An edge AI-driven integrated wind, solar, and energy storage control system

Abstract

This invention relates to an edge AI-driven integrated wind-solar-storage coordinated control system, belonging to the field of new energy coordinated control technology. The system includes: a multi-source data sensing unit for collecting and normalizing electrical quantity data from photovoltaic, wind turbines, and energy storage, and adaptively generating a standardized dataset through a protocol; an edge computing and coordination control unit integrating a lightweight neural network to achieve ultra-short-term prediction of wind and solar power generation, and issuing coordinated control commands based on multi-objective optimization and a priority dynamic allocation mechanism; a wind-solar-storage adaptive learning unit that adjusts model parameters and control thresholds online based on command execution feedback and ambient temperature to achieve system adaptive optimization; and a security protection and communication unit that ensures the physical and data transmission security of the system through hardware encryption and a dual-mode redundant communication architecture. This system achieves on-site intelligent coordination, dynamic optimization, and safe and reliable operation of the wind-solar-storage system, improving energy utilization efficiency and power supply stability.

Description

Technical Field

[0001] This invention belongs to the field of new energy collaborative control technology, specifically relating to an integrated wind, solar and energy storage collaborative control system driven by edge AI. Background Technology

[0002] With the increasing proportion of new energy sources such as photovoltaic and wind power in the power system, the significant intermittency, volatility, and randomness of wind and solar power generation pose a severe challenge to the stable operation of the power grid due to their large-scale grid connection. Combining energy storage systems with wind and solar power generation to form integrated wind-solar-storage systems, and using coordinated regulation to smooth power fluctuations and improve absorption capacity, has become a key technological direction for ensuring the safe and stable operation of the new power system.

[0003] Currently, wind, solar, and energy storage coordinated regulation mainly relies on centralized energy management systems (EMS). These systems are typically deployed in the cloud or at a master station, collecting data from all sites via wide-area communication networks for centralized prediction and optimization decisions. This model has several inherent drawbacks: First, the remote transmission of massive amounts of data leads to significant communication latency and bandwidth pressure, making it difficult to meet the demands of high-frequency, fast-response real-time regulation. It also suffers from insufficient timeliness in the face of power fluctuations on the order of seconds or minutes. Second, the centralized model has limited generalization capabilities, making it difficult to adapt to differences in site topology, equipment models, and environmental characteristics. Furthermore, the model parameters are fixed and cannot be adjusted online to keep pace with slow time-varying factors such as equipment aging and performance degradation, resulting in a gradual decrease in regulation accuracy over long-term operation. Third, existing systems often focus on the network layer for security protection, lacking end-to-end protection from physical devices and embedded hardware to data transmission, making them vulnerable to increasingly severe industrial control security threats. In addition, most systems have insufficient redundancy in their communication link design; a single link failure can easily lead to overall regulation failure, affecting power supply reliability.

[0004] Therefore, how to achieve a rapid-response, highly adaptive, safe, and reliable on-site coordinated control system for wind, solar, and energy storage has become a technical problem that needs to be solved in this field. There is an urgent need for a distributed solution that can deeply integrate artificial intelligence prediction, multi-objective optimization decision-making, and edge computing capabilities, and possess online learning and end-to-end security protection capabilities, in order to improve the operational efficiency, autonomy, and resilience of the wind-solar-energy storage combined system. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides an edge AI-driven integrated wind, solar, and energy storage collaborative control system. The objective of this invention can be achieved through the following technical solutions: An edge AI-driven integrated wind, solar and energy storage coordinated control system includes: a multi-source data sensing unit, an edge computing and coordination control unit, a wind, solar and energy storage adaptive learning unit, and a security protection and communication unit. The multi-source data sensing unit acquires electrical quantity data from local photovoltaic inverters, wind turbine frequency converters, and liquid-cooled energy storage systems, normalizes the electrical quantity data, and generates a standardized dataset; it is configured with an expandable acquisition interface, automatically identifies and adapts equipment through industrial communication protocols, and generates a communication protocol adaptation list. The edge computing and coordination control unit integrates a lightweight neural network model, performs ultra-short-term prediction of local wind and solar power generation based on the standardized dataset, and optimizes the model structure to adapt to the prediction cycle; it constructs a multi-objective collaborative optimization mechanism, calculates and issues wind, solar and energy storage coordinated control commands, and builds a command closed-loop link; it sets a priority dynamic allocation mechanism, coordinates command conflicts through power supply stability levels, and generates a priority adjustment record table. The wind-solar-storage adaptive learning unit extracts equipment status features and adjusts the parameters of the lightweight neural network model based on the wind-solar-storage collaborative control command to adapt to changes in equipment aging performance; it also synchronously links environmental temperature sensor data to match model parameters, calibrates control thresholds to adapt to temperature fluctuations, presets a performance feedback mechanism, compares command execution results with preset targets, optimizes the learning rate by combining the priority adjustment record table, and generates a model parameter optimization report. The security protection and communication unit is equipped with sealed protection and dustproof components, has a built-in heat dissipation module and optimized heat dissipation channel design; it integrates a hardware encryption chip and an embedded firewall to perform end-to-end encryption on uplink and downlink data and establish a transmission security verification mechanism; it adopts a dual-mode redundant communication architecture to monitor the link operation status and generate a communication status report.

[0006] Specifically, the process of generating the standardized dataset is as follows: Electrical quantity data is categorized based on the type of acquisition equipment. Normalized value ranges are set for each data category. Normalization processing is performed on the electrical quantity data according to the ranges. Simultaneously, the processed data is integrated based on the data acquisition time sequence to form a structured and standardized dataset.

[0007] Specifically, the process of automatically identifying and adapting devices through industrial communication protocols is as follows: Scan the communication identification information of the access device, match the protocol type of the industrial communication protocol, configure the communication parameters of the scalable acquisition interface based on the matching result, adapt the communication link between the interface and the device, verify the communication stability of the adapted link, and generate the correspondence information between the device and the protocol.

[0008] Specifically, the process of making ultra-short-term predictions of local wind and solar power generation is as follows: The standardized dataset is input into a lightweight neural network model. The standardized dataset is then split into two data subsets: photovoltaic power generation and wind power generation. The power change characteristics and time series correlation patterns of the two data subsets are extracted. The number of neurons in the model is adjusted to match the ultra-short-term prediction cycle. The power generation of wind and solar power is predicted for each time period. The prediction deviation is calibrated by comparing the prediction with historical data of wind and solar power generation in the same period, and the local ultra-short-term prediction results of wind and solar power generation are generated.

[0009] Specifically, the process of constructing the multi-objective collaborative optimization mechanism is as follows: Multiple objective parameters for wind-solar-storage coordinated regulation are preset and corresponding weight coefficients are configured. Based on the multiple objective parameters and weight coefficients, a multi-objective coordinated optimization function is constructed. The ultra-short-term forecast results of wind and solar power generation, the communication protocol compatibility list, and the real-time operation data of the energy storage system are imported into the optimization function to generate a set of regulation parameters. The set of regulation parameters is verified for equipment operation constraints and adjusted by gradient optimization to generate regulation parameters that meet the equipment operation requirements.

[0010] Specifically, the process of coordinating command conflicts through power supply stability levels is as follows: The system divides power supply stability into levels, assigns priority values ​​to each level according to the importance of load power supply, and sets power supply guarantee standards. It monitors the issuance status of wind-solar-storage coordinated control commands in real time, performs conflict judgment on the issued commands, and records relevant information of conflict commands. It extracts the priority values ​​corresponding to the conflict commands, filters and executes the command with the best priority value through value comparison, and generates a priority adjustment record table.

[0011] Specifically, the process of extracting device state features and adjusting the parameters of the lightweight neural network model is as follows: Based on the wind-solar-storage coordinated control command, the full data of equipment operation is retrieved, the characteristics of equipment operation status are extracted, and the performance deviation characteristics are located by comparing them with the equipment benchmark operation characteristics. An incremental learning strategy is adopted to adjust the parameters of the model layer corresponding to the deviation characteristics, verify the matching degree between the model output and the actual operation status of the equipment, and adjust the model parameters quantitatively based on the matching metric.

[0012] Specifically, the calibration and adjustment threshold adapts to temperature fluctuations, and the specific process is as follows: Ambient temperature sensor data is acquired, and temperature ranges are divided based on the temperature change amplitude. Preset control threshold benchmark values ​​are set for the temperature ranges. The current ambient temperature range is compared in real time, and the corresponding threshold benchmark value is retrieved. The threshold benchmark value is corrected in combination with the real-time operating status of the equipment. The corrected threshold value is substituted into the model to carry out operation verification. Based on the verification results, control threshold values ​​are generated and updated to the lightweight neural network model.

[0013] Specifically, the process of generating the model parameter optimization report is as follows: Record the values ​​of lightweight neural network model parameters before and after adjustment, key data of equipment status feature extraction, and calibration values ​​of temperature adaptation control thresholds and corresponding temperature ranges; statistically analyze the key operating indicators of the model after parameter adjustment, calculate the changes in the key operating indicators before and after adjustment, and label the indicator evaluation dimensions; record the key nodes of parameter adjustment and node triggering conditions, and synchronously record the adaptation judgment results of equipment operating status after each node adjustment; arrange various types of information according to the logic of data recording, indicator statistics, and node registration, configure exclusive data identifiers for information, and label the data collection time, equipment operating conditions, and control scenarios to form a model parameter optimization report.

[0014] Specifically, the optimized heat dissipation channel design involves the following process: The system acquires and categorizes data on the distribution of heat dissipation points during equipment operation. Based on the required airflow for heat dissipation, it presets channel cross-sectional dimensions and optimizes the internal structural parameters of the channels. It adjusts the heat dissipation distribution parameters based on the corresponding channel segments for each heat dissipation point, presets temperature sensing nodes at key locations within the channels, calls real-time monitoring data from these nodes, generates ventilation ratio parameters for the corresponding channel segments, iteratively adjusts the heat dissipation channel design, and verifies the heat dissipation adaptability of the design. The internal structural parameters include channel bending angle, inner wall smoothness, and channel length ratio. The real-time monitoring data includes real-time temperature, ventilation velocity, and heat dissipation efficiency for each channel segment.

[0015] Specifically, the process of performing end-to-end encryption on uplink and downlink data is as follows: Encryption keys are allocated to the uplink and downlink data transmission links. Based on the key synchronization mechanism configured at both ends of the communication, the uplink and downlink data before transmission are processed in blocks. Encryption operations are performed on each data block. A unique verification identifier is added to the encrypted data block. The data receiving end calls the synchronization key to decrypt the data block. The received encrypted data block is decrypted. The consistency between the verification identifier and the decrypted data is compared. After the verification is passed, the data restoration operation is performed, and a full-process encryption and decryption record and key synchronization log are generated.

[0016] Specifically, the dual-mode redundant communication architecture operates as follows: Construct dual-mode redundant links for wired and wireless industrial communication, configure the communication parameters of the primary and backup links and perform link adaptation and debugging, set the transmission status judgment threshold for link switching, monitor the transmission rate and connectivity status of the primary link in real time, trigger the link switching command when the primary link fails to meet the transmission standard, enable the backup link to carry out data transmission, configure the status monitoring module at the link node, and record the triggering conditions and execution time of link switching.

[0017] The beneficial effects of this invention are as follows: (1) This invention integrates a lightweight neural network model by deploying a local edge computing and coordination control unit. It directly performs ultra-short-term power prediction and multi-objective optimization decision-making based on the standardized dataset provided by the multi-source data sensing unit. This avoids the communication delay caused by the remote transmission of massive data to the cloud and achieves rapid closed-loop control at the millisecond to second level, which can effectively smooth out the rapid fluctuations in wind and solar power. At the same time, the wind-solar-storage adaptive learning unit can adjust the model parameters and calibrate the control threshold online according to the instruction execution feedback, equipment status characteristics and changes in ambient temperature. This enables the system to continuously adapt to slow time-varying factors such as equipment performance degradation and environmental fluctuations, realizing the transformation from "fixed strategy" to "autonomous evolution". It maintains high-precision control in the long term, thereby improving the overall energy utilization efficiency, power supply quality and autonomous operation capability of the wind-solar-storage system.

[0018] (2) A comprehensive and in-depth security protection system from the physical layer to the data layer has been constructed, greatly enhancing the system's reliability, resilience, security, and robustness. This invention integrates hardware encryption chips, embedded firewalls, end-to-end encryption, and a dual-mode redundant communication architecture through independent security protection and communication units. This achieves end-to-end encryption and security verification of instructions and data from generation and transmission to execution, effectively resisting network attacks and data theft risks. Optimized physical sealing, heat dissipation, and dustproof design ensure stable operation of the hardware in harsh industrial environments. The dual-mode redundant communication mechanism ensures that the system can seamlessly switch to a backup link when a single communication link is interrupted, maintaining the continuous and reliable issuance of control instructions and avoiding system failure caused by single-point failures. This comprehensive protection combining hardware and software at multiple levels reduces system security risks and ensures the continuous, stable, and reliable operation of critical energy infrastructure in complex network and physical environments. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 This is an architecture diagram of an integrated wind-solar-storage collaborative control system based on edge AI driven according to the present invention. Figure 2 This is a timing diagram of an edge AI-driven integrated wind, solar, and energy storage control system according to the present invention. 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.

[0022] Please see Figure 1-2A wind-solar-storage integrated collaborative control system based on edge AI, comprising: a multi-source data sensing unit, an edge computing and coordination control unit, a wind-solar-storage adaptive learning unit, and a security protection and communication unit; The multi-source data sensing unit acquires electrical quantity data from local photovoltaic inverters, wind turbine frequency converters, and liquid-cooled energy storage systems, normalizes the electrical quantity data, and generates a standardized dataset; it is configured with an expandable acquisition interface, automatically identifies and adapts equipment through industrial communication protocols, and generates a communication protocol adaptation list. The edge computing and coordination control unit integrates a lightweight neural network model, performs ultra-short-term prediction of local wind and solar power generation based on the standardized dataset, and optimizes the model structure to adapt to the prediction cycle; it constructs a multi-objective collaborative optimization mechanism, calculates and issues wind, solar and energy storage coordinated control commands, and builds a command closed-loop link; it sets a priority dynamic allocation mechanism, coordinates command conflicts through power supply stability levels, and generates a priority adjustment record table. The wind-solar-storage adaptive learning unit extracts equipment status features and adjusts the parameters of the lightweight neural network model based on the wind-solar-storage collaborative control command to adapt to changes in equipment aging performance; it also synchronously links environmental temperature sensor data to match model parameters, calibrates control thresholds to adapt to temperature fluctuations, presets a performance feedback mechanism, compares command execution results with preset targets, optimizes the learning rate by combining the priority adjustment record table, and generates a model parameter optimization report. The security protection and communication unit is equipped with sealed protection and dustproof components, has a built-in heat dissipation module and optimized heat dissipation channel design; it integrates a hardware encryption chip and an embedded firewall to perform end-to-end encryption on uplink and downlink data and establish a transmission security verification mechanism; it adopts a dual-mode redundant communication architecture to monitor the link operation status and generate a communication status report.

[0023] Specifically, the process of generating the standardized dataset is as follows: Electrical quantity data is categorized based on the type of acquisition equipment. Normalized value ranges are defined for each data category, and normalization processing is performed on the electrical quantity data according to these ranges. Simultaneously, the processed data is integrated based on the data acquisition time sequence to form a structured, standardized dataset. Specifically, the electrical quantity data includes core operating parameters for three types of equipment: Photovoltaic inverter electrical quantity data: photovoltaic array input voltage, photovoltaic array input current, inverter output voltage, inverter output current, inverter output active power, inverter output reactive power, and inverter conversion efficiency; Wind turbine inverter electrical quantity data: wind turbine hub speed, wind turbine pitch angle, ambient input wind speed, inverter output voltage, inverter output current, wind turbine output active power, and generator winding temperature; Liquid-cooled energy storage system electrical quantity data: battery cluster individual cell voltage, battery cluster total voltage, battery cluster charge / discharge current, battery cluster SOC (remaining capacity), battery cluster charge / discharge active power, battery cluster charge / discharge reactive power, battery cluster temperature, liquid-cooled system inlet water temperature, and liquid-cooled system outlet water temperature.

[0024] Specifically, the process of automatically identifying and adapting devices through industrial communication protocols is as follows: Scan the communication identification information of the access device, match the protocol type of the industrial communication protocol, configure the communication parameters of the scalable acquisition interface based on the matching result, adapt the communication link between the interface and the device, verify the communication stability of the adapted link, and generate the correspondence information between the device and the protocol.

[0025] In this embodiment, a wind-solar-storage microgrid containing a liquid-cooled energy storage system of a certain brand is used as the application scenario. The microgrid is configured with a 500kW photovoltaic inverter, an 800kW wind turbine inverter, a 1MWh liquid-cooled energy storage system, and an edge computing gateway. The data acquisition cycle is set to 1 minute, and the protocol adaptation cycle is set to 5 minutes. The specific implementation process is as follows: The automatic identification and device adaptation method for industrial communication protocols is implemented as follows: After the edge computing gateway is started, it actively scans the communication identification information of the access devices through the scalable acquisition interface, identifying the communication identifier of the photovoltaic inverter as MB-SLAVE-001, the communication identifier of the wind turbine inverter as PN-MASTER-002, and the communication identifier of the liquid-cooled energy storage system cluster management unit as IEC-61850-003; based on the pre-stored protocol library, it matches the protocol type and determines that the photovoltaic inverter is adapted to the ModbusRTU protocol, the wind turbine inverter is adapted to the Profinet protocol, and the liquid-cooled energy storage system cluster management unit is adapted to... The IEC61850 protocol is used. Based on the matching results, interface communication parameters are configured: the photovoltaic inverter interface baud rate is set to 9600bps and the data bits to 8 bits; the wind turbine inverter interface IP is set to 192.168.1.100 and the subnet mask to 255.255.255.0; and the energy storage cluster-level management unit interface APPID is set to 1001. After parameter configuration, the communication link status is continuously monitored for 5 minutes, and the link packet loss rate is statistically analyzed. If the packet loss rate is confirmed to be below 0.1%, the communication stability is deemed to meet the requirements. Finally, a table showing the correspondence between devices and protocols is generated and stored in the local storage module of the edge computing gateway.

[0026] Based on the above equipment adaptation results, the method for generating standardized datasets is executed, with the following steps: Electrical quantity data for three types of equipment are collected through the adapted communication links. Data is categorized according to the type of equipment collected: photovoltaic data includes photovoltaic array voltage, current, and output power; wind turbine data includes wind turbine speed, pitch angle, and output power; and liquid-cooled energy storage data includes battery cluster voltage, current, SOC, and charging / discharging power. A normalized numerical range is set for each data type: photovoltaic output power 0-500kW corresponds to the range 0-1; wind turbine output power 0-800kW corresponds to the range 0-1; and energy storage SOC 0-100% corresponds to the range 0-1. Linear normalization processing is performed on the collected raw electrical quantity data according to the set range. The calculation formula is as follows: ,in : Dimensionless data after normalization, with values ​​ranging from [0,1], used as input for subsequent models; : Raw collected values ​​of electrical quantities of equipment; The minimum value of the electrical quantity corresponding to the equipment is determined by the equipment's rated parameters; The maximum value of the corresponding electrical quantity of the equipment is determined by the equipment's rated parameters. Taking a photovoltaic inverter with an output power of 250kW at a certain moment as an example, substituting into the formula yields: =(250-0) / (500-0)=0.5; Based on the 1-minute acquisition time series, the normalized data of photovoltaic, wind turbine, and energy storage at the same acquisition time are integrated into a time series data record. Each record contains three fields: acquisition timestamp, equipment type identifier, and normalized electrical quantity value; 24-hour data is continuously collected and processed to form a structured and standardized dataset containing 1440 time series records, which is synchronously uploaded to the edge computing and coordination control unit to provide data support for subsequent ultra-short-term prediction of wind and solar power generation.

[0027] This embodiment achieves automatic protocol adaptation and standardized dataset generation for wind, solar, and energy storage microgrid devices through the above steps. The adaptation process requires no manual intervention, and the dataset has a unified format that can be directly used for subsequent model training and control command calculation.

[0028] Specifically, the process of making ultra-short-term predictions of local wind and solar power generation is as follows: The standardized dataset is input into a lightweight neural network model. The standardized dataset is then split into two data subsets: photovoltaic power generation and wind power generation. The power change characteristics and time series correlation patterns of the two data subsets are extracted. The number of neurons in the model is adjusted to match the ultra-short-term prediction cycle. The power generation of wind and solar power is predicted for each time period. The prediction deviation is calibrated by comparing the prediction with historical data of wind and solar power generation in the same period, and the local ultra-short-term prediction results of wind and solar power generation are generated.

[0029] Specifically, the process of constructing the multi-objective collaborative optimization mechanism is as follows: Multiple objective parameters for wind-solar-storage coordinated regulation are preset and corresponding weight coefficients are configured. Based on the multiple objective parameters and weight coefficients, a multi-objective coordinated optimization function is constructed. The ultra-short-term forecast results of wind and solar power generation, the communication protocol compatibility list, and the real-time operation data of the energy storage system are imported into the optimization function to generate a set of regulation parameters. The set of regulation parameters is verified for equipment operation constraints and adjusted by gradient optimization to generate regulation parameters that meet the equipment operation requirements.

[0030] Specifically, the process of coordinating command conflicts through power supply stability levels is as follows: The system divides power supply stability into levels, assigns priority values ​​to each level according to the importance of load power supply, and sets power supply guarantee standards. It monitors the issuance status of wind-solar-storage coordinated control commands in real time, performs conflict judgment on the issued commands, and records relevant information of conflict commands. It extracts the priority values ​​corresponding to the conflict commands, filters and executes the command with the best priority value through value comparison, and generates a priority adjustment record table.

[0031] In this embodiment, a 1.3MW wind-solar-storage microgrid in an industrial park is used as the application scenario. This microgrid is configured with a 500kW photovoltaic inverter, an 800kW wind turbine inverter, a 1MWh liquid-cooled energy storage system cluster management unit, and an edge computing gateway. It primarily supplies power to three types of loads: the park's monitoring center, production workshops, and lighting systems. The specific implementation is as follows: I. Short-term forecast of wind and solar power generation In this embodiment, the lightweight neural network model of the present invention can adopt, but is not limited to, a CNN-LSTM hybrid architecture. Lightweight convolutional neural networks (CNN), lightweight long short-term memory networks (LSTM), lightweight Transformers, and other models adapted to edge computing power can also be selected. This embodiment uses a CNN-LSTM hybrid lightweight neural network model as an example to detail the implementation process of the present invention. This model adapts to the computing power limitations of edge computing gateways, with the number of model parameters controlled within 100,000, supporting local edge computing without cloud computing power support. The model structure consists of four layers: an input layer, a feature extraction layer, a temporal prediction layer, and an output layer. The number of neurons in each layer is dynamically adjusted according to a 15-minute ultra-short-term prediction period. The specific structure and parameters are as follows: Model structure and parameter configuration Input layer: The number of neurons is set to 128, the input time sequence length is 60 minutes, corresponding to 60 1-minute standardized data points, and the input data dimension is [60,2], including normalized power values ​​of two data subsets: photovoltaic and wind turbine. The input data must meet the integrity check, and the proportion of missing data points must not exceed 5%; otherwise, the data completion mechanism will be triggered.

[0032] Feature extraction layer: A CNN convolutional layer is used, with 2 convolutional kernels, each with a kernel size of 3×1 and a stride of 1. The ReLU activation function is used to extract key features such as the slope of power change, the time of peak and valley occurrence, and the amplitude of power fluctuation.

[0033] Temporal prediction layer: An LSTM layer is used, and the number of neurons is adjusted according to the prediction period. The number of neurons is set to 64 in a 15-minute prediction period, and the forget gate threshold is set to 0.8 to capture the temporal correlation of power data and avoid the gradient vanishing problem.

[0034] Output layer: The number of neurons is set to 15, corresponding to the power prediction value of each time period within a 15-minute prediction cycle. The output data dimension is [15,2]. The output value needs to be mapped to the rated power range of the device to ensure that the physical meaning is valid.

[0035] Data processing and prediction process Time-series splitting: The standardized dataset was divided into 15-minute time intervals, and photovoltaic and wind turbine power data for the 60 minutes preceding the prediction time were extracted to form two data subsets. Outlier data points were removed using the 3σ principle; data exceeding the mean ± 3 times the standard deviation were considered outliers and were supplemented using linear interpolation.

[0036] Feature extraction: The two data subsets are convolutionally processed by CNN convolutional layers to generate feature vectors of dimension [58,2], which are then stored in the model cache. The feature extraction time is controlled within 200ms to meet the real-time requirements.

[0037] Time-series prediction: The LSTM layer performs time-series prediction based on feature vectors, outputting the predicted power values ​​of photovoltaics and wind turbines for every minute within 15 minutes. The prediction process adopts a batch processing method, with a single batch processing data volume of 32 sets.

[0038] Deviation calibration: Retrieve historical wind and solar power generation data for the same period (within the last 30 days), and calculate the mean square error between the predicted value and the historical value. The formula for calculating the mean square error is: ,in Historical values Let n be the predicted value and n be the number of data points. An error threshold of 5% is set; if the error exceeds this threshold, a linear correction is applied to the predicted value using the following formula: k is a correction coefficient with a value of 0.3. The final ultra-short-term prediction results of local wind and solar power generation are generated and transmitted to the multi-objective collaborative optimization module. The average absolute error of the prediction results is controlled within 3%. The above parameter settings refer to the requirements of the "Technical Specification for Operation and Control of New Energy Microgrids".

[0039] II. Construction of Multi-Objective Collaborative Optimization Mechanism The target parameters and weight coefficients are configured with three types of core target parameters, and weight coefficients are configured according to the microgrid's operating priority. The sum of the weight coefficients is 1, as detailed below: Objective 1: Maximize wind and solar power generation efficiency, with a weighting coefficient of 0.3. The evaluation index is the ratio of total wind and solar power generation to theoretical power generation, which is calculated based on environmental data such as irradiance and wind speed.

[0040] Objective 2: Achieve SOC balance for the energy storage system, with a weighting factor of 0.2. The evaluation criterion is to maintain the energy storage SOC within the range of 30%-70%. When the SOC is below 30%, prioritize charging; when it is above 70%, prioritize discharging.

[0041] Objective 3: Power supply stability, weighting factor 0.5. The evaluation index is that the load voltage fluctuation rate is controlled within ±2%, and the voltage fluctuation rate is calculated using the following formula: Where U is the actual voltage, U NThe rated voltage is used. A weighted summation type multi-objective collaborative optimization function is constructed: F = 0.3f1 + 0.2f2 + 0.5f3, where f1, f2, and f3 are the normalized evaluation values ​​of the three types of objectives, respectively, and the values ​​range from [0,1]. The higher the value, the better the achievement of the objective.

[0042] Model Input / Output and Parameter Verification Input parameters: 15-minute wind and solar power generation prediction results, communication protocol compatibility list, and real-time operating data of the liquid-cooled energy storage system. The energy storage operation data includes 12 parameters such as battery cluster SOC, charge and discharge power, single cell voltage, battery temperature, and inlet and outlet water temperatures of the liquid cooling system. The data sampling frequency is 1Hz.

[0043] Output parameters: Control parameter group, including photovoltaic inverter power limit coefficient (value range 0-1), wind turbine pitch angle adjustment step size (value range 0.5°-2°), energy storage charging and discharging power command (value range -200kW to 200kW, negative value is charging, positive value is discharging), a total of 3 categories and 15 specific parameters.

[0044] Constraint Verification: The output control parameter set is subjected to equipment operation constraint verification. The constraints are set as follows: photovoltaic inverter output power ≤ 500kW, wind turbine pitch angle adjustment range 0°-30°, energy storage charging and discharging power ≤ 200kW, and battery cell voltage ≥ 3.2V and ≤ 3.65V. Gradient descent is used to perform gradient optimization adjustments on parameters exceeding the constraints, with an adjustment step size of 5kW / time and no more than 10 iterations, until all parameters meet the operational requirements. The verified control parameter set is then distributed to each device via the protocols listed in the communication protocol compatibility list.

[0045] III. Power Supply Stability Level Command Conflict Coordination Power supply stability levels and priority configurations are divided into three levels based on the importance of the load. Priority values ​​and power supply protection standards are configured, with higher priority values ​​indicating higher power supply priority, as detailed below. Tier 1 (Priority value 3): Critical loads, including microgrid monitoring systems and energy storage cluster management units. Power supply guarantee standard is 99.99%, allowable interruption time ≤ 5 minutes / year, and adopts a dual-power redundant power supply mode.

[0046] Tier 2 (Priority Value 2): Critical loads, including equipment in the factory production workshop. Power supply guarantee standard is 99.5%, allowable interruption time ≤ 8h / year, and backup power supply is configured.

[0047] Tier 3 (Priority Value 1): General load, including factory lighting system. Power supply guarantee standard is 95%, allowable interruption time ≤72h / year, no backup power supply.

[0048] The edge computing gateway monitors the issuance status of control commands every 5 seconds. When a command conflict occurs, a priority comparison process is executed. This embodiment uses a typical conflict scenario as an example: an energy storage system charging command (adapting to level 3 load power supply, priority value 1) and a wind turbine power limiting command (adapting to level 1 load voltage stability, priority value 3) are issued simultaneously. The gateway extracts the priority values ​​corresponding to the two commands, compares the values, and selects the wind turbine power limiting command with the highest priority for execution, pausing the energy storage charging command. Simultaneously, it records information such as the type of conflicting command, issuance time, priority value, and cause of the conflict, generating a priority adjustment record table. The record table includes fields: record number, conflict occurrence time, command 1 type, command 1 priority, command 2 type, command 2 priority, executed command, pause command, and processing result. The record table is stored in the edge computing gateway's local database for a 3-month storage period for subsequent system optimization analysis.

[0049] This embodiment achieves ultra-short-term power prediction, multi-objective coordinated regulation, and command conflict coordination for wind-solar-storage microgrids through the above process. The model input and output parameters are clear, and the data processing cycle matches the real-time regulation requirements of the microgrid, which can effectively improve the energy utilization efficiency and power supply stability of the microgrid.

[0050] Specifically, the process of extracting device state features and adjusting the parameters of the lightweight neural network model is as follows: Based on the wind-solar-storage coordinated control command, the full data of equipment operation is retrieved, the characteristics of equipment operation status are extracted, and the performance deviation characteristics are located by comparing them with the equipment benchmark operation characteristics. An incremental learning strategy is adopted to adjust the parameters of the model layer corresponding to the deviation characteristics, verify the matching degree between the model output and the actual operation status of the equipment, and adjust the model parameters quantitatively based on the matching metric.

[0051] In this embodiment, a 1.3MW wind-solar-storage microgrid including a liquid-cooled energy storage system of a certain brand is used as the application scenario. A 500kW photovoltaic inverter, an 800kW wind turbine frequency converter, a 1MWh liquid-cooled energy storage system cluster-level management unit, and an edge computing gateway are configured. The control command response cycle is set to 1 minute, the equipment operation data retrieval cycle to 1 minute, and the model parameter incremental adjustment cycle to 15 minutes. These cycle settings are based on the requirement in the "Technical Specification for Operation Control of New Energy Microgrids" that "the response delay of wind-solar-storage coordinated control commands shall not exceed 2 minutes."

[0052] After the edge computing and coordination control unit issues the wind-solar-storage coordinated control command, the wind-solar-storage adaptive learning unit immediately initiates the equipment operation data retrieval process, retrieving 18 full-scale operation data items, including voltage, current, output power, equipment temperature, and energy storage SOC, from the local databases of the three types of equipment. These data items are selected based on the core operation monitoring indicators in the equipment's factory technical manual, covering key dimensions of the equipment's operating status. Core operating status features of the three types of equipment are extracted using feature engineering algorithms: for the photovoltaic side, conversion efficiency and power fluctuation rate are extracted; for the wind turbine side, power generation efficiency and pitch angle response speed are extracted; and for the energy storage side, charge / discharge response delay and SOC change rate are extracted. The feature extraction algorithm uses the industry-standard random forest feature importance ranking method to ensure the effectiveness of the features.

[0053] The extracted real-time status features are compared with the baseline operating features calibrated at the factory to pinpoint performance deviations. In this embodiment, the comparison revealed that the charge / discharge response delay of the liquid-cooled energy storage system increased from the baseline value of 50ms to 80ms, which was determined to be a critical performance deviation feature. This baseline value is derived from the factory inspection report of the cluster-level management unit of the liquid-cooled energy storage system.

[0054] To address this performance deviation, an online incremental learning strategy was employed to adjust the parameters of the lightweight neural network model. The specific steps followed the deep learning model fine-tuning technical specifications and were performed as follows: The parameters of the model's input layer and feature extraction layer are frozen, while only the parameters of the time series prediction layer are allowed to be fine-tuned. This operation is based on the industry consensus that "incremental learning should only adjust the model layer corresponding to new features to avoid damaging the model's original feature extraction capabilities," thus preventing a decrease in the model's generalization ability due to a full parameter adjustment.

[0055] The incremental learning rate was set to 0.001, with 32 iteration batches. The learning rate was selected based on the incremental learning parameter range of similar lightweight CNN-LSTM models. The iteration batches were adapted to the edge gateway's computing power (2 TOPS) to ensure that the time taken for a single adjustment did not exceed 10 seconds. The offset data of the energy storage charging and discharging response delay was used as new training samples and input into the time series prediction layer for local training.

[0056] During training, the matching degree between the predicted energy storage charging and discharging power output by the model and the actual charging and discharging power of the equipment is used as the evaluation index. The matching degree is calculated using the following formula: In the formula, the predicted value is the energy storage charging and discharging power output by the model, while the actual value is the real operating data collected by the equipment. A matching degree threshold of 97% is set. This threshold is based on the accuracy requirements for microgrid control command execution; a matching degree below 97% will lead to a mismatch between the control commands and the actual operating state of the equipment.

[0057] After completing the first round of incremental training, the matching degree was verified to be 92%, which did not meet the threshold requirement. Based on this matching metric, the model parameters were quantitatively adjusted, increasing the learning rate of the time series prediction layer to 0.003. This adjustment followed the parameter adjustment rule of "increasing the learning rate by 0.0005 for every 1% decrease in matching degree." Simultaneously, the number of iteration batches was increased to 48, and offset data was continued for incremental training. This parameter adjustment and training process was repeated until the model output matched the actual operating state of the device to 97.5%, meeting the preset threshold requirement, at which point parameter adjustment was stopped.

[0058] Specifically, the calibration and adjustment threshold adapts to temperature fluctuations, and the specific process is as follows: Ambient temperature sensor data is acquired, and temperature ranges are divided based on the temperature change amplitude. Preset control threshold benchmark values ​​are set for the temperature ranges. The current ambient temperature range is compared in real time, and the corresponding threshold benchmark value is retrieved. The threshold benchmark value is corrected in combination with the real-time operating status of the equipment. The corrected threshold value is substituted into the model to carry out operation verification. Based on the verification results, control threshold values ​​are generated and updated to the lightweight neural network model.

[0059] Specifically, the process of generating the model parameter optimization report is as follows: Record the values ​​of lightweight neural network model parameters before and after adjustment, key data of equipment status feature extraction, and calibration values ​​of temperature adaptation control thresholds and corresponding temperature ranges; statistically analyze the key operating indicators of the model after parameter adjustment, calculate the changes in the key operating indicators before and after adjustment, and label the indicator evaluation dimensions; record the key nodes of parameter adjustment and node triggering conditions, and synchronously record the adaptation judgment results of equipment operating status after each node adjustment; arrange various types of information according to the logic of data recording, indicator statistics, and node registration, configure exclusive data identifiers for information, and label the data collection time, equipment operating conditions, and control scenarios to form a model parameter optimization report.

[0060] Specifically, the optimized heat dissipation channel design involves the following process: The system acquires and categorizes data on the distribution of heat dissipation points during equipment operation. Based on the required airflow for heat dissipation, it presets channel cross-sectional dimensions and optimizes the internal structural parameters of the channels. It adjusts the heat dissipation distribution parameters based on the corresponding channel segments for each heat dissipation point, presets temperature sensing nodes at key locations within the channels, calls real-time monitoring data from these nodes, generates ventilation ratio parameters for the corresponding channel segments, iteratively adjusts the heat dissipation channel design, and verifies the heat dissipation adaptability of the design. The internal structural parameters include channel bending angle, inner wall smoothness, and channel length ratio. The real-time monitoring data includes real-time temperature, ventilation velocity, and heat dissipation efficiency for each channel segment.

[0061] This embodiment uses a 1.3MW wind-solar-storage microgrid, which includes a liquid-cooled energy storage system from a certain brand, as an application scenario. The safety protection and communication unit housing has an IP55 protection rating and needs to be compatible with a wide operating temperature range of -35~60℃. The heat dissipation channel is optimized according to the "Industrial Equipment Heat Dissipation Design Specification". The specific implementation process is as follows: Thermal simulation analysis of the equipment was initiated, identifying the hardware encryption chip, embedded firewall, and communication module within the unit as core heat dissipation hotspots, with a maximum operating temperature of 85℃. A dual-channel heat dissipation system was planned: the main channel is laid out along the hotspot distribution direction, responsible for dissipating the core heat sources; the secondary channel is laid out around the inner side of the casing to help balance the temperature difference within the cavity. Based on the heat dissipation airflow requirements, the heat load calculation formula was used: Q: Total heat load of the equipment, in kJ, used to calculate the required airflow for the heat dissipation channel. In this embodiment, Q = 55.625 kJ is calculated. c: Specific heat capacity of the main material of the equipment. In this invention, the equipment shell and core module are made of aluminum alloy, and c = 0.89 kJ / (kg・℃), determined according to the "Aluminum Alloy Thermophysical Property Parameter Handbook". m: Total mass of the core heating module of the equipment. In this embodiment, m = 2.5 kg, determined according to the hardware configuration list of the safety protection and communication unit. ΔT: Allowable temperature rise of the equipment. The upper limit of the normal operating temperature of the equipment is 85℃. Under a high temperature environment of 60℃, the temperature rise needs to be controlled to not exceed 25℃, therefore ΔT = 25℃. The main channel cross-sectional dimensions are determined to be 80mm × 50mm, and the secondary channel is 50mm × 30mm. These dimensions can meet the airflow requirements for low-temperature start-up at -35℃ and high-temperature operation at 60℃.

[0062] An arc-shaped airflow guide structure is arranged within the channel to reduce airflow resistance, with the drag coefficient controlled below 0.2. Aluminum alloy heat dissipation fins are added to the main channel section corresponding to hot spots, with a fin spacing of 8mm to enhance heat transfer efficiency. Temperature sensing nodes are configured at the intersection of the main and secondary channels, with a sensing range covering -40~120℃ to accurately match the ambient temperature variation range.

[0063] Based on sensor node monitoring data, the ventilation ratio is dynamically set. When the ambient temperature is below -10℃, the ventilation ratio is adjusted to 3:1 to reduce the amount of cold air entering; when the ambient temperature is above 45℃, the ventilation ratio is adjusted to 1:1 to improve heat dissipation efficiency. Testing shows that the optimized unit operates without condensation at -35℃ and has a cavity temperature difference of ≤5℃ at 60℃, meeting the requirements for wide-temperature operation.

[0064] Specifically, the process of performing end-to-end encryption on uplink and downlink data is as follows: Encryption keys are allocated to the uplink and downlink data transmission links. Based on the key synchronization mechanism configured at both ends of the communication, the uplink and downlink data before transmission are processed in blocks. Encryption operations are performed on each data block. A unique verification identifier is added to the encrypted data block. The data receiving end calls the synchronization key to decrypt the data block. The received encrypted data block is decrypted. The consistency between the verification identifier and the decrypted data is compared. After the verification is passed, the data restoration operation is performed, and a full-process encryption and decryption record and key synchronization log are generated.

[0065] Specifically, the dual-mode redundant communication architecture operates as follows: Construct dual-mode redundant links for wired and wireless industrial communication, configure the communication parameters of the primary and backup links and perform link adaptation and debugging, set the transmission status judgment threshold for link switching, monitor the transmission rate and connectivity status of the primary link in real time, trigger the link switching command when the primary link fails to meet the transmission standard, enable the backup link to carry out data transmission, configure the status monitoring module at the link node, and record the triggering conditions and execution time of link switching.

[0066] In this embodiment, a 1.3MW wind-solar-storage microgrid including a liquid-cooled energy storage system of a certain brand is used as the application scenario. The safety protection and communication unit adopts an IP55 protection level sealed shell, which is higher than the IP54 requirement. The shell is equipped with a dustproof and waterproof sealing ring and a heat dissipation grille to meet the requirements of harsh outdoor operating environments. The unit integrates a national cryptographic SM4 hardware encryption chip and an embedded firewall module, and communicates with the upper-level control system through dual links of fiber optic and industrial Ethernet. The implementation process of the method described in claims 11 and 12 is described in detail.

[0067] I. End-to-end encryption of uplink and downlink data The encryption process in this embodiment is implemented based on a hardware encryption chip, and the parameter settings are in accordance with the "Basic Requirements for Cryptographic Applications in Information Security Technology". The specific steps are as follows: Key distribution and synchronization: The hardware encryption chip built into the security protection and communication unit, through an asymmetric key negotiation mechanism with the encryption chip of the upper-level control system, assigns a unique session key to the uplink and downlink data transmission links. The key length is set to 128 bits, which complies with the requirements of the national cryptographic algorithm. A key synchronization module at both ends of the communication is configured based on a timestamp synchronization mechanism, with a synchronization period of 5 minutes and a synchronization deviation threshold of ≤1 second to avoid decryption failure due to clock skew.

[0068] Data segmentation and encryption: Uplink and downlink data are segmented before transmission, with a segment size of 1024 bytes. This value is based on the industrial communication data frame length standard, balancing encryption efficiency and real-time transmission. The hardware encryption chip uses the SM4 algorithm to perform encryption operations on each data block, with encryption time ≤1ms / block, meeting the low-latency requirements of microgrid control commands.

[0069] Verification and Decryption Restoration: A CRC32-specific verification identifier is added to the encrypted data block, and this identifier is bound to the data block during transmission. The data receiving end calls the synchronization session key to decrypt the encrypted data block through the hardware encryption chip, compares the verification identifier with the decrypted data, and if the verification fails, discards the data block and requests a retransmission. If the verification passes, the data is reassembled and restored.

[0070] Log generation: The embedded firewall automatically retains the entire process of encryption and decryption records and key synchronization logs. The logs include information such as data transmission timestamps, data block numbers, and key synchronization status. The log storage period is 6 months, which meets the data security audit requirements.

[0071] II. Construction of Dual-Mode Redundant Communication Architecture This embodiment constructs a dual-mode redundant link for both fiber optic wired communication and industrial Ethernet wireless communication. The primary and backup link switching thresholds are set according to the "Reliability Requirements for Industrial Communication Networks". The specific steps are as follows: Link Setup and Adaptation: Fiber optic cable is used as the primary communication link with a transmission rate of 1000Mbps for transmitting massive amounts of operational data. Industrial Ethernet is used as the backup link with a transmission rate of 100Mbps for transmitting control commands. Configure the communication parameters for the primary and backup links. The fiber optic link uses single-mode fiber with a transmission distance ≤20km. The Ethernet link uses the TCP / IP protocol, and the network segment is set to the dedicated industrial network segment 192.168.2.0 / 24. Complete link adaptation and debugging.

[0072] Switching threshold setting and monitoring: Set a threshold for determining the transmission status of link switching. A switching command is triggered when the main link transmission rate is below 100Mbps or the packet loss rate is above 1%. This threshold is set based on the real-time requirements of microgrid control data transmission. Configure a status monitoring module at the link node to monitor the main link transmission rate and connectivity status at 1-second intervals.

[0073] Link Switching and Recording: When the primary link fails to meet transmission standards, the security and communication unit immediately triggers a link switching command, activating the industrial Ethernet backup link for data transmission. The switching time is ≤50ms, with no data transmission interruption. The status monitoring module synchronously records the triggering conditions, switching time, and backup link operating status of the link switching, generating a communication status report and uploading it to the upper-level system.

[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any 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 wind-solar-storage integrated collaborative control system based on edge AI, characterized in that, include: Multi-source data sensing unit, edge computing and coordination control unit, wind, solar and energy storage adaptive learning unit, security protection and communication unit; The multi-source data sensing unit acquires electrical quantity data from local photovoltaic inverters, wind turbine frequency converters, and liquid-cooled energy storage systems, and performs normalization processing on the electrical quantity data to generate a standardized dataset. Configure an expandable data acquisition interface, automatically identify and adapt devices through industrial communication protocols, and generate a list of communication protocol compatibility. The edge computing and coordination control unit integrates a lightweight neural network model, performs ultra-short-term prediction of local wind and solar power generation based on the standardized dataset, and optimizes the model structure to adapt to the prediction cycle; it constructs a multi-objective collaborative optimization mechanism, calculates and issues wind, solar and energy storage coordinated control commands, and builds a command closed-loop link; it sets a priority dynamic allocation mechanism, coordinates command conflicts through power supply stability levels, and generates a priority adjustment record table. The wind-solar-storage adaptive learning unit extracts equipment status features and adjusts the parameters of the lightweight neural network model based on the wind-solar-storage collaborative control command to adapt to changes in equipment aging performance; it also synchronously links environmental temperature sensor data to match model parameters, calibrates control thresholds to adapt to temperature fluctuations, presets a performance feedback mechanism, compares command execution results with preset targets, optimizes the learning rate by combining the priority adjustment record table, and generates a model parameter optimization report. The security protection and communication unit is equipped with sealed protection and dustproof components, has a built-in heat dissipation module and optimized heat dissipation channel design; it integrates a hardware encryption chip and an embedded firewall to perform end-to-end encryption on uplink and downlink data and establish a transmission security verification mechanism; it adopts a dual-mode redundant communication architecture to monitor the link operation status and generate a communication status report.

2. The system according to claim 1, characterized in that, The specific process for generating the standardized dataset is as follows: Electrical quantity data is categorized based on the type of acquisition equipment. Normalized value ranges are set for each data category. Normalization processing is performed on the electrical quantity data according to the ranges. Simultaneously, the processed data is integrated based on the data acquisition time sequence to form a structured and standardized dataset.

3. The system according to claim 1, characterized in that, The process of automatically identifying and adapting devices through industrial communication protocols is as follows: Scan the communication identification information of the access device, match the protocol type of the industrial communication protocol, configure the communication parameters of the scalable acquisition interface based on the matching result, adapt the communication link between the interface and the device, verify the communication stability of the adapted link, and generate the correspondence information between the device and the protocol.

4. The system according to claim 1, characterized in that, The specific process for making ultra-short-term predictions of local wind and solar power generation is as follows: The standardized dataset is input into a lightweight neural network model. The standardized dataset is then split into two data subsets: photovoltaic power generation and wind power generation. The power change characteristics and time series correlation patterns of the two data subsets are extracted. The number of neurons in the model is adjusted to match the ultra-short-term prediction cycle. The power generation of wind and solar power is predicted for each time period. The prediction deviation is calibrated by comparing the prediction with historical data of wind and solar power generation in the same period, and the local ultra-short-term prediction results of wind and solar power generation are generated.

5. The system according to claim 1, characterized in that, The specific process for constructing the multi-objective collaborative optimization mechanism is as follows: Multiple target parameters for wind-solar-storage coordinated regulation are preset and corresponding weight coefficients are configured; a multi-objective coordinated optimization function is constructed based on the multi-objective parameters and weight coefficients, and the ultra-short-term prediction results of wind and solar power generation, the communication protocol adaptation list and the real-time operation data of the energy storage system are imported into the optimization function to generate a set of regulation parameters; The control parameter group is subjected to equipment operation constraint verification and gradient optimization adjustment to generate control parameters that meet the equipment operation requirements.

6. The system according to claim 1, characterized in that, The specific process of coordinating command conflicts through power supply stability levels is as follows: The system divides power supply stability into levels, assigns priority values ​​to each level according to the importance of load power supply, and sets power supply guarantee standards. It monitors the issuance status of wind-solar-storage coordinated control commands in real time, performs conflict judgment on the issued commands, and records relevant information of conflict commands. It extracts the priority values ​​corresponding to the conflict commands, filters and executes the command with the best priority value through value comparison, and generates a priority adjustment record table.

7. The system according to claim 1, characterized in that, The specific process of extracting device status features and adjusting the parameters of the lightweight neural network model is as follows: Based on the wind-solar-storage coordinated control command, the full data of equipment operation is retrieved, the characteristics of equipment operation status are extracted, and the performance deviation characteristics are located by comparing them with the equipment benchmark operation characteristics. An incremental learning strategy is adopted to adjust the parameters of the model layer corresponding to the deviation characteristics, verify the matching degree between the model output and the actual operation status of the equipment, and adjust the model parameters quantitatively based on the matching metric.

8. The system according to claim 1, characterized in that, The calibration and adjustment threshold adapts to temperature fluctuations, and the specific process is as follows: Acquire ambient temperature sensor data, divide temperature ranges based on temperature change amplitude, and preset control threshold benchmark values ​​for the temperature ranges; compare the current ambient temperature range in real time, retrieve the corresponding threshold benchmark value, and correct the threshold benchmark value in combination with the real-time operating status of the equipment. The corrected threshold is substituted into the model for runtime verification. Based on the verification results, a control threshold is generated and updated to the lightweight neural network model.

9. The system according to claim 1, characterized in that, The specific process for generating the model parameter optimization report is as follows: Obtain the values ​​of lightweight neural network model parameters before and after adjustment, key data of equipment status features, and calibration values ​​of temperature adaptation control thresholds and corresponding temperature ranges; based on the key operating indicators of the model after parameter adjustment, calculate the changes in the key operating indicators before and after adjustment and label the indicator evaluation dimensions; arrange various types of information according to the logic of data recording, indicator statistics, and node registration, configure exclusive data identifiers for the information, and label the data collection time, equipment operating conditions, and control scenarios to form a model parameter optimization report.

10. The system according to claim 1, characterized in that, The optimized heat dissipation channel design is specifically implemented as follows: Acquire and classify the heat dissipation point distribution data of the equipment operation, preset the channel cross-sectional size parameters based on the heat dissipation air volume requirements, and optimize the internal structural parameters of the channel; adjust the heat dissipation distribution parameters based on the channel segments corresponding to the heat dissipation points, preset temperature sensing nodes at key locations in the channel, call the real-time monitoring data of the sensing nodes, generate ventilation ratio parameters for the channel segments corresponding to the heat dissipation points, iteratively adjust the heat dissipation channel design scheme, and verify the heat dissipation adaptability of the design scheme.

11. The system according to claim 1, characterized in that, The specific process of performing end-to-end encryption on uplink and downlink data is as follows: Encryption keys are allocated to the uplink and downlink data transmission links. Based on the key synchronization mechanism configured at both ends of the communication, the uplink and downlink data before transmission are processed in blocks. Encryption operations are performed on each data block. A unique verification identifier is added to the encrypted data block. The data receiving end calls the synchronization key to decrypt the data block. The received encrypted data block is decrypted. The consistency between the verification identifier and the decrypted data is compared. After the verification is passed, the data restoration operation is performed, and a full-process encryption and decryption record and key synchronization log are generated.

12. The system according to claim 1, characterized in that, The specific process of the dual-mode redundant communication architecture is as follows: Construct dual-mode redundant links for wired and wireless industrial communication, configure the communication parameters of the primary and backup links and perform link adaptation and debugging, set the transmission status judgment threshold for link switching, monitor the transmission rate and connectivity status of the primary link in real time, trigger the link switching command when the primary link fails to meet the transmission standard, enable the backup link to carry out data transmission, configure the status monitoring module at the link node to obtain the triggering conditions and execution time of link switching.

Images