A method and system for optimizing a photovoltaic cleaning strategy based on a Markov chain, and an electronic device

Abstract

The embodiment of the application provides a photovoltaic cleaning strategy optimization method and system based on a Markov chain, and an electronic device, the method comprising: obtaining historical meteorological data and photovoltaic operation data, performing data training after preprocessing, determining a prediction model of photovoltaic theoretical power and photovoltaic dust loss; discretizing the dust level in the photovoltaic dust loss prediction model, determining the corresponding dust state, and based on the MDP framework, combining the random event probability, constructing the transition probability matrix of the dust, and further determining the revenue function; integrating the prediction data of the photovoltaic theoretical power and the random event, rolling solving the MDP model through an iterative algorithm, and adaptively adjusting and optimizing the time domain, taking the maximization of the revenue function as the iteration target, and generating the photovoltaic cleaning strategy of the current time domain; simulating the operation of the photovoltaic cleaning strategy and the remaining benchmark strategy, and calculating the multi-dimensional evaluation index in the simulation operation process.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and in particular to a method, system and electronic device for optimizing photovoltaic cleaning strategies based on Markov chains. Background Technology

[0002] As a crucial component of clean energy, photovoltaic (PV) power generation has seen continuous growth in installed capacity globally. However, in key development areas such as desert and Gobi regions, dust accumulation on PV module surfaces is particularly prominent, becoming a core challenge restricting power generation efficiency and operation and maintenance management. Desert and Gobi areas possess abundant solar resources, but their extremely harsh environments, including strong winds, high ultraviolet radiation, and drastic temperature differences, exacerbate dust accumulation. Dust cover not only reduces light transmittance by blocking, reflecting, and absorbing light, but can also trigger localized hot spot effects, accelerating module aging. Studies show that in these areas, dust accumulation can cause PV module output power to decrease by more than 30% in the short term; if not cleaned for a long period, the loss can even reach over 65%, far exceeding the 8% loss level in typical areas, posing a serious threat to the economic viability of power plants.

[0003] Traditional cleaning strategies rely primarily on fixed cycles or experience-based judgment. However, in arid desert regions where water resources are extremely scarce, frequent cleaning is not only costly (accounting for 10%–70% of total operation and maintenance costs) but also difficult to adapt to dynamically changing weather conditions. For example, ineffective rainfall may exacerbate dust accumulation, while sudden sandstorms can quickly negate the cleaning effect, and fixed-cycle cleaning often leads to resource waste or insufficient cleaning. To address this challenge, intelligent cleaning technologies such as drone cleaning, self-cleaning components, and data analysis platforms are gradually being applied, enabling precise operation and maintenance through data-driven approaches. For instance, drone cleaning technology can clean 6,000 components per day, with the efficiency equivalent to seven manual laborers, while avoiding damage to the underlying ecosystem. Although these technologies have shown initial effectiveness, they still lack the dynamic response capability to weather uncertainties, particularly in the face of the random cleaning effect of rainfall and the dust-accumulating effect of sandstorms. Summary of the Invention

[0004] To address the problems existing in the prior art, embodiments of the present invention provide a photovoltaic cleaning strategy optimization method and system based on Markov chains.

[0005] This invention provides a photovoltaic cleaning strategy optimization method based on Markov chains, the method comprising:

[0006] Historical meteorological data and photovoltaic operation data are acquired, preprocessed, and then used for data training to determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss;

[0007] The dust accumulation level in the photovoltaic dust accumulation loss prediction model is discretized to determine the corresponding dust accumulation state. Based on the MDP framework and combined with the probability of random events, a dust accumulation transition probability matrix is ​​constructed. Then, the revenue function is determined by combining photovoltaic dust accumulation loss and cleaning cost.

[0008] By integrating the theoretical power of photovoltaics and the forecast data of random events, the MDP model is solved in a rolling manner through an iterative algorithm, and the time domain is adaptively adjusted and optimized according to the stability of random events. The photovoltaic cleaning strategy in the current time domain is generated with the maximization of the revenue function as the iterative objective.

[0009] The photovoltaic cleaning strategy was simulated and run against other benchmark strategies. Multi-dimensional evaluation indicators were calculated during the simulation process. The optimization benefits of the photovoltaic cleaning strategy were quantitatively verified based on the multi-dimensional evaluation indicators.

[0010] In one embodiment, the method further includes:

[0011] Set state variables, set Indicates the first The state of dust accumulation in the sky, ,in:

[0012] The dust accumulation loss is 0.

[0013] This represents the maximum dust accumulation loss;

[0014] Set the probability of rainfall in the random event probability. The probability of a sandstorm event, and the probability of a normal day. ;

[0015] Define the action space ,in Indicates no cleaning. 1 indicates cleaning;

[0016] when 1 o'clock;

[0017]

[0018] when hour;

[0019]

[0020] in, This represents the transition in dust state caused by sandstorms, reflecting the intense depositional effect of dust. This represents a random variable representing the normal daily accumulation of dust.

[0021] In one embodiment, the method further includes:

[0022] Calculating daily electricity revenue includes:

[0023]

[0024] in: For electricity price; For the first The theoretical power of photovoltaic power in a day For state Corresponding photovoltaic dust accumulation loss;

[0025] Obtain cleaning costs Determine the daily instant profit function, including:

[0026]

[0027] The Bellman equation for the payoff function, which aims to maximize the expected total payoff, includes:

[0028]

[0029] in, For state s in time The optimal expected return at the end of the period.

[0030] In one embodiment, the method further includes:

[0031] Initialize terminal conditions, including:

[0032] ;

[0033] For each state The function to update its value includes:

[0034] ;

[0035] in, For day t, state The maximum expected total return that can be obtained at the beginning and end of the term;

[0036] Record the optimal actions, including:

[0037] ;

[0038] in, The optimal action for each state at each time step;

[0039] Observe whether the random event occurs, and perform state transitions based on the actual random event. For the k-th step in the prediction time domain, the calculation method using the exponentiation of the transition matrix includes:

[0040] .

[0041] In one embodiment, the method further includes:

[0042] During the state transition process, at each transition step, the probability of the occurrence of random events in the future time period is reacquired, and the transition matrix is ​​dynamically updated.

[0043] The duration of the future time period is dynamically adjusted based on the degree of meteorological stability.

[0044] In one embodiment, the multi-dimensional evaluation metrics include:

[0045] Annual net income indicators Its calculation formula includes:

[0046] ;

[0047] in, Let t be the electricity price on day t. Let t be the actual power generation after dust accumulation. Total number of cleanings per year Cost per cleaning session;

[0048] Average power loss rate Its calculation formula includes:

[0049] ;

[0050] in, This represents the theoretical power of photovoltaic power. This refers to the actual power or the power after data cleaning during simulation.

[0051] Cleaning cost percentage;

[0052] And the results of scenario sensitivity and robustness assessment for the random events.

[0053] In one embodiment, the historical meteorological data includes:

[0054] Temperature, ambient humidity, wind speed, irradiance, atmospheric pressure, and rainfall;

[0055] The photovoltaic operation data includes:

[0056] Actual power generation and dust pollution index;

[0057] The preprocessing includes data cleaning and data normalization. The data cleaning includes missing value imputation and outlier identification and imputation.

[0058] In one embodiment, the method further includes:

[0059] The photovoltaic theoretical power prediction model is a photovoltaic theoretical power prediction model based on CEEMDAN-CNN-LSTM, and the photovoltaic dust accumulation loss prediction model is a photovoltaic dust accumulation loss prediction model based on BO-LSTM.

[0060] This invention provides a photovoltaic cleaning strategy optimization system based on Markov chains, the system comprising:

[0061] The acquisition module is used to acquire historical meteorological data and photovoltaic operation data, preprocess the data, and then train the data to determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss.

[0062] The module is used to discretize the dust accumulation level in the photovoltaic dust accumulation loss prediction model, determine the corresponding dust accumulation state, and construct the dust accumulation transition probability matrix based on the MDP framework and the probability of random events. Then, the revenue function is determined by combining photovoltaic dust accumulation loss and cleaning cost.

[0063] The rolling module is used to integrate the theoretical power of photovoltaics and the forecast data of random events, solve the MDP model in a rolling manner through an iterative algorithm, and adaptively adjust the optimization time domain according to the stability of random events. With the maximization of the revenue function as the iterative objective, it generates the photovoltaic cleaning strategy for the current time domain.

[0064] The verification module is used to simulate the photovoltaic cleaning strategy with other benchmark strategies, calculate multi-dimensional evaluation indicators during the simulation process, and quantitatively verify the optimization benefits of the photovoltaic cleaning strategy based on the multi-dimensional evaluation indicators.

[0065] This invention provides an electronic device, including a processor and a memory;

[0066] The processor is connected to the memory;

[0067] The memory is used to store executable program code;

[0068] The processor runs a program corresponding to the executable program code stored in the memory to perform the methods described in one or more embodiments.

[0069] This invention provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described photovoltaic cleaning strategy optimization method based on Markov chains.

[0070] In view of the above, in one or more embodiments of this specification, historical meteorological data and photovoltaic operation data are acquired, preprocessed, and then used for data training to determine a prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss; the dust accumulation level in the photovoltaic dust accumulation loss prediction model is discretized to determine the corresponding dust accumulation state, and based on the MDP framework, combined with the probability of random events, a dust accumulation transition probability matrix is ​​constructed, and then combined with photovoltaic dust accumulation loss and cleaning cost to determine the benefit function; the photovoltaic theoretical power and random event forecast data are integrated, and the MDP model is solved in a rolling manner through an iterative algorithm, and the optimization time domain is adaptively adjusted according to the stability of random events, with the maximization of the benefit function as the iterative objective, to generate a photovoltaic cleaning strategy for the current time domain; the photovoltaic cleaning strategy is simulated and run with other benchmark strategies, and multi-dimensional evaluation indicators are calculated during the simulation process, and the optimization benefit of the photovoltaic cleaning strategy is quantitatively verified based on the multi-dimensional evaluation indicators. By discretizing the state space and probabilistically transforming the transition matrix, frameworks such as Markov Decision Process (MDP) can embed the uncertainty of random events, such as rainfall and sandstorms, into the decision-making process, thereby maximizing long-term benefits. This provides strong decision support for achieving refined management of photovoltaic power plant operation and maintenance and maximizing benefits throughout the entire life cycle. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a flowchart of a photovoltaic cleaning strategy optimization method based on Markov chains, provided in one embodiment of this specification.

[0073] Figure 2 This is a flowchart of another embodiment of the photovoltaic cleaning strategy optimization method based on Markov chains provided in this specification.

[0074] Figure 3 This is a schematic diagram of a photovoltaic cleaning strategy optimization system based on Markov chains, provided in one embodiment of this specification.

[0075] Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0076] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0077] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0078] like Figure 1 As shown, this embodiment of the invention provides a photovoltaic cleaning strategy optimization method based on Markov chains, including:

[0079] Step S102: Obtain historical meteorological data and photovoltaic operation data, preprocess them, and then train the data to determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss.

[0080] Specifically, historical meteorological data can be obtained from meteorological stations or reanalysis databases (such as ERA5), including at least physical environmental variables that directly affect photovoltaic output, such as temperature, ambient humidity, wind speed, irradiance, atmospheric pressure, and rainfall. Photovoltaic operation data can be obtained from the power station monitoring system, with the core being the actual power generation and the dust pollution index reflecting the cleanliness of the component surface (such as an index calculated based on series resistance, infrared imaging, or power comparison before and after daily cleaning).

[0081] Then, the above data undergoes preprocessing steps including data cleaning and data normalization. Data cleaning may include: Missing value handling: Identifying data breakpoints. For short-term missing values, linear interpolation is used to fill them (assuming continuous change); for long-term missing values ​​(such as equipment failure), they are filled using the mean or weighted mean based on similar days (same season, same weather type). Outlier handling: Anomalies are identified by combining physical constraints (such as nighttime irradiance should be 0, power cannot be negative, and power limits are limited by installed capacity). For example, power close to zero at noon on a sunny day is an anomaly, which can be removed and filled using the above methods. Data normalization: To eliminate differences in the dimensions of different features and accelerate model convergence, the Min-Max normalization method is used to map each feature data to the [0,1] interval.

[0082] Furthermore, the theoretical power of photovoltaics refers to the photovoltaic power generation under ideal dust-free conditions, while photovoltaic dust loss refers to the difference between the actual power generation and the theoretical power caused by dust accumulation on the surface of photovoltaic modules. A photovoltaic theoretical power prediction model based on CEEMDAN-CNN-LSTM and a photovoltaic dust loss prediction model based on BO-LSTM can be established respectively. The specific steps include:

[0083] The process of establishing a photovoltaic theoretical power prediction model includes: using preprocessed meteorological data (temperature, irradiance, etc.) as input features (X), and using the corresponding, filtered "dust-free / freshly cleaned" actual photovoltaic power data as the target value (Y). Here, Y is approximated as "theoretical power." Then, the input meteorological feature sequence or target power sequence is decomposed. The CEEMDAN algorithm adaptively decomposes the complex fluctuating original sequence into a series of intrinsic mode functions (IMFs) arranged from high to low frequency and residual terms representing long-term trends. Feature extraction and time-series modeling are then performed. The decomposed IMF subsequences and residual terms are recombined into multidimensional input data, such as CNN layers, which act on the multidimensional features at each time step (e.g., IMF1 to IMFk and residual values ​​at a given moment). Convolutional kernels extract the local correlations and spatial features between these components (essentially complex coupling relationships between features). An LSTM layer receives the feature sequence processed by the CNN and uses its gating mechanisms (forget gate, input gate, output gate) to capture the long-term dependencies of each subsequence in the time dimension. For example, understanding how morning irradiance patterns affect afternoon power output. Finally, the output of the LSTM layer is integrated through fully connected layers to ultimately output the theoretical power prediction for a future time period (e.g., the next hour, the next day). This determines the performance benchmark, i.e., "how much power should be generated if the components are completely clean."

[0084] The process of establishing a photovoltaic dust accumulation loss prediction model includes: 1. Calculating the actual dust accumulation loss: Dust accumulation loss value = theoretical power prediction value - actual power generation value. The theoretical power prediction value can be obtained through a photovoltaic theoretical power prediction model; 2. Preparing meteorological data (especially wind speed, humidity, rainfall, and dust-related indices) and time-cumulative features (such as the number of consecutive rainless days) as input features (X), and using the calculated dust accumulation loss value as the target value (Y); 3. Model construction and hyperparameter optimization (BO-LSTM), including defining the LSTM network structure search space, running the training iteration and convergence process in Bayesian optimization (BO), and finally determining the trained model. This determines the performance degradation, i.e., "how much the power generation is discounted due to the current dust and weather."

[0085] Step S104: Discretize the dust accumulation level in the photovoltaic dust accumulation loss prediction model, determine the corresponding dust accumulation state, and construct the dust accumulation transition probability matrix based on the MDP framework and the probability of random events. Then, combine the photovoltaic dust accumulation loss and cleaning cost to determine the revenue function.

[0086] Specifically, a photovoltaic cleaning optimization strategy model considering the uncertainties of random events, namely rainfall and sandstorms, is introduced, employing a dynamic stochastic optimization method based on Markov decision processes (MDPs). The model generates an adaptive cleaning strategy by quantifying dust accumulation state transitions, power generation revenue, and cleaning costs, with the objective of maximizing long-term expected returns. Rainfall events can reset dust accumulation on photovoltaic panels to a clean state, while sandstorm events cause a surge in dust accumulation; both are embedded in the state transition process in a probabilistic manner.

[0087] First, the continuous dust accumulation problem is discretized, transforming it into a standard, solvable MDP problem. The discrete state space captures the main characteristics of the dust accumulation dynamics while avoiding the computational complexity of the continuous state space, including:

[0088] The dust accumulation level is discretized into a finite state space to simplify calculations and capture dust accumulation dynamics; state variables are defined: let Indicates the first The state of dust accumulation in the sky, ,in:

[0089] : Completely clean state (dust loss is 0);

[0090] : Severe dust accumulation condition (maximum dust loss);

[0091] The state interval is divided by historical dust accumulation data, such as based on dust density or power loss ratio.

[0092] Wherein, the state mapping is: state Corresponding theoretical power generation loss ratio ,in The loss function model is obtained by training with historical loss data of photovoltaic dust accumulation.

[0093] Furthermore, by embedding uncertain meteorological events into the state transition process in probabilistic form, the model can quantify the impact of these uncertainties on dust accumulation, thereby making more robust decisions. Rainfall and sandstorms, as random events, have their probabilities obtained through historical meteorological data statistics or forecasting models: Rainfall event probability. : The probability of rainfall occurring daily (the rainfall threshold can be set to ≥0.3mm, since light rain has no cleaning effect).

[0094] Sandstorm event probability The probability of a sandstorm occurring daily.

[0095] Normal day probability:

[0096]

[0097] The state transition probability matrix is ​​defined, and state transitions are determined by both actions and random events. The action space is defined. ,in Indicates no cleaning. 1 indicates cleaning (cost is fixed). Transition probability as follows:

[0098] when 1 (Cleaning): Regardless of the event, the dust accumulation state is reset to clean, independent of random events.

[0099]

[0100] when (Without cleaning), state transitions are affected by random events:

[0101]

[0102] in, This represents the transition in dust accumulation caused by sandstorms, reflecting the intense depositional effect of sand and dust. Let be a random variable representing the daily accumulation of dust, whose parameters are fitted using historical data. The transition probability matrix is:

[0103]

[0104] in, These are the normally accumulated transfer nuclei.

[0105] Daily revenue consists of electricity generation revenue minus cleaning costs. Electricity generation revenue:

[0106]

[0107] in: The market electricity price (RMB / kWh); For the first The theoretical daily power generation (kWh) is the output of the CEEMDAN-CNN-LSTM model. For state The corresponding power loss.

[0108] Cleaning costs This includes labor and equipment costs, and is set as a fixed value.

[0109] The immediate revenue function states that the total daily revenue is the electricity generation revenue minus the cleaning cost (if cleaning is performed):

[0110]

[0111] The dynamic optimization objective function is: to maximize the expected total return over the finite period T, the Bellman equation is established as follows:

[0112]

[0113] in: For state s in time The optimal expected return at the end of the period;

[0114] Step S106: Integrate the theoretical photovoltaic power and the prediction data of random events, perform rolling solution on the MDP model through an iterative algorithm, and adaptively adjust the optimization time domain according to the stability of random events. With the maximization of the revenue function as the iterative objective, generate the photovoltaic cleaning strategy for the current time domain.

[0115] Specifically, the optimal policy is solved using value iteration or policy iteration algorithms to find the payoff function. This involves mapping the state to the optimal action (whether to clean or not). The optimization process requires integrating real-time weather forecasts (probability of rainfall / sandstorms) and electricity price data to achieve rolling updates.

[0116] The Bellman equation in step S104 is solved using a value iteration algorithm, and the steps are as follows:

[0117] Initialization: Set terminal conditions;

[0118]

[0119] Reverse iteration (from) arrive For each state Update its value function, recording the state from day t. The maximum expected total return that can be obtained at the beginning and end of the term:

[0120]

[0121] Simultaneously, record the optimal action and store the optimal action for each state at each time step, given the current information:

[0122]

[0123] Rolling execution: Taking action at the current time t Observe actual state transitions (such as whether there is rainfall or sandstorm) and update to Moving to time t+1, the updated weather forecast and electricity price data are retrieved again, and the above steps are repeated. Detecting whether the probabilistic events (rainfall / sandstorm) predicted by the model actually occur will cause a deviation between the actual state transition and the prediction. Updating the state through observation is equivalent to "state correction" of the system using real-world data, avoiding the infinite accumulation of model errors.

[0124] For the k-th step (k≥2) in the prediction time domain, the result is obtained through exponentiation of the transition matrix:

[0125]

[0126] Clarifying the calculation method for multi-step transition probabilities is the theoretical basis for calculating future expected returns in value iteration.

[0127] Weather forecast update: The probability of rainfall / sandstorm is re-acquired every time the forecast reaches day T. The transition matrix P is dynamically updated.

[0128] To reduce computational complexity, an adaptive time-domain adjustment strategy is introduced:

[0129] When weather is stable (e.g., no rain forecast for several consecutive days), extend T to 14 days to capture long-term benefits. Taking full advantage of the predictability of stable periods for longer-term planning may reveal more economical strategies such as "delaying cleansing to wait for natural rainfall."

[0130] When weather changes abruptly (e.g., a sandstorm is forecast for the next two days), shorten the time horizon (T) to 3-5 days to focus on short-term decision-making. Focus on responding to impending high-risk or high-opportunity events. Long-term forecasts are extremely unreliable during periods of sudden change; shortening the time horizon avoids ineffective planning based on erroneous information and reduces computational load for faster response.

[0131] Step S108: Simulate the photovoltaic cleaning strategy with other benchmark strategies, calculate multi-dimensional evaluation indicators during the simulation, and quantitatively verify the optimization benefits of the photovoltaic cleaning strategy based on the multi-dimensional evaluation indicators.

[0132] Specifically, other benchmark strategies include, for example, the fixed-period cleaning method: a widely used benchmark strategy. This method, based on historical experience or general recommendations, sets a fixed cleaning interval (e.g., 15 days, 30 days), without considering dynamic changes in actual weather conditions, dust accumulation rates, and market electricity prices. Static optimization methods: This method is more advanced than the fixed-period method. It typically determines an "optimal" static cleaning threshold or cycle that remains constant throughout the assessment period based on historical data (e.g., average dust accumulation rate, average rainfall probability). While it optimizes for historical patterns, it cannot respond to real-time changes.

[0133] The photovoltaic cleaning strategy was then simulated against other benchmark strategies. For example, a full year's worth of historical data was prepared for simulation, which could include daily weather data, daily theoretical power generation, and daily electricity prices. To comprehensively quantify the performance of each method, the following multi-dimensional evaluation index system was established.

[0134] 1. Economic Benefit Assessment: Economic benefit is the core criterion for evaluating the success or failure of a cleaning strategy. The primary assessment focuses on annual net income. The calculation formula is as follows:

[0135]

[0136] in, Let t be the electricity price on day t. Let t be the actual power generation after dust accumulation. Total number of cleanings per year This represents the cost per cleaning cycle. The target method is expected to achieve the highest annual net profit. Simultaneously, the rate of increase in revenue is calculated:

[0137]

[0138] By quantifying the degree of improvement relative to the best benchmark method, we can intuitively demonstrate the economic benefits brought about by intelligent strategies.

[0139] 2. Technical Performance Evaluation: Technical performance focuses on maintaining the level of power generation efficiency. The average power loss rate is calculated. :

[0140]

[0141] in, The theoretical power predicted by the S1 model. This represents the actual power output or the power output after considering the cleaning decision in the simulation. The lower this value, the better the strategy performs in maintaining high power generation efficiency.

[0142] 3. Operation and Maintenance Efficiency Assessment: This assesses resource input for operation and maintenance. The core indicator is the total number of cleaning operations per year. Under the premise of ensuring power generation revenue, fewer cleaning operations mean lower consumption of manpower and water resources, and lower operation and maintenance costs. Additionally, the percentage of cleaning costs is calculated.

[0143]

[0144] The smaller the ratio, the lower the proportion of cleaning costs in total revenue, and the higher the operational efficiency.

[0145] 4. Scene Sensitivity and Robustness Assessment: This assessment aims to test the method's ability to cope with uncertainty. For example, data is segmented according to meteorological characteristics (e.g., rainy season, dry season, sandstorm season), and the performance of each strategy is statistically analyzed under different seasons. During the rainy season, it is observed whether the target strategy reduces the number of manual cleaning operations (utilizing natural rainfall) and whether the returns are maintained. During the sandstorm season, it is observed whether the target strategy cleans in a timely manner before or after sandstorms to reduce losses. This verifies the dynamic response capability of the target strategy. The performance of fixed-period and static optimization strategies may fluctuate greatly in different seasons, while the target strategy should be able to adaptively adjust and maintain good performance under various meteorological conditions, demonstrating robustness.

[0146] Therefore, through multi-indicator, multi-dimensional comparison, it is demonstrated that the intelligent cleaning strategy based on MDP rolling optimization in this embodiment is superior to traditional methods in terms of economy, technology, and operational efficiency. Furthermore, the system model's state can be further updated based on actual results, such as... Figure 2 As shown, in Figure 2 In the process of the photovoltaic cleaning strategy optimization method based on Markov chain, after theoretical power prediction and dust loss prediction are input into the Markov decision model, rolling optimization is performed to obtain the optimal decision and run the simulation. The model can be updated in real time according to the actual results. After obtaining the running data, the data is used as the data input for the next iteration update to achieve further refinement and accuracy of the model.

[0147] This invention provides a photovoltaic (PV) cleaning strategy optimization method based on Markov chains. The method acquires historical meteorological data and PV operation data, preprocesses them, and trains the data to determine prediction models for theoretical PV power and PV dust accumulation loss. The dust accumulation level in the prediction model is discretized to determine the corresponding dust accumulation state. Based on the MDP framework and combined with random event probabilities, a dust transition probability matrix is ​​constructed. Then, combining PV dust accumulation loss and cleaning costs, a revenue function is determined. Integrating PV theoretical power and random event forecast data, the MDP model is solved iteratively using an iterative algorithm. The optimization time domain is adaptively adjusted based on the stability of random events, with the maximization of the revenue function as the iteration objective, generating a PV cleaning strategy for the current time domain. The PV cleaning strategy is simulated against other benchmark strategies, and multi-dimensional evaluation indicators are calculated during the simulation. The optimization revenue of the PV cleaning strategy is quantitatively verified based on these multi-dimensional evaluation indicators. This innovative approach combines CEEMDAN-CNN-LSTM theoretical power prediction, BO-LSTM dust accumulation loss prediction, and a Markov chain cleaning optimization model that considers the uncertainties of rainfall and sandstorms. This constructs a complete technology chain from accurate power prediction to intelligent cleaning decision-making, effectively solving the problem that traditional fixed cleaning strategies are ill-suited to complex weather changes and fluctuating market electricity prices. By dynamically characterizing the stochastic transition characteristics of dust accumulation states using the Markov decision process, the natural cleaning effect of rainfall and the dust accumulation aggravation effect of sandstorms are cleverly transformed into state transition probabilities. Furthermore, a Bellman equation with the goal of maximizing long-term expected returns is constructed using market electricity price signals, enabling the cleaning strategy to adapt to changes in weather forecasts and electricity prices. The method exhibits fluctuating and adaptive adjustments, significantly improving the economic efficiency of photovoltaic power plants in real-world operating environments. Furthermore, by employing a rolling optimization mechanism to integrate the latest weather forecasts and power prediction data in real time, it dynamically re-optimizes the decision sequence within a limited time domain, executing only the first decision to ensure the timeliness and reliability of the strategy. This effectively overcomes the shortcomings of static optimization methods in responding to sudden weather events. Finally, through comparative verification with fixed-period cleaning methods and static optimization methods in year-round simulations, this method demonstrates significant superiority in key indicators such as annual net income, average power loss rate, and cleaning cost ratio. This provides strong decision support for achieving refined management of photovoltaic power plant operation and maintenance and maximizing life-cycle benefits.

[0148] Please see Figure 3 , Figure 3 This is a schematic diagram of a photovoltaic cleaning strategy optimization system based on a Markov chain, provided in an embodiment of this application. Figure 3 As shown, the system includes:

[0149] The acquisition module S302 is used to acquire historical meteorological data and photovoltaic operation data, preprocess the data, train the data, and determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss.

[0150] The module S304 is used to discretize the dust accumulation level in the photovoltaic dust accumulation loss prediction model, determine the corresponding dust accumulation state, and construct the dust accumulation transition probability matrix based on the MDP framework and the probability of random events. Then, the revenue function is determined by combining photovoltaic dust accumulation loss and cleaning cost.

[0151] The rolling module S306 is used to integrate the theoretical power of photovoltaic and the prediction data of random events, solve the MDP model in a rolling manner through an iterative algorithm, and adaptively adjust the optimization time domain according to the stability of random events. With the maximization of the revenue function as the iterative objective, the photovoltaic cleaning strategy in the current time domain is generated.

[0152] The verification module S308 is used to simulate the photovoltaic cleaning strategy with other benchmark strategies, calculate multi-dimensional evaluation indicators during the simulation process, and quantitatively verify the optimization benefits of the photovoltaic cleaning strategy based on the multi-dimensional evaluation indicators.

[0153] Those skilled in the art will clearly understand that the technical solutions of the embodiments of this application can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware that can independently complete or cooperate with other components to complete a specific function, wherein the hardware may be, for example, a field-programmable gate array (FPGA), an integrated circuit (IC), etc.

[0154] Each processing unit and / or module in the embodiments of this application can be implemented by an analog circuit that implements the functions described in the embodiments of this application, or by software that executes the functions described in the embodiments of this application.

[0155] See Figure 4 It shows a schematic diagram of the structure of an electronic device according to an embodiment of this application, which can be used to implement... Figure 1 The method in the illustrated embodiment. (As shown) Figure 4 As shown, the electronic device 400 may include: at least one processor 401, at least one network interface 404, user interface 403, memory 405, and at least one communication bus 402.

[0156] The communication bus 402 is used to enable communication between these components.

[0157] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0158] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0159] The processor 401 may include one or more processing cores. The processor 401 connects to various parts within the electronic device 400 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 405, and by calling data stored in the memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, without being integrated into the processor 401.

[0160] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. Figure 4 As shown, the memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0161] exist Figure 4 In the electronic device 400 shown, the user interface 403 is mainly used to provide an input interface for the user and acquire user input data; while the processor 401 can be used to call the image-based interactive application stored in the memory 405 and specifically perform the following operations: acquire historical meteorological data and photovoltaic operation data, preprocess them and then train the data to determine the prediction model of photovoltaic theoretical power and photovoltaic dust accumulation loss; discretize the dust accumulation level in the photovoltaic dust accumulation loss prediction model, determine the corresponding dust accumulation state, and construct the dust accumulation transition probability matrix based on the MDP framework and combined with the probability of random events, and then determine the benefit function by combining photovoltaic dust accumulation loss and cleaning cost; integrate the photovoltaic theoretical power and random event forecast data, perform rolling solution of the MDP model through an iterative algorithm, and adaptively adjust the optimization time domain according to the stability of random events, with the maximization of the benefit function as the iterative objective, to generate the photovoltaic cleaning strategy in the current time domain; simulate the photovoltaic cleaning strategy with other benchmark strategies, calculate the multi-dimensional evaluation index during the simulation operation, and quantitatively verify the optimization benefit of the photovoltaic cleaning strategy based on the multi-dimensional evaluation index.

[0162] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0163] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0164] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0165] In the several embodiments provided in this application, it should be understood that the disclosed apparatus 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 service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0166] 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.

[0167] Furthermore, the functional units in the various embodiments of this application 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0168] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0169] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0170] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A photovoltaic cleaning strategy optimization method based on Markov chains, characterized in that, The method includes: Historical meteorological data and photovoltaic operation data are acquired, preprocessed, and then used for data training to determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss; The dust accumulation level in the photovoltaic dust accumulation loss prediction model is discretized to determine the corresponding dust accumulation state. Based on the MDP framework and combined with the probability of random events, a dust accumulation transition probability matrix is ​​constructed. Then, the revenue function is determined by combining photovoltaic dust accumulation loss and cleaning cost. The process of discretizing the dust accumulation level in the photovoltaic dust accumulation loss prediction model to determine the corresponding dust accumulation state, and constructing a dust accumulation transition probability matrix based on the MDP framework and combining random event probabilities, includes: Set state variables, set Indicates the first The state of dust accumulation in the sky, ,in: The dust accumulation loss is 0. This represents the maximum dust accumulation loss; Set the probability of rainfall in the random event probability. Probability of sandstorm events And the probability of a normal day, ; Define the action space ,in Indicates no cleaning. 1 indicates cleaning; when At 1 o'clock, ， when hour, ， in, This represents the transition in dust state caused by sandstorms, reflecting the intense depositional effect of dust. This represents a random variable representing the normal daily accumulation of dust. The transition probability matrix is: ， in, These are normally accumulated transfer nuclei; By integrating the theoretical power of photovoltaics and the forecast data of random events, the MDP model is solved in a rolling manner through an iterative algorithm, and the time domain is adaptively adjusted and optimized according to the stability of random events. The photovoltaic cleaning strategy in the current time domain is generated with the maximization of the revenue function as the iterative objective. The photovoltaic cleaning strategy was simulated and run against other benchmark strategies. Multi-dimensional evaluation indicators were calculated during the simulation process. The optimization benefits of the photovoltaic cleaning strategy were quantitatively verified based on the multi-dimensional evaluation indicators.

2. The method according to claim 1, characterized in that, The determination of the revenue function by combining photovoltaic dust accumulation loss and cleaning costs includes: Calculating daily electricity revenue includes: ， in: For electricity price; For the first The theoretical power of photovoltaic power in a day For state Corresponding photovoltaic dust accumulation loss; Obtain cleaning costs Determine the daily instant profit function, including: ， The Bellman equation for the payoff function, which aims to maximize the expected total payoff, includes: ， in, For state s in time The optimal expected return at the end of the period.

3. The method according to claim 2, characterized in that, The method of integrating the theoretical photovoltaic power and the forecast data of random events, and solving the MDP model using an iterative algorithm, includes: Initialize terminal conditions, including: ， For each state The function to update its value includes: ， in, For day t, state The maximum expected total return that can be obtained at the beginning and end of the term; Record the optimal actions, including: ， in, The optimal action for each state at each time step; Observe whether the random event occurs, and perform state transitions based on the actual random event. For the k-th step in the prediction time domain, the calculation method using the exponentiation of the transition matrix includes: 。 4. The method according to claim 3, characterized in that, The method further includes: During the state transition process, at each transition step, the probability of the occurrence of random events in the future time period is reacquired, and the transition matrix is ​​dynamically updated. The duration of the future time period is dynamically adjusted based on the degree of meteorological stability.

5. The method according to claim 3, characterized in that, The multi-dimensional evaluation indicators include Annual net income indicators Its calculation formula includes: ， in, Let t be the electricity price on day t. Let t be the actual power generation after dust accumulation. Total number of cleanings per year Cost per cleaning session; Average power loss rate Its calculation formula includes: ， in, This represents the theoretical power of photovoltaic power. This refers to the actual power or the power after data cleaning during simulation. Cleaning cost percentage; And the results of scenario sensitivity and robustness assessment for the random events.

6. The method according to claim 1, characterized in that, The historical meteorological data includes: Temperature, ambient humidity, wind speed, irradiance, atmospheric pressure, and rainfall; The photovoltaic operation data includes: Actual power generation and dust pollution index; The preprocessing includes data cleaning and data normalization. The data cleaning includes missing value imputation and outlier identification and imputation.

7. The method according to claim 1, characterized in that, The method further includes: The prediction model corresponding to the theoretical photovoltaic power is a photovoltaic theoretical power prediction model based on CEEMDAN-CNN-LSTM, and the prediction model for photovoltaic dust accumulation loss is a photovoltaic dust accumulation loss prediction model based on BO-LSTM.

8. A photovoltaic cleaning strategy optimization system based on Markov chains, characterized in that, The system includes; The acquisition module is used to acquire historical meteorological data and photovoltaic operation data, preprocess the data, and then train the data to determine the prediction model for photovoltaic theoretical power and photovoltaic dust accumulation loss. The module is used to discretize the dust accumulation level in the photovoltaic dust accumulation loss prediction model, determine the corresponding dust accumulation state, and construct the dust accumulation transition probability matrix based on the MDP framework and the probability of random events. Then, the revenue function is determined by combining photovoltaic dust accumulation loss and cleaning cost. The process of discretizing the dust accumulation level in the photovoltaic dust accumulation loss prediction model to determine the corresponding dust accumulation state, and constructing a dust accumulation transition probability matrix based on the MDP framework and combining random event probabilities, includes: Set state variables, set Indicates the first The state of dust accumulation in the sky, ,in: The dust accumulation loss is 0. This represents the maximum dust accumulation loss; Set the probability of rainfall in the random event probability. Probability of sandstorm events And the probability of a normal day, ; Define the action space ,in Indicates no cleaning. 1 indicates cleaning; when At 1 o'clock, ， when hour, ， in, This represents the transition in dust state caused by sandstorms, reflecting the intense depositional effect of dust. This represents a random variable representing the normal daily accumulation of dust. The transition probability matrix is: ， in, These are normally accumulated transfer nuclei; The rolling module is used to integrate the theoretical power of photovoltaics and the forecast data of random events, solve the MDP model in a rolling manner through an iterative algorithm, and adaptively adjust the optimization time domain according to the stability of random events. With the maximization of the revenue function as the iterative objective, it generates the photovoltaic cleaning strategy for the current time domain. The verification module is used to simulate the photovoltaic cleaning strategy with other benchmark strategies, calculate multi-dimensional evaluation indicators during the simulation process, and quantitatively verify the optimization benefits of the photovoltaic cleaning strategy based on the multi-dimensional evaluation indicators.

9. An electronic device, characterized in that, Including the processor and memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code stored in the memory to perform the method as described in any one of claims 1-7.

Images