A control method and system for a wind-solar-hydrogen-ammonia-ethanol system based on edge computing

Abstract

This application relates to the field of intelligent control technology for energy systems, specifically to a control method and system for a wind-solar-hydrogen-ammonia-methanol system based on edge computing. The method is executed by a locally deployed edge computing node and includes: real-time acquisition and preprocessing of operational data from various devices within a jurisdictional area; based on the preprocessed data and current local strategies, synchronously performing device status assessment and load forecasting, and generating and executing local autonomous control commands accordingly; simultaneously uploading key data to a cloud control platform, receiving global optimization commands from the cloud, and adaptively adjusting the commands based on local real-time assessment results before execution; evaluating the local control effect based on command execution feedback and uploading it to the cloud for global optimization; receiving and applying optimized updated strategy parameters from the cloud to complete iterative updates of the local control strategy. This invention achieves efficient collaboration between rapid local response and cloud-based global optimization.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for energy systems, specifically to a control method and system for a wind-solar-hydrogen-ammonia-ethanol system based on edge computing. Background Technology

[0002] Integrated wind-solar-hydrogen-ammonia-methanol systems are an important approach to achieving efficient conversion and storage of renewable energy. These systems typically include distributed wind and solar power generation, electrolytic hydrogen production, ammonia / methanol synthesis, and energy storage equipment. They are characterized by a large number of devices, geographical dispersion, complex operating conditions, and extremely high requirements for real-time and reliable control.

[0003] Currently, the control of such systems mostly adopts a traditional centralized architecture, where the operating data of all equipment is remotely transmitted to a central cloud control platform for processing, analysis, and decision-making, and then the cloud distributes control commands to various locations for execution. This model has significant drawbacks: First, the control loop has large latency; the remote transmission and centralized processing of massive amounts of data leads to slow command response, making it difficult to cope with second- or minute-level fluctuations in wind and solar power and sudden equipment failures, posing a risk of system instability. Second, network and computing resources are under concentrated pressure; the continuous uploading of large amounts of raw data consumes a significant amount of bandwidth, placing enormous pressure on the cloud center for real-time data processing and storage. Third, the system's reliability is fragile; a failure in the cloud center or communication network will cause the entire system to lose control, lacking local autonomy.

[0004] Edge computing technology makes it possible to perform real-time processing and control locally, close to the data source. However, there is currently no mature solution to deeply integrate it with the control requirements of complex multi-energy flow systems such as wind, solar, hydrogen, ammonia, and alcohol, so as to systematically solve the aforementioned problems of latency, bandwidth, reliability, and flexibility. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a control method and system for a wind-solar-hydrogen-ammonia-ethanol system based on edge computing.

[0006] In a first aspect, the present invention provides a control method for a wind-solar-hydrogen-amine-ethanol system based on edge computing, wherein the method is executed by edge computing nodes deployed within the wind-solar-hydrogen-amine-ethanol system, comprising: S1. Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment within its jurisdiction; S2. Based on the preprocessed data and in accordance with the current local control strategy, simultaneously perform equipment status assessment and local load forecasting, and generate and execute local autonomous control commands accordingly. S3. Upload the preprocessed key data to the cloud control platform; receive real-time global optimization control instructions from the cloud control platform, and execute the real-time global optimization control instructions after adaptively adjusting them according to the local real-time evaluation results. S4. Based on the system feedback data after the command is executed, evaluate the local control effect and upload it to the cloud control platform for global optimization; S5. Receive the globally optimized updated control strategy parameters issued by the cloud control platform, and update the current local control strategy in response to the received parameters.

[0007] As a further limitation of the technical solution of the present invention, step S1 includes: S11. Through the sensor network and device communication interface connected by the edge computing node, the real-time power of the wind and solar power generation equipment, the pressure and flow rate of the hydrogen production equipment, the temperature and reaction status of the ammonia and alcohol synthesis equipment, and the state of charge and charging and discharging current of the energy storage equipment are collected concurrently. S12. Perform outlier detection and removal on the collected data, and align time series data from different devices with different sampling frequencies to the same time reference through interpolation. S13. Extract key feature parameters for state assessment and load forecasting from the aligned data, and perform feature selection on the high-dimensional data to generate standardized feature vectors.

[0008] As a further limitation of the technical solution of the present invention, step S13 includes: S131. Construct at least one set of the following feature parameters from the aligned data: The characteristic set of wind and solar power generation includes the rate of change of active power per unit time, the deviation rate between current power and predicted power, and the intensity of power fluctuation within a short time window. Hydrogen production equipment feature set: including real-time hydrogen production efficiency, electrolyzer operating load rate, and the dominant frequency amplitude of the pressure sequence spectrum; The characteristic set of ammonia-methanol synthesis includes the cumulative deviation of reaction temperature from the set value, the gradient of key component concentration changes, and the variance of reactor pressure fluctuations; Energy storage device feature set: including acceleration of state of charge change, cumulative energy throughput of charge-discharge cycles, and peak factor of charge-discharge current; S132. Input the constructed set of feature parameters and their corresponding historical operating status data into the feature importance evaluation algorithm based on the tree model; calculate the contribution score of each feature parameter to the equipment status evaluation and load prediction results, and sort them according to the score. S133. Based on the preset contribution threshold, select the top N features from the sorted list to form the optimal feature subset for model input. S134. For each feature parameter in the optimal feature subset, perform maximum-minimum value normalization to map all feature values ​​to a uniform numerical range and generate a standardized feature vector.

[0009] As a further limitation of the technical solution of the present invention, step S2 includes: S21. The standardized feature vectors are input in parallel into the pre-trained equipment status assessment model and the local load prediction model; the equipment status assessment model outputs the real-time health score, failure probability and recommended operating conditions of each device; the local load prediction model outputs the load demand curve and its confidence interval for a future preset period. S22. Based on the decision rules in the current local control strategy, the real-time health score, failure probability, recommended operating conditions, load demand curve and real-time wind and solar power generation data are used as inputs. Through the rule engine, a local autonomous control instruction set containing specific equipment control target values ​​is generated. S23. Before the instruction is executed, the local autonomous control instruction set is verified for safety and feasibility based on the failure probability and real-time equipment constraints. After the verification is passed, the instruction is distributed to the local controllers of the corresponding wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment and energy storage equipment for execution.

[0010] As a further limitation of the technical solution of the present invention, in S21, the specific implementation process of inputting the standardized feature vector into the pre-trained device state evaluation model includes: The input standardized feature vector is mapped to a high-dimensional feature space through a multilayer perceptron encoder to obtain the device comprehensive state representation vector; The device's overall state representation vector is simultaneously input into three parallel fully connected neural network branches: Health score branch: Output a continuous value in the range [0,1] as the real-time health score, where 1 represents the optimal state; Failure probability branch: Outputs a continuous value in the interval [0,1] as an estimate of the probability of functional failure occurring within a preset time period in the future; Recommended operating condition branch: Outputs a multi-dimensional vector containing recommended setpoints or adjustment ranges for key parameters such as equipment operating power, pressure, and temperature; The outputs of the health score branch and the failure probability branch are calibrated using the Sigmoid function based on historical equipment failure data; the output of the recommended operating condition branch is corrected for exceeding limits based on the equipment physical constraint model, and finally, the real-time health score, failure probability and recommended operating condition of each device are generated.

[0011] As a further limitation of the technical solution of the present invention, in S21, the process of inputting the standardized feature vector into the pre-trained local load prediction model includes: A standardized feature vector containing multiple current and historical time points is input into a temporal feature encoding layer, which is composed of a one-dimensional long short-term memory network, and is used to extract deep temporal dependency features related to load changes. The time-dependent features are input into a sequence generation network, which outputs the load demand prediction values ​​for each time point within a preset future period through fully connected layers or deconvolution layers, forming a preliminary load demand curve. The time-dependent features are simultaneously input into a parallel uncertainty estimation network, which outputs the standard deviation or quantile information of the load forecast values ​​at each future time point; based on the forecast values ​​and the standard deviation or quantile information, the confidence interval of the load demand curve is calculated.

[0012] As a further limitation of the technical solution of the present invention, step S3 includes: S31. Extract key data packets containing equipment health status, local load forecast curves, and real-time energy balance status from the preprocessed data; encrypt and timestamp the key data packets and then upload them to the cloud control platform via asynchronous communication. S32. Receive a real-time global optimization control command issued by the cloud control platform, the command including the target device control target value; S33. Compare the target control value of the target equipment with the current feasible control domain determined based on the local real-time evaluation results and the equipment physical constraint model and the local load balance boundary; verify whether the target control value falls entirely within the feasible control domain; S34. If the verification finds that some or all of the control target values ​​exceed the feasible control domain, a correction mechanism is activated. For each out-of-limit target value, it is scaled into the feasible domain according to the distance it exceeds the boundary of the feasible domain by a preset correction coefficient. At the same time, based on the failure probability in the equipment status assessment, the fusion weight of the local autonomous control command and the scaled cloud command in the final execution command is dynamically adjusted, wherein the higher the failure probability, the greater the weight of the local autonomous control command. S35. Generate the final execution instruction set after the fusion weight adjustment, and perform a rapid simulation review of the final execution instruction set based on the device physical constraint model and the local load balance boundary. After confirming that there are no new conflicts or risks, distribute it to the corresponding device execution controller.

[0013] As a further limitation of the technical solution of the present invention, step S4 includes: S41. Within the preset monitoring period after the command is executed, collect real-time operation feedback data of wind and solar power generation equipment, energy storage equipment, hydrogen production equipment and ammonia-methanol synthesis equipment. S42. Based on the feedback data, calculate in parallel multiple dimensions of control effect quantification indicators, including: Energy balance index: the root mean square error between the actual net power and the target net power at the local level; Equipment operating efficiency indicators: the ratio of the actual hydrogen production rate to the rated hydrogen production rate of the hydrogen production equipment, or the conversion rate of key reactants in the ammonia-methanol synthesis equipment; Command follow-up performance index: The integral of the dynamic follow-up error between the actual charging and discharging power of the energy storage device and the power required by the command; Equipment safety status indicators: The percentage of time that all controlled equipment operates within the safe operating range during the monitoring period; S43. Assign a preset weight coefficient to each of the quantitative indicators, and generate a comprehensive local control effect score by weighted summation based on their calculated values. S44. Package the quantitative indicator values ​​of multiple dimensions, local control effect scores, corresponding control command identifiers, and evaluation time window information to generate a structured evaluation report. S45. When the triggering conditions are met, the evaluation report is uploaded to the cloud control platform; the triggering conditions include: reaching a preset periodic upload time, the local control effect score being lower than a preset performance threshold, or receiving an active pull request from the cloud control platform.

[0014] As a further limitation of the technical solution of the present invention, step S5 includes: S51. Receive the updated control strategy parameter package sent by the cloud control platform. The parameter package contains parameters for updating the device physical constraint model, local load balancing boundary, or decision rules in the current local control strategy. S52. Before the application update, based on the device physical constraint model and the current operating status, perform security pre-verification on the key parameter values ​​in the parameter package; at the same time, create a complete backup of the current local control policy and set a clear condition-triggered automatic rollback mechanism. S53. If the pre-verification passes, perform the update operation according to the update type specified in the parameter package: If it is a hot update, the new parameters will be dynamically loaded into the running local control strategy module, replacing the original parameters, and ensuring a smooth transition of control logic without interrupting real-time control. If it is a cold update, the current policy module will be stopped, the new parameters will be loaded, and the policy service will be restarted while waiting for the current control cycle to end or when entering the maintenance window. S54. After the update is completed, monitor key system indicators within the preset observation period; if the indicators are normal and better than or equal to the level before the update, confirm the update is successful and send an update success confirmation and initial performance feedback to the cloud control platform; if the rollback mechanism is triggered, automatically restore to the backup strategy and send an update failure alarm and rollback log to the cloud control platform.

[0015] Secondly, the technical solution of the present invention also provides a wind-solar-hydrogen-ammonia-ethanol control system based on edge computing, including: a cloud control platform and at least one edge computing node; The edge computing nodes are deployed within the wind-solar-hydrogen-ammonia-methanol system and are communicatively connected to the wind and solar power generation equipment, hydrogen production equipment, ammonia-methanol synthesis equipment, and energy storage equipment to perform localized control. The edge computing node is configured to perform the following operations: Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment within its jurisdiction; Based on the preprocessed data and in accordance with the current local control strategy, equipment status assessment and local load forecasting are performed simultaneously, and local autonomous control commands are generated and executed accordingly. The preprocessed key data is uploaded to the cloud control platform; real-time global optimization control instructions are received from the cloud control platform, and the instructions are adaptively adjusted and executed based on the local real-time evaluation results. Based on the system feedback data after the command is executed, the local control effect is evaluated and uploaded to the cloud control platform for global optimization. The system receives globally optimized updated control strategy parameters from the cloud control platform and updates the current local control strategy in response to the received parameters.

[0016] As can be seen from the above technical solutions, this application has the following advantages: By offloading data processing and control functions to edge computing nodes deployed locally on the system, a two-tier architecture of local autonomous control and cloud-based global optimization and collaboration is constructed. Edge computing nodes perform real-time data acquisition, preprocessing, intelligent equipment status assessment, load forecasting, and rapid autonomous control, reducing the transmission latency of control commands and achieving millisecond-level rapid response to wind and solar power fluctuations and sudden equipment issues. Simultaneously, the local data processing of edge computing nodes significantly reduces the amount of data uploaded to the cloud, effectively alleviating network bandwidth pressure and cloud processing burden. Through the edge-cloud collaboration mechanism, the cloud performs global optimization and distributes strategies, while edge computing nodes combine local real-time conditions for security verification and adaptive adjustments. The final decision-making power of the local nodes ensures equipment safety and operational reliability, solving the problems of low reliability and excessive reliance on the cloud in traditional centralized control. Attached Figure Description

[0017] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the method provided in an embodiment of the present invention.

[0019] Figure 2 A block diagram of a system provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0022] like Figure 1 As shown, this embodiment of the invention provides a control method for a wind-solar-hydrogen-amine-ethanol system based on edge computing. The method is executed by edge computing nodes deployed within the wind-solar-hydrogen-amine-ethanol system, and includes: S1. Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia-methanol synthesis equipment, and energy storage equipment within its jurisdiction; this step specifically includes: S11. Through the sensor network and device communication interface connected by the edge computing node, the real-time power of the wind and solar power generation equipment, the pressure and flow rate of the hydrogen production equipment, the temperature and reaction status of the ammonia and alcohol synthesis equipment, and the state of charge and charging and discharging current of the energy storage equipment are collected concurrently. S12. Perform outlier detection and removal on the collected data, and align time series data from different devices with different sampling frequencies to the same time reference through interpolation. S13. Extract key feature parameters for state assessment and load forecasting from the aligned data, and perform feature selection on the high-dimensional data to generate standardized feature vectors. Specifically, this includes: S131. Construct at least one set of the following feature parameters from the aligned data: The characteristic set of wind and solar power generation includes the rate of change of active power per unit time, the deviation rate between current power and predicted power, and the intensity of power fluctuation within a short time window. Hydrogen production equipment feature set: including real-time hydrogen production efficiency, electrolyzer operating load rate, and the dominant frequency amplitude of the pressure sequence spectrum; The characteristic set of ammonia-methanol synthesis includes the cumulative deviation of reaction temperature from the set value, the gradient of key component concentration changes, and the variance of reactor pressure fluctuations; Energy storage device feature set: including acceleration of state of charge change, cumulative energy throughput of charge-discharge cycles, and peak factor of charge-discharge current; S132. Input the constructed set of feature parameters and their corresponding historical operating status data into the feature importance evaluation algorithm based on the tree model; calculate the contribution score of each feature parameter to the equipment status evaluation and load prediction results, and sort them according to the score; specifically including: Using historical operational data and their corresponding device status labels or actual load values, train a random forest model consisting of multiple decision trees, and obtain the structural information of each decision tree, including the features used for segmentation in each internal node and the amount of impurity reduction it brings. For each decision tree, traverse all its nodes, and weight the reduction in impurity resulting from splitting each feature at each node according to the proportion of samples covered by that node, and sum them up to obtain the contribution of that feature in that tree; average the contributions of the same feature in all decision trees to obtain the global contribution score of that feature in the random forest model. Based on the calculated global contribution scores of each feature, all feature parameters are sorted from high to low to generate the feature importance ranking list.

[0023] S133. Based on the preset contribution threshold, select the top N features from the sorted list to form the optimal feature subset for model input. S134. For each feature parameter in the optimal feature subset, perform maximum-minimum value normalization to map all feature values ​​to a uniform numerical range and generate a standardized feature vector.

[0024] S2. Based on the preprocessed data and according to the current local control strategy, simultaneously perform equipment status assessment and local load forecasting, and generate and execute local autonomous control commands accordingly; this step includes: S21. The standardized feature vectors are input in parallel into the pre-trained equipment status assessment model and the local load prediction model; the equipment status assessment model outputs the real-time health score, failure probability and recommended operating conditions of each device; the local load prediction model outputs the load demand curve and its confidence interval for a future preset period. The training process of the pre-trained device state assessment model specifically includes: Collect multi-dimensional operational data sequences of wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment during normal operation, performance degradation, and failure during historical operating cycles, and label each data sequence with a corresponding real label; the real label includes: the equipment health score for that period determined by an expert system or post-analysis, a binary identifier indicating whether a failure has occurred, and the optimal operating condition parameters of the equipment during that period. Construct a composite loss function for jointly training the three parallel branches, the composite loss function being a weighted sum of the following three terms: Health score loss: The mean squared error loss is used to calculate the difference between the predicted health score and the labeled health score; Failure probability loss: The difference between the predicted failure probability and the labeled failure indicator is calculated using binary cross-entropy loss. Recommended operating condition loss: Smooth L1 loss is used to calculate the difference between the recommended operating condition parameters and the labeled optimal operating condition parameters; The historical dataset is divided into a training set and a validation set in chronological order. Using the training set data and the composite loss function, the parameters of the equipment condition assessment model are iteratively optimized through the backpropagation algorithm. After each iteration, the validation set is used to evaluate the overall performance of the model in terms of health score accuracy, failure prediction precision, and working condition recommendation rationality, and the optimal model parameters are saved until the model converges.

[0025] The training process of the pre-trained local load prediction model specifically includes: Collect historical continuous standardized feature vector sequences and their corresponding real load value sequences for a predetermined future time period to form supervised training sample pairs; Construct a composite loss function for jointly optimizing the sequence generation network and the uncertainty estimation network. This function is a weighted sum of the following two parts: Prediction accuracy loss: The difference between the load demand curve output by the sequence generation network and the actual load value sequence is calculated using quantile loss or mean square error. Uncertainty calibration loss: Using negative log-likelihood loss or calibration scoring rules, the degree of matching between the confidence interval of the uncertainty estimation network output and the distribution of the true load value sequence is calculated; The training dataset is input into the local load prediction model. By minimizing the bi-objective loss function, the parameters of the temporal feature encoding layer, sequence generation network, and uncertainty estimation network are simultaneously updated using the backpropagation algorithm until the overall prediction performance of the model on the independent validation set converges.

[0026] S22. Based on the decision rules in the current local control strategy, the real-time health score, failure probability, recommended operating conditions, load demand curve, and real-time wind and solar power generation data are used as inputs. Through the rule engine, a local autonomous control instruction set containing specific equipment control target values ​​is generated. This specifically includes the following steps: The real-time health score, failure probability, load demand curve, recommended operating conditions, and real-time wind and solar power generation data are converted into input facts that the rule engine can recognize according to a predefined semantic template and loaded into the working memory of the rule engine; wherein, the load demand curve is discretized into load demand facts at multiple future time points; The rule engine performs pattern matching between the input facts in the working memory and the pre-set production rule set in the knowledge base. The production rule adopts the form IF<condition>THEN<action>, where the condition part combines the equipment health threshold, failure probability threshold, load demand range, and wind and solar power status, and the action part defines the control target value adjustment logic of the equipment. When multiple rules are activated at the same time, conflict resolution is performed according to the predefined rule priority, rule timeliness, or equipment safety weight to determine the final rule sequence to be executed. The action parts of each rule in the final rule sequence are executed sequentially. The operations performed in the action parts include: assigning or calculating intermediate control variables, calling the built-in local energy balance optimization function, and fine-tuning the target values ​​of the equipment according to the recommended operating conditions. By combining the execution results of all rule actions, a set of specific control target values ​​is generated, including the target power of wind and solar power generation equipment, the target load rate of hydrogen production equipment, the target reaction intensity of ammonia and alcohol synthesis equipment, and the target charge and discharge power of energy storage equipment. The specific control target value is combined with the corresponding device identifier and control interface protocol to encapsulate and generate a structured local autonomous control instruction set for subsequent security verification and execution.

[0027] S23. Before the instruction is executed, the local autonomous control instruction set is verified for safety and feasibility based on the failure probability and real-time equipment constraints. After the verification is passed, the instruction is distributed to the local controllers of the corresponding wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment and energy storage equipment for execution.

[0028] It should be noted that, based on the aforementioned failure probability and real-time device constraints, the security and feasibility verification of the local autonomous control instruction set specifically includes the following steps: a. Layered verification initialization: The local autonomous control instruction set is parsed into independent control action sequences for each device; a two-level verification hierarchy is established, with the first level being single-device security verification and the second level being system-level coupling feasibility verification; b. Single-device safety verification: For each device's control action, perform the following checks: Failure probability correlation verification: If the real-time failure probability of the device is higher than the preset high-risk threshold, the power or load increase command in its control action will be forcibly modified to maintain the current state or a safe load reduction command. Equipment physical constraint verification: The target parameter value of the control action (such as target power, target pressure, target SOC) is compared with the upper and lower limits of the equipment operating parameters defined in the equipment physical constraint model. If the limit is exceeded, the target parameter value is corrected to the closest limit value. c. System-level coupling feasibility verification: After all single-device instructions are corrected, based on the verified instruction set, simulate the execution of a short look-ahead time window (e.g., the next 10 seconds): Load balancing simulation: Calculate the algebraic sum of the predicted wind and solar power, the corrected power commands of each device, and the local base load within this time window, and verify whether it continuously meets the local load balancing boundary. Critical resource conflict detection: Check if multiple device commands are competing for the same critical resource (such as grid connection capacity or maximum cooling water flow) at the same time. If so, fine-tune the command execution sequence according to the preset device priority. d. Verification result generation and instruction release: If the system-level coupling feasibility verification passes, a verification pass flag is generated, and the final modified instruction set is marked as an executable version and distributed to the local controllers of each device; if the verification fails, a verification failure alarm is generated, and the emergency rollback mechanism of the local control policy is triggered to execute the security instruction set verified in the previous cycle.

[0029] Furthermore, in S21, the specific implementation process of inputting the standardized feature vector into the pre-trained device state assessment model includes: The input standardized feature vector is mapped to a high-dimensional feature space through a multilayer perceptron encoder to obtain the device comprehensive state representation vector; The device's overall state representation vector is simultaneously input into three parallel fully connected neural network branches: Health score branch: Output a continuous value in the range [0,1] as the real-time health score, where 1 represents the optimal state; Failure probability branch: Outputs a continuous value in the interval [0,1] as an estimate of the probability of functional failure occurring within a preset time period in the future; Recommended operating condition branch: Outputs a multi-dimensional vector containing recommended setpoints or adjustment ranges for key parameters such as equipment operating power, pressure, and temperature; The outputs of the health score branch and the failure probability branch are calibrated using the Sigmoid function based on historical equipment failure data; the output of the recommended operating condition branch is corrected for exceeding limits based on the equipment physical constraint model, and finally, the real-time health score, failure probability and recommended operating condition of each device are generated.

[0030] Accordingly, the process of inputting the standardized feature vector into the pre-trained local load prediction model includes: A standardized feature vector containing multiple current and historical time points is input into a temporal feature encoding layer, which is composed of a one-dimensional long short-term memory network, and is used to extract deep temporal dependency features related to load changes. The time-dependent features are input into a sequence generation network, which outputs the load demand prediction values ​​for each time point within a preset future period through fully connected layers or deconvolution layers, forming a preliminary load demand curve. The time-dependent features are simultaneously input into a parallel uncertainty estimation network, which outputs the standard deviation or quantile information of the load forecast values ​​at each future time point; based on the forecast values ​​and the standard deviation or quantile information, the confidence interval of the load demand curve is calculated.

[0031] S3. Upload the preprocessed key data to the cloud control platform; receive real-time global optimization control commands from the cloud control platform, and execute the commands after adaptively adjusting them based on local real-time evaluation results; this step specifically includes: S31. Extract key data packets containing equipment health status, local load forecast curves, and real-time energy balance status from the preprocessed data; encrypt and timestamp the key data packets and then upload them to the cloud control platform via asynchronous communication. S32. Receive a real-time global optimization control command issued by the cloud control platform, the command including the target device control target value; S33. Compare the target control value of the target equipment with the current feasible control domain determined based on the local real-time evaluation results and the equipment physical constraint model and the local load balance boundary; verify whether all the target control values ​​fall within the feasible control domain; specifically including: S331. Based on the latest health score and failure probability output by the equipment status assessment model, dynamically adjust the safety boundary parameters in the equipment physical constraint model to generate the dynamic safe operating range of each device at the current moment; at the same time, based on the latest local load forecast data and real-time wind and solar power, calculate the system power feasible range that satisfies the local load balance boundary; the dynamic safe operating range of all devices and the system power feasible range together constitute a multi-dimensional current feasible control domain. S332. Project the target device control target value vector in the real-time global optimization control command issued from the cloud onto each dimension of the current feasible control domain: Device-dimensional projection: compares the target value for each device with the corresponding dynamic safe operating range for that device; System dimension projection: Substitute the target values ​​of all devices into the system power balance equation, calculate the net system power, and compare it with the feasible range of system power. S333, Feasibility Status Determination and Conflict Identification: If the target values ​​for all device dimensions fall within their respective dynamic safe operating ranges, and the calculated net system power falls within the feasible system power range, then it is determined that all the control target values ​​fall within the feasible control domain, and the verification is successful. If any target value for any device dimension exceeds its dynamic safe operating range, or if the system net power exceeds the system power feasible range, a conflict is determined, and the specific device target value or system balance condition that violated the constraint is recorded, along with the value and direction of the deviation from the boundary.

[0032] S34. If the verification finds that some or all of the control target values ​​exceed the feasible control domain, a correction mechanism is activated. For each out-of-limit target value, it is scaled into the feasible domain according to the distance it exceeds the boundary of the feasible domain by a preset correction coefficient. At the same time, based on the failure probability in the equipment status assessment, the fusion weight of the local autonomous control command and the scaled cloud command in the final execution command is dynamically adjusted, wherein the higher the failure probability, the greater the weight of the local autonomous control command. In this embodiment of the invention, if verification reveals that some or all of the control target values ​​exceed the feasible control domain, a correction mechanism is initiated, which specifically includes the following steps: (34a) For each target value that exceeds the boundary of the feasible control domain, calculate the excess amount. Based on the equipment type and the direction of exceeding the limit associated with the target value, select the corresponding correction coefficient from the preset correction strategy table. (0< ≤ 1); Set the target value Revised to ,in The boundary value that exceeds the target value, and ensure that It falls within the feasible control domain; (34b) Read the real-time failure probability of each device. According to the preset weight mapping function Calculate local autonomous control commands Fusion weights in the final execution instruction ,in For a function that is monotonically increasing, such that The higher, Larger; cloud-scaled instructions The weights are then ; (34c) For each device, if its cloud instructions are modified, the final executed instruction value is calculated according to the following formula: If its cloud commands do not exceed the limits, then ; (34d) Calculate the final execution instruction values ​​of all devices after fusion, and resubmit them to steps (34b) and (34c) for rapid feasibility verification; if new conflicts arise due to weight allocation (such as the disruption of system power balance), fine-tune the weight mapping function according to preset rules. The parameters or weights of individual devices can be manually overridden and recalculated until the fusion result meets the feasible control domain constraints.

[0033] S35. Generate the final execution instruction set after the fusion weight adjustment, and perform a rapid simulation review of the final execution instruction set based on the device physical constraint model and the local load balance boundary. After confirming that there are no new conflicts or risks, distribute it to the corresponding device execution controller.

[0034] A rapid simulation review of the final execution instruction set is performed to confirm that there are no new conflicts or risks. This includes the following steps: (35a) Based on the current equipment physical constraint model parameters, local load balance boundary and current system state snapshot, initialize a simplified, deterministic system dynamics simulation model; the simulation model focuses on the transmission and balance of electrical power, thermal power and key material flow within a preset short simulation step; (35b) Using the final execution instruction set as input, drive the fast simulation model to execute multiple simulation steps; within each step, calculate and check in real time: Device-level instantaneous constraints: Whether the simulation state variables (such as power, temperature, pressure) of each device instantaneously exceed the dynamic safety boundary of the device's physical constraint model; System-level coupling constraints: Whether the supply-demand difference of the total electrical power and total thermal power of the system exceeds the instantaneous allowable fluctuation band defined by the local load balance boundary; Key process variables: Simulate the state of charge change trajectory of energy storage devices to predict whether they will trigger overcharge or over-discharge protection boundaries; (35c) Summarize all constraint violation events throughout the entire simulation cycle. If a violation occurs, record the time, device, variable, and degree of violation of the first violation. Calculate a comprehensive risk assessment index based on the severity and duration of the violation. (35d) Review of decisions and distribution of instructions: If the comprehensive risk assessment index is lower than the preset safety execution threshold and no key process variables trigger the protection boundary, the review is deemed to have passed. If the risk assessment index exceeds the standard or the protection boundary is triggered, the review will be deemed unsuccessful, a detailed review failure report will be generated, and the local instruction rollback mechanism will be triggered immediately: the local autonomous control instruction set of the previous cycle that has passed the S23 security verification will be used, or the preset system security shutdown sequence will be executed, and a high-risk alarm will be sent to the cloud control platform at the same time.

[0035] In this embodiment of the invention, the method pre-sets a device physical constraint model and a local load balancing boundary; wherein, the device physical constraint model defines the hard safety range of the operating parameters of each controlled device, and the local load balancing boundary defines the allowable fluctuation range of the instantaneous power of the local system. The device physical constraint model includes at least one of the following constraints: For wind and solar power generation equipment: upper and lower limits of active power output and power ramp-up rate restrictions; For hydrogen production equipment: safe range of electrolyzer operating current / voltage, and upper limit of hydrogen production pressure; For ammonia-methanol synthesis equipment: safe temperature range of the reactor core and feed flow rate limits; For energy storage devices: upper and lower limits for state-of-charge operation, and maximum charging and discharging power limits.

[0036] The local load balance boundary is defined as follows: within a preset control period, the absolute value of the local system's net power must not exceed a preset percentage threshold of the system's rated capacity. The net power is the algebraic sum of wind and solar power generation, energy storage power, hydrogen production / synthesis load power, and other local load power.

[0037] S4. Based on the system feedback data after command execution, evaluate the local control effect and upload it to the cloud control platform for global optimization; this step specifically includes: S41. Within the preset monitoring period after the command is executed, collect real-time operation feedback data of wind and solar power generation equipment, energy storage equipment, hydrogen production equipment and ammonia-methanol synthesis equipment. S42. Based on the feedback data, calculate in parallel multiple dimensions of control effect quantification indicators, including: Energy balance index: the root mean square error between the actual net power and the target net power at the local level; Equipment operating efficiency indicators: the ratio of the actual hydrogen production rate to the rated hydrogen production rate of the hydrogen production equipment, or the conversion rate of key reactants in the ammonia-methanol synthesis equipment; Command follow-up performance index: The integral of the dynamic follow-up error between the actual charging and discharging power of the energy storage device and the power required by the command; Equipment safety status indicators: The percentage of time that all controlled equipment operates within the safe operating range during the monitoring period; S43. Assign a preset weight coefficient to each of the quantitative indicators, and generate a comprehensive local control effect score by weighted summation based on their calculated values; this is specifically achieved through the following steps: (M1) The actual calculated value for each quantitative indicator Preprocessing is performed based on its physical meaning: Benefit-oriented indicators (such as equipment operating efficiency indicators): directly calculate their ratio or degree of completion relative to the benchmark or rated value; Cost-based / error-based indicators (such as energy balance indicators and error values ​​of command follow-up performance indicators): First, calculate the deviation from the ideal value (usually 0), and then apply a preset saturation function. , The sensitivity coefficient is used to map it to the interval [0, 1]. The smaller the value (the smaller the error), the closer the mapped value is to 1. Safety status indicator (running percentage): Its value is itself a percentage and is directly used as input in the range [0, 1]. Each indicator after processing receives a normalized score. (0 ≤ ≤ 1), where 1 represents the optimal performance; (M2) Based on the performance of the aforementioned indicators during this evaluation period and their historical trends, dynamically calculate or adjust their weighting coefficients. : Base weight: Each indicator has a pre-set base weight. This reflects its importance in long-term optimization; Performance penalty / reward factor: If the normalized score of a certain indicator... If it is continuously below its historical moving average, then its current weight... Will Add a penalty increment to the base. This allows us to focus more on this weakness in subsequent optimizations; conversely, if the performance is excellent and stable, the weight can be slightly reduced. Constraints: All weight coefficients The sum is 1, and each No less than the preset minimum weight ; (M3) Normalize the scores of each indicator Its corresponding dynamic weight coefficient Multiply and then sum to generate a comprehensive local control effectiveness score. ,in This represents the summation of all indicators; (M4) Based on the completeness of data collection and the volatility of key indicators during this evaluation period, calculate an evaluation confidence factor C (0 ≤ C ≤ 1); and then use the original comprehensive score... Associated with confidence factor C, or as The calibrated score is output in the form of [data], and the confidence level information is sent synchronously to the cloud control platform.

[0038] S44. Package the quantitative indicator values ​​of multiple dimensions, local control effect scores, corresponding control command identifiers, and evaluation time window information to generate a structured evaluation report. S45. When the triggering conditions are met, the evaluation report is uploaded to the cloud control platform; the triggering conditions include: reaching a preset periodic upload time, the local control effect score being lower than a preset performance threshold, or receiving an active pull request from the cloud control platform.

[0039] S5. Receive globally optimized updated control policy parameters from the cloud control platform, and update the current local control policy in response to the received parameters. This step specifically includes: S51. Receive the updated control strategy parameter package sent by the cloud control platform. The parameter package contains parameters for updating the device physical constraint model, local load balancing boundary, or decision rules in the current local control strategy. S52. Before the application update, based on the device physical constraint model and the current operating status, perform security pre-verification on the key parameter values ​​in the parameter package; at the same time, create a complete backup of the current local control policy and set a clear condition-triggered automatic rollback mechanism. S53. If the pre-verification passes, perform the update operation according to the update type specified in the parameter package: If it is a hot update, the new parameters will be dynamically loaded into the running local control strategy module, replacing the original parameters, and ensuring a smooth transition of control logic without interrupting real-time control. If it is a cold update, the current policy module will be stopped, the new parameters will be loaded, and the policy service will be restarted while waiting for the current control cycle to end or when entering the maintenance window. S54. After the update is completed, monitor key system indicators within the preset observation period; if the indicators are normal and better than or equal to the level before the update, confirm the update is successful and send an update success confirmation and initial performance feedback to the cloud control platform; if the rollback mechanism is triggered, automatically restore to the backup strategy and send an update failure alarm and rollback log to the cloud control platform.

[0040] In S52, based on the physical constraint model of the device and the current operating state, a security pre-verification is performed on the key parameter values ​​in the parameter package, specifically including the following steps: (N1) Parse the updated control strategy parameter package, identify the parameter type to be updated and the strategy module it affects, including the equipment physical constraint model, local load balancing boundary or local control strategy rule; for each parameter to be updated, extract its target value to be updated, and identify the set of controlled equipment directly and indirectly affected by it. (N2) For all parameter updates affecting the physical constraint model of the device, perform the following verification: Boundary relaxation verification: If the update relaxes the safe operating boundary of a device (such as increasing the power limit or expanding the temperature range), it is allowed directly because it reduces the control restrictions; Boundary tightening verification: If the update is to tighten the safety boundary, immediately check the real-time parameter values ​​of the equipment in its current operating state. If the current value exceeds the new boundary to be tightened, generate a "tightening conflict" warning and suspend this update, or mark it as needing to be implemented in stages after adjustments based on equipment operating conditions; (N3) For all parameters affecting the control logic or optimization objective (such as weight coefficients and optimization function parameters), construct a temporary policy simulation copy containing the parameter values ​​to be updated; use the current system state as initial conditions to drive this simulation copy to run for a short control cycle (such as the next 1-5 minutes), and monitor it: Command generation feasibility: Whether the simulated commands generated based on the new parameters are within the physical constraint model of the device after static verification; System stability prediction: Under the action of simulated commands, whether the changing trends of key system state variables (such as bus voltage and energy storage SOC) are smooth, or whether there are signs of divergence or violent oscillation; (N4) Summarize the results of static verification and dynamic simulation: If all boundary tightening verifications pass, and the dynamic simulation shows that the system is stable and the instructions are feasible, then a security pre-verification pass conclusion is generated. If a tightening conflict warning is issued or an instability risk is detected during dynamic simulation, a security pre-verification failure conclusion will be generated, along with the specific conflicting device, risk type, and quantified risk level, to trigger subsequent update rejection or rollback mechanisms.

[0041] like Figure 2 As shown, this embodiment of the invention also provides a wind-solar-hydrogen-ammonia-ethanol control system based on edge computing, including: a cloud control platform and at least one edge computing node; The edge computing nodes are deployed within the wind-solar-hydrogen-ammonia-methanol system and are communicatively connected to the wind and solar power generation equipment, hydrogen production equipment, ammonia-methanol synthesis equipment, and energy storage equipment to perform localized control. The edge computing node is configured to perform the following operations: Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment within its jurisdiction; Based on the preprocessed data and in accordance with the current local control strategy, equipment status assessment and local load forecasting are performed simultaneously, and local autonomous control commands are generated and executed accordingly. The preprocessed key data is uploaded to the cloud control platform; real-time global optimization control instructions are received from the cloud control platform, and the instructions are adaptively adjusted and executed based on the local real-time evaluation results. Based on the system feedback data after the command is executed, the local control effect is evaluated and uploaded to the cloud control platform for global optimization. The system receives globally optimized updated control strategy parameters from the cloud control platform and updates the current local control strategy in response to the received parameters.

[0042] In some embodiments, the edge computing node includes: The data acquisition and preprocessing module is configured to: concurrently acquire real-time power data from wind and solar power generation equipment, pressure and flow rate data from hydrogen production equipment, temperature and reaction status data from ammonia and methanol synthesis equipment, and state of charge and charging / discharging current data from energy storage equipment via a connected sensor network and equipment communication interface; detect and remove outliers from the acquired data; and uniformly align time-series data from different equipment with different sampling frequencies to the same time reference using interpolation methods; extract key feature parameters for state assessment and load forecasting from the aligned data; and perform feature selection on high-dimensional data to generate standardized feature vectors.

[0043] In some embodiments, the data acquisition and preprocessing module is further configured to perform the following operations: Construct at least one set of the following feature parameters from the aligned data: The characteristic set of wind and solar power generation includes the rate of change of active power per unit time, the deviation rate between current power and predicted power, and the intensity of power fluctuation within a short time window. Hydrogen production equipment feature set: including real-time hydrogen production efficiency, electrolyzer operating load rate, and the dominant frequency amplitude of the pressure sequence spectrum; The characteristic set of ammonia-methanol synthesis includes the cumulative deviation of reaction temperature from the set value, the gradient of key component concentration changes, and the variance of reactor pressure fluctuations; Energy storage device feature set: including acceleration of state of charge change, cumulative energy throughput of charge-discharge cycles, and peak factor of charge-discharge current; The constructed set of feature parameters and their corresponding historical operating status data are input into a tree-based feature importance evaluation algorithm; the contribution score of each feature parameter to the equipment status evaluation and load prediction results is calculated, and the parameters are sorted according to their scores. Based on a preset contribution threshold, the top N features are selected from the sorted list to form the optimal feature subset for model input. For each feature parameter in the optimal feature subset, perform maximum-minimum normalization to map all feature values ​​to a uniform numerical range and generate a standardized feature vector.

[0044] In some embodiments, the edge computing node further includes: The local intelligent decision-making and control module is configured as follows: The standardized feature vectors are input in parallel into a pre-trained equipment condition assessment model and a local load forecasting model; the equipment condition assessment model outputs the real-time health score, failure probability, and recommended operating conditions for each device; the local load forecasting model outputs the load demand curve and its confidence interval for a future preset period. Based on the decision rules in the current local control strategy, the real-time health score, failure probability, recommended operating conditions, load demand curve and real-time wind and solar power generation data are used as inputs. Through the rule engine, a local autonomous control instruction set containing specific equipment control target values ​​is generated. Before the instruction is executed, the local autonomous control instruction set is verified for safety and feasibility based on the failure probability and real-time equipment constraints. After the verification is passed, the instruction is distributed to the local controllers of the corresponding wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment and energy storage equipment for execution.

[0045] In some embodiments, the pre-trained device state assessment model includes: A multilayer perceptron encoder is used to map the input standardized feature vector to a high-dimensional feature space to obtain a comprehensive state representation vector of the device. Three parallel fully connected neural network branches, including: The health score branch outputs a continuous value in the range [0,1] as the real-time health score, where 1 represents the optimal state. The failure probability branch is used to output a continuous value in the interval [0,1] as an estimate of the probability of functional failure occurring within a preset time period in the future. The recommended operating condition branch is used to output a multi-dimensional vector containing recommended setpoints or adjustment ranges for key parameters such as equipment operating power, pressure, and temperature. In addition, a calibration and correction unit is used to calibrate the outputs of the health score branch and the failure probability branch using the Sigmoid function based on historical equipment failure data; and to correct the outputs of the recommended operating condition branch for exceeding limits based on the equipment physical constraint model.

[0046] In some embodiments, the pre-trained local load prediction model includes: The temporal feature encoding layer, composed of a one-dimensional long short-term memory network, is used to receive standardized feature vectors containing multiple current and historical time points and extract deep temporal dependency features related to load changes. A sequence generation network is used to receive the time-dependent features and output the predicted load demand values ​​at each time point within a preset future time period through a fully connected layer or a deconvolution layer, forming a preliminary load demand curve. An uncertainty estimation network is used to receive the time-dependent features in parallel and output the standard deviation or quantile information of the load forecast values ​​at each future time point; based on the forecast values ​​and the standard deviation or quantile information, the confidence interval of the load demand curve is calculated.

[0047] In some embodiments, the edge computing node further includes: The edge-cloud collaboration and command adjustment module is configured as follows: Extract key data packets containing equipment health status, local load forecast curves, and real-time energy balance status from the preprocessed data; encrypt and timestamp the key data packets and then upload them to the cloud control platform via asynchronous communication. Receive real-time global optimization control instructions issued by the cloud control platform, the instructions including target control values ​​for the target device; The target control value of the target device is compared with the current feasible control domain determined based on the local real-time evaluation results and the physical constraint model of the device and the local load balance boundary; it is verified whether the target control value falls entirely within the feasible control domain. If the verification finds that some or all of the control target values ​​exceed the feasible control domain, a correction mechanism is activated. For each out-of-limit target value, it is scaled into the feasible domain according to the distance it exceeds the boundary of the feasible domain by a preset correction coefficient. At the same time, based on the failure probability in the equipment status assessment, the fusion weight of the local autonomous control command and the scaled cloud command in the final execution command is dynamically adjusted, wherein the higher the failure probability, the greater the weight of the local autonomous control command. The final execution instruction set after weight adjustment is generated, and the final execution instruction set is quickly simulated and verified based on the device physical constraint model and local load balance boundary. After confirming that there are no new conflicts or risks, it is distributed to the corresponding device execution controller.

[0048] In some embodiments, the edge computing node further includes: The control effect evaluation module is configured as follows: Within the preset monitoring period after the command is executed, real-time operational feedback data of wind and solar power generation equipment, energy storage equipment, hydrogen production equipment, and ammonia-methanol synthesis equipment are collected. Based on the feedback data, multiple dimensions of control effect quantitative indicators are calculated in parallel. The quantitative indicators include: energy balance indicators, equipment operating efficiency indicators, command follow-up performance indicators, and equipment safety status indicators. Each quantitative indicator is assigned a preset weight coefficient, and a comprehensive local control effect score is generated by weighted summation based on its calculated values. The quantitative indicator values ​​of multiple dimensions, local control effect scores, corresponding control command identifiers, and evaluation time window information are packaged to generate a structured evaluation report. When the triggering conditions are met, the evaluation report will be uploaded to the cloud control platform. The triggering conditions include: reaching a preset periodic upload time, the local control effect score being lower than a preset performance threshold, or receiving an active pull request from the cloud control platform.

[0049] In some embodiments, the edge computing node further includes: The policy update and security management module is configured as follows: Receive the updated control strategy parameter package sent by the cloud control platform. The parameter package contains parameters for updating the device physical constraint model, local load balancing boundary, or decision rules in the current local control strategy. Before the application update, based on the device physical constraint model and the current operating status, the key parameter values ​​in the parameter package are pre-verified for security; at the same time, a complete backup of the current local control policy is created, and a clear condition-triggered automatic rollback mechanism is set. If the pre-verification passes, the update operation will be performed according to the update type specified in the parameter package: either a hot update or a cold update will be performed. After the update is completed, key system indicators are monitored during the preset observation period. If the indicators are normal and better than or equal to the level before the update, the update is confirmed to be successful, and an update success confirmation and initial performance feedback are sent to the cloud control platform. If the rollback mechanism is triggered, the system is automatically restored to the backup strategy, and an update failure alarm and rollback log are sent to the cloud control platform.

[0050] In some embodiments, the cloud control platform is configured as follows: Receive and integrate key data and evaluation reports uploaded from various edge computing nodes; Based on global data, global optimization calculations are performed to generate the real-time global optimization control commands and send them to the corresponding edge computing nodes. Based on the global optimization objective and historical data, the control strategy is iteratively optimized globally, generating the updated control strategy parameters and distributing them to each edge computing node.

[0051] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or other media capable of storing program code. It includes several instructions to cause a computer terminal (which may be a personal computer, server, or a second terminal, network terminal, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0052] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0054] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0055] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A control method for a wind-solar-hydrogen-amine-ethanol system based on edge computing, characterized in that, The method is executed by edge computing nodes deployed within the wind-solar-hydrogen-amine-ethanol system, and includes: S1. Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment within its jurisdiction; S2. Based on the preprocessed data and in accordance with the current local control strategy, simultaneously perform equipment status assessment and local load forecasting, and generate and execute local autonomous control commands accordingly. S3. Upload the preprocessed key data to the cloud control platform; receive real-time global optimization control instructions from the cloud control platform, and execute the real-time global optimization control instructions after adaptively adjusting them according to the local real-time evaluation results. S4. Based on the system feedback data after the command is executed, evaluate the local control effect and upload it to the cloud control platform for global optimization; S5. Receive the globally optimized updated control strategy parameters issued by the cloud control platform, and update the current local control strategy in response to the received parameters.

2. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 1, characterized in that, The steps in S1 include: S11. Through the sensor network and device communication interface connected by the edge computing node, the real-time power of the wind and solar power generation equipment, the pressure and flow rate of the hydrogen production equipment, the temperature and reaction status of the ammonia and alcohol synthesis equipment, and the state of charge and charging and discharging current of the energy storage equipment are collected concurrently. S12. Perform outlier detection and removal on the collected data, and align time series data from different devices with different sampling frequencies to the same time reference through interpolation. S13. Extract key feature parameters for state assessment and load forecasting from the aligned data, and perform feature selection on the high-dimensional data to generate standardized feature vectors.

3. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 2, characterized in that, The steps in S13 include: S131. Construct at least one set of the following feature parameters from the aligned data: The characteristic set of wind and solar power generation includes the rate of change of active power per unit time, the deviation rate between current power and predicted power, and the intensity of power fluctuation within a short time window. Hydrogen production equipment feature set: including real-time hydrogen production efficiency, electrolyzer operating load rate, and the dominant frequency amplitude of the pressure sequence spectrum; The characteristic set of ammonia-methanol synthesis includes the cumulative deviation of reaction temperature from the set value, the gradient of key component concentration changes, and the variance of reactor pressure fluctuations; Energy storage device feature set: including acceleration of state of charge change, cumulative energy throughput of charge-discharge cycles, and peak factor of charge-discharge current; S132. Input the constructed set of feature parameters and their corresponding historical operating status data into the feature importance evaluation algorithm based on the tree model; calculate the contribution score of each feature parameter to the equipment status evaluation and load prediction results, and sort them according to the score. S133. Based on the preset contribution threshold, select the top N features from the sorted list to form the optimal feature subset for model input. S134. For each feature parameter in the optimal feature subset, perform maximum-minimum value normalization to map all feature values ​​to a uniform numerical range and generate a standardized feature vector.

4. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 3, characterized in that, The steps in S2 include: S21. The standardized feature vectors are input in parallel into the pre-trained equipment status assessment model and the local load prediction model; the equipment status assessment model outputs the real-time health score, failure probability and recommended operating conditions of each device; the local load prediction model outputs the load demand curve and its confidence interval for a future preset period. S22. Based on the decision rules in the current local control strategy, the real-time health score, failure probability, recommended operating conditions, load demand curve and real-time wind and solar power generation data are used as inputs. Through the rule engine, a local autonomous control instruction set containing specific equipment control target values ​​is generated. S23. Before the instruction is executed, the local autonomous control instruction set is verified for safety and feasibility based on the failure probability and real-time equipment constraints. After the verification is passed, the instruction is distributed to the local controllers of the corresponding wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment and energy storage equipment for execution.

5. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 4, characterized in that, In S21, the specific implementation process of inputting the standardized feature vector into the pre-trained device state assessment model includes: The input standardized feature vector is mapped to a high-dimensional feature space through a multilayer perceptron encoder to obtain the device comprehensive state representation vector; The device's overall state representation vector is simultaneously input into three parallel fully connected neural network branches: Health score branch: Output a continuous value in the range [0,1] as the real-time health score, where 1 represents the optimal state; Failure probability branch: Outputs a continuous value in the interval [0,1] as an estimate of the probability of functional failure occurring within a preset time period in the future; Recommended operating condition branch: Outputs a multi-dimensional vector containing recommended setpoints or adjustment ranges for key parameters such as equipment operating power, pressure, and temperature; The outputs of the health score branch and the failure probability branch are calibrated using the Sigmoid function based on historical equipment failure data; the output of the recommended operating condition branch is corrected for exceeding limits based on the equipment physical constraint model, and finally, the real-time health score, failure probability and recommended operating condition of each device are generated.

6. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 4, characterized in that, In S21, the process of inputting the standardized feature vector into the pre-trained local load prediction model includes: A standardized feature vector containing multiple current and historical time points is input into a temporal feature encoding layer, which is composed of a one-dimensional long short-term memory network, and is used to extract deep temporal dependency features related to load changes. The time-dependent features are input into a sequence generation network, which outputs the load demand prediction values ​​for each time point within a preset future period through fully connected layers or deconvolution layers, forming a preliminary load demand curve. The time-dependent features are simultaneously input into a parallel uncertainty estimation network, which outputs the standard deviation or quantile information of the load forecast values ​​at each future time point; based on the forecast values ​​and the standard deviation or quantile information, the confidence interval of the load demand curve is calculated.

7. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 6, characterized in that, The steps in S3 include: S31. Extract key data packets containing equipment health status, local load forecast curves, and real-time energy balance status from the preprocessed data; encrypt and timestamp the key data packets and then upload them to the cloud control platform via asynchronous communication. S32. Receive a real-time global optimization control command issued by the cloud control platform, the command including the target device control target value; S33. Compare the target control value of the target equipment with the current feasible control domain determined based on the local real-time evaluation results and the equipment physical constraint model and local load balance boundary; verify whether the target control value falls entirely within the feasible control domain; S34. If the verification finds that some or all of the control target values ​​exceed the feasible control domain, a correction mechanism is activated. For each out-of-limit target value, it is scaled into the feasible domain according to the distance it exceeds the boundary of the feasible domain by a preset correction coefficient. At the same time, based on the failure probability in the equipment status assessment, the fusion weight of the local autonomous control command and the scaled cloud command in the final execution command is dynamically adjusted, wherein the higher the failure probability, the greater the weight of the local autonomous control command. S35. Generate the final execution instruction set after the fusion weight adjustment, and perform a rapid simulation review of the final execution instruction set based on the device physical constraint model and the local load balance boundary. After confirming that there are no new conflicts or risks, distribute it to the corresponding device execution controller.

8. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 7, characterized in that, The steps in S4 include: S41. Within the preset monitoring period after the command is executed, collect real-time operation feedback data of wind and solar power generation equipment, energy storage equipment, hydrogen production equipment and ammonia-methanol synthesis equipment. S42. Based on the feedback data, calculate in parallel multiple dimensions of control effect quantification indicators, including: Energy balance index: the root mean square error between the actual net power and the target net power at the local level; Equipment operating efficiency indicators: the ratio of the actual hydrogen production rate to the rated hydrogen production rate of the hydrogen production equipment, or the conversion rate of key reactants in the ammonia-methanol synthesis equipment; Command follow-up performance index: The integral of the dynamic follow-up error between the actual charging and discharging power of the energy storage device and the command-required power; Equipment safety status indicators: The percentage of time that all controlled equipment operates within the safe operating range during the monitoring period; S43. Assign a preset weight coefficient to each of the quantitative indicators, and generate a comprehensive local control effect score by weighted summation based on their calculated values. S44. Package the quantitative indicator values ​​of multiple dimensions, local control effect scores, corresponding control command identifiers, and evaluation time window information to generate a structured evaluation report. S45. When the triggering conditions are met, the evaluation report is uploaded to the cloud control platform; the triggering conditions include: reaching a preset periodic upload time, the local control effect score being lower than a preset performance threshold, or receiving an active pull request from the cloud control platform.

9. The edge computing-based control method for a wind-solar-hydrogen-ammonia-ethanol system according to claim 6, characterized in that, The steps in S5 include: S51. Receive the updated control strategy parameter package sent by the cloud control platform. The parameter package includes parameters for updating the device physical constraint model, local load balancing boundary, or decision rules in the current local control strategy. S52. Before the application update, based on the device physical constraint model and the current operating status, perform security pre-verification on the key parameter values ​​in the parameter package; at the same time, create a complete backup of the current local control policy and set a clear condition-triggered automatic rollback mechanism. S53. If the pre-verification passes, perform the update operation according to the update type specified in the parameter package: If it is a hot update, the new parameters will be dynamically loaded into the running local control strategy module, replacing the original parameters, and ensuring a smooth transition of control logic without interrupting real-time control. If it is a cold update, the current policy module will be stopped, the new parameters will be loaded, and the policy service will be restarted while waiting for the current control cycle to end or when entering the maintenance window. S54. After the update is completed, monitor key system indicators within the preset observation period; if the indicators are normal and better than or equal to the level before the update, confirm the update is successful and send an update success confirmation and initial performance feedback to the cloud control platform; if the rollback mechanism is triggered, automatically restore to the backup strategy and send an update failure alarm and rollback log to the cloud control platform.

10. A wind-solar-hydrogen-amine-ethanol control system based on edge computing, characterized in that, include: A cloud-based control platform and at least one edge computing node; The edge computing nodes are deployed within the wind-solar-hydrogen-ammonia-methanol system and are communicatively connected to the wind and solar power generation equipment, hydrogen production equipment, ammonia-methanol synthesis equipment, and energy storage equipment to perform localized control. The edge computing node is configured to perform the following operations: Real-time collection and preprocessing of operational data from wind and solar power generation equipment, hydrogen production equipment, ammonia and alcohol synthesis equipment, and energy storage equipment within its jurisdiction; Based on the preprocessed data and in accordance with the current local control strategy, equipment status assessment and local load forecasting are performed simultaneously, and local autonomous control commands are generated and executed accordingly. The preprocessed key data is uploaded to the cloud control platform; real-time global optimization control instructions are received from the cloud control platform, and the instructions are adaptively adjusted and executed based on the local real-time evaluation results. Based on the system feedback data after the command is executed, the local control effect is evaluated and uploaded to the cloud control platform for global optimization. The system receives globally optimized updated control strategy parameters from the cloud control platform and updates the current local control strategy in response to the received parameters.

Images