anomaly status is written. When the instruction version changes, only the execution period and target regulation power corresponding to the currently valid instruction version are used to calculate the execution evaluation. If a subsidy settlement record with the same task identifier and instruction version already exists and its settlement status is already generated, a new subsidy settlement record will not be generated again. When the metering data anomaly time exceeds the configured duration, the settlement status of the subsidy settlement record is written as pending review.

[0113] The execution and feedback module updates the production process status record based on the actual execution start time, actual execution end time, and actual adjustment amount. When production equipment remains running but reduces power during peak shaving, the actual execution time is included in the power reduction runtime, and the continuous runtime is updated based on equipment parameter records. When production equipment stops running during peak shaving, the actual execution time is included in the cumulative downtime, and the remaining restart preheating time is generated based on the actual execution end time and restart preheating time. After the production equipment resumes operation, the continuous runtime is recalculated. The number of adjustments performed in the current scheduling cycle is incremented by 1 after the task is completed.

[0114] The adjustment capability update record includes task identifier, producer-consumer identifier, production equipment identifier, updated current operating status, updated continuous operating duration, updated cumulative downtime, updated remaining restart / warm-up time, updated number of adjustments performed, updated remaining adjustable power, updated remaining adjustable time, and status update time. The remaining adjustable power is calculated based on the actual adjustment amount, producer-consumer elastic resource parameters, and current production equipment status; the remaining adjustable time is calculated based on the maximum adjustable time, actual execution time, and recovery conditions.

[0115] In one implementation, a peak-shaving instruction package for an electric calcined coal production equipment records a target regulating power of 600kW, an execution time of 18:00 to 18:30, a baseline load of 2400kW, and an allowable deviation of 60kW. During execution, the electricity metering device collects the actual load at 5-minute metering cycles; the execution and feedback module calculates the actual regulating power based on the difference between the baseline load and the actual load for each metering period, and summarizes them to form the actual regulating amount. The actual execution ends at 18:30, and the production equipment subsequently enters the preheating state; the execution and feedback module updates the remaining preheating time to 1 hour based on the parameter records, and forms a regulating capacity update record.

[0116] In another implementation, a candidate regulation task has generated a peak-shaving instruction package, but the producer-consumer terminal has not returned an instruction receipt within the receipt period. The instruction generation and issuance module marks the task as an unconfirmed instruction, deducts the corresponding target regulation power from the effective regulation amount, and updates the remaining peak-shaving gap record; the remaining peak-shaving gap processing module selects a backup candidate regulation task, re-executes the production process status acquisition and executability verification; after the supplementary task passes the verification, a new peak-shaving instruction package is generated.

[0117] The S400 ultimately generates execution evaluation records, subsidy settlement records, and adjustment capacity update records. The execution evaluation records are stored as execution data for this round of peak-shaving tasks; the subsidy settlement records are sent to the transaction settlement system; and the adjustment capacity update records are written back to the production process status records and used as input to the S100 when the next target peak-shaving period arrives or new peak-shaving demands are issued. In the next round, the S100 regenerates dynamic adjustment boundary records based on the updated continuous operating time, cumulative downtime, remaining restart preheating time, number of adjustments already performed, and remaining adjustable power.

[0118] Example 2: Figure 2 This diagram illustrates a structural block diagram of a system based on a neural algorithm, involving producer-consumer participation in auxiliary peak shaving, according to an embodiment of the present invention. Figure 2 As shown, the structure may include:

[0119] Information acquisition module 01 is used to acquire the producer-consumer type, target peak-shaving period, production process status record, peak-shaving demand, real-time load, baseline load, and actual load of multiple producer-consumers. Specifically, the information acquisition module receives the producer-consumer type, production process status record, real-time load, and actual load provided by multiple producer-consumer terminals, and receives the target peak-shaving period, peak-shaving demand, and baseline load corresponding to a peak-shaving task. The production process status record includes producer-consumer identifier, production equipment identifier, current process stage, continuous running time, restart preheating status, and status acquisition time. The real-time load and the actual load are associated with the producer-consumer identifier, production equipment identifier, and acquisition time, respectively. The baseline load is associated with the producer-consumer identifier, target peak-shaving period, and metering period. The information acquisition module matches input objects according to producer-consumer identifiers and production equipment identifiers, and performs time matching of peak-shaving demand, real-time load, and baseline load according to the target peak-shaving period. When the production process status record lacks the current process stage, continuous running time, or restart preheating status, the corresponding production process status record is marked as invalid. When the acquisition time of the real-time load does not fall within the data acquisition range corresponding to the target peak-shaving period, the real-time load is not transmitted to the peak-shaving task generation module. The information acquisition module provides the producer-consumer type and the production process status record to the adjustment boundary generation module, the target peak-shaving period, the peak-shaving demand, and the real-time load to the peak-shaving task generation module, and the baseline load and the actual load to the instruction execution and feedback module. The information acquisition module also receives the adjustment capacity update record returned by the instruction execution and feedback module and writes the adjustment capacity update record into the production process status record subsequently acquired by the corresponding producer-consumer.

[0120] The adjustment boundary generation module 02, connected to the information acquisition module, is used to match the producer-consumer elastic resource parameters corresponding to the producer-consumer type. It calculates the adjustable power range, adjustable duration, and allowed execution status based on the current process stage, continuous running time, and restart preheating status in the production process status record, generating a dynamic adjustment boundary record. It is also used to update the dynamic adjustment boundary record based on the production process status record re-acquired after the candidate adjustment task record is generated. Specifically, the adjustment boundary generation module receives the producer-consumer type and production process status record provided by the information acquisition module, and matches the producer-consumer elastic resource parameters according to the producer-consumer type. The producer-consumer elastic resource parameters include the correspondence between the adjustable power lower limit, adjustable power upper limit, maximum adjustable duration, minimum continuous running time, restart preheating time, and allowed execution status. The adjustment boundary generation module matches the current process stage with the corresponding relationship to form an allowed execution state corresponding to the current process stage; compares the continuous running time with the minimum continuous running time to form the remaining adjustable duration; matches the restart preheating state with the restart preheating time to form an allowed execution time interval; and then calculates the adjustable power range based on the real-time load, the adjustable power lower limit, and the adjustable power upper limit. The dynamic adjustment boundary record includes producer-consumer identifier, production equipment identifier, target peak shaving period, adjustable power range, adjustable duration, allowed execution state, and status acquisition time; when the producer-consumer type does not match the producer-consumer elastic resource parameter, the adjustment boundary generation module records the corresponding allowed execution state as invalid; when the status acquisition time of the production process status record is earlier than the currently used production process status record, the current dynamic adjustment boundary record is retained, and the earlier production process status record is not used for updating. After the peak shaving task generation module generates candidate adjustment task records, the information acquisition module re-acquires the production process status records corresponding to the candidate adjustment task records. The adjustment boundary generation module recalculates the adjustable power range, adjustable duration, and allowed execution status based on the re-acquired current process stage, continuous running time, and restart preheating status, and updates the dynamic adjustment boundary record with the recalculated results. The adjustment boundary generation module provides the initially generated dynamic adjustment boundary record to the peak shaving task generation module and the updated dynamic adjustment boundary record to the executability verification and remaining peak shaving gap processing module.

[0121] The peak-shaving task generation module 03, connected to both the information acquisition module and the adjustment boundary generation module, is used to generate a peak-shaving task context by binding the peak-shaving demand and the real-time load with task identifiers based on the dynamic adjustment boundary record. The peak-shaving task context is then input into a neural network model to generate candidate adjustment task records with pending verification and standby states. Specifically, the peak-shaving task generation module receives the target peak-shaving period, peak-shaving demand, and real-time load provided by the information acquisition module, and receives the dynamic adjustment boundary record provided by the adjustment boundary generation module. The peak-shaving task generation module filters prosumers based on the allowed execution states in the dynamic adjustment boundary record; it binds the prosumer identifier, production equipment identifier, adjustable power range, adjustable duration, target peak-shaving period, peak-shaving demand, and real-time load with valid allowed execution states to the task identifier, generating a peak-shaving task context. The task identifier corresponds to a single peak-shaving demand and maintains association with the target peak-shaving period, the prosumer identifier, and the production equipment identifier. The peak-shaving task generation module organizes the peak-shaving task context according to the input order of the neural network model, and inputs the peak-shaving demand, the real-time load, the adjustable power range, the adjustable duration, and the target peak-shaving period into the neural network model; the neural network model outputs candidate adjustment power, candidate adjustment duration, candidate execution period, and candidate priority. The peak-shaving task generation module sorts multiple candidate adjustment tasks according to the candidate priority, and according to the correspondence between the cumulative value of the candidate adjustment power and the peak-shaving demand, marks the candidate adjustment tasks that enter the first round of comparison as pending verification, marks the remaining candidate adjustment tasks as standby, and generates candidate adjustment task records. The candidate adjustment task record includes task identifier, producer-consumer identifier, production equipment identifier, candidate adjustment power, candidate adjustment duration, candidate execution period, candidate priority, and candidate task status; when the candidate adjustment power, candidate adjustment duration, or candidate execution period output by the neural network model is empty, no corresponding candidate adjustment task record is generated; when the candidate execution period does not belong to the target peak-shaving period, the corresponding candidate adjustment task is marked as standby. The peak shaving task generation module provides the candidate adjustment task record to the executability verification and remaining peak shaving gap processing module; after the candidate adjustment task record is generated, the peak shaving task generation module calls the information acquisition module to re-acquire the production process status record of the corresponding production equipment, and the adjustment boundary generation module forms an updated dynamic adjustment boundary record.

[0122] The executability verification and remaining peak-shaving gap processing module 04 is connected to the adjustment boundary generation module and the peak-shaving task generation module, respectively. It compares the candidate adjustment task records in the verification state with the updated dynamic adjustment boundary records to determine the effective adjustment amount and the verification failure amount. Based on the peak-shaving demand and the effective adjustment amount, it generates remaining peak-shaving gap records. When the remaining peak-shaving gap is greater than the allowable threshold, it filters the candidate adjustment task records in the standby state and compares them again to generate an executable task set. Specifically, the executability verification and remaining peak-shaving gap processing module receives candidate adjustment task records provided by the peak-shaving task generation module and receives updated dynamic adjustment boundary records provided by the adjustment boundary generation module. The executability verification and remaining peak-shaving gap processing module matches the candidate adjustment task records in the verification state with the corresponding updated dynamic adjustment boundary records based on the producer-consumer identifier and the production equipment identifier; it compares the candidate adjustment power with the adjustable power range, compares the candidate adjustment duration with the adjustable duration, and matches the candidate execution period with the allowed execution state. When the candidate adjustment power is within the adjustable power range, the candidate adjustment duration is not greater than the adjustable duration, and the allowed execution state is valid, the candidate adjustment power is included in the effective adjustment amount. When the candidate adjustment power is greater than the upper limit of the adjustable power range, the upper limit of the adjustable power range is included in the effective adjustment amount, and the difference between the candidate adjustment power and the upper limit of the adjustable power range is included in the verification failure amount. When the allowed execution state is invalid or the candidate adjustment duration is greater than the adjustable duration and there is no usable execution period, the candidate adjustment power is included in the verification failure amount. The executability verification and remaining peak shaving gap processing module summarizes the effective adjustment amounts corresponding to multiple candidate adjustment tasks and generates a remaining peak shaving gap record based on the difference between the peak shaving demand and the effective adjustment amount. The remaining peak shaving gap record includes task identifier, peak shaving demand, effective adjustment amount, verification failure amount, remaining peak shaving gap, and allowable threshold. When the remaining peak-shaving gap exceeds the allowable threshold, the executability verification and remaining peak-shaving gap processing module filters candidate adjustment task records in the standby state according to candidate priority, and compares the filtered candidate adjustment task records with their corresponding updated dynamic adjustment boundary records again; the effective adjustment amount formed after the comparison is added to the original effective adjustment amount, and the remaining peak-shaving gap record is updated. Candidate adjustment tasks lacking updated dynamic adjustment boundary records or whose producer-consumer identifiers do not match the production equipment identifiers are not included in the executable task set, and their candidate adjustment power is included in the verification failure amount. The executability verification and remaining peak-shaving gap processing module summarizes the candidate adjustment tasks that meet the updated dynamic adjustment boundary records into an executable task set, and provides the executable task set to the instruction execution and feedback module.

[0123] The instruction execution and feedback module 05 is connected to the information acquisition module and the executability verification and remaining peak-shaving gap processing module, respectively. It is used to generate peak-shaving instruction packages based on the set of executable tasks and send them to the producer-consumer terminals; determine the actual adjustment amount based on the baseline load and the actual load in the peak-shaving instruction package, and generate an execution evaluation record and an adjustment capacity update record. Specifically, the instruction execution and feedback module receives the set of executable tasks provided by the executability verification and remaining peak-shaving gap processing module, and receives the baseline load and actual load provided by the information acquisition module. The instruction execution and feedback module generates peak-shaving instruction packages based on the task identifier, producer-consumer identifier, production equipment identifier, target adjustment power, and execution period in the set of executable tasks, and writes the baseline load into the peak-shaving instruction package corresponding to the task identifier; the peak-shaving instruction package includes the task identifier, producer-consumer identifier, production equipment identifier, target adjustment power, execution start time, execution end time, and baseline load. The instruction execution and feedback module sends the peak-shaving instruction package to the corresponding producer-consumer terminal according to the producer-consumer identifier, and receives the instruction receipt returned by the producer-consumer terminal. If the instruction receipt is not received within the receipt period, the corresponding target regulation power is deducted from the effective regulation amount, and the effective regulation amount after deduction is provided to the executability verification and remaining peak-shaving gap processing module to update the remaining peak-shaving gap record. For peak-shaving instruction packages that have returned instruction receipts, the information acquisition module obtains the actual load according to the execution start time and execution end time, and provides the actual load to the instruction execution and feedback module. The instruction execution and feedback module compares the baseline load and the actual load according to the same metering period to form the actual regulation power; forms the actual regulation amount according to the actual regulation power and the metering period; and associates the target regulation power, the actual regulation amount, the execution start time, and the execution end time with the task identifier to generate an execution evaluation record. When the actual load is missing or the acquisition time of the actual load does not fall between the execution start time and the execution end time, the corresponding actual load is not included in the actual regulation amount. The instruction execution and feedback module updates the continuous running time, restart preheating status, and remaining adjustable power according to the actual adjustment amount and actual execution period, and generates an adjustment capability update record; the adjustment capability update record is returned to the information acquisition module, and the information acquisition module writes it into the subsequent production process status record, so that the adjustment boundary generation module can call it in the next target peak shaving period.

Claims

1. A method for auxiliary peak shaving based on neural algorithms involving prosumers and consumers, characterized in that, include: S100: Obtain records of producer-consumer types, target peak-shaving periods, and production process status for multiple producers-consumers; Match the elastic resource parameters of producers and consumers, calculate the adjustable power range, adjustable duration and allowed execution status based on the current process stage, continuous running time and restart preheating status, and generate dynamic adjustment boundary records; S200. Based on the dynamic adjustment boundary record, the peak-shaving demand and real-time load are bound with the task identifier to generate a peak-shaving task context; the peak-shaving task context is input into the neural network model to generate candidate adjustment task records with a state to be verified and a standby state. S300: Reacquire the production process status record and update the dynamic adjustment boundary record; Compare the candidate adjustment task records in the pending verification state with the updated dynamic adjustment boundary records to determine the effective adjustment amount and the verification failure amount; generate the remaining peak shaving gap record based on the peak shaving demand and the effective adjustment amount; when the remaining peak shaving gap is greater than the allowable threshold, filter the candidate adjustment tasks in the standby state and compare them again to generate an executable task set. S400: Generate a peak shaving instruction package based on the set of executable tasks and send it to the consumer terminal; The actual adjustment amount is determined based on the baseline load and actual load in the peak shaving instruction package, and an execution evaluation record and a regulation capacity update record are generated.

2. The method according to claim 1, characterized in that, The production process status record includes producer / consumer identifier, production equipment identifier, current process stage, current operating status, continuous running time, cumulative downtime, remaining restart / preheating time, number of adjustments executed in the current scheduling cycle, and status acquisition time. The status acquisition time of the production process status record re-acquired in S300 is later than the generation time of the candidate adjustment task record.

3. The method according to claim 2, characterized in that, In S100, the allowed execution state is matched according to the current process stage, and the adjustable power range, adjustable duration and resumption conditions are calculated according to the continuous running time, the cumulative downtime, the remaining restart preheating time and the existing peak shaving tasks. In S300, the adjustable power range, the adjustable duration, the allowed execution state, and the boundary version are updated based on the reacquired current process stage, continuous running time, cumulative downtime, and remaining restart preheating time.

4. The method according to claim 1, characterized in that, In S200, the following information is obtained: given load forecast data, peak-shaving price ceiling for ancillary services market, data cutoff time, boundary version, market rule version, and model version. Match the peak shaving demand, real-time load, given load forecast data, and the peak shaving price ceiling of the ancillary service market according to the target peak shaving period, and bind the matching result, the data cutoff time, the boundary version, the market rule version, and the model version with the task identifier to generate the peak shaving task context.

5. The method according to claim 1, characterized in that, The neural network model generates candidate producers and consumers, candidate adjustment power, candidate adjustment duration, candidate execution period, and candidate priority based on the context of the peak shaving task. Candidate adjustment tasks with priority within the first priority range are marked as pending verification, and the remaining candidate adjustment tasks are marked as standby. The candidate adjustment task record includes task identifier, producer-consumer identifier, production equipment identifier, candidate adjustment power, candidate adjustment duration, candidate execution period, candidate priority, candidate task status, model version, and boundary version.

6. The method according to claim 5, characterized in that, In S300, the candidate adjustment power is compared with the updated adjustable power range, the candidate adjustment duration is compared with the updated adjustable duration, the candidate execution period is matched with the updated allowed execution state and recovery operation conditions, and the boundary version of the candidate adjustment task record is verified to be the updated boundary version. Based on the comparison, matching, and version verification results, candidate adjustment tasks are determined to be in a pass, fix, or fail state.

7. The method according to claim 6, characterized in that, For candidate regulation tasks that are in the pass state, the candidate regulation power is included in the effective regulation amount; For candidate regulation tasks in the correction state, the regulation power corrected by the adjustable power range is included in the effective regulation amount, and the difference between the candidate regulation power and the corrected regulation power is included in the verification failure amount. For candidate regulation tasks that are in a failed state, the candidate regulation power is included in the verification failure amount; The remaining peak-shaving gap record is generated based on the difference between the peak-shaving demand and the effective adjustment amount; Candidate adjustment tasks in the standby state are selected according to candidate priority and the comparison, matching and version verification as described in claim 6 are performed until the remaining peak-shaving gap is not greater than the allowable threshold or the maximum replenishment round is reached.

8. The method according to claim 1, characterized in that, The peak shaving instruction package includes a task identifier, producer-consumer identifier, production equipment identifier, target regulation power, execution start time, execution end time, baseline load, allowable deviation, boundary version, instruction version, and receipt deadline; If no instruction response is received within the response period, the target regulation power is deducted from the effective regulation amount, the remaining peak shaving gap record is updated, and candidate regulation tasks in standby status are selected and executed in step S300.

9. The method according to claim 8, characterized in that, Obtain the actual load from the start time to the end time of execution based on the task identifier; The actual adjustment amount is determined based on the difference between the baseline load and the actual load, and the metering period. The execution compliance rate is generated based on the actual adjustment amount and the target adjustment amount. The execution compliance rate, the peak-shaving price ceiling of the ancillary service market and the market rule version are bound to the task identifier to generate a subsidy settlement record. The adjustment capability update record is generated by updating the continuous running time, cumulative downtime, remaining restart and warm-up time, number of adjustments performed, and remaining adjustable power based on the actual execution start time, actual execution end time, and actual adjustment amount.

10. A system for auxiliary peak shaving based on neural algorithms involving prosumers and consumers, characterized in that, include: The system comprises an information acquisition module, an adjustment boundary generation module, a peak shaving task generation module, an executability verification and remaining peak shaving gap processing module, and an instruction execution and feedback module; the system is used to implement the method described in any one of claims 1-9.

Images