A park intelligent energy load coordination scheduling optimization system

Abstract

The application discloses a park intelligent energy load collaborative scheduling optimization system and relates to the technical field of intelligent energy and power system scheduling.The data acquisition and preprocessing module is used for collecting multi-source operation data of the power generation side, the power transmission side, the power distribution side, the load side and the energy storage side of the park in real time and completing standardization processing, the digital twin mapping module is used for constructing a full-factor virtual mirror to realize real-time dynamic mapping of the multi-energy flow network, the fault prediction module with the built-in improved space-time attention deep learning model is used for identifying early-stage abnormalities of equipment and dividing four risk grades, the active self-healing scheduling module is used for constructing a multi-objective optimization model to solve an optimal scheduling strategy, and after the digital twin virtual verification, the execution control module is used for decomposing and issuing instructions to each terminal for execution, and finally, the visual monitoring module is used for realizing global state display and manual intervention, so that the complete closed loop of park electric-thermal-cold-gas multi-energy flow collaborative scheduling and fault self-healing is realized.

Description

Technical Field

[0001] This invention relates to the field of smart energy and power system dispatching technology, specifically a smart energy load collaborative dispatching optimization system for industrial parks. Background Technology

[0002] With the rapid development of my country's industrial park economy, energy demand within these parks is becoming increasingly diversified. Multiple energy sources, including electricity, heat, air conditioning, and natural gas, coexist and operate within these parks, forming complex multi-energy flow coupled energy systems. Traditional park energy management typically employs an independent operation and separate scheduling approach for each energy subsystem, lacking a holistic consideration of the coupling relationships between various energy sources. Simultaneously, the large-scale integration of new energy devices such as distributed power sources and energy storage systems has made the operating conditions of park energy systems more complex, placing higher demands on the real-time, accurate, and coordinated nature of energy scheduling. Against this backdrop, how to achieve coordinated scheduling of multiple energy flows within parks to improve energy utilization efficiency and system operational reliability has become a key technical issue in the construction of smart parks.

[0003] However, existing energy dispatching technologies for industrial parks still have many shortcomings. On the one hand, traditional dispatching systems lack the ability to predict equipment failures in the early stages, typically responding only passively after a failure occurs. They cannot identify abnormal equipment signals in advance and generate differentiated risk level assessments, leading to delayed fault handling and affecting the power supply reliability of critical loads in the park. On the other hand, traditional technologies mostly optimize dispatching for single power systems, failing to fully consider the coupling characteristics between multiple energy flows such as electricity, heat, cooling, and gas. They cannot utilize the heat and cold storage capacity of heating and cooling networks, or gas turbines for energy complementarity and transfer during power system failures, resulting in energy waste and high dispatching costs. Furthermore, traditional dispatching schemes lack a virtual verification process before implementation, making it difficult to guarantee the security and effectiveness of the strategies. The systems also do not support mutual assistance and purchase with external energy networks, limiting the improvement of overall energy utilization efficiency.

[0004] In summary, existing park energy dispatching technologies have significant shortcomings in fault prediction capabilities, multi-energy flow collaborative dispatching, dispatching strategy security verification, and energy interoperability between the park and external energy networks. These shortcomings make it difficult to meet the high reliability, high economy, and high synergy requirements of smart parks for energy systems. Therefore, there is an urgent need for a smart park energy load collaborative dispatching and optimization system that can achieve early fault prediction and tiered response, support multi-energy flow collaborative optimization dispatching, possess digital twin virtual verification capabilities, and can interoperate with external energy networks to overcome the deficiencies of existing technologies. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a smart energy load collaborative scheduling and optimization system for industrial parks. This system can collect multi-source operational data from the power generation, transmission, distribution, load, and energy storage sides of the park in real time through a data acquisition and preprocessing module, and complete standardized processing. It utilizes a digital twin mapping module to construct a virtual mirror of all elements to achieve real-time dynamic mapping of the multi-energy flow network. A fault prediction module with a built-in improved spatiotemporal attention deep learning model identifies early equipment anomalies and classifies them into four risk levels. An active self-healing scheduling module constructs a multi-objective optimization model to solve for the optimal scheduling strategy, which is then virtually verified by the digital twin. The execution control module decomposes and distributes instructions to each terminal for execution. Finally, a visualization monitoring module displays the global status and allows for manual intervention, thereby achieving a complete closed loop of collaborative scheduling and fault self-healing for multi-energy flow (electricity, heat, and air conditioning) in the industrial park.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a smart energy load collaborative scheduling and optimization system for industrial parks, which includes the following components: Data acquisition and preprocessing module: Collects multi-source operation data from each node of the park's energy system in real time and completes standardized processing, and synchronously transmits the processed data to the digital twin mapping module and the fault prediction module; Digital twin mapping module: Constructs a virtual mirror of all elements of the park's energy system to realize real-time dynamic mapping and operational status simulation of multi-energy flow networks; Fault prediction module: Based on processed operating data and historical fault data, it identifies early abnormal signals of the equipment, predicts key fault information, and generates a risk assessment report; Active self-healing scheduling module: Constructs a multi-objective optimization model based on fault warning signals and solves for the optimal self-healing scheduling strategy; Execution control module: decomposes and issues scheduling instructions, executes corresponding operations, and provides feedback on the execution results; Visual monitoring module: Provides a human-computer interaction interface to display the system's operating status and scheduling information in real time, and supports manual intervention by dispatchers in the scheduling process.

[0007] Furthermore, the data acquisition and preprocessing module deploys IoT sensors on the generation, transmission, distribution, load, and energy storage sides of the park's energy system. The sampling frequency of the sensors is set to 0.1Hz-10Hz according to the importance of the equipment. The collected data includes the output power and operating parameters of the distributed power source, the voltage and current of the energy storage system, the oil temperature and level and load rate of the SOC transformer, the current, voltage and power factor of the lines, the electricity load data of users, and the temperature, humidity, and wind speed data of the environment. The module uses the 3σ criterion to remove outliers in the data, uses linear interpolation to fill missing data, and uses the minimum-maximum normalization method to map all data to the 0-1 range. The processed data is transmitted to the subsequent modules through a 5G private network or optical fiber. The transmission process uses the AES-256 encryption algorithm to ensure data security.

[0008] Furthermore, the digital twin mapping module first constructs a three-dimensional geometric model based on the CAD drawings and on-site survey data of the park's energy system, restoring the spatial layout and equipment form of the system. Then, it adds the physical parameters and operating characteristics of each piece of equipment, including the rated capacity efficiency curve, loss characteristics, and fault threshold of the equipment, to construct a physical model. Next, it trains the behavior model of the equipment based on the park's historical operating data over the past 5 years, simulating the response characteristics of the equipment under different operating conditions. Finally, through a real-time data interface, it synchronizes the real-time operating data output by the data acquisition and preprocessing module to the digital twin, with an update frequency set to 1 second, to achieve real-time synchronization between the virtual system and the physical system, providing a simulation environment for fault prediction and scheduling verification.

[0009] Furthermore, the fault prediction module incorporates an improved spatiotemporal attention deep learning model. The specific network structure of this model is as follows: the input layer receives multi-source time-series data, which then sequentially passes through a temporal attention layer, a spatial attention layer, a fusion layer, two fully connected layers, and an output layer. The model employs the Adam optimizer, with an initial learning rate of 0.001, a batch size of 64, and a training epoch count not exceeding 200 epochs. The early stopping strategy is based on validation set loss, and the model's loss function uses the following formula: Where L is the total loss value of the model, This is the standard cross-entropy loss function, used to measure the accuracy of the model's fault classification. The loss balance coefficient ranges from 0.1 to 0.5. The optimal value is determined on the training set using a grid search method. This is the fault risk attenuation loss term, used to penalize the model for missing high-risk faults. Its calculation method is as follows: Where N is the total number of predefined fault categories, The impact weight of the i-th type of fault is determined based on the load loss and repair cost caused by the fault. The fault risk attenuation coefficient, with a value of 0.05, is determined based on the fault propagation speed of the park's energy system. The time interval between the predicted occurrence time of the i-th type of fault and the current time is expressed in hours (h).

[0010] Furthermore, the fault prediction module divides the fault risk level into four levels, each corresponding to different fault occurrence probabilities and impact ranges. When the fault occurrence probability is less than 30% and the impact range is less than 1MW, it is considered a low-risk level, which only provides text prompts on the visual monitoring interface without triggering other operations. When the fault occurrence probability is between 30% and 60% or the impact range is between 1MW and 5MW, it is considered a medium-risk level, which sends SMS and email warnings to dispatchers to remind them to pay attention to the equipment's operating status. When the fault occurrence probability is between 60% and 90% or the impact range is between 5MW and 20MW, it is considered a high-risk level, which automatically triggers the proactive self-healing dispatch process, generates and executes a preliminary dispatch strategy. When the fault occurrence probability is higher than 90% or the impact range is greater than 20MW, it is considered an emergency risk level, which immediately executes emergency load transfer and equipment isolation operations, prioritizing the power supply to critical loads.

[0011] Furthermore, the multi-objective optimization comprehensive objective function constructed by the active self-healing scheduling module adopts the following formula: Where F is the comprehensive objective function value, and the optimization objective is to minimize F. The total operating cost of the system includes the generation cost of distributed power sources, the charging and discharging loss cost of the energy storage system, and the cost of purchasing electricity from the grid. The cost of load loss due to a fault is calculated based on the capacity of the disconnected load and the unit load loss. This is a penalty item for equipment health, used to prevent aging equipment from operating under overload conditions. Its value is inversely proportional to the equipment's health level. , , These are weighting coefficients, determined using the analytic hierarchy process (AHP), and adjusted according to the park's operational needs. The default values ​​are 0.4, 0.35, and 0.25.

[0012] Furthermore, after generating the initial scheduling strategy, the proactive self-healing scheduling module sends the strategy parameters to the digital twin mapping module. The entire execution process of the scheduling strategy is simulated in the digital twin, with the simulation time step set to 1 minute. System operation data during the simulation process is acquired, including the voltage and current of each node, the load rate of the equipment, the power supply status of the load, and the operating cost of the system. The simulation results are compared with the expected target, and the deviation values ​​of each indicator are calculated. When the deviation value of any indicator exceeds 5%, the proactive self-healing scheduling module adjusts the constraints and weight coefficients of the optimization model according to the simulation results, regenerates the scheduling strategy, and performs virtual verification again until the deviation values ​​of all indicators are within the allowable range. Finally, the final scheduling strategy is sent to the execution control module for execution.

[0013] Furthermore, the execution control module receives the global scheduling instructions generated by the active self-healing scheduling module, and decomposes the instructions into specific control instructions for each energy unit and load terminal, including output adjustment instructions for distributed power sources, charging and discharging instructions for energy storage systems, opening and closing instructions for circuit breakers, and load adjustment instructions for users. The instructions are sent to each terminal device through Modbus OPCUA or MQTT communication protocols. The module collects the operating status and instruction execution status of each device in real time, with the collection frequency set to 1Hz, and synchronously transmits the feedback data to the digital twin mapping module and the active self-healing scheduling module. When the deviation between the actual execution value and the instruction value is detected to exceed 5%, a deviation signal is automatically sent to the active self-healing scheduling module to trigger dynamic adjustment of the scheduling strategy.

[0014] Furthermore, the visualization monitoring module is developed based on Web technology, supporting access from both PC and mobile devices. It provides a visual display interface for the overall system operation status, displays the power flow distribution of multiple energy sources (electricity, heat, and cold) in real time, the operating parameters of each device, and the system's performance indicators. Fault warning information is pushed out via pop-up windows and sound prompts, while also displaying the type, location, estimated occurrence time, and impact range of the fault. The module displays the execution progress of the self-healing scheduling scheme and the completion status of each step, saves all historical operating data and scheduling records, and supports querying and statistics by time and device fault type. Scheduling personnel can manually adjust scheduling parameters and issue control commands through the interface, and the system restricts the scope of operation based on user permissions.

[0015] Furthermore, the system supports the coordinated scheduling of multiple energy flows, including electricity, heating, cooling, and gas. It comprehensively considers the coupling characteristics of these multiple energy flows during fault prediction and proactive self-healing. When a power system failure occurs, it not only adjusts the power system's operating status but also optimizes the load distribution of the heating, cooling, and gas supply systems. It utilizes the heat and cold storage capabilities of the heating and cooling networks for energy transfer, compensating for power supply gaps. Simultaneously, it coordinates the output of the gas turbines to convert gas energy into electricity, achieving mutual complementarity and support among multiple energy flows. The system also supports energy exchange with external power grids, heating networks, and gas networks. When energy is insufficient within the park, it purchases energy from external networks and sells energy to external networks when there is a surplus, further enhancing the system's reliability and economy.

[0016] Compared with existing technologies, this smart energy load collaborative scheduling and optimization system for industrial parks has the following advantages: I. This system, through its built-in improved spatiotemporal attention deep learning model, can identify early abnormal signals of equipment and predict key fault information. It also categorizes fault risks into four levels: low, medium, high, and emergency, corresponding to differentiated response strategies such as text prompts, SMS / email alerts, automatic self-healing scheduling, emergency load transfer, and equipment isolation. After generating a strategy, the proactive self-healing scheduling module first performs virtual verification in a digital twin, with a simulation time step set to 1 minute. When any indicator deviation exceeds 5%, the model is automatically adjusted and optimized, and re-verified to ensure the scheduling strategy is safe and reliable before being issued for execution. This achieves a complete closed loop of prediction-decision-verification-execution.

[0017] Second, this system comprehensively considers the coupling characteristics of multiple energy flows during fault prediction and proactive self-healing. When a power system fault occurs, it not only adjusts the power system's operating status but also optimizes the load distribution of heating, cooling, and gas supply systems. It utilizes the heat and cold storage capabilities of the heating and cooling networks to transfer energy and compensate for power supply gaps. Simultaneously, it coordinates gas turbines to convert gas energy into electrical energy, achieving complementary and mutually supportive multi-energy flows. Furthermore, the system supports energy exchange with external power grids, heating networks, and gas networks. When internal energy is insufficient, it purchases energy from external sources; when there is a surplus, it sells energy to external sources. By combining a multi-objective optimization model to comprehensively minimize operating costs, fault loss costs, and equipment health penalties, the system's economic efficiency is effectively improved.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0020] Figure 1 A flowchart of an overall system for collaborative scheduling and optimization of smart energy loads in a park. Figure 2 A flowchart for fault prediction and four-level risk classification response of a smart energy load collaborative scheduling and optimization system for a park. Figure 3 This is a flowchart of an active self-healing scheduling and digital twin virtual verification process for a smart energy load collaborative scheduling and optimization system in a park. Detailed Implementation

[0021] 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.

[0022] Example This embodiment is applied to an equipment manufacturing industrial park with a total power load of approximately 35MW. The park is equipped with a 20MW distributed photovoltaic power station, a 5MW / 10MWh lithium battery energy storage system, three 10MVA gas turbines, two 5MW electric chiller units, and a total capacity of 8000m³ / h. 3 The system includes a cold storage tank that is connected to both the external 110kV power grid and the municipal gas pipeline network. The scenario for this implementation was during the summer peak electricity consumption period. Due to prolonged high-load operation, the No. 3 outgoing line bay of the park's 10kV power distribution system experienced a localized overheating fault. The system completed the entire process of dispatching from fault prediction to proactive self-healing.

[0023] The data acquisition and preprocessing module deploys IoT sensors on the power generation, transmission, distribution, load, and energy storage sides of the industrial park. It collects real-time data on the output power and inverter operating parameters of distributed photovoltaic systems, the voltage, current, and SOC of energy storage systems, the oil temperature, oil level, and load factor of transformers, the current, voltage, and power factor of distribution lines, the power load data of each production workshop, and environmental data such as temperature, humidity, and wind speed. The module uses the 3σ criterion to remove outliers from the collected data, employs linear interpolation to fill in missing data caused by brief sensor downtime, and uses a minimum-maximum normalization method to map all data to the 0-1 range. The processed data is then synchronously transmitted via a 5G private network to the digital twin mapping module and the fault prediction module. AES-256 encryption is used during transmission to ensure data security.

[0024] The digital twin mapping module first constructs a three-dimensional geometric model based on the CAD drawings and on-site survey data of the park's energy system, fully restoring the spatial layout and equipment form of the park's substations, distribution lines, distributed power sources, energy storage systems, and cold and heat energy stations. Then, it adds physical parameters such as the rated capacity, efficiency curves, loss characteristics, and fault thresholds of each device to construct a physical model. Next, it trains the behavior model of each device based on the park's historical operating data over the past 5 years, simulating the response characteristics of the equipment under different loads and ambient temperatures. Finally, it synchronizes the real-time operating data output by the data acquisition and preprocessing module to the digital twin through a real-time data interface, realizing real-time dynamic mapping and operating status simulation of the park's multi-energy flow network.

[0025] The fault prediction module receives the preprocessed operational data and calls the built-in improved spatiotemporal attention deep learning model to analyze the operational status of the No. 3 exit interval; the model's loss function uses the following formula: Where L is the total loss value of the model, The standard cross-entropy loss function, This is the loss balance coefficient. This is the fault risk attenuation loss term, used to penalize the model for missing high-risk faults. Its calculation method is as follows: ,in The influence weight of the i-th type of fault, This is the failure risk attenuation coefficient. The time interval between the predicted occurrence of the i-th type of fault and the current time is used. During model training and inference, a loss function including the standard cross-entropy loss function and a fault risk attenuation loss term is used for optimization, focusing on penalizing missed detections of high-risk faults. In the initial stage, the model identifies an abnormal increase in the conductor temperature of this interval, predicting a fault probability of 25% and an impact range of approximately 0.8MW, classifying it as a low-risk fault, and only providing a text prompt on the visual monitoring interface. Two hours later, the conductor temperature of this interval continues to rise to 92℃, and the model recalculates the fault probability to 72%, with an impact range of approximately 8MW, classifying it as a high-risk fault, and automatically triggering the proactive self-healing scheduling process.

[0026] After receiving an advanced risk warning signal, the proactive self-healing scheduling module constructs a multi-objective optimization comprehensive objective function; the formula is: Where F is the value of the comprehensive objective function, The total operating cost of the system, Costs of load loss due to faults, This is a penalty item for device health. , , Using weighted coefficients, and taking the total system operating cost, load loss cost due to failure, and equipment health penalty as optimization objectives, an initial self-healing scheduling strategy is generated based on the current energy supply and demand situation in the park. The module sends the initial strategy parameters to the digital twin mapping module, which simulates the complete execution process of the scheduling strategy within the digital twin. During the simulation, the module obtains the voltage and current of each node, equipment load rate, load power supply status, and system operating cost. The simulation results are compared with the expected targets, and the deviation values ​​of each indicator are calculated. The initial simulation results show that the energy storage system's discharge power deviation is 7.3%, exceeding the allowable range of 5%. Based on the simulation results, the module adjusts the constraints and weighted coefficients of the optimization model, regenerates the scheduling strategy, and performs virtual verification again until the deviation values ​​of all indicators are within the allowable range. Finally, the final scheduling strategy is sent to the execution control module.

[0027] After receiving the global scheduling command, the execution control module decomposes it into specific control commands for each energy unit and load terminal. These commands include transferring the 2MW non-core production load from line bay 3 to line bays 1 and 2, discharging the energy storage system at full load (4MW), increasing the gas turbine output to 8MW, reducing the electric chiller output by 20%, and starting the cold storage tank for cooling. The commands are sent to each terminal device via Modbus and OPCUA communication protocols. The module collects the operating status of each device and the command execution status in real time, synchronously transmitting the feedback data to the digital twin mapping module and the active self-healing scheduling module. During execution, if a deviation of 6.1% is detected between the actual load adjustment value and the command value in a certain workshop, the module automatically sends a deviation signal to the active self-healing scheduling module. The module then dynamically adjusts the scheduling strategy, transferring the remaining load requiring adjustment in that workshop to other adjustable load terminals to ensure the scheduling objective is achieved.

[0028] The visualization monitoring module is developed based on web technology, supporting access from both PC and mobile devices. It displays the real-time power flow distribution of electricity, heat, and cold in the park, the operating parameters of each device, and the system's performance indicators. When a fault occurs, it pushes early warning information via pop-up windows and sound alerts, simultaneously displaying the fault type as local overheating of the distribution line, the fault location as the No. 3 outgoing line bay of the 10kV distribution room, the estimated occurrence time as 1.5 hours, and the affected area as approximately 8MW. The module displays the execution progress and completion status of the self-healing dispatching scheme throughout the entire process, saves all historical operating data and dispatching records, and supports querying and statistics by time, device, and fault type. Dispatchers can manually fine-tune dispatching parameters through the interface, and the system restricts the scope of operation based on user permissions.

[0029] During this dispatching process, the system fully leveraged its multi-energy flow coordinated dispatching capabilities. While adjusting the power system's operating status, it optimized the load distribution of the heating and cooling systems, utilized the cold storage capacity of the cold storage tanks to transfer cooling loads, and compensated for the power supply gap in the power system. Simultaneously, it coordinated with gas turbines to increase output and convert gas energy into electricity. During the fault handling period, the core production loads in the park did not experience power outages. The total system operating cost was reduced by 18% compared to traditional fault handling methods, the impact of the fault was controlled within 2MW, and no equipment damage occurred.

[0030] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart energy load collaborative scheduling and optimization system for industrial parks, characterized in that, The system includes the following components: Data acquisition and preprocessing module: Collects multi-source operation data from each node of the park's energy system in real time and completes standardized processing, and synchronously transmits the processed data to the digital twin mapping module and the fault prediction module; Digital twin mapping module: Constructs a virtual mirror of all elements of the park's energy system to realize real-time dynamic mapping and operational status simulation of multi-energy flow networks; Fault prediction module: Based on processed operating data and historical fault data, it identifies early abnormal signals of the equipment, predicts key fault information, and generates a risk assessment report; Active self-healing scheduling module: Constructs a multi-objective optimization model based on fault warning signals and solves for the optimal self-healing scheduling strategy; Execution control module: decomposes and issues scheduling instructions, executes corresponding operations, and provides feedback on the execution results; Visual monitoring module: Provides a human-computer interaction interface to display the system's operating status and scheduling information in real time, and supports manual intervention by dispatchers in the scheduling process.

2. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The data acquisition and preprocessing module deploys IoT sensors on the power generation, transmission, distribution, load, and energy storage sides of the park's energy system. The collected data includes the output power and operating parameters of distributed power sources, the voltage and current of energy storage systems, the oil temperature and level and load rate of SOC transformers, the current, voltage, and power factor of lines, user electricity load data, and environmental temperature, humidity, and wind speed data. The module uses the 3σ criterion to remove outliers from the data, uses linear interpolation to fill missing data, and uses the minimum-maximum normalization method to map all data to the 0-1 range. The processed data is transmitted to subsequent modules via a 5G private network or optical fiber, and the AES-256 encryption algorithm is used to ensure data security during the transmission process.

3. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The digital twin mapping module first constructs a three-dimensional geometric model based on the CAD drawings and on-site survey data of the park's energy system to restore the spatial layout and equipment form of the system. Then, it adds the physical parameters and operating characteristics of each piece of equipment, including the rated capacity efficiency curve, loss characteristics and fault threshold of the equipment, to build a physical model. Next, it trains the behavior model of the equipment based on the park's historical operating data over the past 5 years to simulate the response characteristics of the equipment under different operating conditions. Finally, it synchronizes the real-time operating data output by the data acquisition and preprocessing module to the digital twin through a real-time data interface.

4. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The fault prediction module incorporates an improved spatiotemporal attention deep learning model. The specific network structure of this model is as follows: the input layer receives multi-source time-series data, which then sequentially passes through a temporal attention layer, a spatial attention layer, a fusion layer, two fully connected layers, and an output layer. The model uses the Adam optimizer, with an initial learning rate of 0.001, a batch size of 64, and a training epoch count not exceeding 200. The early stopping strategy is based on validation set loss, and the model's loss function uses the following formula: Where L is the total loss value of the model, The standard cross-entropy loss function, This is the loss balance coefficient. This is the fault risk attenuation loss term, used to penalize the model for missing high-risk faults. Its calculation method is as follows: ,in The influence weight of the i-th type of fault, This is the failure risk attenuation coefficient. Let be the time interval between the predicted occurrence time of the i-th type of fault and the current time.

5. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The fault prediction module classifies fault risk levels into four levels, each corresponding to different fault occurrence probabilities and impact ranges. When the fault occurrence probability is below 30% and the impact range is less than 1MW, it is considered a low-risk level, which only provides text prompts on the visual monitoring interface without triggering other operations. When the fault occurrence probability is between 30% and 60% or the impact range is between 1MW and 5MW, it is considered a medium-risk level, which sends SMS and email alerts to dispatchers to remind them to pay attention to the equipment's operating status. When the fault occurrence probability is between 60% and 90% or the impact range is between 5MW and 20MW, it is considered a high-risk level, which automatically triggers an active self-healing dispatch process, generates and executes a preliminary dispatch strategy. When the fault occurrence probability is above 90% or the impact range is greater than 20MW, it is considered an emergency risk level, which immediately executes emergency load transfer and equipment isolation operations.

6. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The multi-objective optimization comprehensive objective function constructed by the active self-healing scheduling module adopts the following formula: Where F is the value of the comprehensive objective function, The total operating cost of the system, Costs of load loss due to faults, This is a penalty item for device health. , , These are the weighting coefficients.

7. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, After generating the initial scheduling strategy, the proactive self-healing scheduling module sends the strategy parameters to the digital twin mapping module. The module simulates the complete execution process of the scheduling strategy within the digital twin, acquiring system operation data during the simulation, including the voltage and current of each node, the load rate of the equipment, the power supply status of the load, and the system's operating costs. The simulation results are compared with the expected targets, and the deviation values ​​of each indicator are calculated. When the deviation value of any indicator exceeds 5%, the proactive self-healing scheduling module adjusts the constraints and weight coefficients of the optimization model based on the simulation results, regenerates the scheduling strategy, and performs virtual verification again until the deviation values ​​of all indicators are within the allowable range. Finally, the final scheduling strategy is sent to the execution control module for execution.

8. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The execution control module receives global scheduling instructions generated by the active self-healing scheduling module, and decomposes the instructions into specific control instructions for each energy unit and load terminal, including output adjustment instructions for distributed power sources, charging and discharging instructions for energy storage systems, opening and closing instructions for circuit breakers, and load adjustment instructions for users. The instructions are sent to each terminal device through Modbus OPCUA or MQTT communication protocols. The module collects the operating status and instruction execution status of each device in real time, and transmits the feedback data synchronously to the digital twin mapping module and the active self-healing scheduling module. When the deviation between the actual execution value and the instruction value is detected to exceed 5%, a deviation signal is automatically sent to the active self-healing scheduling module to trigger dynamic adjustment of the scheduling strategy.

9. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The visualization monitoring module is developed based on Web technology, supporting access from both PC and mobile devices. It provides a visual display interface for the overall system operation status, displays the power flow distribution of electric, heating, and cooling systems in real time, the operating parameters of each device, and the system's performance indicators. Fault warning information is pushed out via pop-up windows and sound prompts, while also displaying the type, location, estimated occurrence time, and impact range of the fault. The module displays the execution progress of the self-healing scheduling scheme and the completion status of each step, saves all historical operating data and scheduling records, and supports querying and statistics by time and device fault type. Scheduling personnel can manually adjust scheduling parameters and issue control commands through the interface, and the system restricts the scope of operation based on user permissions.

10. The smart energy load collaborative scheduling and optimization system for industrial parks according to claim 1, characterized in that, The system supports the coordinated scheduling of multiple energy flows, including electricity, heating, cooling, and gas. It comprehensively considers the coupling characteristics of multiple energy flows during fault prediction and proactive self-healing. When a power system fault occurs, it not only adjusts the operating status of the power system but also optimizes the load distribution of the heating, cooling, and gas supply systems. It utilizes the heat and cold storage capacity of the heating and cooling networks to transfer energy, making up for the power supply gap in the power system. At the same time, it coordinates the output of the gas turbine to convert gas energy into electrical energy. The system also supports energy exchange with the power grid, heating network, and gas network outside the park. When there is a shortage of energy within the park, it purchases energy from the external network and sells energy to the external network when there is a surplus of energy.

Images