Intelligent park source network load storage and charging integrated scheduling method based on AI

Abstract

The invention discloses an AI-based intelligent park source network load storage and charging integrated scheduling method, and relates to the technical field of power grid scheduling, and the method comprises the steps: carrying out the multi-modal fusion and semantic packaging of multi-source energy process data, and outputting a park standardized data package; inputting the park standardized data packet into a pre-trained digital twinborn model for multi-energy flow coupling state simulation, and generating a multi-energy flow collaborative diagnosis report; performing dynamic balance and strategy matching on the real-time electricity price information and the multi-energy flow collaborative diagnosis report to form a collaborative optimization scheduling scheme; carrying out distributed issuing and execution on the collaborative optimization scheduling scheme, and obtaining scheduling execution effect data; and analyzing and comparing the performance index deviation of the collaborative optimization scheduling scheme and the scheduling execution effect data, and outputting an optimization scheduling strategy. According to the invention, through deep coupling of multi-energy flow collaborative diagnosis and a distributed negotiation mechanism, accurate perception, collaborative decision and continuous optimization of intelligent park source network load storage and charging integrated scheduling are realized.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching technology, and in particular to an AI-based intelligent park integrated dispatching method for power generation, grid, load, storage and charging. Background Technology

[0002] In recent years, smart park energy management has been undergoing a paradigm shift from single-device monitoring to multi-energy flow collaborative optimization. With the large-scale deployment of distributed photovoltaics, energy storage, and electric vehicle charging piles, the integration of power generation, grid, load, storage, and charging has become the core architecture of smart park energy. In the field of power grid dispatching technology, static models based on historical data are commonly used for energy dispatching, combined with fixed threshold mechanisms to achieve basic control, forming a traditional method path of data acquisition, model calculation, and command issuance. The introduction of digital twin methods has further improved the accuracy of power grid dispatching simulation, supported multi-energy flow coupled state analysis, and formed a methodological framework centered on data acquisition, model construction, and centralized optimization, providing preliminary intelligent support for smart park energy dispatching.

[0003] However, existing methods have shortcomings in dynamic collaborative optimization and process constraint embedding. Traditional optimization algorithms struggle to efficiently handle the complex game relationships between multiple devices and multiple objectives, resulting in insufficient global optimality of scheduling strategies. Static models and fixed threshold mechanisms cannot adapt to rapid changes in the park's operating status, making it difficult to achieve real-time dynamic balance of "source-grid-load-storage-charging." The lack of embedded processing of rigid constraints on production processes makes it difficult to balance energy scheduling schemes with production continuity. Furthermore, the absence of a closed-loop optimization mechanism of diagnosis-control-verification hinders the simultaneous improvement of energy economy and reliability in smart parks. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an AI-based intelligent park source-grid-load-storage-charging integrated scheduling method to address the shortcomings in dynamic collaborative optimization and process constraint embedding.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an AI-based integrated scheduling method for energy sources, grid, load, storage, and charging in smart industrial parks. The method includes: collecting multi-source energy process data from the smart park; performing multi-modal fusion and semantic encapsulation on the multi-source energy process data to output standardized data packets for the park; inputting the standardized data packets into a pre-trained digital twin model for multi-energy flow coupling state simulation to output multi-energy flow coupling status data; evaluating performance indicators and predicting load on the multi-energy flow coupling status data; integrating and generating a multi-energy flow collaborative diagnostic report; acquiring real-time electricity price information; dynamically balancing and matching the real-time electricity price information and the multi-energy flow collaborative diagnostic report through a distributed negotiation mechanism to output preliminary scheduling instructions; performing process verification and energy efficiency optimization on the preliminary scheduling instructions to form a collaboratively optimized scheduling scheme; distributing and executing the collaboratively optimized scheduling scheme in a distributed manner; monitoring the scheduling effect of the collaboratively optimized scheduling scheme and acquiring scheduling execution effect data; analyzing and comparing the performance indicator deviations between the collaboratively optimized scheduling scheme and the scheduling execution effect data; outputting performance indicator deviation data; optimizing and adjusting the performance indicator deviation data to output an optimized scheduling strategy.

[0008] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method of the present invention, the specific steps for outputting standardized park data packets are as follows:

[0009] Perform format conversion and time-series alignment on multi-source energy process data to output unified spatiotemporal reference data;

[0010] The coupling and correlation relationships between various modal data in the unified spatiotemporal reference data are analyzed, and feature-level fusion is performed to form multimodal fused data;

[0011] The multimodal fusion data is tagged, encapsulated, and packaged in a standardized manner to generate a standardized data package for the park.

[0012] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method of the present invention, the specific steps for outputting multi-energy flow coupling situation data are as follows:

[0013] The standardized data package of the park is input into the pre-trained digital twin model for digital simulation, and multi-energy flow simulation data is output.

[0014] Extract the coupling and interaction features between energy flows in the multi-energy flow simulation data and integrate them to form multi-energy flow dynamic interaction parameters;

[0015] The dynamic interaction parameters of multi-energy flow are integrated and structured to output multi-energy flow coupled situational data.

[0016] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method described in this invention, the specific steps of evaluating performance indicators and predicting loads from multi-energy flow coupling situation data, and integrating and generating a multi-energy flow collaborative diagnostic report, are as follows:

[0017] Perform quantitative evaluation of performance indicators on multi-energy flow coupling situation data and output performance indicator evaluation data;

[0018] Acquire historical energy process operation data, combine multi-energy flow coupling situation data with historical energy process operation data to extrapolate load trends, and output load forecast data;

[0019] The performance index evaluation data and load forecast data are used to diagnose energy efficiency bottlenecks and conduct correlation analysis of operational risks, and then integrated to generate a multi-energy flow collaborative diagnostic report.

[0020] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method described in this invention, the specific steps of performing load trend extrapolation by combining multi-energy flow coupling situation data with historical energy process operation data to output load forecast data are as follows:

[0021] Time-series feature extraction is performed on multi-energy flow coupling situation data and historical energy process operation data to generate load time-series feature data;

[0022] By using pattern matching to deduce the mapping relationship between historical operating patterns and real-time status in load time series characteristic data, load change trajectory data is output.

[0023] The confidence level of the load change trajectory data is calibrated and the error is corrected to output load forecast data.

[0024] As a preferred embodiment of the AI-based intelligent park power generation, grid, load, storage, and charging integrated scheduling method of the present invention, the specific steps of outputting the preliminary scheduling instruction are as follows:

[0025] Semantically align real-time electricity price information with multi-energy flow collaborative diagnostic reports to generate negotiation input packages;

[0026] The negotiation input packets are distributed according to the responsibility domain of the smart park, and the proposals are constructed from the negotiation input packets through a distributed negotiation mechanism to output the initial scheduling proposal;

[0027] Initial scheduling proposals are exchanged through a communication network, and consistency constraints are verified on the exchanged initial scheduling proposals to generate a consistent scheduling plan.

[0028] The consistency scheduling plan is verified for safety boundaries and dynamically adjusted for power balance, and preliminary scheduling instructions are output.

[0029] As a preferred embodiment of the AI-based intelligent park integrated scheduling method for power generation, grid, load, storage, and charging as described in this invention, the following steps are taken: distributing the negotiation input packets according to the responsibility domain of the intelligent park, constructing proposals from the negotiation input packets through a distributed negotiation mechanism, and outputting an initial scheduling proposal.

[0030] The real-time electricity price information and multi-energy flow collaborative diagnostic report in the negotiated input package are converted into a set of operational feasibility space descriptions.

[0031] The feasible space description set is decomposed according to the responsibility domain to generate a set of responsibility domain runtime subspaces;

[0032] Each operational constraint description item in the operational subspace of the responsibility domain is verified, and operational constraint description items that are inconsistent with the process rigid constraints, equipment availability status and safety boundaries in the multi-energy flow collaborative diagnostic report are removed. The items are then summarized and encapsulated into a responsible domain operational subspace.

[0033] The boundary descriptions of the subspaces that can be committed to operation by each responsibility domain are exchanged through a communication network. The exchanged boundary descriptions are then overlaid and compared with the boundary descriptions of the subspaces that can be committed to operation by the current responsibility domain to generate the intersection of the multi-responsibility domain operation spaces.

[0034] Consistent boundaries are solidified for the intersection of the multi-responsibility domain operating spaces, and an initial scheduling proposal is output.

[0035] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method of the present invention, the specific steps for obtaining scheduling execution effect data are as follows:

[0036] The collaborative optimization scheduling scheme is distributed to each execution terminal in the smart park through the communication network to generate localized scheduling instructions;

[0037] Each execution terminal performs power adjustment and control operations on its assigned energy source according to localized scheduling instructions, and monitors the scheduling effect of the power adjustment and control operations in real time, and integrates and generates scheduling execution effect data.

[0038] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method of the present invention, the specific steps of analyzing and comparing the performance index deviations of the collaborative optimization scheduling scheme and the scheduling execution effect data, and outputting the performance index deviation data, are as follows:

[0039] Extract the actual performance index values ​​corresponding to the collaborative optimization scheduling scheme from the scheduling execution effect data, and integrate them to generate actual performance index data;

[0040] The predicted performance index values ​​in the collaborative optimization scheduling scheme are quantitatively compared with the actual performance index data and standardized and coded to output the performance index deviation data.

[0041] As a preferred embodiment of the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method of the present invention, the specific steps of outputting the optimized scheduling strategy are as follows:

[0042] Perform strategy rule matching analysis on performance index deviation data, identify the direction of strategy adjustment, and generate a set of strategy adjustment parameters;

[0043] Based on the strategy adjustment parameter set, the collaborative optimization scheduling scheme is modified in terms of strategy parameters and updated in terms of constraints, and the iterative scheduling strategy scheme is output.

[0044] The scheduling strategy iteration scheme is tested for scenario stability and convergence, and an optimized scheduling strategy is output.

[0045] The beneficial effects of this invention are as follows: Through the deep coupling of multi-energy flow collaborative diagnosis and distributed negotiation mechanism, accurate perception, collaborative decision-making, and continuous optimization of integrated scheduling of power generation, grid, load, storage, and charging in intelligent parks are achieved. Multi-energy flow collaborative diagnosis, based on multi-energy flow coupling situation data output from a digital twin model, integrates performance index evaluation and load forecasting to provide high-value state awareness for scheduling. The distributed negotiation mechanism distributes real-time electricity price information and multi-energy flow collaborative diagnosis reports to various local decision centers. Through local optimization, multi-round proposal interaction, and alternating direction multiplier method verification, dynamic strategy matching and global consistency among multiple entities are achieved while meeting power balance, tie-line constraints, and process rigidity requirements. The strategy optimization and adjustment stage constructs a closed-loop optimization capability of "execution-evaluation-learning-iteration" by analyzing the deviation between the scheduling execution effect and expected performance, realizing refined collaborative allocation of electricity, heat, cooling, and storage resources, and significantly improving energy utilization efficiency, operational economy, and reliability. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an AI-based intelligent park integrated scheduling method for power generation, grid, load, storage, and charging.

[0048] Figure 2 This is a flowchart for outputting multi-energy flow coupling situational data.

[0049] Figure 3This is a flowchart for outputting load forecast data.

[0050] Figure 4 This is a flowchart for outputting preliminary scheduling instructions. Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0054] Reference Figures 1-4 This is one embodiment of the present invention, which provides an AI-based intelligent park source-grid-load-storage-charging integrated scheduling method, including the following steps:

[0055] Collect multi-source energy process data of smart parks, perform multi-modal fusion and semantic encapsulation of multi-source energy process data, and output standardized data packets for parks.

[0056] Collect multi-source energy process data of the smart park, perform format conversion and time-series alignment processing on the multi-source energy process data, and output unified spatiotemporal reference data.

[0057] Specifically, multi-source energy process data of the smart park is collected. This is achieved through smart meters, temperature and humidity sensors, current and voltage transformers, energy storage battery management interfaces, photovoltaic inverter monitoring signals, electric vehicle charging pile status monitoring points, and production management data output ports deployed within the smart park. This allows for the real-time acquisition of multi-dimensional information covering photovoltaic power generation, energy storage status of charge and charging / discharging power, power load of each production line and equipment, heat output of heat pumps and electric boilers, cooling capacity of the cooling station, environmental temperature and humidity parameters, production line start / stop signals, equipment operating conditions, and the impact of process flow on temperature or humidity. This constitutes the multi-source energy process data of the smart park.

[0058] The system performs format conversion and time-series alignment on the multi-source energy process data of the smart park. Format conversion involves reconstructing the field structure of data from different sources according to a unified JSON Schema specification, unifying the unit system (e.g., power to kW, temperature to ℃, and time to ISO 8601 format), and performing structured encoding on discrete event data (e.g., equipment start-up and shutdown) and continuous time-series data (e.g., power curves). Time-series alignment uses seconds as the base time granularity, upsampling the multi-source energy process data using linear interpolation, and calibrating the timestamps of all data sources according to GPS timing or NTP network time protocol to ensure that all multi-source energy process data are strictly aligned under the same time axis and spatial topological coordinates, outputting unified spatiotemporal reference data.

[0059] The coupling and correlation relationships between various modal data in the unified spatiotemporal reference data are analyzed, and feature-level fusion is performed to form multimodal fused data.

[0060] Specifically, for different modal data such as photovoltaic power generation, energy storage state of charge, power load, heat output, cooling supply, ambient temperature and humidity, production line start-up and shutdown signals, and equipment operating conditions in the unified spatiotemporal reference data, the data is sliced ​​according to the same time window. The mean, rate of change, and state identifier of each modality within the current window are extracted as basic features to form a time-aligned feature sequence. Based on the time-aligned feature sequence, the synchronicity and sequence of changes in the features of each modality are compared through a sliding window to identify coupling modes such as "load power increases after production line starts" and "energy storage discharge response when photovoltaic output decreases". The identified coupling modes are used as association rules to splice the basic features of each modality to form multimodal fusion data.

[0061] The multimodal fusion data is tagged, encapsulated, and packaged in a standardized manner to generate a standardized data package for the park.

[0062] Specifically, semantic labels are assigned to each feature field in the multimodal fusion data. The label content includes modality type (such as photovoltaic, energy storage and load), physical quantity name (such as power, state of charge and temperature), unit (such as kW, % and ℃) and the identification of the production area or equipment. The labeled multimodal fusion data is organized into a hierarchical JSON structure, where the top-level fields include timestamp, spatial location and data version number, and the lower level is grouped and nested feature vectors according to energy type and process category. The entire JSON structure is UTF-8 encoded and compressed into GZIP format to generate a standardized data package for the park.

[0063] The standardized data package of the park is input into the pre-trained digital twin model to simulate the multi-energy flow coupling state, output multi-energy flow coupling situation data, evaluate the performance index and predict the load of the multi-energy flow coupling situation data, and integrate to generate a multi-energy flow collaborative diagnostic report.

[0064] The standardized data package of the park is input into the pre-trained digital twin model for digital simulation, and multi-energy flow simulation data is output.

[0065] Specifically, structured fields such as timestamps, spatial locations, power / flow values ​​for each energy type (electricity, heat, and cooling), energy storage state of charge, equipment operating status, and process constraints are parsed from the standardized data packets of the industrial park. These structured fields are then mapped according to the input interface format defined by the pre-trained digital twin model and assigned to the initial state variables in the pre-trained digital twin model, including the voltage amplitude and phase angle of the power grid bus, the supply and return water temperatures of the heating network, the cooling capacity setpoint of the chiller plant, the current SOC of the energy storage equipment, and the start / stop flags of the production line. The simulation model uses seconds as the simulation step size. Within each time step, it sequentially performs power flow calculation, dynamic heat transfer calculation, and cooling supply and demand balance calculation, and transmits coupling variables between energy sources (such as the conversion of electrical power consumed by electric boilers into heat output, and heat pumps affecting both electrical and thermal loads). After completing a rolling simulation of a scheduling cycle, the pre-trained digital twin model summarizes the state variables of each energy flow at each time step, including the active / reactive power of each branch, the temperature of the heat medium in the pipeline, the cooling capacity distribution ratio, the charging and discharging power of energy storage, and the equipment operating state sequence, and outputs multi-energy flow simulation data.

[0066] Furthermore, the pre-trained digital twin model utilizes historical data accumulated by the smart park under various operating conditions, including photovoltaic power generation, electricity load, energy storage state of charge, thermal and cooling output, environmental parameters, and production process status. This historical data is organized into input-output pairs according to a time series, where the input is the park's operating state at a certain moment, and the output is the actual observed state at the next moment. With the goal of minimizing the time-series error between the simulated state and the actual observed state, a gradient descent algorithm is used to directly optimize all learnable parameters within the pre-trained digital twin model. This enables the pre-trained digital twin model to accurately reproduce the dynamic evolution of the park's multi-energy flow under different operating conditions. Through iterative adjustments to the pre-trained digital twin model's parameters, the state prediction error on the validation set converges (meaning the state prediction error of the pre-trained digital twin model on the validation set stabilizes with increasing training iterations and no longer decreases), resulting in a pre-trained digital twin model with high-fidelity simulation capabilities.

[0067] During the training of the pre-trained digital twin model, the training sample size is set to no less than 10,000 effective time steps to cover the park's operating status under different scenarios such as seasons, weekdays / holidays, and different production intensities, ensuring that the digital twin model has sufficient generalization ability; the time window length is set to 900 seconds (15 minutes), consistent with the park's scheduling cycle, so that the digital twin model can learn the dynamic response characteristics under the scheduling scale; the batch size is set to 64, which ensures the stability of gradient estimation while taking into account training efficiency, avoiding convergence oscillations due to too small a batch or memory overflow due to too large a batch.

[0068] Extract the coupling and interaction features between energy flows in the multi-energy flow simulation data and integrate them to form multi-energy flow dynamic interaction parameters.

[0069] Specifically, numerical sequences of power branch power, heating network supply and return water temperatures, cooling capacity distribution ratio, and energy storage charging and discharging power at each simulation time step are extracted from multi-energy flow simulation data. Based on the time-aligned numerical sequences, the ratio of input electrical power to output thermal power of electro-thermal coupling devices (such as electric boilers) is statistically analyzed as the instantaneous conversion efficiency. The energy ratio of heat input to cold output of thermal-cold coupling devices (such as absorption chillers) is calculated, as well as the response delay between electrical energy input and thermal energy release during the energy storage charging and discharging process. Instantaneous conversion efficiency, energy ratio, and response delay are used as coupling correlation interaction features, and the data is analyzed according to time steps (time step refers to the time step used by the pre-trained digital twin model during digital simulation). The fixed simulation step size is usually 1 second. The time step is used to discretize the continuous physical process during the simulation. Each time step corresponds to a simulation output moment. The multi-energy flow simulation data is generated point by point according to this 1-second interval. Therefore, the coupled and related interaction features are also extracted and organized in the same 1-second time step unit and organized into a vector form. The parameter features in the vector form (referring to the numerical vector formed by the three coupled and related interaction features of instantaneous conversion efficiency, energy ratio and response delay organized in 1-second time steps at each time step) are normalized and grouped and aggregated according to energy type to form multi-energy flow dynamic interaction parameters that characterize the real-time energy conversion relationship and dynamic response characteristics between electricity, heat and cold.

[0070] The dynamic interaction parameters of multi-energy flow are integrated and structured to output multi-energy flow coupled situational data.

[0071] Specifically, the instantaneous conversion efficiency, energy ratio, and response delay characteristics in the dynamic interaction parameters of multi-energy flow are organized by time step and grouped according to three coupling relationships: electricity-heat, heat-cold, and electricity-storage. The mean, rate of change, and extreme values ​​of each group of parameters are calculated within a continuous time window to form high-order features reflecting the current operating trend. The high-order features are converted into fixed-dimensional structured vectors, where each position of the structured vector represents the coupling type identifier, efficiency value, response delay duration, and direction of change trend (rising, stable, or falling). The structured vectors corresponding to all time steps are arranged in sequence and appended with the corresponding timestamps and spatial location information to output multi-energy flow coupling status data.

[0072] Perform quantitative evaluation of performance indicators on multi-energy flow coupling situation data and output performance indicator evaluation data.

[0073] Specifically, efficiency values, response delay durations, and trends at each time step are extracted from multi-energy flow coupling situation data. Combined with the rated performance parameters of the park's energy equipment, three performance indicators—comprehensive energy efficiency ratio, multi-energy synergy, and dynamic response deviation—are calculated. The comprehensive energy efficiency ratio is the ratio of actual output heat / cold energy to input electrical energy. Multi-energy synergy is measured by the simultaneous activation ratio of the three coupling relationships of electricity, heat, and cold. The dynamic response deviation is the absolute difference between the actual response delay duration and the baseline response time. The three performance indicators—comprehensive energy efficiency ratio, multi-energy synergy, and dynamic response deviation—are organized into a structured numerical sequence according to time steps, with corresponding coupling type identifiers and timestamps added, and the performance indicator evaluation data is output.

[0074] Historical energy process operation data is acquired, and time-series features are extracted from the multi-energy flow coupling situation data and historical energy process operation data to generate load time-series feature data.

[0075] Specifically, acquiring historical energy process operation data involves retrieving time-series data covering at least one full year from the park's long-term operation records, including electricity load power, heat output, cooling supply, photovoltaic power generation, energy storage charge status, and corresponding production line start-up and shutdown signals; aligning the multi-energy flow coupling situation data with the historical energy process operation data on the time axis to form a joint time-series sequence; performing a unified time-series feature extraction operation on the joint time-series sequence, including calculating the mean, standard deviation, first-order difference maximum value, periodic peak position, and trend slope for each energy type within a sliding window, and extracting context labels such as weekdays / holidays, seasons, and production shifts; simultaneously, extracting lag values ​​from multiple historical moments as autoregressive features for load-related dimensions (such as electricity load power and heat output), and organizing all extracted time-series features into a structured feature vector according to time steps to generate load time-series feature data.

[0076] By using pattern matching to deduce the mapping relationship between historical operating patterns and real-time status in load time-series characteristic data, load change trajectory data is output.

[0077] Specifically, the current load time-series feature data (including statistical features and context labels of dimensions such as power consumption and heat output) is used as the real-time situation feature vector. In the load time-series feature library constructed from historical energy process operation data, the current real-time situation feature vector is used as the query target. The corresponding feature vectors for all historical time periods are traversed. The dynamic time warping algorithm is used to calculate the similarity of feature sequences within local windows of each historical time period. That is, the dynamic time warping algorithm constructs a non-linear alignment path between the current real-time situation feature vector and the load time-series feature sequence of each historical time period, and calculates the cumulative distance of the non-linear alignment path within the local window. The current real-time situation feature vector (usually a short sequence or a single moment's upper and lower bounds) is used as the real-time situation feature vector. The system performs point-by-point distance measurements (such as Euclidean distance) between the extended historical sequence and the subsequences within the sliding window in the historical sequence to find the minimum cumulative distance that minimizes the total cumulative distance. The minimum cumulative distance is used as the matching distance between the current historical period and the current situation. The historical period with the smallest matching distance is selected as the most similar historical operating mode. For each selected historical operating mode, the actual load evolution sequence of subsequent time steps is extracted as a preliminary reference trajectory for future load changes. Combining the context of the current real-time situation (such as whether it is a workday, the season, and the production shift status), the reference trajectory is scaled or offset to eliminate the structural differences between the historical pattern and the current conditions, and the load change trajectory data is output.

[0078] The confidence level of the load change trajectory data is calibrated and the error is corrected to output load forecast data.

[0079] Specifically, load change trajectory data and corresponding actual observations generated under the same contextual conditions (such as the same season, workday / holiday type, and production shift status) are extracted from historical energy process operation data. The deviation between the historical energy process operation data and the load change trajectory data at each time step is calculated to form a historical error sequence. Based on the historical error sequence, the actual accuracy at different prediction confidence levels is statistically analyzed. The confidence level implied in the current load change trajectory data is adjusted using a temperature scaling method to align the confidence level with the true accuracy. The average deviation value under the same contextual conditions in the above historical error sequence is used as the correction amount for the current load change trajectory data at each time step. The correction amount is removed from the original trajectory, and the load prediction data after confidence calibration and error correction is output.

[0080] Furthermore, the implicit confidence level refers to the predictive reliability of the current load change trajectory data, which is not explicitly output during generation but can be inferred from the matching process. The minimum matching distance obtained by the dynamic time warping algorithm can be mapped to the initial confidence level (the smaller the distance, the higher the confidence level). The temperature scaling method introduces a learnable positive real scalar parameter (i.e., T) to perform nonlinear calibration on the original confidence level: the original confidence level is regarded as the softmax output of the probability distribution, and the sharpness is adjusted by T. When T>1, the probability distribution is smoother (reducing overconfidence), and when T<1, the probability distribution is more concentrated (enhancing high confidence). The temperature parameter is determined by fitting the historical error sequence under the same context conditions, so that the calibrated average confidence level is equal to the actual observation accuracy. The calibration process continues to iterate until the confidence level and the true accuracy are aligned in each confidence interval.

[0081] The performance index evaluation data and load forecast data are used to diagnose energy efficiency bottlenecks and conduct correlation analysis of operational risks, and then integrated to generate a multi-energy flow collaborative diagnostic report.

[0082] Specifically, the system aligns the comprehensive energy efficiency ratio, multi-energy synergy, and dynamic response deviation in the performance evaluation data with the future electricity load power, heat output, and cooling demand in the load forecast data, establishing a one-to-one correspondence in the time dimension. Based on the equipment's rated efficiency threshold and response time limit requirements, it identifies periods where the comprehensive energy efficiency ratio is consistently lower than the equipment's rated efficiency threshold and the load forecast value is on an upward trend, thus determining them as energy efficiency bottlenecks. Simultaneously, when the dynamic response deviation exceeds the safety margin and the load forecast shows abrupt changes, it is marked as an operational risk. The system further analyzes the propagation correlation between energy efficiency bottlenecks and operational risks along the electricity-heat-cooling coupling path. For example, low multi-energy synergy accompanied by high electricity load forecasts may trigger transformer overload risks, and outputs diagnostic conclusions. The diagnostic conclusions are organized in chronological order, with corresponding performance index values, load forecast curves, bottleneck types (such as energy storage response lag and insufficient thermoelectric conversion efficiency), and risk levels, integrating them to generate a multi-energy flow synergy diagnostic report.

[0083] Furthermore, the rated efficiency threshold and response time limit requirements for equipment are determined jointly by the equipment's factory parameters and historical operational measurement data from the park: the rated efficiency threshold is based on the equipment's nominal conversion efficiency under rated operating conditions, and calibrated by combining the 90th percentile of the measured efficiency of the current equipment during its stable operation in the park in the past, ensuring that the rated efficiency threshold reflects both the equipment's design capabilities and its actual operating performance; the response time limit requirements are based on the dynamic response performance indicators provided by the equipment manufacturer (such as the time required for the energy storage device to reach the target power from receiving the command), and refer to the park's actual tolerance limit for energy switching speed in typical production scenarios (such as the thermal energy supply needing to respond to load changes within 90 seconds), comprehensively forming the upper limit of the response time limit for various types of equipment.

[0084] By inputting standardized data packages from the park into a pre-trained digital twin model, high-fidelity dynamic simulation of the coupling state of multiple energy flows (electricity, heat, and cold) is achieved. This is combined with historical operating patterns and real-time situational awareness to complete quantitative evaluation of performance indicators and high-precision load forecasting. Furthermore, energy efficiency bottleneck diagnosis and operational risk correlation analysis are conducted, generating structured and interpretable multi-energy flow collaborative diagnostic reports. This approach overcomes the limitations of traditional scheduling relying on a single energy perspective, achieving deep fusion of multi-source heterogeneous data under a unified spatiotemporal benchmark, dynamic reconstruction of multi-energy flow coupling mechanisms, and accurate prediction of future situations. It significantly improves the smart park's ability to perceive complex operating conditions, the accuracy of identifying energy efficiency shortcomings, and the level of predicting operational risks, providing a highly reliable decision-making basis for subsequently generating safe, economical, and efficient collaborative optimization scheduling schemes.

[0085] The system acquires real-time electricity price information, dynamically balances and matches the real-time electricity price information with multi-energy flow collaborative diagnostic reports through a distributed negotiation mechanism, outputs preliminary scheduling instructions, performs process verification and energy efficiency optimization on the preliminary scheduling instructions, and forms a collaborative optimization scheduling scheme.

[0086] Obtain real-time electricity price information, semantically align the real-time electricity price information with the multi-energy flow collaborative diagnostic report, and generate a negotiation input package.

[0087] Specifically, time-of-use electricity price data is obtained in real time from the electricity market interface, and the timestamp, price value, and price type label are extracted. Performance index evaluation results, load forecast curves, energy efficiency bottleneck periods, and operational risk levels within the same timestamp are extracted from the multi-energy flow collaborative diagnostic report. The real-time electricity price information and the multi-energy flow collaborative diagnostic report are aligned on the time axis according to a unified time granularity, and semantic labels are added to the price values ​​and diagnostic elements, including electricity price-peak-valley attributes, energy efficiency-bottleneck type, and risk-coupling path. The aligned electricity price information and diagnostic elements are organized into a set of key-value pairs according to the responsibility domain structure. Each key-value pair contains a time step, energy type, decision attribute, and value, and is encapsulated into a standardized negotiation input package.

[0088] The real-time electricity price information and multi-energy flow collaborative diagnostic report in the negotiated input package are converted into a set of operational feasibility space descriptions.

[0089] Specifically, the process involves extracting the electricity price, price type label, performance evaluation results, load forecast, energy efficiency bottleneck type, and operational risk level from the negotiated input package. Based on the physical characteristics of the equipment and process constraints, the electricity price information is transformed into an economically driven power adjustment direction (e.g., suppressing electricity load during periods of high electricity prices and prioritizing energy storage charging during periods of low electricity prices). Energy efficiency bottlenecks and operational risks are transformed into insurmountable operational boundaries (e.g., limiting the output of heat pumps during periods of low energy efficiency and setting upper limits for tie-line power during periods of transformer overload risk). Combined with the load forecast, the feasible supply and demand balance range for each energy source (electricity, heat, and cooling) at each time step is determined. The adjustment direction, operational boundaries, and feasible supply and demand balance range are then uniformly expressed as a mathematical description set composed of inequality constraints and upper and lower limits of state variables, forming an operational feasible space description set.

[0090] Furthermore, the feasible range for supply and demand balance is defined by taking the power load, heat demand, and cooling demand at the corresponding time step in the load forecast data as the lower limit benchmark values, and adding an allowable adjustment margin (such as ±5% to cope with forecast errors) on top of these. At the same time, it does not exceed the maximum supply capacity of dispatchable resources in the park (such as the maximum output of photovoltaic power, the rated thermal power of electric boilers, and the upper limit of cooling station cooling) and the minimum technical output (such as the minimum load required for safe operation of equipment). The basis for the values ​​includes: the load forecast value itself, the capacity boundary specified by the equipment nameplate parameters, the output limitation requirements of energy efficiency bottlenecks and operational risks marked in the multi-energy flow collaborative diagnosis report, and the rigid constraints of production processes on the continuity and stability of energy supply.

[0091] The feasible space description set is decomposed according to the responsibility domain to generate a set of responsibility domain runtime subspaces.

[0092] Specifically, based on the physical topology and management affiliation of the smart park, the inequality constraints and state variables involved in the feasible operational space description are divided into corresponding responsibility domains according to the power production and consumption area, heat supply area, cold supply area, and energy storage scheduling. For each responsibility domain, the constraints directly related to the managed equipment and energy flow (such as the capacity limit of a transformer in a certain area, the upper and lower limits of the output of local electric boilers, and the boundary of the energy storage SOC) are extracted, as well as the load forecast value and adjustment direction that the responsibility domain needs to respond to. These are then organized into subsets that contain only the variables and constraints of the responsibility domain. Each subset independently describes the feasible operational range of the current responsibility domain at each time step, forming a responsibility domain operational subspace set.

[0093] Each operational constraint description item in the operational subspace of the responsibility domain is verified, and operational constraint description items that are inconsistent with the process rigid constraints, equipment availability status and safety boundaries in the multi-energy flow collaborative diagnostic report are removed. These items are then summarized and encapsulated into a responsible domain operational subspace that can be committed to.

[0094] Specifically, each operational constraint description item in the responsibility domain's operational subspace set is traversed, and the specified variable range, action direction, or time window is logically compared with the process rigid constraints (such as the production line's power supply voltage must not be lower than 95% of the rated value), equipment availability status (such as the electric boiler currently being under maintenance and out of service), and safety boundaries (such as the energy storage's charge status must be maintained within the 20%–90% range) explicitly marked in the multi-energy flow collaborative diagnostic report. If the action or status required by the current operational constraint description item violates any of the process rigid constraints, equipment availability status, or safety boundaries explicitly marked in the multi-energy flow collaborative diagnostic report, then the current operational constraint description item is removed from the responsibility domain's operational subspace set. All verified operational constraint description items are retained and reorganized into a structured set according to time step and energy type, forming an operational capability description that only contains the operational capability description that the current responsibility domain can actually fulfill under the premise of physical feasibility, process compliance, and equipment availability, i.e., the operational subspace that the responsibility domain can commit to.

[0095] The boundary descriptions of the subspaces that can be committed to operation by each responsibility domain are exchanged through a communication network. The exchanged boundary descriptions are then overlaid and compared with the boundary descriptions of the subspaces that can be committed to operation by the current responsibility domain to generate the intersection of the multi-responsibility domain operation spaces.

[0096] Specifically, each responsibility domain broadcasts boundary descriptions (such as maximum output power, minimum acceptable heat load, and allowable cooling capacity interaction range) of coupled variables involving tie-line power, shared energy storage scheduling, and inter-regional heat supply in its operational subspace to other responsibility domains via a communication network. After receiving the boundary descriptions from other responsibility domains, each responsibility domain compares the boundary descriptions with the boundary descriptions of the corresponding coupled variables in its own operational subspace at the same time step, taking the overlapping portion as the feasible intersection (e.g., if this responsibility domain can output 100–200kW and the other responsibility domain can accept 150–250kW, then the intersection is 150–200kW). After completing the numerical comparison of all coupled variables at all time steps, the results are integrated into a unified set of constraints that satisfies the common feasible conditions of all responsibility domains, i.e., the intersection of the multi-responsibility domain operational spaces.

[0097] Consistent boundaries are solidified for the intersection of the multi-responsibility domain operating spaces, and an initial scheduling proposal is output.

[0098] Specifically, the feasible intersection intervals of each time step and each coupling variable in the multi-responsibility domain operating space are converted into deterministic scheduling boundary values, using the interval median or the anchor value selected according to the principle of economic priority as the benchmark; the solidified scheduling boundary values ​​are organized into a structured instruction sequence according to energy type, time step and responsibility domain affiliation, clarifying the power exchange quantity, start-stop sequence and operating level that each responsibility domain can execute under the collaborative framework; the structured instruction sequence is the initial scheduling proposal that satisfies global physical consistency, coupling constraint enforceability and multilateral feasibility.

[0099] Initial scheduling proposals are exchanged through a communication network, and consistency constraints are verified on the exchanged initial scheduling proposals to generate a consistent scheduling plan.

[0100] Specifically, each responsibility domain sends its generated initial scheduling proposal to other responsibility domains via the communication network. After receiving the initial scheduling proposal from other responsibility domains, each responsibility domain extracts the scheduling instructions (such as tie-line power, shared energy storage charging and discharging plans, and cross-regional thermal interaction quantities) related to its own coupling interfaces and compares them step-by-step with the scheduling instructions of the corresponding interfaces in its own initial scheduling proposal. If the deviation of the scheduling instructions of all responsibility domains at the same time step and the same coupling variables is less than the preset tolerance threshold (such as power deviation ≤ 1kW, temperature deviation ≤ 0.5℃), it is determined that the consistency constraint is met. The verified scheduling instructions are integrated into a globally consistent scheduling scheme according to a unified time axis to form a consistent scheduling plan.

[0101] Furthermore, the preset tolerance thresholds are jointly set based on the accuracy level of the measuring equipment in the smart park, the control resolution of the execution terminal, and the physical continuity requirements of the multi-energy flow coupling process. The power deviation tolerance threshold is set to 1kW, based on the minimum metering resolution of the smart meters deployed in the park being 0.5kW, plus the ±0.5kW fluctuation margin allowed by the communication and control links. The temperature deviation tolerance threshold is set to 0.5℃, based on the typical measurement error of the heating network temperature sensor being ±0.2℃ and the execution accuracy of the cooling station outlet temperature control being ±0.3℃, and the conservative upper limit is taken after superposition. The values ​​also take into account the small imbalances that naturally exist between the responsibility domains of the coupled variables under normal operating conditions in historical operating data, ensuring that the tolerance thresholds can filter communication noise and execution jitter without masking substantial scheduling conflicts, thereby ensuring the engineering feasibility and physical rationality of consistency constraint verification.

[0102] The consistency scheduling plan is verified for safety boundaries and dynamically adjusted for power balance, and preliminary scheduling instructions are output.

[0103] Specifically, the equipment output, tie-line power, energy storage charging and discharging power, and heat and cold supply at each time step in the consistent dispatch plan are compared one by one with the safety boundaries (such as transformer rated capacity, line thermal stability limit, energy storage charging state safety range, and heating network pressure and temperature limits) defined in the multi-energy flow collaborative diagnosis report. If there are any items exceeding the limits, the corresponding loads are reduced or transferred according to priority rules. For the verified dispatch plan, the difference between the total supply and total demand of electricity, heat, and cold energy within the park at each time step is calculated. By adjusting the compensation imbalance of adjustable resources (such as energy storage charging and discharging power, electric boiler output, and cooling station cooling capacity), the energy flow meets the dynamic power balance. The dispatch instructions after completing safety verification and balance adjustment are structured and encapsulated according to responsibility domain and time step to form preliminary dispatch instructions.

[0104] The initial scheduling instructions are verified for process and optimized for energy efficiency to form a collaborative optimization scheduling scheme.

[0105] Specifically, the equipment start-up and shutdown plans, power setpoints, and energy allocation schemes at each time step in the initial dispatch instructions are compared item by item with the rigid process constraints (such as uninterrupted continuous power supply to the production line, the temperature and humidity maintenance range of key processes, and the minimum heat load requirements of heat treatment processes) specified in the multi-energy flow collaborative diagnostic report. If conflicts exist, the relevant dispatch instructions are adjusted according to the production process priority. Under the premise of satisfying all process constraints, with the goal of maximizing the comprehensive energy efficiency ratio, the operating time and output level of adjustable equipment (such as electric boilers, heat pumps, chillers, and energy storage) are locally rescheduled. High-efficiency equipment is arranged to operate during peak load periods or low electricity price periods, and the number of energy conversions in the coupling path is reduced. The dispatch instructions after completing the process compliance correction and energy efficiency improvement adjustment are integrated according to a unified time axis and responsibility domain structure to form a collaborative optimization dispatch scheme.

[0106] The collaborative optimization scheduling scheme is distributed and executed in a distributed manner, and the scheduling effect of the collaborative optimization scheduling scheme is monitored to obtain scheduling execution effect data.

[0107] The collaborative optimization scheduling scheme is distributed to each execution terminal in the smart park through the communication network to generate localized scheduling instructions.

[0108] Specifically, the collaborative optimization scheduling scheme is broken down into sub-task instruction sets for different execution terminals according to energy type and spatial location. Each sub-task instruction set includes the power regulation target, start-stop sequence, operating mode and allowable deviation to be executed by the corresponding execution terminal. Each sub-task instruction set is sent to the corresponding execution terminal, such as energy storage converter, electric boiler controller, chiller station management interface or production line power supply interface, through the communication network. After receiving the sub-task instruction set, each execution terminal performs local adaptation and control based on its own equipment's current status (such as state of charge, outlet temperature and operating level) to generate localized scheduling instructions that can directly drive equipment actions, including specific voltage setpoints, valve opening, start-stop signals or charging / discharging current instructions.

[0109] Each execution terminal performs power adjustment and control operations on its assigned energy source according to localized scheduling instructions, and monitors the scheduling effect of the power adjustment and control operations in real time, and integrates and generates scheduling execution effect data.

[0110] Specifically, the energy storage converter adjusts its output power according to the charging and discharging current command, the electric boiler controller adjusts the number of heating elements put into operation according to the start / stop signal and power target, the chiller station management interface regulates the refrigerant flow through the valve opening set value, and the production line power supply interface provides stable power supply according to the voltage set value. During the execution process, each execution terminal collects the actual output power, temperature, current, voltage and equipment status feedback in real time at a second-level sampling frequency, and compares them with the target values ​​in the localized scheduling command to calculate the deviation statistics. The actual operating data during the execution process is aligned by timestamp, and the command execution identifier, equipment location information and deviation statistics are added to integrate and generate structured scheduling execution effect data.

[0111] Analyze and compare the performance index deviations of the collaborative optimization scheduling scheme and the scheduling execution effect data, output the performance index deviation data, optimize and adjust the strategy based on the performance index deviation data, and output the optimized scheduling strategy.

[0112] Extract the actual performance index values ​​corresponding to the collaborative optimization scheduling scheme from the scheduling execution effect data, and integrate them to generate actual performance index data.

[0113] Specifically, the actual operation records corresponding to the scheduling instructions at each time step in the collaborative optimization scheduling scheme are identified in the scheduling execution effect data, and precise matching is performed based on timestamps and equipment location information. The underlying measurement values ​​related to the three performance indicators of comprehensive energy efficiency ratio, multi-energy synergy, and dynamic response deviation are extracted from the actual operation records, including actual input electrical energy, output heat / cold energy, synchronous operation status of electric-heat-cold coupled equipment, and the time interval from instruction issuance to equipment response completion. Based on the underlying measurement values, the actual values ​​of the performance indicators at each time step are recalculated according to the same calculation logic as the performance indicator evaluation data. The actual values ​​of the performance indicators at all time steps are organized in chronological order, and the corresponding coupling type identifier and timestamp are attached to integrate and generate actual performance indicator data.

[0114] The predicted performance index values ​​in the collaborative optimization scheduling scheme are quantitatively compared with the actual performance index data and standardized and coded to output the performance index deviation data.

[0115] Specifically, the expected performance index values ​​at each time step in the aligned and coordinated optimization scheduling scheme are matched with the actual performance index values ​​at the corresponding time steps in the actual performance index data, ensuring complete matching in timestamp, coupling type identifier, and spatial location. For the three indicators of comprehensive energy efficiency ratio, multi-energy synergy, and dynamic response deviation, the absolute and relative deviations between the expected and actual values ​​are calculated respectively. The deviation values ​​are standardized, including mapping the deviations to a unified dimension range, labeling the deviation direction (positive or negative), and associating them with the corresponding energy coupling paths. All standardized deviation information is organized into structured records by time step, with each record containing a timestamp, coupling type, indicator name, expected value, actual value, and standardized deviation value, outputting the performance index deviation data.

[0116] Perform strategy rule matching analysis on performance index deviation data, identify the direction of strategy adjustment, and generate a set of strategy adjustment parameters.

[0117] Specifically, each structured record in the performance index deviation data is matched item by item with a preset strategy rule base. The strategy rule base contains several condition-action rules, such as "if the comprehensive energy efficiency ratio deviation is negative and the dynamic response deviation is positive, then increase the energy storage response priority" or "if the multi-energy synergy continues to decrease, then increase the scheduling weight of the electric-thermal coupling equipment". When a structured record in the performance index deviation data meets the precondition of the current strategy rule, the adjustment action corresponding to the current rule is triggered, and the corresponding strategy adjustment direction is marked, such as "increase the output ratio of the heat pump", "shorten the energy storage charging and discharging response delay tolerance threshold" or "reduce the adjustability margin of uninterrupted load". All triggered adjustment actions are summarized, and the adjustable parameters involved and their changes are extracted, including equipment scheduling weight, response time limit upper limit, efficiency offset and load transfer elasticity coefficient, to form a structured set of strategy adjustment parameters.

[0118] Furthermore, the pre-defined strategy rule base is a set of condition-action rules constructed based on the historical energy process operation data, equipment operation boundaries, and energy efficiency optimization goals of the smart park before the deployment of the scheduling method. The strategy rule base contains response logic for typical performance index deviation scenarios such as comprehensive energy efficiency ratio deviation, multi-energy synergy decline, and dynamic response delay. Each rule consists of a clear trigger condition and a corresponding adjustment action (such as "increasing the scheduling priority of electric boilers" or "tightening the energy storage load state adjustment range"). The rules cover adjustable parameters such as scheduling weight adjustment, response time limit correction, and load transfer elasticity coefficient update of electric, heat, and cold coupled equipment, and are solidified in the scheduling process in a structured form to guide the generation of strategy adjustment parameter sets.

[0119] The collaborative optimization scheduling scheme is modified and its constraints are updated according to the set of adjustment parameters, and the iterative scheduling strategy scheme is output.

[0120] Specifically, adjustable parameters and their changes are read item by item from the strategy adjustment parameter set. The corresponding equipment scheduling weights, response time limits, efficiency offsets, and load transfer elasticity coefficients in the collaborative optimization scheduling scheme are numerically replaced or incrementally adjusted. At the same time, according to the constraint-type adjustments involved in the strategy adjustment parameter set (such as "tightening the energy storage state of charge adjustment range"), the corresponding operating boundary conditions in the collaborative optimization scheduling scheme are updated synchronously. For example, the upper and lower limits of the state of charge of energy storage equipment and the power transmission limit of tie lines are modified. After all parameters and constraints are updated, the overall structure and time series framework of the collaborative optimization scheduling scheme are retained, and only the adjusted parts are replaced to form a scheduling strategy iteration scheme.

[0121] The scheduling strategy iteration scheme is tested for scenario stability and convergence, and an optimized scheduling strategy is output.

[0122] Specifically, the iterative scheduling strategy is input into a pre-trained digital twin model, and simulations are run under multiple test scenarios covering typical operating conditions, extreme load fluctuations, and sudden changes in electricity prices. The stability of each energy flow state variable is assessed by observing whether it remains bounded and free from continuous oscillations during the simulation. Simultaneously, for the iterative optimization parts of the scheduling strategy (such as distributed negotiation or power balance adjustment), the solution process is repeatedly executed, recording the changes in the objective function value and energy flow state variables in consecutive iterations. Convergence is determined when both the change in the objective function value and the magnitude of variable updates are less than a preset small positive threshold. If the iterative scheduling strategy meets the stability requirements in all test scenarios and the relevant optimization process passes the convergence test, it is confirmed as the final result, and the optimized scheduling strategy is output.

[0123] This embodiment also provides a computer device applicable to the AI-based intelligent park integrated scheduling method for power generation, grid, load, storage, and charging, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the AI-based intelligent park integrated scheduling method for power generation, grid, load, storage, and charging as proposed in the above embodiment.

[0124] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0125] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the AI-based intelligent park source-grid-load-storage-charging integrated scheduling method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0126] In summary, this invention achieves precise perception, collaborative decision-making, and continuous optimization of integrated power generation, grid, load, storage, and charging scheduling in smart parks through deep coupling of multi-energy flow collaborative diagnosis and distributed negotiation mechanisms. Multi-energy flow collaborative diagnosis, based on multi-energy flow coupling situational data output from a digital twin model, integrates performance index evaluation and load forecasting to provide high-value state awareness for scheduling. The distributed negotiation mechanism distributes real-time electricity price information and multi-energy flow collaborative diagnosis reports to local decision centers. Through local optimization, multi-round proposal interaction, and alternating direction multiplier method verification, it achieves dynamic strategy matching and global consistency among multiple entities while meeting power balance, tie-line constraints, and process rigidity requirements. The strategy optimization and adjustment phase constructs a closed-loop optimization capability of "execution-evaluation-learning-iteration" by analyzing the deviation between scheduling execution effects and expected performance, achieving refined collaborative allocation of electricity, heat, cooling, and storage resources, significantly improving energy utilization efficiency, operational economy, and reliability.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An AI-based intelligent park integrated scheduling method for power generation, grid, load, storage, and charging, characterized in that: include, Collect multi-source energy process data of smart parks, perform multi-modal fusion and semantic encapsulation of multi-source energy process data, and output standardized data packets for parks; The standardized data package of the park is input into the pre-trained digital twin model to simulate the multi-energy flow coupling state, output multi-energy flow coupling situation data, evaluate the performance index and predict the load of the multi-energy flow coupling situation data, and integrate to generate a multi-energy flow collaborative diagnostic report. The system acquires real-time electricity price information, dynamically balances and matches the real-time electricity price information with multi-energy flow collaborative diagnostic reports through a distributed negotiation mechanism, outputs preliminary scheduling instructions, performs process verification and energy efficiency optimization on the preliminary scheduling instructions, and forms a collaborative optimization scheduling scheme. Distribute and execute the collaborative optimization scheduling scheme in a distributed manner, monitor the scheduling effect of the collaborative optimization scheduling scheme, and obtain scheduling execution effect data; Analyze and compare the performance index deviations of the collaborative optimization scheduling scheme and the scheduling execution effect data, output the performance index deviation data, optimize and adjust the strategy based on the performance index deviation data, and output the optimized scheduling strategy.

2. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The specific steps for outputting the standardized data packet for the campus are as follows: Perform format conversion and time-series alignment on multi-source energy process data to output unified spatiotemporal reference data; The coupling and correlation relationships between various modal data in the unified spatiotemporal reference data are analyzed, and feature-level fusion is performed to form multimodal fused data; The multimodal fusion data is tagged, encapsulated, and packaged in a standardized manner to generate a standardized data package for the park.

3. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The specific steps for outputting multi-energy flow coupling situational data are as follows: The standardized data package of the park is input into the pre-trained digital twin model for digital simulation, and multi-energy flow simulation data is output. Extract the coupling and interaction features between energy flows in the multi-energy flow simulation data and integrate them to form multi-energy flow dynamic interaction parameters; The dynamic interaction parameters of multi-energy flow are integrated and structured to output multi-energy flow coupled situational data.

4. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The specific steps for evaluating performance indicators and predicting loads from multi-energy flow coupling situation data, and integrating this data to generate a multi-energy flow collaborative diagnostic report, are as follows: Perform quantitative evaluation of performance indicators on multi-energy flow coupling situation data and output performance indicator evaluation data; Acquire historical energy process operation data, combine multi-energy flow coupling situation data with historical energy process operation data to extrapolate load trends, and output load forecast data; The performance index evaluation data and load forecast data are used to diagnose energy efficiency bottlenecks and conduct correlation analysis of operational risks, and then integrated to generate a multi-energy flow collaborative diagnostic report.

5. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 4, characterized in that: The specific steps for performing load trend extrapolation by combining multi-energy flow coupling situation data with historical energy process operation data and outputting load forecast data are as follows: Time-series feature extraction is performed on multi-energy flow coupling situation data and historical energy process operation data to generate load time-series feature data; By using pattern matching to deduce the mapping relationship between historical operating patterns and real-time status in load time series characteristic data, load change trajectory data is output. The confidence level of the load change trajectory data is calibrated and the error is corrected to output load forecast data.

6. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The specific steps for outputting the initial scheduling instruction are as follows: Semantically align real-time electricity price information with multi-energy flow collaborative diagnostic reports to generate negotiation input packages; The negotiation input packets are distributed according to the responsibility domain of the smart park, and the proposals are constructed from the negotiation input packets through a distributed negotiation mechanism to output the initial scheduling proposal; Initial scheduling proposals are exchanged through a communication network, and consistency constraints are verified on the exchanged initial scheduling proposals to generate a consistent scheduling plan. The consistency scheduling plan is verified for safety boundaries and dynamically adjusted for power balance, and preliminary scheduling instructions are output.

7. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 6, characterized in that: The process involves distributing negotiation input packets according to the responsibility domains of the smart park, constructing proposals from the negotiation input packets through a distributed negotiation mechanism, and outputting an initial scheduling proposal. The specific steps are as follows: The real-time electricity price information and multi-energy flow collaborative diagnostic report in the negotiated input package are converted into a set of operational feasibility space descriptions. The feasible space description set is decomposed according to the responsibility domain to generate a set of responsibility domain runtime subspaces; Each operational constraint description item in the operational subspace of the responsibility domain is verified, and operational constraint description items that are inconsistent with the process rigid constraints, equipment availability status and safety boundaries in the multi-energy flow collaborative diagnostic report are removed. The items are then summarized and encapsulated into a responsible domain operational subspace. The boundary descriptions of the subspaces that can be committed to operation by each responsibility domain are exchanged through a communication network. The exchanged boundary descriptions are then overlaid and compared with the boundary descriptions of the subspaces that can be committed to operation by the current responsibility domain to generate the intersection of the multi-responsibility domain operation spaces. Consistent boundaries are solidified for the intersection of the multi-responsibility domain operating spaces, and an initial scheduling proposal is output.

8. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The specific steps for obtaining the scheduling execution effect data are as follows: The collaborative optimization scheduling scheme is distributed to each execution terminal in the smart park through the communication network to generate localized scheduling instructions; Each execution terminal performs power adjustment and control operations on its assigned energy source according to localized scheduling instructions, and monitors the scheduling effect of the power adjustment and control operations in real time, and integrates and generates scheduling execution effect data.

9. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The analysis compares the performance index deviations between the collaborative optimization scheduling scheme and the scheduling execution effect data, and outputs the performance index deviation data. The specific steps are as follows: Extract the actual performance index values ​​corresponding to the collaborative optimization scheduling scheme from the scheduling execution effect data, and integrate them to generate actual performance index data; The predicted performance index values ​​in the collaborative optimization scheduling scheme are quantitatively compared with the actual performance index data and standardized and coded to output the performance index deviation data.

10. The AI-based intelligent park source-grid-load-storage-charging integrated scheduling method as described in claim 1, characterized in that: The output optimization scheduling strategy comprises the following steps: Perform strategy rule matching analysis on performance index deviation data, identify the direction of strategy adjustment, and generate a set of strategy adjustment parameters; Based on the strategy adjustment parameter set, the collaborative optimization scheduling scheme is modified in terms of strategy parameters and updated in terms of constraints, and the iterative scheduling strategy scheme is output. The scheduling strategy iteration scheme is tested for scenario stability and convergence, and an optimized scheduling strategy is output.

Images