Quantum-inspired optimization-based park-level energy edge gateway scheduling method and system

Abstract

The application discloses a park-level energy edge gateway scheduling method and system based on quantum heuristic optimization, and belongs to the gateway scheduling field, which comprises the following steps: collecting real-time operation data of multi-source heterogeneous energy equipment through edge gateway rolling, when the data source is interrupted or out of limit, dynamic completion is carried out, and confidence information is generated; reading the performance attenuation parameters of each device in the device state record, dynamically adjusting the constraint boundary according to the attenuation parameters, converting part of the deterministic constraint into a probability constraint according to the confidence information, and establishing an optimization problem to be solved; dynamically adjusting the calculation complexity of the algorithm according to the real-time calculation load, outputting the control instruction and performing conflict checking, and feeding back the execution result to the local database after the checking; and sliding the time window by one step to form a rolling optimization control. The application solves the technical problems that the existing energy scheduling technology is difficult to adapt to the uncertainty of data, resulting in high safety risk of device operation and insufficient continuous operation stability.

Description

Technical Field

[0001] This invention relates to the field of gateway scheduling, and more specifically to a campus-level energy edge gateway scheduling method and system based on quantum-inspired optimization. Background Technology

[0002] Against the backdrop of the accelerated implementation of new power system construction, park-level integrated energy systems have become the core carrier for realizing energy cascade utilization and reducing costs and carbon emissions. The edge gateways deployed on the park side, as the core units for local data collection, optimization decision-making, and real-time control, directly determine the operational safety, economy, and stability of the park's energy system. However, in practical engineering applications, existing park-level energy edge gateway scheduling technologies generally have difficulty adapting to multiple uncertainties on-site, which can easily lead to distortion of the input data of the optimization model. This not only reduces scheduling accuracy but also easily causes safety risks such as equipment overload and shutdown. Furthermore, once cloud communication is interrupted, scheduling will become uncontrollable. Summary of the Invention

[0003] This application provides a campus-level energy edge gateway scheduling method and system based on quantum-inspired optimization, aiming to solve the technical problems that existing energy scheduling technologies cannot adapt to data uncertainty, resulting in high equipment operation safety risks and insufficient continuous operation stability.

[0004] In view of the above problems, this application provides a campus-level energy edge gateway scheduling method and system based on quantum-inspired optimization.

[0005] The first aspect disclosed in this application provides a campus-level energy edge gateway scheduling method based on quantum-inspired optimization, the method comprising: S1: Real-time operating data of multi-source heterogeneous energy devices is collected by an edge gateway deployed on the park side. When any data source is interrupted or exceeds its limit, a prediction model is started for dynamic completion, and confidence information representing the reliability of the completion is generated. S2: Based on the collected and completed data, the device status records stored locally are read. The status records contain the performance degradation parameters of each device. The constraint boundaries of each device in the optimization model are dynamically adjusted according to the degradation parameters. Based on the confidence information, some deterministic constraints in the optimization model are converted into probabilistic constraints to establish the optimization problem to be solved. S3: The optimization problem is solved using a quantum heuristic optimization algorithm. During the solution process, the computational complexity of the algorithm is dynamically adjusted according to the real-time computing load of the edge gateway, and the control command for the current time slot is output. S4: The control command is conflict checked. After the check passes, it is sent to the corresponding device for execution, and the execution result is fed back to the local database. S5: The time window is moved forward by one step and returned to S1 to form a rolling optimization control closed loop.

[0006] Another aspect disclosed in this application provides a campus-level energy edge gateway scheduling system based on quantum-inspired optimization, the system comprising: The acquisition module, used in S1, continuously collects real-time operating data of multi-source heterogeneous energy devices through an edge gateway deployed on the park side. When any data source is interrupted or exceeds its limit, it initiates a prediction model for dynamic completion and generates confidence information characterizing the reliability of the completion. The reading module, used in S2, reads locally stored device status records based on the collected and completed data. The status records contain performance degradation parameters of each device. It dynamically adjusts the constraint boundaries of each device in the optimization model according to the degradation parameters and converts some deterministic constraints in the optimization model into probabilistic constraints according to the confidence information, establishing the optimization problem to be solved. The calculation module, used in S3, uses a quantum heuristic optimization algorithm to solve the optimization problem. During the solution process, it dynamically adjusts the computational complexity of the algorithm according to the real-time computing load of the edge gateway and outputs the control command for the current time slot. The execution module, used in S4, performs conflict verification on the control command. After the verification is successful, it sends the command to the corresponding device for execution and feeds back the execution result to the local database. The optimization module, used in S5, slides the time window forward by one step and returns to S1, forming a rolling optimization control closed loop.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: By deploying edge gateways on the park side, rolling data collection of multi-source heterogeneous energy equipment operation is achieved. For data sources that are interrupted or exceed limits, dynamic completion is completed through a hybrid model of time series decomposition and lightweight gradient booster, and confidence information is generated synchronously. The constraint boundary of the optimization model is dynamically adjusted based on the equipment performance degradation parameters. At the same time, the deterministic constraints corresponding to low reliability data are converted into probabilistic constraints by combining the confidence of the completed data. An optimization problem to be solved is constructed that fits the real-time operating conditions of the equipment and the reliability of the data. The optimization solution is completed by a quantum heuristic optimization algorithm that can dynamically adjust the computational complexity according to the real-time computing load of the edge gateway. The output control commands are issued and executed after conflict verification, and a rolling optimization control closed loop is formed through a sliding time window. The online update and autonomous scheduling of the prediction model are completed locally, realizing the highly reliable, continuous and stable optimized operation of the scheduling system, and achieving the dual optimization effect of minimizing operating costs and carbon emissions.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1A flowchart illustrating a campus-level energy edge gateway scheduling method based on quantum-inspired optimization is provided for embodiments of this application; Figure 2 A schematic diagram of a campus-level energy edge gateway scheduling system based on quantum-inspired optimization is provided for the embodiments of this application.

[0010] Explanation of reference numerals in the attached diagram: Acquisition module 11, Reading module 12, Calculation module 13, Execution module 14, Optimization module 15. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] The overall concept of the technical solution provided in this application is as follows: This application provides a campus-level energy edge gateway scheduling method and system based on quantum-inspired optimization. It achieves rolling data acquisition of multi-source heterogeneous energy equipment operation data through edge gateways deployed on the campus side. For interrupted or over-limit data sources, a hybrid model combining time-series decomposition and lightweight gradient booster is used to dynamically complete the data and simultaneously generate confidence information. The constraint boundaries of the optimization model are dynamically adjusted based on equipment performance degradation parameters. Simultaneously, the deterministic constraints corresponding to low-reliability data are converted into probabilistic constraints by combining the confidence of the completed data. This constructs an optimization problem to be solved regarding the real-time operating conditions of the equipment and data reliability. A quantum-inspired optimization algorithm that dynamically adjusts computational complexity based on the real-time computing load of the edge gateway is used to complete the optimization solution. The output control commands are issued and executed after conflict verification, and a rolling optimization control closed loop is formed through a sliding time window. Furthermore, it can automatically switch to local autonomous mode when communication with the cloud is interrupted, completing online updates and autonomous scheduling of the prediction model locally.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0014] Example 1, as Figure 1 As shown in the embodiments of this application, a campus-level energy edge gateway scheduling method based on quantum-inspired optimization is provided. The method includes: S1: Real-time operating data of multi-source heterogeneous energy devices are collected in a rolling manner through the edge gateway deployed on the park side. When any data source is interrupted or exceeds the limit, the prediction model is started to perform dynamic completion and generate confidence information that characterizes the reliability of the completion.

[0015] Specifically, the edge gateway deployed on the park side first performs rolling data collection. Rolling data collection refers to continuously and cyclically collecting operational data from target devices within the park at preset fixed time steps, forming a continuous time-series data chain. This provides real-time, synchronous data input for subsequent rolling optimization loops. The objects of the data collection are multi-source heterogeneous energy devices within the park. These devices refer to various production, consumption, and storage devices belonging to different energy categories such as electricity, heat, and cooling, using different industrial communication protocols, and whose data formats differ significantly from the native sampling frequencies. Typical examples include photovoltaic modules, energy storage batteries, central air conditioning units, industrial controllable loads, and new energy charging piles. These are the core control targets of the entire park's energy dispatching system. During the rolling data collection process, the edge gateway... The edge gateway will synchronously perform anomaly detection on the collected data sources, including two types of anomalies. The first type is data source interruption. The judgment rule is that if the edge gateway does not receive a valid data packet from the corresponding device data source for more than two sampling periods, it is judged as a communication interruption of the data source. The judgment threshold of two sampling periods is set to avoid misjudgment caused by single network jitter and instantaneous packet loss, and to balance the sensitivity of anomaly identification and the stability of system operation. The second type is value exceeding the limit. If the device operation value received by the gateway exceeds the physical range of the device's factory preset or calibrated by operation and maintenance, it is judged as a value exceeding the limit. The physical range refers to the upper and lower limits of the operating parameters that the device can achieve under compliant and safe operating conditions. Values ​​exceeding this range are invalid abnormal data and cannot be directly used for subsequent optimization calculations.

[0016] Subsequently, for any abnormal data source determined to be interrupted or exceeding limits, the edge gateway immediately initiates a predictive model to perform a dynamic completion process. The predictive model used is a hybrid model based on time-series decomposition and a lightweight gradient booster, adapted to the limited computing resources of the edge gateway. This model first uses a time-series decomposition method to decompose the valid historical data sequence of the data source marked as abnormal within a preset historical time window into three independent parts: trend component, periodic component, and residual component. The trend component represents the long-term trend of equipment operation data, including the decay trend of energy storage battery capacity with the number of cycles and the power decay trend of photovoltaic modules with the service life. The periodic component represents the regularity of data over time. The complex fluctuation characteristics include the daily cycle of photovoltaic output changing with the alternation of day and night and the variation of solar irradiance, and the weekly cycle of air conditioning load in the park changing with weekdays / restdays. The residual component is the random fluctuation part remaining after removing the trend and cycle characteristics, corresponding to environmental sudden disturbances, temporary load fluctuations and other influencing factors without clear rules. While completing the dynamic data completion, the edge gateway will simultaneously generate confidence information that represents the reliability of the completion. Among them, the data completion confidence is the core indicator that quantifies the probability of deviation between the predicted value and the actual value obtained by the completion and the reliability level. Its value range is fixed from 0 to 1. The larger the value, the higher the reliability of the completion result and the stronger the matching degree with the actual operating state of the equipment.

[0017] S2: Based on the collected and completed data, read the device status records stored locally. The status records contain the performance degradation parameters of each device. Adjust the constraint boundaries of each device in the optimization model according to the degradation parameters. Convert some deterministic constraints in the optimization model into probabilistic constraints according to the confidence information to establish the optimization problem to be solved.

[0018] Specifically, based on the time-series operation data of multi-source heterogeneous energy equipment in the park, which is generated by rolling collection and dynamic anomaly completion, and labeled with confidence level, the edge gateway deployed on the park side first reads the equipment status records that are continuously maintained and updated in real time in the local time-series database. The equipment status records are the full life cycle operation ledger of all energy equipment included in the scheduling and management in the park. The ledger includes the performance degradation parameters of each equipment. These parameters are quantitative indicators of the deviation of the equipment's operating performance, rated output capacity, and safe operating boundary from the factory nominal value due to factors such as component aging, mechanical wear, and environmental corrosion during long-term operation. They include parameters that are strongly correlated with the safe operation and optimized scheduling of equipment, such as the cumulative cycle number and state of health (SOH) of energy storage batteries, the power degradation rate of photovoltaic modules, and the cumulative operating time and energy efficiency ratio reduction coefficient of air conditioning compressors.

[0019] After completing the reading of device status records, the edge gateway traverses all energy devices to be optimized within the current scheduling cycle. Based on the read performance degradation parameters, it calculates the constraint boundaries of each device under the current actual operating conditions. The constraint boundaries are the upper and lower limits of the set adjustable decision variables. These boundaries will be dynamically adjusted as the device performance degrades. The upper limit of the charging and discharging power of the energy storage battery and the available capacity range will be adjusted in sync with the decrease of the battery health status (SOH). The upper limit of the maximum output of the photovoltaic module will be adjusted in sync with the increase of the power degradation rate. The upper limit of the maximum load of the air conditioning unit will be dynamically narrowed as the energy efficiency ratio decrease coefficient increases, thereby ensuring that the constraint boundaries always conform to the real-time operating capabilities of the device.

[0020] During the dynamic adjustment of constraint boundaries, the edge gateway simultaneously performs safety verification on the degree of device attenuation. When the attenuation factor of the device to be optimized is lower than the preset safety threshold, the edge gateway will temporarily remove the corresponding device from the set of decision variables to be optimized. The set of decision variables refers to the set of device operating parameters in the optimization model that can be actively controlled by scheduling commands, including energy storage charging and discharging power, air conditioning set temperature, controllable load switching status, photovoltaic grid connection point power, etc. The removed device will be forced to switch to the preset safe operation mode. At the same time, the edge gateway will generate operation and maintenance alarms to avoid safety risks such as overload and shutdown caused by severely aged and performance deterioration devices participating in optimization scheduling.

[0021] While dynamically adjusting the constraint boundaries, the edge gateway dynamically transforms the constraint conditions in the optimization model based on the confidence information of the supplementary data generated in the previous steps. Deterministic constraints are rigid constraints that must be 100% satisfied in the optimization model, including the physical safety operating limits of equipment, the hard grid connection rules, and the minimum energy consumption requirements of the park. The validity of these constraints does not change with the reliability of the data. Probabilistic constraints, on the other hand, are flexible constraints that allow the constraint conditions to be satisfied within a preset confidence probability range. During the transformation process, the confidence level of the probabilistic constraints is set synchronously based on the confidence value of the supplementary data. The lower the confidence level of the supplementary data, the lower the confidence level that the corresponding probabilistic constraints need to satisfy, and vice versa.

[0022] After completing the dynamic adjustment of the constraint boundary and the conversion of the constraint type, the edge gateway uses the adjusted upper limit of the actual constraint as the new feasible domain boundary of the decision variables of each device. The feasible domain refers to the legal range of values ​​that the decision variables in the optimization model are allowed to take, which is jointly limited by all constraints. Then, the converted probabilistic constraints are merged with the unconverted deterministic constraints to form constraints. At the same time, the objective function is constructed with minimizing the park's operating cost and minimizing carbon emissions as the dual optimization objectives. The objective function is a quantitative indicator used to measure the quality of the scheduling scheme in the optimization problem, forming the optimization problem to be solved.

[0023] S3: The optimization problem is solved using a quantum heuristic optimization algorithm. During the solution process, the computational complexity of the algorithm is dynamically adjusted according to the real-time computing load of the edge gateway, and the control command for the current time slot is output.

[0024] Specifically, for the optimization problem with dynamically adjusted equipment constraints and dual objectives of minimizing park operating costs and carbon emissions, the edge gateway employs a quantum-heuristic optimization algorithm adapted to the characteristics of edge computing power to solve the problem locally. This algorithm is an intelligent optimization algorithm that integrates quantum mechanical tunneling effects, potential well characteristics, and chaotic search mechanisms. It has the advantages of not needing to solve for the gradient of the objective function, adapting to non-convex nonlinear multi-constraint optimization problems, being less prone to getting trapped in local optima, and flexibly adjusting computing power overhead. The solution process first randomly initializes the positions and velocities of N particles within the feasible space of the decision variables. The position vector of each particle corresponds to a complete set of multi-time-slot scheduling control sequences, and each dimension of the vector corresponds to an independent decision variable. The particle's velocity corresponds to its search step size during iterative updates within the feasible space. After initialization, the algorithm evaluates the fitness of each particle. After a single round of fitness evaluation, the algorithm updates the individual optimal position of each particle based on the comparison between its current fitness and the historical best fitness, and simultaneously updates the global optimal position based on the optimal fitness of the entire population. The algorithm then calculates the diversity index of the current population; when the calculated diversity index is lower than a preset diversity threshold, the algorithm linearly increases the quantum tunneling probability. For particles that have not undergone tunneling, the algorithm uses quantum behavior position update rules to perform position iterations, simultaneously calculating the potential well feature length. During the iteration process, the algorithm continuously monitors the progress towards the global optimal solution. In the convergence state, if the rate of change of the global optimal fitness for three consecutive iterations is lower than a preset rate of change threshold, it indicates that the convergence has entered the late stage, and the global optimal solution is close to the theoretical optimal value. At this time, the chaotic local search subroutine is triggered, generating m chaotic perturbation points through chaotic mapping, evaluating the fitness of each perturbation point one by one, and replacing the current global optimal position if a position with better fitness is found. The algorithm iteratively executes the process of fitness evaluation, optimal position update, population diversity monitoring, quantum tunneling operation, quantum behavior position update, and chaotic local search until the preset maximum number of iterations is reached or the preset convergence condition is met, and then terminates the iteration, outputting the position vector of the optimal particle as the optimal control sequence. The optimal control sequence is the entire rolling optimization time window. The multi-slot scheduling instruction set of the port is based on the rolling optimization logic of model predictive control. The algorithm only takes the first element of the optimal control sequence as the control instruction to be executed in the current slot. In the entire algorithm solution, the edge gateway will continuously monitor its own real-time computing load. The real-time computing load refers to the current CPU utilization, memory utilization, and the computing power consumption ratio of parallel tasks such as data acquisition, communication interaction, and conflict verification of the edge gateway. The algorithm's computational complexity will be dynamically adjusted according to the load. When the real-time computing load of the gateway is higher than the preset high load threshold, the number of initial particles N, the maximum number of iterations, the number of chaotic disturbance points m, and the population diversity threshold and convergence change rate threshold will be increased simultaneously to trigger the iteration termination in advance.When the real-time computing load of the gateway falls below a preset low-load threshold, the number of particles, the maximum number of iterations, and the number of chaotic perturbation points are increased synchronously to reduce the convergence rate threshold, improve search accuracy, and obtain a better scheduling scheme.

[0025] S4: Perform conflict verification on the control command. If the verification passes, send it to the corresponding device for execution and feed the execution result back to the local database.

[0026] Specifically, the edge gateway initiates full-dimensional conflict verification through its built-in conflict resolution module. The verification dimensions include three types of constraints: internal device consistency constraints, inter-device coupling constraints, and security protocol constraints. After completing the full-dimensional verification, if the control command to be executed fully complies with all three types of constraints, the verification is considered successful. The edge gateway then encapsulates the control command according to the industrial bus protocol corresponding to the target device and sends it to the corresponding energy device for execution. The industrial bus protocol is a standardized communication and interaction rule between industrial field devices in the park and the edge gateway, adaptable to multi-source heterogeneous energy devices of different brands, types, and communication interfaces. During the command issuance process, the edge gateway synchronously monitors the command issuance status in real time. If a command issuance failure occurs, a preset retry mechanism is immediately triggered, and the command is reissued according to the set retry interval and maximum number of retries. If the command still fails after reaching the maximum number of retries, the edge gateway will mark the corresponding device as offline and force the device to switch to a preset safe operating mode. Regardless of whether the command is issued and executed normally, executed after degradation, or switched to a safe mode after failure, the edge gateway will collect the actual operation feedback data of the target device after the command execution cycle ends. The command issuance value, actual execution value of the device, command execution timestamp, execution status markers including normal execution, degradation execution, issuance failure, offline, and other full-link information are synchronously written to the local time-series database. The local time-series database is a database deployed locally by the edge gateway to store time-stamped device operation and scheduling data. It has the characteristics of high write performance, high data compression ratio, and efficient time-series query, and retains the execution data of the entire rolling optimization closed loop.

[0027] S5: Slide the time window forward by one step, return to S1, and form a rolling optimization control closed loop.

[0028] Specifically, the edge gateway first performs integrity verification and exception handling for the current scheduling cycle, confirming that each step of S1 data collection and completion, S2 optimization problem construction, S3 algorithm solution, and S4 command verification and issuance has been completed and closed. Simultaneously, it checks for any unprocessed device offline markers, command issuance failure events, device performance degradation exceeding the standard maintenance alarms, or other anomalies. If any unresolved anomalies exist, the complete information of the anomaly event is first archived to the local maintenance event database and the device safe operation status marker is updated synchronously before entering the formal time window sliding process.

[0029] Subsequently, the edge gateway slides the preset rolling optimization time window forward by a fixed sliding step. Here, the rolling optimization time window refers to the look-ahead optimization time interval covered by the quantum heuristic optimization algorithm in each optimization solution step S3, consisting of multiple equal-length scheduling slots. This is used to achieve multi-step look-ahead optimization, fundamentally avoiding the scheduling shortsightedness problem caused by single-slot optimization focusing only on the current operating conditions. The sliding step refers to the fixed duration by which the time window advances each time. This duration is consistent with the fixed sampling period and single-slot scheduling period of the edge gateway in step S1 of the scheme. For example, when the rolling optimization time window... Assuming the optimization solution covers the next 5 scheduling slots and the single-slot scheduling cycle (sliding step) is set to 15 minutes, the optimization solution in step S3 covers 5 consecutive slots from T0 to T4. Only the control instructions for slot T0 are issued and executed. After the execution process of slot T0 is completed, the time window slides forward by one step of 15 minutes. The new optimization time window is updated synchronously to cover 5 consecutive slots from T1 to T5, always maintaining a fixed length of look-ahead optimization interval. At the same time, the completed slot T0 is discarded, and only the unexecuted slots from T1 to T4 are retained as part of the look-ahead interval of the new window. Together with the newly added slot T5, a complete new optimization window is formed.

[0030] After completing the sliding operation of the time window, the edge gateway synchronously performs parameter initialization and dynamic update of data boundaries for the new scheduling cycle. That is, based on the actual execution data and operating status feedback of the devices written to the local time-series database in the previous cycle, it completes the update of two major parameters. The first is to update the device status records and performance degradation parameters stored locally, including updating the cumulative number of cycles and the SOH value of the energy storage battery based on the actual execution data of this charge and discharge cycle, updating the cumulative running time and energy efficiency ratio reduction coefficient based on the actual running time of the air conditioning unit, and updating the power degradation rate based on the actual output data of the photovoltaic module. The second is to update the historical data sequence of the prediction model used for data completion in step S1, supplementing the effective measured data of the devices collected in the previous cycle into the training sliding window of the prediction model, and simultaneously removing old historical data that exceeds the time window.

[0031] After completing all parameter updates and new cycle initialization configuration, the edge gateway automatically jumps back to step S1 and starts a new scheduling cycle of rolling data acquisition from multi-source heterogeneous energy devices. This initiates a new round of rolling optimization control loop, which involves continuously updating on-site measured data with a fixed sliding step size, reconstructing optimization problems tailored to the current operating conditions in each cycle, and executing only the optimal control commands for the current time slot. Through continuous real-time operational feedback and iterative optimization, it continuously adapts to various uncertainties such as real-time changes in the park's energy system, long-term performance degradation of equipment, random disturbances in the external environment, and communication link anomalies. Simultaneously, during the entire time window sliding and iteration process, the edge gateway continuously monitors its communication status with the cloud management platform. If the communication link is normal, it can synchronously upload the rolling optimization operation data, scheduling effects, equipment status information, and maintenance alarms to the cloud management platform according to the preset reporting cycle. If communication is interrupted, it maintains complete local autonomous operation. All parameter updates, process jumps, and cycle iterations are completed locally on the edge gateway, without relying on any cloud command issuance or computing power support, ensuring the continuous and stable operation of the entire control process.

[0032] Furthermore, in the method provided in the application embodiments, during the execution of S1-S4, the method further includes: The system continuously monitors the communication status between the edge gateway and the cloud. When a communication interruption is detected, it automatically switches to local autonomous mode and updates the prediction model online locally using recent measured data. The updated prediction model is not transmitted to the cloud.

[0033] Specifically, throughout the entire scheduling cycle from S1 to S4, the edge gateway deployed on the campus side continuously monitors the bidirectional communication link status between the edge gateway and the cloud management platform through a built-in communication status monitoring module using a heartbeat detection mechanism. The heartbeat detection mechanism refers to the edge gateway sending a simplified heartbeat data packet carrying the gateway's unique identifier and a precise timestamp to the cloud management platform at a preset fixed interval of milliseconds, while simultaneously waiting for a confirmation response packet returned by the cloud. The monitoring module is simultaneously set with dual anti-false judgment communication interruption judgment rules, forming a unified anomaly identification system with the dual sampling period judgment logic for data source interruption in step S1. If the edge gateway fails to receive a valid response packet from the cloud after continuously sending a preset number of heartbeat packets, it is judged as an uplink communication link interruption. If the edge gateway fails to receive any valid downlink data packets such as configuration updates, scheduling instructions, or parameter calibrations from the cloud within multiple preset consecutive scheduling cycles, it is judged as a downlink communication link interruption. As long as any one of the judgment conditions is met, and bidirectional normal communication cannot be restored after a preset short-term reconnection and retry process, it is ultimately judged as a communication interruption between the edge gateway and the cloud.

[0034] Once the communication interruption determination takes effect, the edge gateway will immediately perform a non-intrusive mode switch, automatically switching to local autonomous mode. Local autonomous mode means that the edge gateway is completely independent of the cloud management platform in terms of computing power, data, and command. All collection, calculation, storage, decision-making, and control actions in the entire scheduling process are completed locally on the campus-side edge gateway. Specifically, this includes: firstly, immediately locking all core operating baseline data in local non-volatile storage, such as the entire lifecycle ledger of all devices, performance degradation parameter table, pre-training weights of the prediction model used in step S1, configuration parameters of the quantum heuristic optimization algorithm, and security operation constraint rules; and stopping receiving and blocking all configurations issued from the cloud. Update instructions are implemented to prevent invalid downlink data from interfering with the stable operation of local scheduling logic. Secondly, all real-time device operation data, scheduling execution logs, and device alarm information that originally needed to be synchronously uploaded to the cloud are written to the local time-series database. The sliding window length for local data storage is simultaneously expanded to ensure sufficient continuous historical data to support model updates and optimization solutions in local autonomous mode. At the same time, all unnecessary cloud communication processes are shut down to release the gateway's CPU and memory resources, prioritizing the computing power supply for the S1 to S4 main scheduling processes, especially the computing power required for solving optimization algorithms. During local autonomous mode operation, the edge gateway will complete the S4 step of each scheduling cycle. After the control command execution results are fully written to the local time-series database, the prediction model used in step S1 is updated online locally using recent measured data. The prediction model is a hybrid prediction model based on time-series decomposition and lightweight gradient booster, used in step S1 for dynamic data anomaly completion. The recent measured data consists of runtime sequence data of multi-source heterogeneous energy devices, environmental characteristic parameter data, and actual execution feedback data of scheduling commands, collected locally by the edge gateway before and after the communication interruption and verified for validity. The data range is limited to the continuous sliding time window maintained locally after the communication interruption. The online update is to complete the incremental training of the model and the update of weight parameters locally on the edge gateway. The update method is as follows: the edge gateway automatically extracts the latest measured effective data within the current sliding window to construct an incremental training dataset, performs incremental online training on the lightweight gradient booster regression module in the hybrid prediction model, updates the regression tree weights and node splitting rules of the model, and recalibrates the fitting parameters of the trend component and periodic component of the time series decomposition based on the latest measured data. This allows the prediction model to continuously adapt to changes in the operating patterns of park energy equipment, fluctuations in environmental parameters, and adjustments in load characteristics during communication interruptions. The entire online update process is completed locally on the edge gateway, and the updated model weights and fitting parameters are stored only in the local encrypted storage unit and are not transmitted to the cloud in any form.

[0035] Furthermore, in the method provided in the application embodiment, real-time operating data of multi-source heterogeneous energy devices are collected in a rolling manner through an edge gateway deployed on the park side. When any data source is interrupted or exceeds its limit, a prediction model is activated for dynamic completion, and confidence information characterizing the reliability of the completion is generated. This includes: the edge gateway polls each multi-source heterogeneous energy device data source at a fixed sampling period. If no valid data packet is received for more than two sampling periods, it is determined to be a communication interruption; if the received value exceeds the preset physical range of the corresponding device, it is determined to be a value exceeding the limit; for data sources determined to be interrupted or exceeding the limit, a dynamic completion process is activated, and the device data is predicted through the prediction model to obtain the predicted value and the corresponding data completion confidence; the predicted value is used for dynamic completion, and the completed data is labeled with confidence to generate confidence information.

[0036] Specifically, the park-side edge gateway, deployed locally within the park and responsible for the core functions of local data collection, calculation, and control, performs cyclic polling collection on all multi-source heterogeneous energy device data sources connected within the park using a fixed sampling period that precisely matches the scheduling rolling step size. The fixed sampling period refers to a pre-set time interval that is completely consistent with the single-time-slot scheduling period. Polling collection involves the edge gateway actively initiating standardized data reading requests to each device data source according to the preset device addressing order, and synchronously receiving the running data packets returned by the devices. The multi-source heterogeneous energy device data sources refer to the data sources of various devices within the park that cover multiple energy categories such as electricity, heat, cooling, and storage, belonging to four major categories: production capacity, energy consumption, energy storage, and energy conversion. These devices use different industrial communication protocols, data formats, and native sampling frequencies with significant differences, including photovoltaic inverters, energy storage battery clusters, central air conditioning units, industrial controllable loads, and new energy charging piles. They are the core control objects of the entire scheduling system.

[0037] During the polling data collection process, the edge gateway synchronously performs a two-dimensional anomaly judgment on the return results of each data source. The first dimension is communication interruption judgment. The judgment rule is that if the edge gateway does not receive a valid data packet from the corresponding data source for more than two consecutive sampling periods, it is judged as a communication interruption of that data source. The judgment threshold of two sampling periods is set to avoid misjudgment caused by single network jitter and instantaneous packet loss, which are common in industrial sites. A valid data packet is a compliant data packet that conforms to the communication protocol specifications of the corresponding device, carries a complete timestamp, has no check bit errors, and contains valid operating values. Empty packets, garbled packets, and failed check packets are excluded. The second dimension is value limit judgment. If the device operating value received by the edge gateway exceeds the pre-set physical range of the device, it is judged as a value limit violation. The physical range refers to the upper and lower limits of the safe operating parameters of the device as specified by the manufacturer and calibrated by on-site operation and maintenance. Values ​​exceeding this range are invalid abnormal data and cannot be directly used for subsequent optimization model solutions.

[0038] For any abnormal data source identified as having a communication interruption or exceeding numerical limits, the edge gateway immediately initiates a pre-set dynamic completion process. This process uses a pre-trained, lightweight prediction model that can run at the edge to predict missing or abnormal operational data, simultaneously obtaining the predicted values ​​and corresponding data completion confidence levels. The prediction model is a hybrid model based on time series decomposition and a lightweight gradient booster. The complete prediction calculation process is as follows: First, the valid historical data sequence of the data source marked as abnormal within a pre-set historical time window is obtained. Then, the time series decomposition method is used to decompose this valid data sequence into three independent parts: a trend component, a periodic component, and a residual component. The trend component represents the long-term trend of the equipment's operational data, including the decline trend of the energy storage battery's health status with the number of cycles and the decline trend of photovoltaic module power with the years of use. The periodic component represents the regular, recurring fluctuations in data over time, including the daily cycle of photovoltaic output varying with day and night sunshine, and the weekly cycle of the park's cooling load varying with weekdays and rest days. The residual component represents the random fluctuations remaining after removing the trend and periodic characteristics, corresponding to unpredictable influencing factors such as sudden environmental disturbances and temporary load fluctuations. The strong regularity of the periodic component is then used for time-series extrapolation to obtain the periodic prediction value for the current missing or abnormal moment. The historical residual components are then input into a lightweight gradient booster regression model, using a preset number of historical residual values ​​before the current missing moment, as well as environmental characteristic parameters at the current moment, such as ambient temperature, real-time light intensity, and the park's baseline load value, as input features. The model outputs the residual prediction value for the current moment. Finally, the periodic prediction value and the residual prediction value are added together to obtain the final prediction value used to replace the abnormal data.

[0039] While generating predicted values, the edge gateway simultaneously calculates the corresponding data completion confidence score. The data completion confidence score is an indicator that quantifies the probability of deviation between the completed predicted value and the actual operating value of the device, representing its reliability level. Its value range is fixed between 0 and 1; a higher value indicates higher reliability and a stronger match with the actual operating state of the device. The calculation process is as follows: First, a fixed-length sliding window continuously records the prediction residuals of the prediction model for historical valid data points. The standard deviation of the prediction residuals within the sliding window is calculated to obtain the historical residual standard deviation, which characterizes the historical prediction error benchmark level of the prediction model on the corresponding device. Then, the variance component of the residual prediction value corresponding to the currently missing completed value is used as the estimated prediction residual value. The edge gateway first assesses the expected fluctuation level of the current prediction result, then calculates the ratio between the standard deviation of the predicted residual value and the standard deviation of the historical residual. Through a preset monotonically decreasing mapping relationship, this ratio is transformed into a confidence value between 0 and 1. Simultaneously, the edge gateway adds a corresponding discrete identifier to the completed data based on the specific confidence value, completing the reliability labeling of the completed data. After generating the predicted value and confidence information, the edge gateway uses the predicted value to dynamically complete abnormal or missing data sources, filling gaps and anomalies in the time series data sequence. At the same time, it synchronously adds a corresponding confidence value and discrete identifier to each completed data point, completing the confidence labeling of the entire data set, and finally generating a complete equipment operation dataset with reliability quantification information.

[0040] Furthermore, in the method provided in the application embodiment, predicting device data using a prediction model to obtain predicted values ​​includes: acquiring a valid data sequence of data sources marked as interrupted or exceeding limits within a preset historical time window; decomposing the valid data sequence into trend components, periodic components, and residual components using a time-series decomposition method; wherein the prediction model is a hybrid model based on time-series decomposition and a lightweight gradient booster; extrapolating the periodic components to obtain the periodic prediction value at the current missing moment; inputting the residual components into a pre-trained lightweight gradient booster regression model, using a preset number of residual values ​​before the current missing moment and environmental feature parameters at the current moment as input features, and outputting the residual prediction value; adding the periodic prediction value and the residual prediction value to obtain the final predicted value.

[0041] Specifically, when the edge gateway on the park side completes the data source anomaly determination and marks a certain device data source as interrupted or exceeding the limit, it obtains the valid data sequence of the marked data source within a preset historical time window. The preset historical time window is a fixed-duration sliding data interval that is pre-set locally by the edge gateway and matches the characteristics of the device's operating cycle. It includes the device's complete daily and weekly operating cycles, which is used to provide sufficient historical data support that fits the current operating pattern for predictive calculations. The valid data sequence is the continuous time-series operating data of the data source within the corresponding time window that has not experienced communication interruption, has not exceeded the value limit, has been verified by the communication protocol check bit, and has a precise synchronization timestamp, completely eliminating outliers, invalid empty packets, and garbled data. Subsequently, the edge gateway employs a time-series decomposition method to decompose the acquired effective data sequence into three independent subsequences: a trend component, a periodic component, and a residual component, each of which can be modeled separately. The time-series decomposition method is an algorithm designed for industrial energy time-series data and adaptable to non-stationary data sequences. It decomposes raw data, which contains a mixture of long-term changes, regular fluctuations, and random disturbances, into subsequences with different feature dimensions. The trend component represents the long-term, monotonous, and smooth trend of equipment operation data over time, unaffected by short-term random fluctuations. This includes the decline trend of the State of Health (SOH) of energy storage batteries with increasing charge-discharge cycles, the decreasing trend of photovoltaic module output power with increasing service life, and the long-term increase or decrease trend of the park's basic energy load with changes in production capacity. The periodic component, also called the seasonal component, represents the regular and repetitive fluctuation characteristics of equipment operation data over a fixed time period, including the variation of photovoltaic output with diurnal solar radiation intensity. The system employs a 24-hour daily cycle, a 7-day weekly cycle for the park's air conditioning cooling load (which varies with workdays and rest days), and an hourly cycle for industrial production load (which varies with shift schedules). The residual component is the random fluctuation portion of the original valid data sequence after removing trend and periodic components, which lacks a clear long-term pattern. This fluctuation is caused by unpredictable random factors such as sudden environmental changes, temporary load adjustments, and instantaneous equipment operation disturbances. The prediction model used is a hybrid model based on time series decomposition and a lightweight gradient booster. This hybrid model uses differentiated prediction strategies for different characteristic components after decomposition. It utilizes the strong regularity of the periodic component to achieve high-precision basic prediction, and uses a machine learning model to fit the nonlinear random characteristics of the residual component. At the same time, the lightweight architecture used in the model has been pruned and optimized with quantization compression, enabling millisecond-level real-time inference to be completed locally without relying on cloud computing power. This perfectly adapts to the low latency and localized operation requirements of the park's edge.

[0042] After completing the time series decomposition, the edge gateway first uses the decomposed periodic components to perform time series extrapolation to obtain the periodic prediction value for the current missing moment. The periodic component extrapolation is based on the fixed repetitive fluctuation pattern of the periodic components. The periodic component values ​​of the same phase moment in the corresponding period within the historical time window are combined with the long-term change trend of the trend component for weighted time series extrapolation. Since the periodic components have extremely strong regularity, the extrapolation calculation error is extremely small, forming the basic main part of the final predicted value. For example, for the missing photovoltaic power output data at 10 am on Tuesday, the periodic component benchmark values ​​at 10 am on all weekdays within the historical window can be extrapolated based on the long-term power decay trend of the photovoltaic modules to obtain the periodic prediction value for the current missing moment.

[0043] After calculating the periodic prediction value, the edge gateway inputs the decomposed historical residual components into a pre-trained lightweight gradient booster regression model. This pre-trained model is trained offline with supervised training using long-term historical operating data of the corresponding device before deployment to the edge gateway, and is optimized through lightweight compression. It is specifically designed to fit the nonlinear and non-stationary random fluctuation characteristics of the residual components. The model's input features include two core types of data: one is a preset number of historical residual values ​​before the current missing time, used to capture the temporal autocorrelation of the residual sequence; the other is environmental feature parameters that can be collected in real time at the current moment, including external parameters that can significantly affect residual fluctuations, such as real-time light intensity, ambient temperature and humidity, park production scheduling status, and baseline load level, which are strongly correlated with the operation of the corresponding device. After the model completes inference based on the input features, it outputs the residual prediction value for the current missing time. Finally, the edge gateway linearly adds the calculated periodic prediction value to the residual prediction value to restore the final predicted value of the device's operating data at the current missing or abnormal time. This value is the dynamic completion value used to fill in data source interruptions or gaps.

[0044] Furthermore, in the method provided in the application embodiment, obtaining the corresponding data completion confidence score includes: recording the prediction residuals of the prediction model for historical valid data points using a sliding window; calculating the standard deviation of the prediction residuals within the sliding window to obtain the historical residual standard deviation; using the variance component of the residual prediction value corresponding to the completed value at the current missing time as the prediction residual estimate; and performing prediction value confidence analysis based on the comparison result between the historical residual standard deviation and the prediction residual estimate to obtain the data completion confidence score.

[0045] Specifically, a sliding window is used to record the prediction residuals of the hybrid prediction model for historical valid data points. The sliding window is an independently configured, equal-length sliding time-series window for each multi-source heterogeneous energy device data source, precisely matched to the device's operating cycle characteristics. The window length is set to cover 168 consecutive sampling points covering the complete daily and weekly operating cycles of the device, adapting to a scheduling step size of 15 minutes fixed sampling period. The window slides forward one step with each sampling cycle, always retaining the latest valid historical data and automatically removing outdated and mismatched data exceeding the window range. The prediction residual refers to the data at the valid sampling moment when the data source is communicating normally and the data has not exceeded the physical range, processed using a hybrid prediction model completely consistent with the anomaly completion process. The edge gateway calculates the algebraic difference between the predicted value obtained from the synchronous backtesting prediction and the actual effective operating value measured by the device. For each effective sampling point of each data source, the edge gateway calculates and stores the corresponding prediction residual in real time. After completing the construction of the historical residual sequence, the edge gateway uses the unbiased standard deviation calculation formula to calculate the standard deviation of all historical prediction residuals within the sliding window to obtain the historical residual standard deviation. The historical residual standard deviation is a quantitative indicator in statistics that characterizes the degree of dispersion of random variables. It eliminates small sample errors by dividing by the sample size minus 1. Its value directly characterizes the historical prediction error benchmark level and fluctuation stability of the hybrid prediction model on the corresponding device. The smaller the value, the higher the historical prediction accuracy of the model and the smaller the error fluctuation.

[0046] Subsequently, the edge gateway uses the variance component of the residual prediction value corresponding to the imputed value at the current time as the estimated value of the predicted residual. Based on the time-series decomposition logic of the hybrid prediction model, almost all of the total prediction error of the imputed value comes from the prediction stage of the residual components. The periodic component is based on strong regular time-series extrapolation, and the trend component is a smooth long-period change. The prediction errors of the two are negligible. Therefore, the uncertainty of the imputed value is entirely determined by the fluctuation characteristics of the residual prediction value. The variance component of the residual prediction value is the variance statistic calculated by the lightweight gradient booster regression model in the hybrid prediction model based on the dispersion of the output results of multiple regression trees in the model ensemble learning while outputting the residual prediction value at the current time. By reasoning about the current input features through multiple regression trees that have been trained, multiple residual prediction candidate values ​​are obtained. The variance of the candidate value sequence is the variance component of the residual prediction value. This variance component directly quantifies the expected fluctuation range of the residual prediction at the current time, which is the expected error amplitude of the imputed value. The edge gateway directly uses this variance component as the estimated value of the predicted residual to complete the real-time prediction of the expected error of the current imputed operation.

[0047] Finally, based on the comparison result between the historical residual standard deviation and the predicted residual estimate, the edge gateway conducts a standardized analysis of the confidence level of the predicted value, obtaining the final confidence level for data completion. Specifically: First, take the square root of the predicted residual estimate, that is, the variance, to obtain the standard deviation of the current predicted residual, achieving a same-dimensional comparison with the historical residual standard deviation. Subsequently, calculate the dimensionless ratio K of the two, that is, the ratio of the current predicted residual standard deviation to the historical residual standard deviation. This ratio K directly reflects the magnification factor of the expected error of the current completion operation relative to the historical benchmark error of the model. When K≥1, it means the current expected error is higher than the historical benchmark level, and the larger K is, the lower the reliability of completion. When 0<K<1, it means the current expected error is lower than the historical benchmark level, and the reliability of completion is higher. The data completion confidence level is a normalized index that quantitatively characterizes the matching degree and trustworthiness between the current predicted completion value and the actual operating value of the device, with a value range in the interval [0,1]. The larger the value, the higher the reliability of the completion result and the stronger the matching degree with the actual working conditions of the device. The edge gateway converts the ratio K into a compliant confidence level value through a preset monotonically decreasing mapping relationship. This mapping relationship specifically adopts a piecewise exponential decay mapping adapted to the industrial control scenario: When K≤1, the confidence level is directly taken as 1, indicating that the current completion expected error is not higher than the historical benchmark, and the confidence level is at the highest level; when K>1, the confidence level: ; where is a preset positive decay coefficient, with a value range of 0.5~2, which can be flexibly adjusted according to the safety importance level of the park equipment and the rigid requirements of scheduling constraints. This mapping relationship ensures that the confidence level decreases monotonically with the increase of K and finally converges to 0. At the same time, the edge gateway will attach discrete reliability identifiers to the calculated continuous confidence level values according to a preset threshold interval, that is, [0,0.3) low confidence level, [0.3,0.7) medium confidence level, [0.7,1] high confidence level, completing the full process generation and annotation of the confidence level.

[0048] Furthermore, in the method provided by the application embodiment, based on the comparison result between the historical residual standard deviation and the predicted residual estimate, a confidence analysis of the predicted value is conducted to obtain the confidence level for data completion, including: comparing the standard deviation of the residual prediction value with the historical residual standard deviation and calculating the ratio; based on the ratio, the edge gateway converts the ratio into a confidence level value within the range of 0 to 1 through a monotonically decreasing mapping relationship, where the larger the value, the more reliable the completion result. The edge gateway attaches a discrete identifier to the completed data according to the specific value of the confidence level.

[0049] Specifically, first, perform a square root operation on the variance component of the residual prediction value, which is the predicted residual value of the prediction residual estimate, to obtain the standard deviation of the residual prediction value. The standard deviation of the residual prediction value is a dimensional statistical indicator of the dispersion degree of the current prediction result of the residual and the expected error amplitude. The historical residual standard deviation refers to the unbiased standard deviation calculated from the prediction residuals of the effective historical data points of the model within the sliding window, and is the benchmark statistic of the prediction accuracy of the hybrid prediction model based on time series decomposition and lightweight gradient boosting machine under the normal operating state of the corresponding device. After the two complete dimensional alignment, the edge gateway calculates the dimensionless ratio of the two according to the preset normalization formula. The specific calculation formula is: Ratio K = standard deviation of the current residual prediction value / historical residual standard deviation, The ratio K here is an indicator that quantifies the degree of amplification or reduction of the expected error of the current completion operation relative to the historical benchmark error of the model. It is clear that when K = 1, it means that the expected error of the current completion is exactly the same as the historical benchmark accuracy of the model; when K > 1, it means that the expected error of the current completion is higher than the historical benchmark, and the larger the value of K, the higher the expected error and the worse the reliability of the completion result; when 0 < K < 1, it means that the expected error of the current completion is lower than the historical benchmark, and the reliability of the completion result is better than the normal level. After calculating the ratio K, the edge gateway converts the ratio K into a confidence value strictly limited within the range of 0 to 1 through a preset monotonically decreasing mapping relationship. The monotonically decreasing mapping relationship means that the input ratio K and the output confidence value show a strictly reverse monotonic change relationship, ensuring that the higher the expected error of the completion, the lower the output confidence value, and at the same time, the output value always falls within the normalized interval of [0, 1]. The mapping relationship uses a piecewise exponential decay mapping function, which is specifically mapped in two intervals: The first interval, when K ≤ 1, the confidence value Conf is directly taken as 1.0, indicating that the expected error of the current completion is not higher than the historical benchmark accuracy of the model, and the completion result has the highest level of reliability; The second interval, when K > 1, the confidence value: ; Where α is a preset positive attenuation coefficient, typically ranging from 0.5 to 2.0, which can be flexibly adjusted according to the security importance level of the corresponding device and the rigidity requirements of scheduling constraints. This mapping function ensures that the confidence value continuously and monotonically decreases with the increase of K and always converges to 0, fully meeting the requirement of the value range of 0 to 1. The larger the value, the more reliable the completion result. After the confidence value is calculated, the edge gateway adds a discrete identifier to the completed data according to the specific confidence value. The discrete identifier is based on a preset confidence threshold range, which is a continuous confidence value. The confidence level values ​​are matched with the corresponding standardized reliability level labels. The specific threshold ranges and labeling rules are as follows: if the confidence level value falls within the range of [0.7, 1.0], a high confidence discrete label is added; if it falls within the range of [0.3, 0.7), a medium confidence discrete label is added; and if it falls within the range of [0.0, 0.3), a low confidence discrete label is added. The thresholds for each range can be flexibly adjusted according to the scheduling safety requirements of the park's energy system. The discrete labels will be added to the completed data synchronously with the confidence level values ​​and output along with the complete dataset to the optimization problem construction.

[0050] Furthermore, in the method provided in the application embodiment, the constraint boundaries of each device in the optimization model are dynamically adjusted according to the attenuation parameters, and some deterministic constraints in the optimization model are converted into probabilistic constraints according to the confidence information to establish the optimization problem to be solved. This includes: the edge gateway locally maintaining a performance attenuation parameter table for each device, wherein the performance attenuation parameters include the cumulative cycle count and state of health (SOH) of the energy storage battery, the power attenuation rate of the photovoltaic module, the cumulative operating time and energy efficiency ratio reduction coefficient of the air conditioning compressor, and in each scheduling cycle, the edge gateway traverses all devices to be optimized and calculates the actual constraints of each device. Upper limit; when the decay factor of the device to be optimized is lower than the preset threshold, the edge gateway will temporarily remove the corresponding device from the set of decision variables to be optimized, force a switch to the preset safe operation mode, and generate an operation and maintenance alarm; at the same time, based on the confidence level of the supplementary data, the edge gateway will convert the physical constraints with confidence levels lower than the threshold in the optimization model from deterministic constraints to probabilistic constraints; the adjusted upper limit of the actual constraints will be used as the new boundary for each device, the converted probabilistic constraints will be merged with the unconverted deterministic constraints, and the final optimization problem to be solved will be constructed with the objective function of minimizing the park's operating costs and minimizing carbon emissions.

[0051] Specifically, the edge gateway on the park side first retrieves the performance degradation parameter table for each device, which is maintained in real time in local non-volatile storage. This performance degradation parameter table is a full life-cycle operation parameter ledger independently established by the edge gateway for each multi-source heterogeneous energy device included in the park's energy dispatch system. The performance degradation parameters in the ledger include the cumulative cycle count and state of health (SOH) of the energy storage battery, the power degradation rate of the photovoltaic module, and the cumulative operating time and energy efficiency ratio reduction coefficient of the air conditioning compressor. Among them, the cumulative cycle count of the energy storage battery refers to the number of complete charge-discharge cycles completed by the battery since it was put into operation, and the state of health (SOH) refers to the current actual usable capacity of the battery compared with the factory rated capacity. The percentage ratio of capacity is an indicator that quantifies the degree of battery aging. The power degradation rate of photovoltaic modules refers to the ratio of the difference between the current actual maximum output power of the module and the factory-rated peak power to the rated value, which characterizes the long-term degradation of the module's photoelectric conversion efficiency. The cumulative running time of the air conditioner compressor refers to the cumulative effective working time of the compressor since it was put into operation. The energy efficiency ratio reduction coefficient refers to the ratio of the compressor's current actual energy efficiency ratio (EER) to the factory-rated energy efficiency ratio, which characterizes the degradation level of the compressor's cooling efficiency. This parameter table will be updated in real time in each scheduling cycle based on the actual operating feedback data of the equipment in the previous cycle to ensure that the parameters are completely matched with the current operating conditions of the equipment.

[0052] After completing the parameter table reading, the edge gateway traverses all devices to be optimized, calculating the actual constraint upper limit for each device. Devices to be optimized refer to controllable devices included in the park's scheduling system, operating normally, and whose operating parameters can be actively adjusted via scheduling commands. These include energy storage battery clusters, photovoltaic grid-connected inverters, adjustable central air conditioning units, controllable industrial loads, and new energy charging piles. The actual constraint upper limit refers to the maximum adjustable boundary that the device can operate safely and stably under its current performance degradation state. The specific calculation uses standardized formulas for different device types: For energy storage batteries: Actual constraint upper limit for charging and discharging power = Factory rated charging and discharging power × Current SOH × Safety margin The coefficients, where the safety margin coefficient is 0.9~0.95, are used to avoid overload risks under extreme operating conditions; at the same time, the upper and lower limits of the available energy storage capacity are updated synchronously as follows: rated capacity × current SOH × preset charge and discharge depth limit; for photovoltaic modules: the actual upper limit of maximum output constraint = nominal peak power of a single module × number of module strings × (1 - power decay rate) × current ambient temperature correction coefficient; for air conditioning compressors, the actual upper limit of operating load constraint = rated cooling power × energy efficiency ratio reduction coefficient × cumulative running time correction coefficient, where the cumulative running time correction coefficient is adjusted downward in a stepwise manner as the cumulative operating time of the compressor increases, to avoid overload operation of aging equipment.

[0053] After calculating the actual constraint upper limits for all devices, the edge gateway synchronously verifies whether the attenuation factor of each device to be optimized is lower than the preset threshold. The attenuation factor is a normalized index defined for each type of device to quantify its performance degradation. Specifically, the attenuation factor for energy storage batteries is the current State of Health (SOH), for photovoltaic modules it is (1 - power degradation rate), and for air conditioning compressors it is the energy efficiency ratio reduction coefficient. The preset threshold is a safety degradation lower limit pre-set for each type of device based on industry operation and maintenance standards, with values ​​of 60% of the SOH threshold for energy storage batteries, 80% for photovoltaic modules, and 70% for air conditioning compressors. This threshold can be flexibly adjusted according to park operation and maintenance requirements. When the attenuation factor of a device to be optimized falls below the corresponding preset threshold, the edge gateway immediately and temporarily removes the device from the set of decision variables to be optimized. The set of decision variables refers to the set of variables corresponding to the operating parameters of the equipment in the optimization model that can be actively adjusted by scheduling commands. These include energy storage charging and discharging power, air conditioning set temperature, controllable load switching ratio, and output limit of photovoltaic grid connection point. The operating parameters of the removed equipment are no longer used as adjustable variables in the optimization model. At the same time, the equipment is forced to switch to a preset safe operating mode. The preset safe operating mode is a fixed operating mode with no adjustment risk that is pre-set for each type of equipment. This includes switching energy storage batteries to float charging standby mode, photovoltaic modules to MPPT full grid connection mode, and air conditioning compressors to fixed energy-saving operating mode. The edge gateway will generate standardized operation and maintenance alarms with unique equipment numbers, attenuation parameter values, and trigger times, write them to the local operation and maintenance event database, and upload them to the cloud operation and maintenance platform when communication with the cloud is normal.

[0054] While dynamically adjusting equipment constraint boundaries, the edge gateway simultaneously performs constraint type conversion operations based on the confidence level of the completed data. First, it iterates through all physical constraints in the optimization model that rely on the completed data. Physical constraints refer to constraints related to equipment operating parameters, park energy supply and demand balance, and grid connection rules, and are divided into two categories: hard safety boundary constraints that do not rely on completed data and operating condition adaptation constraints that rely on completed data. Then, it compares the confidence level of the completed data with a preset constraint conversion threshold. This threshold is typically 0.7 and can be flexibly adjusted according to the robustness requirements of park scheduling. When the confidence level of the completed data upon which a physical constraint depends is lower than this threshold, the edge gateway... The gateway converts the constraint from a deterministic constraint to a probabilistic constraint. A deterministic constraint is a rigid constraint that must be 100% satisfied during the optimization process, expressed as g(x)≤0 and must be strictly true. The probabilistic constraint here is also called a chance constraint, which is a flexible constraint that allows the constraint to be satisfied under a preset confidence probability, expressed as P(g(x)≤0)≥β, where β is the constraint confidence level, and its value is exactly equal to the confidence level of the corresponding completed data. That is, the lower the confidence level of the completed data, the lower the confidence level that the constraint needs to satisfy. For physical constraints that depend on the confidence level of the completed data being higher than the threshold, and hard safety boundary constraints that do not depend on the completed data, the deterministic constraint remains unchanged. After completing the dynamic adjustment of constraint boundaries and the conversion of constraint types, the edge gateway uses the calculated actual constraint upper limits of each device as the new feasible region boundary of the corresponding decision variables in the optimization model, replacing the traditional fixed factory-rated boundary. Simultaneously, it merges the converted probabilistic constraints with the unconverted deterministic constraints to form a complete constraint set for the optimization model. This constraint set includes four main categories: equipment safe operation constraints, real-time power balance constraints within the park, grid-connected power limit constraints, and carbon emission quota constraints. Subsequently, an objective function is constructed with minimizing park operating costs and carbon emissions as its dual core objectives. This objective function adopts a standardized and reproducible mathematical form, and the dual objective function expression is as follows: ; in The full-cycle scheduling and operation cost of the park is specifically composed of: =Σ(Power purchased by the power grid × Time-of-use price for the corresponding time period) +Σ(Operation and maintenance cost of each equipment unit × Operating power of the corresponding equipment) + Penalty for curtailment of solar power; The total carbon emissions for the entire lifecycle of the park are specifically composed of: =Σ(Power purchased by the power grid × corresponding time period power grid carbon emission factor) +Σ(Park self-consumption of fossil energy × corresponding energy carbon emission factor); The bi-objective function can be transformed into a single-objective optimization problem through a linear weighting method. The weight coefficients can be flexibly adjusted according to the park's priorities for economic and low-carbon objectives. Ultimately, a standardized optimization problem with dynamically updated equipment constraint boundaries, mixed determinism, and probabilistic constraints is formed and output to a quantum heuristic optimization algorithm for solution.

[0055] Furthermore, in the method provided in the application embodiment, a quantum heuristic optimization algorithm is used to solve the optimization problem. During the solution process, the computational complexity of the algorithm is dynamically adjusted according to the real-time computing load of the edge gateway, and control instructions for the current time slot are output. This includes: randomly initializing the positions and velocities of N particles within the feasible space of the decision variables, wherein the feasible space is defined by the dynamic constraint boundary after performance decay parameter adjustment; evaluating the fitness of each particle, where fitness includes the objective function value, deterministic constraint violation penalty, and probabilistic constraint penalty; updating the individual optimal position and the global optimal position based on the fitness of each particle; calculating the diversity index of the current population, and linearly increasing quantum tunneling when the diversity index is lower than a preset diversity threshold. For each particle, a random number is generated. If the random number is less than the quantum tunneling probability, a tunneling operation is performed, and the particle's position is randomly jumped to a position outside the neighborhood of the global optimal position. For particles that do not undergo tunneling, quantum behavior is used to update the position and calculate the potential well characteristic length. If the rate of change of the global optimal fitness for three consecutive iterations is lower than a preset rate of change threshold, a chaotic local search subroutine is triggered. With the current global optimal position as the center, m chaotic perturbation points are generated and the fitness of each perturbation point is evaluated. If a better solution is found, the global optimal position is replaced. This process is repeated until the maximum number of iterations is reached or the convergence condition is met. The position vector of the optimal particle is output as the optimal control sequence, and the first element of the optimal control sequence is taken as the control instruction to be executed in the current time slot.

[0056] Specifically, the edge gateway employs a quantum-heuristic optimization algorithm adapted to the characteristics of edge computing power to perform the solution locally. This algorithm is an improved intelligent optimization algorithm that integrates the quantum mechanical tunneling effect, the probability distribution characteristics of quantum potential wells, and a chaotic traversal search mechanism. Throughout the algorithm's execution, the edge gateway continuously monitors its real-time computing load. The computing load refers to the edge gateway's current CPU utilization, memory utilization, and the proportion of computing power overhead for parallel tasks such as data acquisition, communication interaction, and conflict verification. The algorithm's computational complexity is dynamically adjusted based on the load: when the gateway's real-time computing load exceeds a preset high-load threshold, the core algorithm parameters are simultaneously reduced to lower computing power overhead; when the gateway's real-time computing load falls below a preset low-load threshold, the core algorithm parameters are simultaneously increased to optimize solution accuracy. The first step in the algorithm's solution is population initialization, which involves randomly initializing N particles within the feasible space of the decision variables. Position and velocity, the feasible space of decision variables refers to the range of legal values ​​of all adjustable equipment operating parameters defined by the dynamic constraint boundary after adjustment of equipment performance degradation parameters. Each dimension corresponds to an independent decision variable, including energy storage charging and discharging power, air conditioning unit load setpoint, controllable load switching ratio, photovoltaic grid connection point output limit, etc. The upper and lower limits of each dimension completely match the real-time actual constraint boundary of the corresponding equipment. Particles are the basic optimization units of intelligent optimization algorithms. The position vector of each particle corresponds to a complete set of scheduling control sequences covering all time slots within the entire rolling optimization time window. Among them, the vector dimension = number of decision variables × number of optimization window time slots. The velocity of the particle is a vector with the same dimension as the position vector. During initialization, the particle position is randomly generated in a uniform distribution within the feasible domain of each dimension, the velocity is initialized to 0, and the initial default value of the particle number N is 30, which can be dynamically adjusted within the range of 10~50 according to the real-time load calculated by the gateway.

[0057] Then, the fitness of each particle is evaluated. Fitness is a comprehensive index that quantitatively measures the quality of the scheduling scheme corresponding to the particle. The smaller the fitness value, the better the scheme. The calculation adopts the penalty function method. Fitness Fit = Weighted normalized value of biobjective function + Penalty term for violation of deterministic constraints + Penalty term for violation of probabilistic constraints; The weighted normalized value of the dual objective function is the value obtained by linear weighting and maximum / minimum normalization of the dual objective functions of park operating cost and carbon emission. The weight coefficients can be flexibly set according to the priority of the park's economic and low-carbon objectives. The weights can be 0.6 for operating cost and 0.4 for carbon emission. The deterministic constraint violation penalty is a high rigid penalty. As long as the corresponding scheme of the particle violates any deterministic constraint, the penalty is calculated by multiplying the square of the violation by the penalty coefficient, which can be calculated by taking the value 1e6. This ensures that particles that violate rigid safety constraints cannot become the optimal solution. The probabilistic constraint violation penalty is a gradient flexible penalty. The penalty is calculated by multiplying the absolute value of the difference between the constraint violation probability and the required confidence level by the gradient penalty coefficient, which can be calculated by taking the value 1e3. This adapts to the input uncertainty brought about by the supplementary data.

[0058] The algorithm then updates the individual optimal position p of each particle based on the comparison between its current fitness and the historical best fitness. best Simultaneously, the global optimal position g is updated based on the optimal fitness of the entire population. best The optimal position p of the individual here best This refers to the position vector with the minimum fitness found by a single particle throughout all iterations. In each iteration, if the fitness of the current particle is less than its p-value... best The corresponding fitness, i.e., updating p best Given the current particle position, the global optimal position g is... best This refers to the position vector with the minimum fitness of the entire particle population found throughout all iterations. In each iteration, if the optimal position vector for the entire population is p... best The corresponding fitness is less than the current g. best The fitness of g, i.e., updating g best For the optimal p best .

[0059] The algorithm then calculates the diversity index of the current population. The diversity index is an indicator that quantifies the dispersion of all particles in the feasible space, and is calculated using the standardized mean Euclidean distance. ; in It is the position vector of the i-th particle. It is the mean vector of the positions of all particles in the population. 2 is the Euclidean distance norm. A smaller value indicates a more concentrated particle distribution, lower population diversity, and a greater likelihood that the algorithm will get trapped in a local optimum problem with premature convergence. N is the total number of particles set during initialization, with an initial default value of 30. The algorithm will calculate the current... With preset diversity threshold Compare them. The value is taken as 20% of the initial population diversity. Below When the quantum tunneling probability increases linearly, this probability is a global search enhancement parameter designed by referencing the tunneling effect in quantum mechanics where microscopic particles can overcome energy barriers. The initial base tunneling probability P0 = 0.05, and the linear increase formula is as follows: ; in The growth coefficient can be 0.3. Then, a uniformly distributed random number between 0 and 1 is generated for each particle in the population. If this random number is less than the current quantum tunneling probability... If the particle is in a certain position, a tunneling operation is performed, which involves randomly jumping the particle's position vector to the globally optimal position g. best Within the feasible space outside the neighborhood of g, where the neighborhood is defined as denoted by g. best The region centered on the feasible domain width within 10% of each dimension; For particles that have not undergone tunneling, position iteration is performed using the quantum behavior position update rule, and the characteristic length of the potential well is calculated simultaneously. The quantum behavior position update is designed based on the probability distribution of quantum particles in the potential well, eliminating the need for velocity iteration and resulting in faster convergence and stronger global search capabilities. The characteristic length of the potential well is a parameter describing the range of motion of a particle within the quantum potential well formed by the individual optimum and the global optimum, determining the particle's local search step size. Its calculation formula is as follows: ; in A uniformly random number between 0 and 1. This represents the optimal position for the current particle. To be the globally optimal position For absolute value operations, the corresponding position update formula is: ; in A uniformly random number between 0 and 1. The particle number is randomly selected with a 50% probability to ensure that the particle performs a probabilistic search within the quantum potential well, further enhancing the algorithm's optimization capability. During the iteration process, the algorithm continuously monitors the convergence state of the global optimal solution. If the rate of change of the global optimal fitness is lower than a preset rate of change threshold for three consecutive iterations, a chaotic local search subroutine is triggered. The rate of change of the global optimal fitness is calculated using a standardized formula. ; in To determine the globally optimal fitness in the t-th iteration, a preset rate of change threshold is defined. The typical value is 1e-4, when < When the algorithm has entered the late convergence stage, the global optimal solution is close to the theoretical optimal value. The chaotic local search subroutine is a local optimization mechanism designed using the randomness, ergodicity, and high sensitivity to initial values ​​of chaotic sequences. It is implemented using Logistic chaotic mapping, specifically: using the current global optimal position... Centered on, firstly The components of each dimension are normalized to the interval [0,1] as the initial values ​​for the chaotic mapping, using the Logistic mapping formula: ; Generate m chaotic sequence points, where The chaos control parameter is set to a fixed value of 4 to ensure the mapping is in a completely chaotic state. m represents the number of chaotic perturbation points, with an initial default value of 10, which can be dynamically adjusted within the range of 5 to 20 based on the real-time computational load of the gateway. Subsequently, the generated m chaotic sequence points are denormalized to the feasible region of each decision variable, resulting in m chaotic perturbation points. The fitness of each chaotic perturbation point is evaluated one by one. If there exists a fitness better than the current... If the perturbation point is not found, then gbest is replaced with the optimal perturbation point, thereby further improving the accuracy of the optimal solution in the later stage of convergence and preventing the algorithm from missing better local optima. Finally, the algorithm iteratively executes the entire process of fitness evaluation, optimal position update, population diversity monitoring, quantum tunneling operation, quantum behavior position update, and chaotic local search until the preset maximum number of iterations is reached or the convergence condition is met. The initial default value of the maximum number of iterations is 50, which can be dynamically adjusted within the range of 20 to 100 according to the real-time computing load of the gateway. The convergence condition is that the rate of change of the global optimal fitness is less than 1e-5 for 5 consecutive iterations. After the iteration terminates, the algorithm outputs the position vector of the optimal particle as the optimal control sequence. The optimal control sequence is a set of multi-slot scheduling instructions for the entire rolling optimization time window. Only the first element of the optimal control sequence is taken as the control instruction to be executed in the current slot.

[0060] Furthermore, in the method provided in the application embodiment, conflict verification is performed on the control command. After the verification is passed, the command is sent to the corresponding device for execution. This includes: the edge gateway has a built-in conflict resolution module to verify whether the control command to be executed violates the internal consistency constraints of the device, the coupling constraints between devices, and the security procedure constraints; if the verification is passed, the command is sent to the device for execution according to the corresponding industrial bus protocol; if a conflict is found during the verification, the command is downgraded to the next best feasible command according to the preset priority rules until the conflict is eliminated, and the corresponding command is sent for execution. After the command is sent, the command value, the actual execution value, and the execution timestamp are written to the local time series database; if the command fails to send, a retry mechanism is triggered. If the command still fails, the corresponding device is marked as offline and switched to a safe operation mode.

[0061] Specifically, the edge gateway on the park side initiates verification and execution control through its built-in conflict resolution module. This module is a pre-fixed embedded functional unit within the edge gateway, operating synchronously with the scheduling cycle. It includes a standardized verification rule library, hierarchical degradation processing logic, and anomaly handling procedures. First, it breaks down the multi-dimensional control commands to be executed into single-device control sub-commands uniquely bound to each target device. Each sub-command carries a unique hardware identifier for the device, the type of control parameters, the target execution value, and a precise execution timestamp. Then, the module checks whether the control commands to be executed violate three categories of constraints in an irreversible priority order: single-device security verification, system-level verification, and compliance red-line verification. The first category is internal device consistency constraints. These constraints are the internal logical mutual exclusion rules that a single energy device must meet for its operation, the real-time safe operation boundary after dynamic adjustment of performance degradation parameters, and the operation and maintenance specifications. Specific verification includes whether there are logical conflicts arising from the simultaneous issuance of mutually exclusive operation modes such as energy storage charging and discharging, and air conditioning cooling and heating; whether the control parameters exceed the actual upper and lower limits of the constraints; whether the device start-up and shutdown operations meet the preset minimum start-up and shutdown interval requirements; and whether the device is under maintenance. The first category is issuing remote control commands to locked devices. The second category is inter-device coupling constraints. These are the collaborative operation rules that must be met between multiple devices within the park that have energy supply and demand correlations, grid interaction bindings, and distribution network capacity sharing. These include whether the total input and output power of the park's grid connection point exceeds the upper and lower limits of the grid-approved grid connection power; whether the total load of the park's distribution transformers exceeds its rated capacity limit; whether the total output of the chiller and heat source units matches the total load demand of the terminal air conditioning equipment; whether the superposition of photovoltaic output and energy storage charging and discharging power exceeds the safe current carrying capacity limit of the distribution network line; and the total controllable load. The third category is safety regulation constraints. Safety regulation constraints are mandatory and rigid red line constraints that must comply with national power safety industry standards and park emergency supply management regulations. Specifically, this includes whether shutdown or power restriction orders are issued for first-level security loads such as fire protection, emergency lighting, and security systems; whether control orders are issued for equipment that has triggered fault alarms; whether illegal orders to feed power back to the grid are issued during grid faults; and whether the supply dispatch regulations during major events in the park are violated. It is necessary to ensure that dispatch orders fully comply with compliance requirements and the red line of safe operation.

[0062] After completing the full-dimensional three-level verification, if the control command to be executed fully complies with all three types of constraints, the verification is deemed successful. The edge gateway immediately encapsulates the control sub-commands according to the industrial bus protocol corresponding to the target device and sends them to the corresponding energy device for execution. The industrial bus protocol here is a mainstream industrial protocol adapted to different types of multi-source heterogeneous energy devices. If any constraint violation is found in the command during the verification process, the edge gateway immediately activates the preset priority rules to downgrade the currently conflicting command to the next best feasible command. The preset priority rules are pre-set and strictly follow the principle of prioritizing safety over economy, supply security over optimization, and critical load over adjustable load in the command classification and downgrading. The priorities are divided from high to low into the non-adjustable rigid red line layer, the secondary rigid safety constraint layer, and the semi-flexible system. The system consists of four layers: a system coupling constraint layer, a fully flexible adjustable parameter layer, and a semi-flexible system coupling constraint layer. Among these, the safety regulations and first-level security load supply requirements of the rigid red line layer are absolutely not allowed to be adjusted. The internal consistency constraints of the equipment in the secondary rigid safety constraint layer can only be slightly adjusted within the safety boundary. The distribution network capacity and grid connection limit of the semi-flexible system coupling constraint layer can be finely adjusted within the compliance range. The non-core controllable load switching ratio, energy storage charging and discharging power timing allocation, air conditioning set temperature fine-tuning, and photovoltaic curtailment ratio adjustment of the fully flexible adjustable parameter layer are the core objects of degradation adjustment. When degradation is executed, it starts from the lowest priority fully flexible adjustable parameter and adjusts the parameter step by step according to the preset fixed step size. After each parameter adjustment is completed, a full-dimensional three-level verification is re-executed until the instruction conflict is completely eliminated. Then, a feasible degradation instruction that has passed the verification is issued, and the entire process log of degradation adjustment is written to the local event database.

[0063] During the command issuance process, the edge gateway synchronously monitors the command issuance status and device response feedback in real time. If there is no response after command issuance or the device returns an execution failure, a preset retry mechanism is triggered. The retry mechanism adopts a fixed interval and limited number of retransmission rules to avoid invalid retransmissions consuming gateway computing power and communication resources. If the command still fails after reaching the maximum number of retry attempts, the edge gateway will immediately mark the corresponding device as offline and force the device to switch to a preset safe operation mode. The preset safe operation modes specifically include switching the energy storage battery to float charging standby mode, switching the photovoltaic modules to MPPT full grid-connected mode, switching the air conditioning compressor to fixed energy-saving operation mode, and controlling the load. Switching to local normal operation mode and exiting remote scheduling avoids the risk of operational loss of control due to device disconnection, ensuring the basic stability of the park's energy system. Regardless of whether the instruction is issued and executed normally, executed after degradation, or switched to a safe mode after issuance failure, the edge gateway will collect the actual operation feedback data of the target device after the execution cycle of the current time slot ends. It will synchronously write the entire link information, including the instruction issuance value, actual device execution value, accurate execution timestamp, device unique identifier, execution status mark, conflict verification and degradation adjustment log, into the local time-series database. The local time-series database is a database deployed locally by the edge gateway, specifically used to store timestamped device operation and scheduling data.

[0064] In summary, the quantum-inspired optimization-based campus-level energy edge gateway scheduling method provided in this application has the following technical effects: By deploying edge gateways on the park side, rolling data collection of multi-source heterogeneous energy equipment operation is achieved. For data sources that are interrupted or exceed limits, dynamic completion is completed through a hybrid model of time series decomposition and lightweight gradient booster, and confidence information is generated synchronously. The constraint boundary of the optimization model is dynamically adjusted based on the equipment performance degradation parameters. At the same time, the deterministic constraints corresponding to low reliability data are converted into probabilistic constraints by combining the confidence of the completed data. An optimization problem to be solved is constructed that fits the real-time operating conditions of the equipment and the reliability of the data. The optimization solution is completed by a quantum heuristic optimization algorithm that can dynamically adjust the computational complexity according to the real-time computing load of the edge gateway. The output control commands are issued and executed after conflict verification, and a rolling optimization control closed loop is formed through a sliding time window. The online update and autonomous scheduling of the prediction model are completed locally, realizing the highly reliable, continuous and stable optimized operation of the scheduling system, and achieving the dual optimization effect of minimizing operating costs and carbon emissions.

[0065] Example 2, based on the same inventive concept as the quantum-inspired optimization-based campus-level energy edge gateway scheduling method in the aforementioned examples, such as... Figure 2 As shown, this application provides a campus-level energy edge gateway scheduling system based on quantum-inspired optimization. The system includes: The acquisition module 11 is used for S1: It continuously acquires real-time operating data of multi-source heterogeneous energy devices through an edge gateway deployed on the park side. When any data source is interrupted or exceeds its limit, it initiates a prediction model for dynamic completion and generates confidence information characterizing the reliability of the completion. The reading module 12 is used for S2: Based on the acquired and completed data, it reads the device status records stored locally. The status records contain the performance degradation parameters of each device. It dynamically adjusts the constraint boundaries of each device in the optimization model according to the degradation parameters, and converts some deterministic constraints in the optimization model into probabilistic constraints according to the confidence information, establishing the optimization problem to be solved. The calculation module 13 is used for S3: It uses a quantum heuristic optimization algorithm to solve the optimization problem. During the solution process, it dynamically adjusts the computational complexity of the algorithm according to the real-time computing load of the edge gateway and outputs the control command for the current time slot. The execution module 14 is used for S4: It performs conflict verification on the control command. After the verification is successful, it sends the command to the corresponding device for execution and feeds back the execution result to the local database. The optimization module 15 is used for S5: It slides the time window forward by one step and returns to S1, forming a rolling optimization control closed loop.

[0066] Furthermore, the system is also used to implement the following functions: The system continuously monitors the communication status between the edge gateway and the cloud. When a communication interruption is detected, it automatically switches to local autonomous mode and updates the prediction model online locally using recent measured data. The updated prediction model is not transmitted to the cloud.

[0067] Furthermore, the acquisition module 11 is also used to perform the following steps: the edge gateway polls the data sources of each multi-source heterogeneous energy device at a fixed sampling period. If no valid data packet is received for more than two sampling periods, it is determined that the communication is interrupted; if the received value exceeds the preset physical range of the corresponding device, it is determined that the value exceeds the limit; for data sources determined to be interrupted or exceeding the limit, a dynamic completion process is started, and the device data is predicted by a prediction model to obtain the predicted value and the corresponding data completion confidence level; the predicted value is used to perform dynamic completion, and the completed data is labeled with confidence level to generate confidence information.

[0068] Furthermore, the acquisition module 11 is also used to perform the following steps: acquiring the valid data sequence of the data source marked as interrupted or out of bounds within a preset historical time window; decomposing the valid data sequence into trend components, periodic components, and residual components using a time series decomposition method, wherein the prediction model is a hybrid model based on time series decomposition and lightweight gradient boosting machine; using the periodic component to extrapolate and obtain the periodic prediction value at the current missing moment; inputting the residual component into a pre-trained lightweight gradient boosting machine regression model, using a preset number of residual values ​​before the current missing moment and the environmental feature parameters at the current moment as input features, and outputting the residual prediction value; adding the periodic prediction value and the residual prediction value to obtain the final prediction value.

[0069] Furthermore, the acquisition module 11 is also used to perform the following steps: record the prediction residuals of the prediction model for historical valid data points using a sliding window, calculate the standard deviation of the prediction residuals within the sliding window, and obtain the historical residual standard deviation; take the variance component of the residual prediction value corresponding to the imputed value at the current missing time as the prediction residual estimate; and perform prediction value confidence analysis based on the comparison result between the historical residual standard deviation and the prediction residual estimate to obtain the data imputed confidence.

[0070] Furthermore, the acquisition module 11 is also used to perform the following steps: compare the standard deviation of the residual prediction value with the standard deviation of the historical residual, and calculate the ratio; based on the ratio, the edge gateway converts the ratio into a confidence value between 0 and 1 through a monotonically decreasing mapping relationship, where the larger the value, the more reliable the completion result is; the edge gateway adds a discrete identifier to the completed data according to the specific value of the confidence.

[0071] Furthermore, the reading module 12 is also used to perform the following steps: the edge gateway locally maintains a performance degradation parameter table for each device, the performance degradation parameters including the cumulative cycle count and state of health (SOH) of the energy storage battery, the power degradation rate of the photovoltaic module, the cumulative running time and energy efficiency ratio reduction coefficient of the air conditioning compressor; in each scheduling cycle, the edge gateway traverses all devices to be optimized and calculates the actual constraint upper limit of each device; when the degradation factor of the device to be optimized is lower than the preset threshold, the edge gateway temporarily removes the corresponding device from the set of decision variables to be optimized, forces a switch to the preset safe operation mode, and generates an operation and maintenance alarm; at the same time, based on the confidence level of the supplementary data, the edge gateway converts the physical constraints with confidence levels lower than the threshold in the optimization model from deterministic constraints to probabilistic constraints; the adjusted actual constraint upper limit is used as the new boundary for each device, the converted probabilistic constraints are merged with the unconverted deterministic constraints, and the final optimization problem to be solved is constructed with the objective function of minimizing park operating costs and minimizing carbon emissions.

[0072] Furthermore, the calculation module 13 is also used to perform the following steps: randomly initialize the positions and velocities of N particles within the feasible space of the decision variables, wherein the feasible space is defined by the dynamic constraint boundary after performance decay parameter adjustment; evaluate the fitness of each particle, the fitness including objective function value, deterministic constraint violation penalty, and probabilistic constraint penalty; update the individual optimal position and global optimal position according to the fitness of each particle; calculate the diversity index of the current population, and when the diversity index is lower than a preset diversity threshold, linearly increase the quantum tunneling probability; generate a random number for each particle; if the random number is less than the quantum tunneling probability, perform tunneling. The operation involves randomly jumping the particle's position to a location outside the neighborhood of the global optimal position. For particles that do not tunnel, quantum behavior is used to update their positions and calculate the potential well characteristic length. If the rate of change of the global optimal fitness for three consecutive iterations is lower than a preset rate of change threshold, a chaotic local search subroutine is triggered. Centered on the current global optimal position, m chaotic perturbation points are generated, and the fitness of each perturbation point is evaluated. If a better solution is found, the global optimal position is replaced. This process is repeated until the maximum number of iterations is reached or the convergence condition is met. The position vector of the optimal particle is output as the optimal control sequence, and the first element of the optimal control sequence is taken as the control instruction to be executed in the current time slot.

[0073] Furthermore, the execution module 14 is also used to perform the following steps: the edge gateway has a built-in conflict resolution module to verify whether the control instruction to be executed violates the internal consistency constraints of the device, the coupling constraints between devices, and the security procedure constraints; if the verification passes, the instruction is sent to the device for execution according to the corresponding industrial bus protocol; if a conflict is found during the verification, the instruction is downgraded to the next best feasible instruction according to the preset priority rules until the conflict is eliminated, and the corresponding execution is sent. After the sending is completed, the instruction value, the actual execution value, and the execution timestamp are written to the local time series database; if the sending fails, a retry mechanism is triggered. If it still fails, the corresponding device is marked as offline and switched to a safe operation mode.

[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A campus-level energy edge gateway scheduling method based on quantum-inspired optimization, characterized in that, include: S1: Real-time operating data of multi-source heterogeneous energy devices are collected in a rolling manner through the edge gateway deployed on the park side. When any data source is interrupted or exceeds the limit, the prediction model is started to perform dynamic completion and generate confidence information that characterizes the reliability of the completion. S2: Based on the collected and completed data, read the device status records stored locally. The status records contain the performance degradation parameters of each device. Adjust the constraint boundaries of each device in the optimization model according to the degradation parameters. Convert some deterministic constraints in the optimization model into probabilistic constraints according to the confidence information to establish the optimization problem to be solved. S3: The optimization problem is solved using a quantum heuristic optimization algorithm. During the solution process, the computational complexity of the algorithm is dynamically adjusted according to the real-time computing load of the edge gateway, and the control command for the current time slot is output. S4: Perform conflict verification on the control command. If the verification passes, send it to the corresponding device for execution and feed the execution result back to the local database. S5: Slide the time window forward by one step, return to S1, and form a rolling optimization control closed loop.

2. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 1, characterized in that, The execution of S1-S4 also includes: The system continuously monitors the communication status between the edge gateway and the cloud. When a communication interruption is detected, it automatically switches to local autonomous mode and updates the prediction model online locally using recent measured data. The updated prediction model is not transmitted to the cloud.

3. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 1, characterized in that, Real-time operational data from multi-source heterogeneous energy devices is collected on a rolling basis through edge gateways deployed on the park side. When any data source is interrupted or exceeds its limits, a predictive model is activated to dynamically complete the data collection and generate confidence information characterizing the reliability of the completion, including: The edge gateway polls the data sources of each multi-source heterogeneous energy device at a fixed sampling period. If no valid data packet is received for more than two sampling periods, it is determined that the communication is interrupted. If the received value exceeds the preset physical range of the corresponding device, it is determined to be a value exceeding the limit; For data sources that are determined to be interrupted or exceed limits, a dynamic completion process is initiated. The device data is predicted using a prediction model to obtain the predicted values ​​and the corresponding data completion confidence levels. The predicted values ​​are used for dynamic completion, and the completed data is labeled with confidence level to generate confidence information.

4. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 3, characterized in that, Predictive models are used to forecast device data, yielding predicted values, including: Obtain the valid data sequence of the data source marked as interrupted or out of limit within a preset historical time window, and use the time series decomposition method to decompose the valid data sequence into trend component, periodic component and residual component. The prediction model is a hybrid model based on time series decomposition and lightweight gradient booster. The periodic prediction value at the current missing moment is obtained by extrapolating the periodic components. The residual components are input into a pre-trained lightweight gradient booster regression model, which uses a preset number of residual values ​​before the current missing time step and environmental feature parameters at the current time step as input features to output the residual prediction value. The predicted period value is added to the predicted residual value to obtain the final predicted value.

5. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 4, characterized in that, Obtain the corresponding data to complete the confidence level, including: The prediction residuals of the prediction model for historical valid data points are recorded using a sliding window. The standard deviation of the prediction residuals within the sliding window is calculated to obtain the standard deviation of the historical residuals. The variance component of the predicted residual corresponding to the imputed value at the current missing time is used as the estimated value of the predicted residual; Based on the comparison between the historical residual standard deviation and the predicted residual estimate, a confidence analysis of the predicted values ​​is performed to obtain the data completion confidence level.

6. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 5, characterized in that, Based on the comparison between the historical residual standard deviation and the estimated predicted residual, a confidence analysis of the predicted values ​​is performed to obtain the data completion confidence level, including: Compare the standard deviation of the predicted residuals with the standard deviation of the historical residuals and calculate the ratio. Based on the ratio, the edge gateway converts the ratio into a confidence value between 0 and 1 through a monotonically decreasing mapping relationship. The larger the value, the more reliable the completion result. The edge gateway adds a discrete identifier to the completed data according to the specific confidence value.

7. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 1, characterized in that, The constraint boundaries of each device in the optimization model are dynamically adjusted according to the attenuation parameters, and some deterministic constraints in the optimization model are converted into probabilistic constraints according to the confidence information, thus establishing the optimization problem to be solved, including: The edge gateway locally maintains a performance degradation parameter table for each device. The performance degradation parameters include the cumulative cycle count and state of health (SOH) of the energy storage battery, the power degradation rate of the photovoltaic module, and the cumulative running time and energy efficiency ratio reduction coefficient of the air conditioning compressor. In each scheduling cycle, the edge gateway traverses all devices to be optimized and calculates the actual constraint upper limit for each device. When the attenuation factor of the device to be optimized is lower than the preset threshold, the edge gateway will temporarily remove the corresponding device from the set of decision variables to be optimized, force it to switch to the preset safe operation mode, and generate an operation and maintenance alarm. At the same time, the edge gateway converts physical constraints with confidence levels below a threshold in the optimization model from deterministic constraints to probabilistic constraints based on the confidence level of the supplementary data. The adjusted upper limit of the actual constraints is used as the new boundary for each device. The transformed probabilistic constraints are merged with the untransformed deterministic constraints. The final optimization problem to be solved is constructed with the objective function of minimizing the park's operating costs and carbon emissions.

8. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 1, characterized in that, A quantum-inspired optimization algorithm is used to solve the optimization problem. During the solution process, the computational complexity of the algorithm is dynamically adjusted according to the real-time computing load of the edge gateway, and control commands for the current time slot are output, including: The positions and velocities of N particles are randomly initialized within the feasible space of the decision variables, wherein the feasible space is defined by the dynamic constraint boundary after adjustment of the performance decay parameter. Each particle undergoes a fitness evaluation, which includes the objective function value, penalties for violating deterministic constraints, and penalties for probabilistic constraints. Update the individual optimal position and the global optimal position based on the fitness of each particle; Calculate the diversity index of the current population. When the diversity index is lower than the preset diversity threshold, linearly increase the quantum tunneling probability and generate a random number for each particle. If the random number is less than the quantum tunneling probability, perform a tunneling operation and randomly jump the particle position to a position outside the neighborhood of the global optimal position. For particles that do not tunnel, quantum behavior is used to update the position and calculate the characteristic length of the potential well; If the rate of change of the global optimal fitness in three consecutive iterations is lower than the preset rate of change threshold, the chaotic local search subroutine is triggered. Taking the current global optimal position as the center, m chaotic perturbation points are generated and the fitness of each perturbation point is evaluated. If there is a better solution, the global optimal position is replaced. Repeat until the maximum number of iterations is reached or the convergence condition is met. Output the position vector of the optimal particle as the optimal control sequence, and take the first element of the optimal control sequence as the control instruction to be executed in the current time slot.

9. The campus-level energy edge gateway scheduling method based on quantum-inspired optimization according to claim 1, characterized in that, The control commands are subjected to conflict checking. If the check passes, the commands are sent to the corresponding devices for execution, including: The edge gateway has a built-in conflict resolution module to verify whether the control command to be executed violates the internal consistency constraints of the device, the coupling constraints between devices, and the security procedure constraints. If the verification passes, the command is sent to the device for execution according to the corresponding industrial bus protocol; If a conflict is found during the verification, the instruction is downgraded to the next best feasible instruction according to the preset priority rules until the conflict is eliminated. Then, the corresponding execution is issued. After the execution is completed, the instruction value, the actual execution value and the execution timestamp are written to the local time series database. If the delivery fails, a retry mechanism will be triggered. If it still fails, the corresponding device will be marked as offline and switched to safe operating mode.

10. A campus-level energy edge gateway scheduling system based on quantum-inspired optimization, characterized in that, The system is used to execute the campus-level energy edge gateway scheduling method based on quantum-inspired optimization as described in any one of claims 1-9, and the system includes: The acquisition module is used for S1: It continuously collects real-time operating data of multi-source heterogeneous energy devices through the edge gateway deployed on the park side. When any data source is interrupted or exceeds the limit, it starts the prediction model to perform dynamic completion and generates confidence information that characterizes the reliability of the completion. The reading module is used for S2: based on the collected and completed data, it reads the device status records stored locally. The status records contain the performance degradation parameters of each device. Based on the degradation parameters, it dynamically adjusts the constraint boundaries of each device in the optimization model, and converts some deterministic constraints in the optimization model into probabilistic constraints based on the confidence information, and establishes the optimization problem to be solved. The computing module, used in S3, employs a quantum-inspired optimization algorithm to solve the optimization problem. During the solution process, it dynamically adjusts the computational complexity of the algorithm based on the real-time computing load of the edge gateway and outputs control commands for the current time slot. The execution module is used in S4 to: perform conflict verification on the control commands, and after the verification is successful, send them to the corresponding devices for execution and feed back the execution results to the local database; The optimization module, used in S5, slides the time window forward by one step and returns to S1, forming a rolling optimization control closed loop.

Images