Cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters

Abstract

This invention relates to the field of photovoltaic power plant operation and maintenance technology, and discloses a cloud-edge-device collaborative intelligent cleaning operation and maintenance management system and method for photovoltaic power plant clusters. The system includes: an end device execution module, an edge decision arbitration module, and a cloud-based global optimization module. It collects and generates multi-source sensing data through the end device layer, and processes this data through the edge computing layer to generate local decision commands and cleaning status data, ensuring comprehensive and real-time perception of the surface status and environmental information of photovoltaic modules. Simultaneously, it rapidly generates cleaning control strategies at the edge based on this data, enabling real-time response to local environmental changes and sudden staining conditions, ensuring the accuracy and timeliness of cleaning decisions, and reducing decision-making biases caused by information lag and omissions. Through the cloud management layer, it integrates multi-source information to generate global strategies, improving overall power generation revenue while considering local operating condition adaptability, ensuring the rationality and efficiency of cleaning task scheduling in complex scenarios.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant operation and maintenance technology, specifically to a cloud-edge-device collaborative intelligent cleaning operation and maintenance management system and method for photovoltaic power plant clusters. Background Technology

[0002] Intelligent cleaning and maintenance of photovoltaic power plant clusters is a key link in ensuring their power generation efficiency and operating revenue. It plays a vital role in enhancing the value of photovoltaic assets throughout their entire life cycle and supporting the stable operation of new power systems. Therefore, accurately sensing the pollution status of power plants and efficiently coordinating and intelligently scheduling cleaning resources has become an extremely complex and crucial task. Its scope includes the collaborative management of environmental perception, status diagnosis, resource scheduling, operation control, and data analysis.

[0003] Currently, due to the spatially dispersed multiple sites and dynamically changing pollution conditions involved in the cleaning and maintenance of photovoltaic power plant clusters, traditional centralized and fully distributed architectures cannot simultaneously achieve both long-term economic optimization at the cluster level and instantaneous accurate response at the plant level when scheduling global cleaning resources and making local real-time decisions. When cluster network conditions are limited or there are sudden changes in the local environment, it can lead to delays in maintenance instructions, improper resource allocation, and misjudgments of power generation losses, thus failing to guarantee the economic efficiency of the overall operation and maintenance of the photovoltaic cluster and the optimization of power generation revenue.

[0004] Therefore, a cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters is proposed to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters. This solves the problems mentioned in the background technology, such as delayed maintenance instructions, improper resource allocation, and misjudgment of power generation losses, which fail to guarantee the economic efficiency of the overall operation and maintenance of the photovoltaic cluster and the optimization of power generation revenue.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters, the method comprising the following steps: S1. Based on the initial state parameters, collect stain images, temperature, humidity, power generation and equipment status data through the terminal device layer to generate multi-source sensing data; S2. Based on the multi-source sensing data, a multi-objective optimization algorithm is used through the edge computing layer to perform decision processing and generate local decision instructions and cleaning status data. S3. Based on the local decision-making instructions and the cleaning status data, a cluster-level cleaning revenue function is constructed through the cloud management layer, and global optimization is performed by integrating weather forecasts and electricity price signals to generate a global strategy. S4. Based on the global strategy and the environmental parameters in the multi-source sensing data, perform environmental change analysis and decision arbitration through the edge computing layer, compare the benefit difference, and generate a decision arbitration result. S5. Based on the decision arbitration result, perform the cleaning operation through the terminal device layer, and feed back the operation data and cleaning effect to generate cleaning operation data; S6. Based on the cleaning operation data, update the model parameters using an incremental learning algorithm through the edge computing layer to generate updated parameters; S7. Based on the updated parameters, the revenue function is dynamically optimized through the cloud management layer to generate an adjusted global strategy. S8. Based on the adjusted global strategy and the updated parameters, an adaptive collaborative optimization process is generated through the strategy execution and effect feedback of the cloud management layer and the edge computing layer.

[0007] Preferably, generating multi-source sensing data in step S1 includes the following steps: S11. By acquiring and processing images of the photovoltaic module surface, stain image data is generated; S12. Temperature data is generated by monitoring and processing the surface temperature of the photovoltaic module and the ambient temperature; S13. Generate humidity data by monitoring and processing ambient air humidity; S14. By measuring and processing the output electrical parameters of the photovoltaic module, power generation data is generated; S15. By collecting and processing the operating parameters of the cleaning robot, equipment status data is generated; S16. By collecting and integrating the stain image data, temperature data, humidity data, power generation data and equipment status data, multi-source sensing data is generated.

[0008] Preferably, the generation of local decision instructions and cleaning status data in step S2 includes the following steps: S21. Based on the multi-source sensing data, a multi-objective optimization algorithm is generated by setting cleaning efficiency, energy consumption, equipment wear and tear and cleaning benefits targets through a reinforcement learning framework. S22. Based on the stain image data and power generation data in the multi-source sensing data, a cleaning priority is generated by analyzing the severity of the stains and the power generation loss rate. S23. Based on the device status data in the multi-source sensing data and the preset component layout information, a cleaning path is generated through a path planning algorithm; S24. Based on the stain image data in the multi-source sensing data, generate cleaning intensity by matching stain type with preset parameters; S25. By integrating the cleaning priority, the cleaning path, and the cleaning intensity, a local decision command is generated; S26. By monitoring the energy consumption and speed information in the device status data, cleaning status data is generated.

[0009] Preferably, generating the global policy in S3 includes the following steps: S31. Based on the local decision-making instructions and the cleaning status data, a cluster-level cleaning revenue function is constructed by defining the sum of power generation gain minus cleaning cost, equipment depreciation cost and communication overhead cost as the target through the cloud management layer. S32. Based on the power generation changes in the cleaning status data, normalization correction is performed using the irradiance prediction model to generate power generation gain data. S33. Obtain future weather information and generate weather forecast data by integrating a weather forecast application programming interface; S34. Obtain real-time electricity price information by connecting to the electricity market data platform and generate electricity price signals; S35. A global strategy is generated by integrating the cluster-level cleaning revenue function, meteorological forecast data, and electricity price signals using an integer programming model.

[0010] Preferably, generating the decision arbitration result in step S4 includes the following steps: S41. Based on the temperature and humidity data in the multi-source sensing data, environmental change trend data is generated through time series analysis; S42. Based on the local decision instruction and the global strategy, a profit difference is generated by comparing the expected economic benefits of executing the two. S43. Based on the environmental change trend data and the profit difference, an arbitration decision is generated through arbitration logic.

[0011] Preferably, the generation of cleaning operation data in step S5 includes the following steps: S51. Based on the decision arbitration result, control the cleaning robot to perform movement and cleaning actions through the terminal device layer; S52. Real-time data collection of energy consumption, work progress and equipment indicators through built-in sensors in the terminal device layer to generate operational data; S53. By comparing and analyzing the differences between surface images before and after cleaning, cleaning effect evaluation data is generated. S54. By integrating and feeding back the operation data and the cleaning effect evaluation data to the edge computing layer, cleaning operation data is generated.

[0012] Preferably, generating the updated parameters in step S6 includes the following steps: S61. Based on the cleaning operation data, an incremental learning algorithm is constructed by training a neural network model based on historical cleaning operation data. S62. Analyze the error of the cleaning operation data through the incremental learning algorithm, and adjust the internal weight coefficients of the multi-objective optimization algorithm. S63. By updating the model parameters using the adjusted weight coefficients, updated parameters are generated.

[0013] Preferably, generating the adjusted global policy in step S7 includes the following steps: S71. Based on the updated parameters, retrain the internal parameters of the cluster-level cleaning revenue function using a machine learning algorithm; S72. By using the retrained profit function, global optimization is performed again to generate an adjusted global policy.

[0014] Preferably, the adaptive collaborative optimization process in S8 includes the following steps: S81. Execute the adjusted global strategy through the edge computing layer and collect actual effect data during the strategy execution process; S82. By comparing and analyzing the differences between the expected and actual results of the strategy through the cloud management layer, effect feedback data is generated. S83. Optimize the model parameters based on the effect feedback data, and simultaneously adjust the strategy generation rules to generate an adaptive collaborative optimization process.

[0015] Preferably, the system includes: The terminal device execution module is used to collect data on surface dirt, temperature, humidity, power generation and equipment status of photovoltaic modules through the environmental sensing unit, generate multi-source sensing data, and execute physical cleaning operations according to cleaning control commands; The edge decision arbitration module receives the multi-source sensing data, processes it in real time through the local data processing unit, generates local decision instructions and cleaning status data using a multi-objective optimization algorithm generation unit, and outputs decision arbitration results through environmental change analysis and arbitration unit. The cloud-based global optimization module receives the cleaning status data, integrates it through the cluster data aggregation unit, uses the revenue function construction and optimization unit to fuse weather forecasts and electricity price signals to generate a global strategy, and feeds it back to the edge decision arbitration module through the strategy distribution unit.

[0016] Compared with existing technologies, this invention provides a cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters, which has the following beneficial effects: 1. In this invention, multi-source sensing data is collected and generated through the terminal device layer, and local decision-making instructions and cleaning status data are generated based on the data through the edge computing layer. This ensures the comprehensiveness and real-time sensing of the surface status and environmental information of photovoltaic modules. At the same time, a cleaning control strategy is quickly generated at the edge based on the data, which can respond in real time to local environmental changes and sudden stains, ensuring the accuracy and timeliness of cleaning decisions and reducing decision-making bias caused by information lag and missing information.

[0017] 2. In this invention, a global strategy is generated by integrating multi-source information through a cloud-based management layer, and a decision arbitration result is generated by an edge computing layer based on the global strategy and local real-time data. This achieves synergy between cluster-level optimization and local rapid response, which can improve overall power generation revenue while taking into account local operating conditions. Furthermore, when conflicts arise between global objectives and local needs, dynamic arbitration and adjustment can be performed according to preset rules to ensure the rationality and efficiency of cleaning task scheduling in complex scenarios.

[0018] 3. In this invention, by continuously updating model parameters based on cleaning operation data and generating adjusted global strategies, an adaptive collaborative optimization process is formed, from data collection and decision execution to effect evaluation and strategy optimization. This process can continuously learn and improve using historical operation data, and achieve continuous optimization of strategies for different power plants and different seasons' climate and stain characteristics. This reduces long-term performance degradation caused by strategy rigidity and improves the intelligence level and long-term stable income of the entire photovoltaic power plant cluster operation and maintenance management. Attached Figure Description

[0019] Figure 1 This is a flowchart of the intelligent cleaning and maintenance management method for photovoltaic power plant clusters based on cloud-edge-device collaboration according to the present invention; Figure 2 This is a diagram illustrating the architecture of the cloud-edge-device collaborative intelligent cleaning and maintenance management system for photovoltaic power plant clusters, as described in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] For specific implementation examples, please refer to: Figure 1-2 A cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters, characterized in that the method includes the following steps: S1. Based on the initial state parameters, collect stain images, temperature, humidity, power generation and equipment status data through the terminal device layer to generate multi-source sensing data; S2. Based on multi-source sensing data, a multi-objective optimization algorithm is used through the edge computing layer to perform decision processing and generate local decision instructions and cleaning status data. S3. Based on local decision-making instructions and cleaning status data, a cluster-level cleaning revenue function is constructed through the cloud management layer, and global strategies are generated by integrating weather forecasts and electricity price signals for global optimization. S4. Based on the environmental parameters in the global strategy and multi-source sensing data, environmental change analysis and decision arbitration are performed through the edge computing layer. The difference in benefits is compared and a decision arbitration result is generated. S5. Based on the decision arbitration result, the cleaning operation is executed through the terminal device layer, and the operation data and cleaning effect are fed back to generate cleaning operation data; S6. Based on the cleaning operation data, update the model parameters using an incremental learning algorithm through the edge computing layer to generate updated parameters; S7. Based on the updated parameters, the revenue function is dynamically optimized through the cloud management layer to generate the adjusted global strategy. S8. Based on the adjusted global strategy and updated parameters, an adaptive collaborative optimization process is generated through the strategy execution and effect feedback of the cloud management layer and the edge computing layer.

[0022] The generation of multi-source sensing data in S1 includes the following steps: S11. By acquiring and processing images of the photovoltaic module surface, stain image data is generated, specifically including the following steps: S111. Obtain raw image data of the photovoltaic module surface through an image acquisition device; S112. Perform grayscale conversion and noise filtering on the original image data to generate a preprocessed image; The expression for grayscale conversion is as follows: ; in, Represents a grayscale image. , , These represent the red, green, and blue channel components, respectively. The noise filtering expression is as follows: ; in, This represents the Gaussian filter kernel. Indicates standard deviation, Represents the natural constant. and Represents coordinate variables; S113. Identify the boundaries of photovoltaic modules using edge detection algorithms to complete image region localization; ; in, Indicates the magnitude of the image gradient. Indicates image brightness exist Rate of change of direction Indicates image brightness exist Rate of change of direction; S114. Extract images of the surface region of the photovoltaic module based on the regional positioning results; S115. Perform stain feature analysis on the regional image to generate stain image data; ; in, Indicates the percentage of the stained area. This indicates the number of pixels in the stained area. Indicates the total number of pixels in the image; S12. Temperature data is generated by monitoring and processing the surface temperature of the photovoltaic module and the ambient temperature; S13. Generate humidity data by monitoring and processing ambient air humidity; S14. By measuring and processing the output electrical parameters of the photovoltaic module, power generation data is generated; S15. By collecting and processing the operating parameters of the cleaning robot, equipment status data is generated; S16. By collecting and integrating stain image data, temperature data, humidity data, power generation data and equipment status data, multi-source sensing data is generated.

[0023] The process of generating local decision commands and cleaning status data in S2 includes the following steps: S21. Based on multi-source sensing data, a multi-objective optimization algorithm is generated by setting cleaning efficiency, energy consumption, equipment wear and tear, and cleaning benefits targets through a reinforcement learning framework. This includes the following steps: S211. Construct optimization functions with cleaning efficiency, energy consumption, equipment wear and tear, and cleaning benefits as objectives respectively; ; in, Represents the overall objective function. , , , Let represent the objective functions for cleaning efficiency, energy consumption, equipment wear and tear, and cleaning revenue, respectively. to Represents the weighting coefficient of the objective; S212. Use a weight allocation method to merge multiple optimization objectives into a single comprehensive optimization objective; S213. Set corresponding weight coefficients for each optimization objective, and the sum of all weight coefficients is a fixed value; ; in, Indicates the first The weighting coefficients of each objective; S214. Use iterative optimization methods to solve the comprehensive optimization objective and obtain the final optimization parameters; ; in, Indicates the first The parameters for the next iteration Indicates the first The parameters for the next iteration Indicates the learning rate. Indicates the objective function in gradient at; S22. Based on the stain image data and power generation data in the multi-source sensing data, a cleaning priority is generated by analyzing the stain severity and power generation loss rate; S23. Based on the device status data in the multi-source sensing data and the preset component layout information, a cleaning path is generated through a path planning algorithm; S24. Based on the stain image data in the multi-source sensing data, the cleaning intensity is generated by matching the stain type with preset parameters; S25. By integrating cleaning priority, cleaning path and cleaning intensity, local decision instructions are generated; S26. By monitoring the energy consumption and speed information in the equipment status data, cleaning status data is generated.

[0024] Generating a global policy in S3 involves the following steps: S31. Based on local decision-making instructions and cleaning status data, a cluster-level cleaning revenue function is constructed by defining the sum of power generation gain minus cleaning cost, equipment depreciation cost and communication overhead cost as the target through the cloud management layer. S32. Based on the changes in power generation in the cleaning status data, and combined with the irradiance prediction model, normalization correction is performed to generate power generation gain data. S33. Obtain future weather information and generate weather forecast data by integrating a weather forecast application programming interface; S34. Obtain real-time electricity price information by connecting to the electricity market data platform and generate electricity price signals; S35. A global strategy is generated by integrating cluster-level cleaning revenue function, meteorological forecast data and electricity price signal through an integer programming model.

[0025] The steps involved in generating the decision arbitration result in S4 are as follows: S41. Based on temperature and humidity data from multi-source sensing data, environmental change trend data is generated through time-series analysis, specifically including the following steps: S411. Collect and arrange temperature and humidity data in chronological order; S412. Calculate the rate of change of temperature and humidity data over time, respectively. ; in, Indicates the rate of temperature change. Indicates the rate of change in humidity. and Representing temperature and humidity The minute changes Indicates time The minute changes; S413. Determine whether the rate of change of temperature and humidity exceeds their respective corresponding threshold values. ; in, Indicates a marker of environmental mutation. Indicates the rate of temperature change. Indicates the rate of change in humidity. Indicates the temperature change threshold. Indicates the threshold for humidity change. Indicates other situations; S414. Based on the judgment result of exceeding the change threshold, identify the environmental status and adjust the priority of the cleaning task accordingly, and output the environmental change analysis result. S42. Based on local decision-making instructions and global strategies, generate a profit difference by comparing the expected economic benefits of executing both, specifically including the following steps: S421. Based on local decision-making instructions and global strategies respectively, simulate the execution of cleaning tasks, predict the corresponding power generation gain, energy consumption cost and operation and maintenance cost, and generate expected benefits of local decision-making and expected benefits of cloud strategy. ; ; in, This represents the net economic benefit in a local decision-making scenario. This represents the net economic benefit in a cloud-based global strategy scenario. This represents the power generation gain in a local decision-making scenario. This represents the power generation gain under a cloud-based global strategy scenario. This represents the energy cost when executing local decision-making instructions. This represents the energy cost when executing a global cloud policy. This indicates the operation and maintenance costs in a local decision-making scenario. This indicates the operational costs in a cloud-based global policy scenario; S422. Generate a revenue difference by calculating the difference between the expected revenue of local decision-making and the expected revenue of cloud strategy; ; in, Indicates the difference in earnings. Indicates the expected benefits of local decision-making. Indicates the expected return of the cloud strategy; S43. Based on environmental change trend data and the benefit difference generated from the above steps, an arbitration decision is generated through arbitration logic, specifically including the following steps: S431. Obtain the absolute value of the profit difference and compare it with a pre-set decision threshold; ; in, Indicates the adaptive decision threshold. The standard deviation of earnings data This represents the mean of the earnings data. and Indicates the target weight coefficient; S432. When the absolute value of the difference is less than or equal to the decision threshold, the cloud strategy is executed. S433. When the absolute value of the difference is greater than the decision threshold and the difference is positive, it is determined that the local decision should be executed first. S434. When the absolute value of the difference is greater than the decision threshold and the difference is negative, a new round of benefit evaluation process is triggered to form a new decision result.

[0026] The steps involved in generating cleaning operation data in S5 are as follows: S51. Based on the decision arbitration result, control the cleaning robot to perform movement and cleaning actions through the terminal device layer; S52. Real-time data collection of energy consumption, work progress and equipment indicators through built-in sensors in the terminal device layer to generate operational data; S53. By comparing and analyzing the differences between surface images before and after cleaning, cleaning effect evaluation data is generated, which specifically includes the following steps: S531. Acquire the baseline image before cleaning and the detection image after cleaning; S532. Perform feature point matching and image registration on the two images; S533. Calculate the difference matrix of the registered images; ; in, Represents the difference matrix, This indicates the image after cleaning. Image before cleaning; S534. Generate cleaning effect evaluation scores based on the difference matrix; ; in, The score indicates the cleaning effectiveness. Indicates the total number of pixels in the image. Represents the difference matrix; S535. By comparing the cleaning effect evaluation score with the re-cleaning threshold, a re-cleaning instruction is triggered when the evaluation score is lower than the threshold. ; in, The score indicates the cleaning effectiveness. Indicates the threshold for re-cleaning; S54. By integrating and feeding back operational data and cleaning effect evaluation data to the edge computing layer, cleaning operation data is generated.

[0027] Generating updated parameters in S6 involves the following steps: S61. Based on cleaning operation data, an incremental learning algorithm is constructed by training a neural network model based on historical cleaning operation data, specifically including the following steps: S611. Obtain historically accumulated cleaning operation data and currently generated cleaning operation data; S612. Select the most recent period of data from all data in chronological order as the model training set; S613. Based on the training set data, calculate the error between the predicted output and the expected output of the current model; ; in, Represents the error function. Indicates the expected output. This represents the model's predicted output. Indicates the number of samples; S614. Based on the calculation error, adjust and update the internal parameters of the model in a targeted manner; ; in, This indicates the updated parameters. Indicates the original parameter. Indicates the learning rate. Represents the gradient of the error function; S615. Save the model with updated parameters to complete this incremental learning; S62. Analyze the error of the cleaning operation data using an incremental learning algorithm, and adjust the internal weight coefficients of the multi-objective optimization algorithm. This includes the following steps: S621. By calculating the difference between the actual output and the expected output of the cleaning operation data, an error evaluation index is generated. ; in, Represents the error function. Indicates the first The actual output value of each sample Indicates the first The expected output value for each sample. Indicates the number of samples; S622. Calculate the partial derivative of the error function with respect to the weight coefficients using the gradient descent method to generate the weight adjustment gradient; ; in, Represents the gradient of the error function. Representing the error function For the Weight coefficients of each objective The partial derivatives; S623. Adjust the gradient according to the weights and update the internal weight coefficients of the multi-objective optimization algorithm; ; in, This represents the original weighting coefficient. This represents the updated weight coefficients. Indicates the learning rate. Represents the gradient of the error function; S63. By updating the model parameters using the adjusted weight coefficients, updated parameters are generated.

[0028] Generating the adjusted global policy in S7 involves the following steps: S71. Based on the updated parameters, retrain the internal parameters of the cluster-level cleaning reward function using a machine learning algorithm, specifically including the following steps: S711. Collect and organize historical data for model training, including cleaning records, environmental parameters, and profit results; S712. Divide the complete dataset into a part for training the model and a part for testing the model; ; in, Represents the training set, This represents the test set. Represents the complete dataset. Symbol for empty set; S713. Using the training set data and the benefit as the guide label, train the benefit function model; ; in, Represents the cross-entropy loss function. Indicates the first The true label of each sample Indicates the first The predicted probability of a sample. Indicates the number of samples. Represents a logarithmic function; S714. Use test set data to evaluate the predictive performance and accuracy of the trained model; ; in, Indicates the model accuracy. Indicates the number of correctly predicted samples. Indicates the total number of samples; S715. Select the performance evaluation model and its parameters as the final model to be adopted; S72. By using the retrained profit function, global optimization is performed again to generate an adjusted global policy.

[0029] The process of generating adaptive collaborative optimization in S8 includes the following steps: S81. Execute the adjusted global strategy through the edge computing layer and collect actual effect data during the strategy execution process; S82. By comparing and analyzing the differences between the expected and actual results of the strategy through the cloud management layer, effect feedback data is generated, which includes the following steps: S821. Record the start and end times of the executed strategy; S822. During strategy execution, collect actual power generation data and related environmental data simultaneously; S823. Based on the actual power generation data collected, calculate the actual power generation revenue generated during the strategy execution period; ; in, Indicates actual electricity generation revenue, Indicates actual power generation. Indicates real-time electricity price. , These represent the start and end times of the strategy execution, respectively. Indicates time The minute changes; S824. Compare the actual power generation revenue with the expected power generation revenue before the strategy was implemented, and calculate the degree of deviation between the two. ; in, Indicates the profit deviation rate. Indicates actual electricity generation revenue, Indicates expected revenue from electricity generation; S825. Based on the calculated degree of profit deviation, generate a quantitative evaluation report on the effectiveness of the strategy. The formula for calculating the degree of profit deviation is as follows: ; in, Indicates the assessment level. This indicates that revenue is biased towards expense ratios; The quantitative assessment report is generated using the following formula: ; in, This indicates the overall evaluation score. Indicates efficiency indicators. Indicates economic indicators, Indicates reliability index, Indicates the weighting coefficient; S83. Optimize model parameters based on performance feedback data, and simultaneously adjust the strategy generation rules to generate an adaptive collaborative optimization process, specifically including the following steps: S831. Generate a quantitative indicator of strategy effectiveness by calculating the deviation rate between actual power generation revenue and expected power generation revenue. ; in, Indicates the profit deviation rate. Indicates actual electricity generation revenue, Indicates expected revenue from electricity generation; S832. By comparing the profit deviation rate with the preset optimization trigger threshold, determine whether the model and strategy need to be adjusted. S833. When it is determined that adjustment is needed, the model parameters are corrected in reverse through effect feedback data, and the decision weights in the strategy generation rules are adjusted accordingly to generate optimized model parameters and strategy rules.

[0030] The system includes: The terminal device execution module is used to collect data on surface dirt, temperature, humidity, power generation and equipment status of photovoltaic modules through the environmental sensing unit, generate multi-source sensing data, and execute physical cleaning operations according to cleaning control commands; The edge decision arbitration module receives multi-source sensing data, processes it in real time through the local data processing unit, generates local decision commands and cleaning status data using a multi-objective optimization algorithm generation unit, and outputs decision arbitration results through environmental change analysis and arbitration unit. The cloud-based global optimization module receives cleaning status data, integrates it through the cluster data aggregation unit, uses the revenue function construction and optimization unit to fuse weather forecasts and electricity price signals to generate a global strategy, and then feeds it back to the edge decision arbitration module through the strategy distribution unit.

[0031] The operation steps of this cloud-edge-device collaborative intelligent cleaning and maintenance management system and method for photovoltaic power plant clusters are as follows: Step 1: Source Sensing Data Generation Based on initial state parameters, multi-source sensing data is generated by collecting stain images, temperature, humidity, power generation, and equipment status data through the end device layer. Specifically, this includes: generating stain image data by acquiring and processing images of the photovoltaic module surface; generating temperature data by monitoring and processing the surface temperature of the photovoltaic module and the ambient temperature; generating humidity data by monitoring and processing the ambient air humidity; generating power generation data by measuring and processing the output electrical parameters of the photovoltaic module; generating equipment status data by collecting and processing the operating parameters of the cleaning robot; and finally, generating multi-source sensing data by collecting, integrating, and processing the stain image data, temperature data, humidity data, power generation data, and equipment status data.

[0032] Step 2: Generation of local decision-making instructions and cleaning status data Based on multi-source sensing data, a multi-objective optimization algorithm is used for decision processing through the edge computing layer to generate local decision commands and cleaning status data. Specifically, this includes: setting cleaning efficiency, energy consumption, equipment wear and tear, and cleaning benefit targets through a reinforcement learning framework to generate a multi-objective optimization algorithm; analyzing the severity of stains and power generation loss rate based on stain image data and power generation data from the multi-source sensing data to generate cleaning priorities; generating cleaning paths through path planning algorithms based on equipment status data and preset component layout information from the multi-source sensing data; matching stain types with preset parameters based on stain image data from the multi-source sensing data to generate cleaning intensity; generating local decision commands by integrating cleaning priorities, cleaning paths, and cleaning intensity; and generating cleaning status data by monitoring energy consumption and speed information in the equipment status data.

[0033] Step 3: Global Strategy Generation Based on local decision-making instructions and cleaning status data, a cluster-level cleaning revenue function is constructed through a cloud-based management layer. This function is then globally optimized by integrating weather forecasts and electricity price signals to generate a global strategy. Specifically, this involves: defining the sum of power generation gain minus cleaning costs, equipment depreciation costs, and communication overhead costs as the objective through the cloud-based management layer; generating power generation gain data by normalizing and correcting the power generation changes in the cleaning status data using an irradiance prediction model; obtaining future weather information through an integrated weather forecast application interface to generate weather forecast data; obtaining real-time electricity price information through a connection to the electricity market data platform to generate electricity price signals; and finally, solving the cluster-level cleaning revenue function, weather forecast data, and electricity price signals using an integer programming model to generate the global strategy.

[0034] Step 4: Generating the Decision Arbitration Result Based on environmental parameters from global strategies and multi-source sensing data, environmental change analysis and decision arbitration are performed through an edge computing layer. The difference in benefits is compared to generate a decision arbitration result. Specifically, this includes: generating environmental change trend data through time-series analysis based on temperature and humidity data from multi-source sensing data; generating a benefit difference by comparing the expected economic benefits of executing local decision commands and global strategies; and finally generating a decision arbitration result through arbitration logic based on environmental change trend data and benefit difference.

[0035] Step 5: Generating Cleaning Operation Data Based on the decision arbitration result, the cleaning operation is executed through the edge device layer, and the operation data and cleaning effect are fed back to generate cleaning operation data. Specifically, this includes: controlling the cleaning robot to perform movement and cleaning actions through the edge device layer; collecting energy consumption, work progress and equipment indicators in real time through the built-in sensors of the edge device layer to generate operation data; analyzing the differences between surface images before and after cleaning through image comparison to generate cleaning effect evaluation data; and finally integrating and feeding back the operation data and cleaning effect evaluation data to the edge computing layer to generate cleaning operation data.

[0036] Step Six: Generating Updated Parameters Based on cleaning operation data, the model parameters are updated using an incremental learning algorithm through the edge computing layer to generate updated parameters. Specifically, this includes: training a neural network model based on historical cleaning operation data, constructing an incremental learning algorithm, analyzing the error of the current cleaning operation data through the incremental learning algorithm, adjusting the internal weight coefficients of the multi-objective optimization algorithm, and finally updating the model parameters by using the adjusted weight coefficients as model parameters.

[0037] Step 7: Generating the Adjusted Global Strategy Based on the updated parameters, the revenue function is dynamically optimized through the cloud management layer to generate an adjusted global strategy. Specifically, this includes: retraining the internal parameters of the cluster-level sweeping revenue function using machine learning algorithms, and then using the retrained revenue function to perform global optimization calculations again to generate an adjusted global strategy.

[0038] Step 8: Generation of Adaptive Collaborative Optimization Process Based on the adjusted global strategy and updated parameters, an adaptive collaborative optimization process is generated through strategy execution and effect feedback at the cloud management layer and the edge computing layer. Specifically, this includes: executing the adjusted global strategy at the edge computing layer and collecting actual effect data during strategy execution; comparing and analyzing the difference between the expected effect and the actual effect data at the cloud management layer to generate effect feedback data; and finally using the effect feedback data to optimize model parameters and strategy generation logic to generate the adaptive collaborative optimization process.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters, characterized by: The method includes the following steps: S1. Based on the initial state parameters, collect stain images, temperature, humidity, power generation and equipment status data through the terminal device layer to generate multi-source sensing data; S2. Based on the multi-source sensing data, a multi-objective optimization algorithm is used through the edge computing layer to perform decision processing and generate local decision instructions and cleaning status data. S3. Based on the local decision-making instructions and the cleaning status data, a cluster-level cleaning revenue function is constructed through the cloud management layer, and global optimization is performed by integrating weather forecasts and electricity price signals to generate a global strategy. S4. Based on the global strategy and the environmental parameters in the multi-source sensing data, perform environmental change analysis and decision arbitration through the edge computing layer, compare the benefit difference, and generate a decision arbitration result. S5. Based on the decision arbitration result, perform the cleaning operation through the terminal device layer, and feed back the operation data and cleaning effect to generate cleaning operation data; S6. Based on the cleaning operation data, update the model parameters using an incremental learning algorithm through the edge computing layer to generate updated parameters; S7. Based on the updated parameters, the revenue function is dynamically optimized through the cloud management layer to generate an adjusted global strategy. S8. Based on the adjusted global strategy and the updated parameters, an adaptive collaborative optimization process is generated through the strategy execution and effect feedback of the cloud management layer and the edge computing layer.

2. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 1, characterized in that: The generation of multi-source sensing data in S1 includes the following steps: S11. By acquiring and processing images of the photovoltaic module surface, stain image data is generated; S12. Temperature data is generated by monitoring and processing the surface temperature of the photovoltaic module and the ambient temperature; S13. Generate humidity data by monitoring and processing ambient air humidity; S14. By measuring and processing the output electrical parameters of the photovoltaic module, power generation data is generated; S15. By collecting and processing the operating parameters of the cleaning robot, equipment status data is generated; S16. By collecting and integrating the stain image data, temperature data, humidity data, power generation data and equipment status data, multi-source sensing data is generated.

3. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 2, characterized in that: The process of generating local decision commands and cleaning status data in step S2 includes the following steps: S21. Based on the multi-source sensing data, a multi-objective optimization algorithm is generated by setting cleaning efficiency, energy consumption, equipment wear and tear and cleaning benefits targets through a reinforcement learning framework. S22. Based on the stain image data and power generation data in the multi-source sensing data, a cleaning priority is generated by analyzing the severity of the stains and the power generation loss rate. S23. Based on the device status data in the multi-source sensing data and the preset component layout information, a cleaning path is generated through a path planning algorithm; S24. Based on the stain image data in the multi-source sensing data, generate cleaning intensity by matching stain type with preset parameters; S25. By integrating the cleaning priority, the cleaning path, and the cleaning intensity, a local decision command is generated; S26. By monitoring the energy consumption and speed information in the device status data, cleaning status data is generated.

4. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 3, characterized in that: The generation of the global policy in S3 includes the following steps: S31. Based on the local decision-making instructions and the cleaning status data, a cluster-level cleaning revenue function is constructed by defining the sum of power generation gain minus cleaning cost, equipment depreciation cost and communication overhead cost as the target through the cloud management layer. S32. Based on the power generation changes in the cleaning status data, normalization correction is performed using the irradiance prediction model to generate power generation gain data. S33. Obtain future weather information and generate weather forecast data by integrating a weather forecast application programming interface; S34. Obtain real-time electricity price information by connecting to the electricity market data platform and generate electricity price signals; S35. A global strategy is generated by integrating the cluster-level cleaning revenue function, meteorological forecast data, and electricity price signals using an integer programming model.

5. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 4, characterized in that: The process of generating the decision arbitration result in S4 includes the following steps: S41. Based on the temperature and humidity data in the multi-source sensing data, environmental change trend data is generated through time series analysis; S42. Based on the local decision instruction and the global strategy, a profit difference is generated by comparing the expected economic benefits of executing the two. S43. Based on the environmental change trend data and the profit difference, an arbitration decision is generated through arbitration logic.

6. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 5, characterized in that: The step of generating cleaning operation data in S5 includes the following steps: S51. Based on the decision arbitration result, control the cleaning robot to perform movement and cleaning actions through the terminal device layer; S52. Real-time data collection of energy consumption, work progress and equipment indicators through built-in sensors in the terminal device layer to generate operational data; S53. By comparing and analyzing the differences between surface images before and after cleaning, cleaning effect evaluation data is generated. S54. By integrating and feeding back the operation data and the cleaning effect evaluation data to the edge computing layer, cleaning operation data is generated.

7. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 6, characterized in that: The steps involved in generating the updated parameters in step S6 are as follows: S61. Based on the cleaning operation data, an incremental learning algorithm is constructed by training a neural network model based on historical cleaning operation data. S62. Analyze the error of the cleaning operation data through the incremental learning algorithm, and adjust the internal weight coefficients of the multi-objective optimization algorithm. S63. By updating the model parameters using the adjusted weight coefficients, updated parameters are generated.

8. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 7, characterized in that: The process of generating the adjusted global policy in S7 includes the following steps: S71. Based on the updated parameters, retrain the internal parameters of the cluster-level cleaning revenue function using a machine learning algorithm; S72. By using the retrained profit function, global optimization is performed again to generate an adjusted global policy.

9. The cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters according to claim 8, characterized in that: The adaptive collaborative optimization process in S8 includes the following steps: S81. Execute the adjusted global strategy through the edge computing layer and collect actual effect data during the strategy execution process; S82. By comparing and analyzing the differences between the expected and actual results of the strategy through the cloud management layer, effect feedback data is generated. S83. Optimize the model parameters based on the effect feedback data, and simultaneously adjust the strategy generation rules to generate an adaptive collaborative optimization process.

10. A cloud-edge-device collaborative intelligent cleaning and maintenance management system for photovoltaic power plant clusters, used to implement the cloud-edge-device collaborative intelligent cleaning and maintenance management method for photovoltaic power plant clusters as described in any one of claims 1-9, characterized in that: The system includes: The terminal device execution module is used to collect data on surface dirt, temperature, humidity, power generation and equipment status of photovoltaic modules through the environmental sensing unit, generate multi-source sensing data, and execute physical cleaning operations according to cleaning control commands; The edge decision arbitration module receives the multi-source sensing data, processes it in real time through the local data processing unit, generates local decision instructions and cleaning status data using a multi-objective optimization algorithm generation unit, and outputs decision arbitration results through environmental change analysis and arbitration unit. The cloud-based global optimization module receives the cleaning status data, integrates it through the cluster data aggregation unit, uses the revenue function construction and optimization unit to fuse weather forecasts and electricity price signals to generate a global strategy, and feeds it back to the edge decision arbitration module through the strategy distribution unit.

Images