Photovoltaic panel adaptive cleaning path planning method and system based on environmental perception

Abstract

The invention provides a photovoltaic panel self-adaptive cleaning path planning method and system based on environmental perception, and relates to the technical field of photovoltaic panel cleaning, and the method comprises the steps: calculating a cleanliness value through obtaining historical weather data and state monitoring data, and triggering cleaning when the cleanliness value is lower than a dynamic cleaning threshold value; optimizing a cleaning path by adopting a multi-pheromone ant colony algorithm and an environment self-adaptive multi-neighborhood simulated annealing algorithm; and subareas are divided based on the dirt density, and multi-robot collaborative cleaning is achieved. The cleaning efficiency is improved, energy consumption is reduced, and intelligentization and precision of photovoltaic panel cleaning are achieved.

Description

Technical Field

[0001] The present invention relates to photovoltaic panel cleaning technology, and in particular to a method and system for adaptively planning a photovoltaic panel cleaning path based on environmental perception. Background Art

[0002] With the widespread adoption of photovoltaic power generation technology, the impact of dust accumulation on panel surfaces on power generation efficiency has become increasingly prominent. Traditionally, panel cleaning relies primarily on manual labor, which is not only labor-intensive and inefficient, but also poses safety risks. With the advancement of automation technology, robotic cleaning systems are increasingly being used in the field of panel cleaning. However, existing automated panel cleaning systems still have many shortcomings in cleaning path planning.

[0003] The main problems with current photovoltaic panel cleaning path planning technology include: First, existing cleaning systems lack an adaptive cleaning decision-making mechanism based on environmental perception. Most of them use fixed cycles or simple thresholds to trigger cleaning, failing to fully consider the impact of environmental factors such as weather changes and dust accumulation rates on cleaning timing, resulting in wasted cleaning resources or power generation losses due to untimely cleaning. Second, existing cleaning path planning algorithms usually only consider a single optimization objective, such as the shortest path length or the largest coverage area, and fail to comprehensively consider multi-dimensional factors such as cleaning efficiency, energy consumption, and time cost, making it difficult to achieve global optimization in complex environments. Finally, the multi-robot collaborative cleaning control strategy is imperfect, lacking an intelligent task allocation mechanism based on dirt distribution and an effective obstacle avoidance strategy, resulting in uneven resource allocation and serious interference between robots during the cleaning process, affecting the overall cleaning efficiency and quality. Summary of the Invention

[0004] The embodiments of the present invention provide a method and system for adaptively planning a photovoltaic panel cleaning path based on environmental perception, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provides an adaptive path planning method for cleaning photovoltaic panels based on environmental perception, including:

[0007] Obtaining historical weather data and status monitoring data of the photovoltaic panels in the area where the photovoltaic panels are located, and calculating a cleanliness value of the photovoltaic panels; generating a cleaning trigger instruction when the cleanliness value is lower than a cleaning threshold dynamically calculated based on power generation loss and cleaning costs;

[0008] After receiving the cleaning trigger instruction, a cleaning path optimization step is executed, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization goals, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones, iteratively calculating the global path through a multi-pheromone ant colony algorithm, and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm;

[0009] Based on the optimized cleaning path, the multi-robot collaborative cleaning control steps are executed, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

[0010] In an optional embodiment,

[0011] Obtaining historical weather data and status monitoring data for the area where the photovoltaic panels are located, and calculating the cleanliness value of the photovoltaic panels include:

[0012] The sliding time window method is used to process historical weather data. The time attenuation factor is introduced into the data in each time window for weighted calculation to obtain the historical weather impact value.

[0013] The historical weather impact value and the status monitoring data are multiplied by the corresponding weight coefficients and summed to obtain the photovoltaic panel cleanliness value, wherein the initial value of each weight coefficient is determined by the fuzzy analytic hierarchy process, and the gradient direction of the weight is dynamically adjusted based on the cleanliness value during the calculation process;

[0014] Establish a historical cleaning record database to record the cleanliness value before each cleaning, cleaning cost and actual power generation loss during the corresponding period. Combined with seasonal and environmental factors, calculate the power generation loss caused by unit cleanliness reduction;

[0015] A cleaning benefit evaluation function is established based on the unit cleaning cost and the power generation loss caused by the unit cleanliness reduction. The corresponding cleaning benefits are calculated by traversing different cleanliness values, and the cleanliness value that maximizes the cleaning benefit is determined as the cleaning threshold for the current time period. The cleaning benefit evaluation function also considers the cleaning resource scheduling cost and the cleaning operation time window constraints.

[0016] In an optional embodiment,

[0017] Dynamically adjust weights based on environmental impact and cleaning energy efficiency ratio, including:

[0018] Establishing an environmental impact function, the environmental impact function is used to characterize the influence of temperature parameters, humidity parameters, wind speed parameters, light intensity parameters, and precipitation probability parameters on the cleaning effect, wherein the temperature parameter is calculated using an exponential function to determine the degree of influence of temperature deviation from the optimal cleaning temperature, and the humidity parameter is calculated using an S-shaped response curve to determine the degree of influence of humidity exceeding a critical value; constructing a coupling coefficient between the temperature parameter and the humidity parameter based on the environmental impact function;

[0019] Divide the grid cells based on the density of dirt distribution on the photovoltaic panel surface, and set the initial weights of cleaning efficiency, energy consumption and time cost;

[0020] Calculating an environmental adjustment coefficient based on the environmental impact function and the coupling coefficient, wherein the environmental adjustment coefficient is positively correlated with the environmental sensitivity of the grid unit; calculating an energy efficiency ratio adjustment coefficient based on the cleaning energy efficiency ratio of the grid unit, wherein the energy efficiency ratio adjustment coefficient is a power function relationship between the current cleaning energy efficiency ratio and the maximum cleaning energy efficiency ratio;

[0021] The product of the environmental adjustment coefficient and the energy efficiency ratio adjustment coefficient is used as the weight adjustment amount, so that the cleaning efficiency weight is negatively correlated with the weight adjustment amount, and the energy consumption weight and the time cost weight are positively correlated with the weight adjustment amount, and the initial weight is dynamically updated.

[0022] In an optional embodiment,

[0023] A genetic algorithm based on dirt density coding is used to optimize the population. The optimized population is mapped to the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. The global path is obtained through iterative calculation of the multi-pheromone ant colony algorithm. The environment-adaptive multi-neighborhood simulated annealing algorithm is used for local optimization, including:

[0024] Initializing and constructing an initial population using a genetic algorithm, wherein the genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism;

[0025] The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and a multi-pheromone ant colony algorithm based on grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone;

[0026] The global optimal cleaning path is locally optimized using an environment-adaptive multi-neighborhood simulated annealing algorithm. The multi-neighborhood includes a reverse neighborhood, an exchange neighborhood, an insertion neighborhood, and a regional reconstruction neighborhood. Local path optimization is performed using an environment-aware temperature regulation mechanism and a dynamic acceptance criterion strategy.

[0027] In an optional embodiment,

[0028] The genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism, including:

[0029] Obtain multi-type dirt features on the photovoltaic panel surface, divide the photovoltaic panel surface into grid cells, and calculate the dirt density of each grid cell;

[0030] The chromosome of the genetic algorithm is constructed as a multi-level coding structure of space sequence coding, resource allocation coding and density association coding, in which the space sequence coding represents the cleaning path, the resource allocation coding represents the cleaning parameters, and the density association coding represents the dirt treatment strategy.

[0031] Constructing a density weight mapping function based on the dirt density, the density weight mapping function simultaneously considering the dirt density of the target grid cell and the dirt density distribution of its adjacent grid cells, and initializing the population according to the density weight mapping function;

[0032] Acquiring environmental parameters and calculating environmental impact factors, dynamically adjusting the crossover probability and mutation probability of the genetic algorithm based on the environmental impact factors and the density weight mapping function, optimizing the cleaning path sequence through the crossover operation, and adjusting the resource allocation strategy through the mutation operation;

[0033] The spatial sequence coding, resource allocation coding and density association coding of the best individual in the optimized population are mapped to the initial pheromone distribution matrix of the ant colony algorithm.

[0034] In an optional embodiment,

[0035] The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and a multi-pheromone ant colony algorithm based on grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone.

[0036] The optimization result of the genetic algorithm is mapped to an initial distribution of multidimensional pheromones, wherein the multidimensional pheromones include cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. Each pheromone component is obtained by a weighted combination of the sequence information corresponding to the optimal individual and the grid dirt density.

[0037] Calculating a grid energy efficiency ratio based on a ratio of the cleaning benefit to the energy consumption of the grid unit, and constructing a grid energy efficiency gradient vector based on the horizontal and vertical partial derivatives of the grid energy efficiency ratio;

[0038] Performing a weighted product operation on the heuristic information of each pheromone component to obtain comprehensive heuristic information, and calculating a state transition probability matrix based on the comprehensive heuristic information, the multi-dimensional pheromone initial distribution, and the grid energy efficiency gradient vector;

[0039] Determining the ant's movement path according to the state transition probability matrix, performing local pheromone updates on the grid cells on the movement path, wherein the local pheromone updates take into account the influence of the pheromone volatility coefficient and the grid energy efficiency ratio, and performing pheromone enhancement on the global optimal path based on the evaluation index of the movement path;

[0040] The variance of each pheromone component is calculated, and a multidimensional convergence judgment is performed based on the relative change rate of the variance. When the multidimensional convergence judgment result is less than a preset convergence threshold, a global optimal cleaning path is output.

[0041] In an optional embodiment,

[0042] The multi-robot collaborative cleaning control steps include:

[0043] Obtaining the dirt density distribution and area of the cleaning area, calculating the workload of the sub-areas, and dividing the cleaning area into multiple sub-areas based on the balance of the workload of the sub-areas and assigning them to multiple cleaning robots;

[0044] The cleaning trajectory in each sub-area is divided into a high-density trajectory segment and a low-density trajectory segment according to the dirt density threshold. A first sliding surface is constructed for the high-density trajectory segment using a weighted combination of velocity error and position error. A second sliding surface is constructed for the low-density trajectory segment using a weighted combination of position error integral term and position error.

[0045] An equivalent control variable is calculated based on the state quantities of the first sliding surface and the second sliding surface, the equivalent control variable being used to compensate for the deterministic dynamic characteristics of the cleaning process; an adaptive switching gain is calculated based on the dirt density, the cleaning resistance, and the state quantities of the sliding surface, the adaptive switching gain being used to suppress uncertain disturbances during the cleaning process; and a piecewise adaptive sliding mode control input is generated by combining the equivalent control variable with a sign function of the adaptive switching gain.

[0046] The proportion of the cleaned area and the distance to the target point of each cleaning robot are calculated to determine the obstacle avoidance priority. When the distance between adjacent cleaning robots is less than the safety distance threshold, the segmented adaptive sliding mode control input is corrected according to the obstacle avoidance priority, and the correction amount is proportional to the difference in the obstacle avoidance priorities of the adjacent robots and the distance gradient.

[0047] According to a second aspect of the embodiments of the present invention,

[0048] Provides an adaptive photovoltaic panel cleaning path planning system based on environmental perception, including:

[0049] The first unit is configured to obtain historical weather data of the area where the photovoltaic panels are located and status monitoring data of the photovoltaic panels, calculate a cleanliness value of the photovoltaic panels, and generate a cleaning trigger instruction when the cleanliness value is lower than a cleaning threshold dynamically calculated based on power generation loss and cleaning cost;

[0050] The second unit is configured to, upon receiving the cleaning trigger instruction, execute a cleaning path optimization step, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization objectives, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into an initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones; iteratively calculating the global path using a multi-pheromone ant colony algorithm; and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm;

[0051] The third unit is used to execute the multi-robot collaborative cleaning control steps according to the optimized cleaning path, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

[0052] According to a third aspect of the embodiments of the present invention,

[0053] An electronic device is provided, comprising:

[0054] processor;

[0055] a memory for storing processor-executable instructions;

[0056] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0057] According to a fourth aspect of the embodiments of the present invention,

[0058] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0059] The beneficial effects of this application are as follows:

[0060] By acquiring historical weather data and status monitoring data for the area where the photovoltaic panels are located, and dynamically calculating the cleanliness value and cleaning threshold, intelligent decision-making on the timing of cleaning is achieved, avoiding unnecessary cleaning operations. This reduces maintenance costs while ensuring power generation efficiency, and improves the overall economic benefits of the photovoltaic system.

[0061] A path planning method that integrates multi-objective optimization strategies and multiple algorithms is adopted. Cleaning efficiency, energy consumption and time cost are taken as optimization targets, and the weights are dynamically adjusted based on environmental impact and cleaning energy efficiency ratio. Through the synergistic effect of genetic algorithm, multi-pheromone ant colony algorithm and environment-adaptive multi-neighborhood simulated annealing algorithm, global optimization and local fine-tuning of the cleaning path are achieved, significantly improving the efficiency and quality of the cleaning operation.

[0062] A multi-robot collaborative cleaning control mechanism is introduced, sub-areas are divided based on dirt density and workload balance, and segmented adaptive sliding mode variable structure control technology is used to achieve precise trajectory tracking. An obstacle avoidance priority strategy based on the proportion of cleaned area and target distance is used to solve the conflict problem in multi-robot collaborative operation, greatly improving the cleaning efficiency and robustness of the system, and adapting to the needs of photovoltaic panel cleaning under various complex environmental conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Schematic diagram of a flow chart of a method for adaptively planning a photovoltaic panel cleaning path based on environmental perception according to an embodiment of the present invention;

[0064] Figure 2 This is a comparison chart of cleaning path optimization rate versus iteration number;

[0065] Figure 3 This is a comparison chart of the performance retention rates of different algorithms under different environmental conditions;

[0066] Figure 4 This is a simulation diagram of the cleaning path of the multi-pheromone ant colony algorithm based on grid energy efficiency gradient of the present invention;

[0067] Figure 5 This is a convergence process analysis diagram of the multi-pheromone ant colony algorithm of the present invention. DETAILED DESCRIPTION

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0069] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0070] Figure 1FIG. 1 is a flow chart of a method for adaptively planning a cleaning path for photovoltaic panels based on environmental perception according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0071] Obtaining historical weather data and status monitoring data of the photovoltaic panels in the area where the photovoltaic panels are located, and calculating a cleanliness value of the photovoltaic panels; generating a cleaning trigger instruction when the cleanliness value is lower than a cleaning threshold dynamically calculated based on power generation loss and cleaning costs;

[0072] After receiving the cleaning trigger instruction, a cleaning path optimization step is executed, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization goals, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones, iteratively calculating the global path through a multi-pheromone ant colony algorithm, and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm;

[0073] Based on the optimized cleaning path, the multi-robot collaborative cleaning control steps are executed, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

[0074] In an optional embodiment, obtaining historical weather data of the area where the photovoltaic panels are located and status monitoring data of the photovoltaic panels, and calculating the cleanliness value of the photovoltaic panels includes:

[0075] The sliding time window method is used to process historical weather data. The time attenuation factor is introduced into the data in each time window for weighted calculation to obtain the historical weather impact value.

[0076] The historical weather impact value and the status monitoring data are multiplied by the corresponding weight coefficients and summed to obtain the photovoltaic panel cleanliness value, wherein the initial value of each weight coefficient is determined by the fuzzy analytic hierarchy process, and the gradient direction of the weight is dynamically adjusted based on the cleanliness value during the calculation process;

[0077] Establish a historical cleaning record database to record the cleanliness value before each cleaning, cleaning cost and actual power generation loss during the corresponding period. Combined with seasonal and environmental factors, calculate the power generation loss caused by unit cleanliness reduction;

[0078] A cleaning benefit evaluation function is established based on the unit cleaning cost and the power generation loss caused by the unit cleanliness reduction. The corresponding cleaning benefits are calculated by traversing different cleanliness values, and the cleanliness value that maximizes the cleaning benefit is determined as the cleaning threshold for the current time period. The cleaning benefit evaluation function also considers the cleaning resource scheduling cost and the cleaning operation time window constraints.

[0079] For example, the system acquires historical weather data for the area where the photovoltaic panels are located, as well as status monitoring data for the panels. The historical weather data includes rainfall, snowfall, wind speed, wind direction, humidity, and temperature, while the status monitoring data includes panel surface temperature, power generation, current, voltage, and irradiance.

[0080] A sliding time window method is used to process historical weather data. Specifically, a time window is set, such as 14 days, and the weather data within the window is processed. A time decay factor is introduced to weight the data at different time points, so that the data closer to the current time has a greater weight. The impact value of different weather factors is calculated separately. For example, for rainfall, the daily rainfall data is first multiplied by the corresponding time decay factor, and then the sum is calculated to obtain the weighted rainfall. The weighted wind speed, weighted humidity, etc. are calculated similarly. Finally, the weighted values of each weather factor are combined to obtain the historical weather impact value.

[0081] The cleanliness value of the photovoltaic panel is calculated by combining the historical weather impact value and the status monitoring data. The weight coefficient of each indicator is set, and the initial weight is determined by the fuzzy hierarchical analysis method. For example, the weight of the historical weather impact value is 0.4, the weight of the power generation efficiency is 0.3, the weight of the surface temperature is 0.2, and the weight of the current and voltage characteristics is 0.1. The cleanliness value of the photovoltaic panel is obtained by multiplying each indicator value by the corresponding weight and summing them. The range is 0-100, and the lower the value, the more serious the pollution. During the calculation process, the system will dynamically adjust the gradient direction of the weight based on the cleanliness value. Specifically, by comparing the difference between the calculated cleanliness value and the actual cleanliness value measured before cleaning, the weight of each indicator is adjusted to make the calculation result more accurate. The adjustment range is proportional to the size of the difference, but there are upper and lower limits to ensure stability.

[0082] For example, if a calculated cleanliness value is 75, while the actual pre-cleaning value is 70, a difference of 5 units, the system will appropriately increase the weights of indicators indicating more severe pollution, such as adjusting the weight of historical weather impact from 0.4 to 0.42, while reducing the weights of other indicators to ensure that the total weight is 1.

[0083] The system establishes a historical cleaning record database, recording the cleanliness value before each cleaning, the cleaning cost, and the actual power generation loss during the corresponding period. Cleaning costs include labor costs, water costs, equipment wear and tear costs, etc. Power generation loss is calculated by comparing power generation efficiency before and after cleaning.

[0084] The power generation loss caused by a reduction in cleanliness per unit is calculated by combining seasonal and environmental factors. The seasonal factor accounts for seasonal variations in solar radiation intensity, for example, a coefficient of 1.2 in summer and 0.8 in winter. The environmental factor accounts for the varying effects of pollutants on power generation efficiency in different environments, for example, a coefficient of 1.3 in industrial areas and 0.9 in rural areas.

[0085] Through historical data analysis, we can determine the power generation loss caused by a decrease in cleanliness per unit (for example, a decrease from 100 to 99). For example, in the summer, when the cleanliness level of a photovoltaic power station drops from 95 to 85, the average daily power generation drops from 500 kWh to 470 kWh. The power generation loss caused by the decrease in cleanliness per unit is (500 - 470) / 10 = 3 kWh per unit.

[0086] A cleaning benefit evaluation function was established based on the unit cleaning cost and the power generation loss caused by a unit cleanliness reduction. The cleaning benefit is equal to the increased power generation revenue after cleaning minus the cleaning cost. The increased power generation revenue is the cleanliness improvement multiplied by the increased power generation per unit cleanliness multiplied by the electricity price.

[0087] Cleaning efficiency assessment also considers the cost of cleaning resource scheduling and the time window constraints of cleaning operations. The cost of cleaning resource scheduling is related to the distance cleaning equipment moves and the difficulty of staffing. The time window constraints take into account factors such as weather conditions and power grid scheduling requirements.

[0088] By traversing different cleanliness values (e.g., from 50 to 90, with a step size of 1), the benefits of cleaning at each cleanliness value are calculated. For example, when the cleanliness is 80, the cleaning cost is 200 yuan. After cleaning, the cleanliness is improved to 98, and the power generation increases by 18 × 3 = 54 kWh. At an electricity price of 0.8 yuan / kWh, the increased revenue is 54 × 0.8 = 43.2 yuan, and the cleaning benefit is 43.2 - 200 = -156.8 yuan, which is a negative value, indicating that cleaning is uneconomical at this time.

[0089] The system finds the cleanliness value that maximizes cleaning benefits and sets it as the cleaning threshold for the current period. When the actual cleanliness falls below this threshold, the system recommends cleaning. For example, if calculations show that a cleanliness value of 65 results in a maximum cleaning benefit of 300 yuan, 65 will be set as the cleaning threshold.

[0090] By comprehensively considering weather impact, pollution level, cleaning cost and power generation efficiency, the present invention realizes accurate assessment of photovoltaic panel cleanliness and intelligent cleaning decision-making, effectively improving the operation and maintenance efficiency and economic benefits of the photovoltaic power generation system.

[0091] In an optional embodiment, dynamically adjusting the weight based on environmental impact and cleaning energy efficiency ratio includes:

[0092] Establishing an environmental impact function, the environmental impact function is used to characterize the influence of temperature parameters, humidity parameters, wind speed parameters, light intensity parameters, and precipitation probability parameters on the cleaning effect, wherein the temperature parameter is calculated using an exponential function to determine the degree of influence of temperature deviation from the optimal cleaning temperature, and the humidity parameter is calculated using an S-shaped response curve to determine the degree of influence of humidity exceeding a critical value; constructing a coupling coefficient between the temperature parameter and the humidity parameter based on the environmental impact function;

[0093] Divide the grid cells based on the density of dirt distribution on the photovoltaic panel surface, and set the initial weights of cleaning efficiency, energy consumption and time cost;

[0094] Calculating an environmental adjustment coefficient based on the environmental impact function and the coupling coefficient, wherein the environmental adjustment coefficient is positively correlated with the environmental sensitivity of the grid unit; calculating an energy efficiency ratio adjustment coefficient based on the cleaning energy efficiency ratio of the grid unit, wherein the energy efficiency ratio adjustment coefficient is a power function relationship between the current cleaning energy efficiency ratio and the maximum cleaning energy efficiency ratio;

[0095] The product of the environmental adjustment coefficient and the energy efficiency ratio adjustment coefficient is used as the weight adjustment amount, so that the cleaning efficiency weight is negatively correlated with the weight adjustment amount, and the energy consumption weight and the time cost weight are positively correlated with the weight adjustment amount, and the initial weight is dynamically updated.

[0096] Exemplarily, an environmental impact function is established to characterize the impact of various environmental parameters on the cleaning effect. The degree of influence of the temperature parameter is calculated by an exponential function. In specific implementation, when the actual temperature is 25°C and the optimal cleaning temperature is 20°C, the temperature deviation value is 5°C. After substituting into the exponential function, the temperature influence coefficient is 0.78. The humidity parameter is calculated by an S-type response curve. For example, when the actual humidity is 75% and the critical humidity value is 60%, the humidity influence coefficient is 0.65 after substituting into the S-type response function. The wind speed parameter is processed using a piecewise function. When the wind speed is lower than 3m / s, the influence coefficient is 0.9. When the wind speed is between 3-8m / s, it decreases linearly to 0.5. When the wind speed exceeds 8m / s, the influence coefficient is 0.3. The light intensity parameter is mapped by a logarithmic function. For example, the light intensity is 800W / m 2 When the temperature is below 20%, the corresponding influence coefficient is 0.85. The precipitation probability parameter uses a threshold judgment method. When the precipitation probability is less than 20%, the influence coefficient is 0.95; when the precipitation probability is between 20% and 50%, the influence coefficient is 0.6; and when the precipitation probability exceeds 50%, the influence coefficient is 0.2. The coupling coefficient between the temperature and humidity parameters is calculated by multiplying the temperature influence coefficient and the humidity influence coefficient by a weighted product of 0.6 and 0.4, respectively. For example, when the temperature influence coefficient is 0.78 and the humidity influence coefficient is 0.65, the calculated coupling coefficient is 0.726.

[0097] Grid cells are divided based on the density of dirt distribution on the photovoltaic panel surface. A standard photovoltaic panel is divided into a 6×8 grid, for a total of 48 grid cells. Initial weights for cleaning efficiency, energy consumption, and time cost are set for each grid cell, with initial values of 0.5, 0.3, and 0.2, respectively.

[0098] The environmental adjustment coefficient is calculated based on the environmental impact function and the coupling coefficient. The environmental adjustment coefficient is positively correlated with the environmental sensitivity of the grid unit, which is determined by the type and degree of dirt. When the environmental sensitivity of a grid unit is 0.8, the calculated result of the environmental impact function is 0.7, and the coupling coefficient is 0.726, the environmental adjustment coefficient is calculated as 0.8×0.7×0.726=0.406. The energy efficiency ratio adjustment coefficient is calculated based on the cleaning energy efficiency ratio of the grid unit. The cleaning energy efficiency ratio is defined as the ratio of cleaning effect to energy consumption and is obtained through real-time monitoring data. For example, the current cleaning energy efficiency ratio of a grid unit is 2.5, the maximum cleaning energy efficiency ratio in the system history is 4.0, and the power function exponent is set to 0.5, then the energy efficiency ratio adjustment coefficient is (2.5 / 4.0) 0.5 =0.791.

[0099] The product of the environmental adjustment coefficient and the energy efficiency ratio adjustment coefficient is used as the weight adjustment. For example, when the environmental adjustment coefficient is 0.406 and the energy efficiency ratio adjustment coefficient is 0.791, the weight adjustment is 0.406 × 0.791 = 0.321. The initial weights are dynamically updated to ensure that the cleaning efficiency weight is negatively correlated with the weight adjustment, and that the energy consumption weight and time cost weight are positively correlated with the weight adjustment. In specific implementation, a linear adjustment formula is used: Cleaning Efficiency Weight = Initial Cleaning Efficiency Weight - Weight Adjustment × Adjustment Factor; Energy Consumption Weight = Initial Energy Consumption Weight + Weight Adjustment × Energy Adjustment Factor; Time Cost Weight = Initial Time Cost Weight + Weight Adjustment × Time Adjustment Factor. The adjustment factor, energy adjustment factor, and time adjustment factor are set to 0.5, 0.3, and 0.2, respectively.

[0100] Using the above example for calculation: adjusted cleaning efficiency weight = 0.5 - 0.321 × 0.5 = 0.3395; adjusted energy consumption weight = 0.3 + 0.321 × 0.3 = 0.3963; adjusted time cost weight = 0.2 + 0.321 × 0.2 = 0.2642.

[0101] In actual application, the system collects environmental data and updates the weights at regular intervals. For example, during a complete cleaning cycle, the system records the following data: initial temperature of 22°C, humidity of 65%, wind speed of 2m / s, and light intensity of 750W / m 2, the probability of precipitation is 15%; in the medium term, the temperature rises to 28℃, the humidity drops to 55%, the wind speed increases to 4m / s, and the light intensity increases to 850W / m 2 , the probability of precipitation rises to 25%; the temperature at the end is 30℃, the humidity is 50%, the wind speed is 5m / s, and the light intensity is 900W / m 2 , the probability of precipitation is 30%.

[0102] Based on these environmental changes, the system calculates environmental adjustment coefficients and energy efficiency ratio adjustment coefficients for each of the 48 grid cells and then updates the weights. For example, for grid cell A containing water-soluble dirt, the initial calculated environmental adjustment coefficient is 0.406, the energy efficiency ratio adjustment coefficient is 0.791, and the weight adjustment is 0.321, resulting in an adjusted weight of [0.3395, 0.3963, 0.2642]. At the mid-term, the environmental adjustment coefficient changes to 0.352, the energy efficiency ratio adjustment coefficient changes to 0.745, and the weight adjustment is 0.262, resulting in an adjusted weight of [0.369, 0.379, 0.252]. At the end, the environmental adjustment coefficient further decreases to 0.301, the energy efficiency ratio adjustment coefficient is 0.712, and the weight adjustment is 0.214, resulting in an adjusted weight of [0.393, 0.364, 0.243].

[0103] Existing photovoltaic panel cleaning systems fail to fully consider the interactive effects of environmental parameters, lack adaptability to the dynamic characteristics of the cleaning process, and fail to establish a systematic correlation between environmental impacts and cleaning energy efficiency. This invention innovatively proposes a multi-environmental parameter coupling mechanism, employing different function models to characterize the impact of each environmental parameter. A temperature-humidity coupling coefficient is constructed to reflect these interactions, enabling a comprehensive quantitative assessment of environmental impacts. A grid-based cleaning energy efficiency assessment method is designed, incorporating the concept of grid cell environmental sensitivity and establishing a power function mapping of the cleaning energy efficiency ratio to achieve refined local cleaning effect evaluation. A two-factor weight dynamic adjustment strategy is proposed, combining the environmental adjustment coefficient with the energy efficiency ratio adjustment coefficient, using a differentiated weight update mechanism to achieve adaptive optimization of the cleaning process. This invention improves cleaning uniformity by 15%-25% under different environmental conditions; reduces energy consumption by 20%-30% for the same cleaning effect; reduces total cleaning time by 25%-35%; and shortens response time to environmental changes by 40%-50%.

[0104] In an optional embodiment, a genetic algorithm based on dirt density coding is used to optimize the population, and the optimized population is mapped to the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. A global path is obtained by iterative calculation using a multi-pheromone ant colony algorithm, and local optimization is performed using an environment-adaptive multi-neighborhood simulated annealing algorithm, including:

[0105] Initializing and constructing an initial population using a genetic algorithm, wherein the genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism;

[0106] The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and a multi-pheromone ant colony algorithm based on grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone;

[0107] The global optimal cleaning path is locally optimized using an environment-adaptive multi-neighborhood simulated annealing algorithm. The multi-neighborhood includes a reverse neighborhood, an exchange neighborhood, an insertion neighborhood, and a regional reconstruction neighborhood. Local path optimization is performed using an environment-aware temperature regulation mechanism and a dynamic acceptance criterion strategy.

[0108] For example, a genetic algorithm is used for initialization and to construct the initial population. In a practical application scenario, assume the cleaning area is a 100m x 100m rectangular area, divided into 10 x 10 grids, each 10m x 10m in size. Each grid has a different dirt density value ranging from 0 to 10, where 0 represents no dirt and 10 represents the highest degree of dirt.

[0109] The genetic algorithm uses a weighted encoding strategy based on dirt density. Each chromosome represents a cleaning path, with a length equal to the total number of grids, i.e., 100. Each gene in the chromosome represents the order in which a grid is visited. The initial population size is set to 50, and 50 different paths are randomly generated as the initial population.

[0110] The environment-aware adaptive crossover-mutation mechanism dynamically adjusts the crossover and mutation probabilities based on dirt density. When the cleaning robot is in a high-density area, the crossover probability is set to 0.85 and the mutation probability is set to 0.15; when in a medium-density area, the crossover probability is set to 0.75 and the mutation probability is set to 0.2; and when in a low-density area, the crossover probability is set to 0.65 and the mutation probability is set to 0.25.

[0111] The roulette wheel selection method was used in the selection process, and the fitness function comprehensively considered path length, dirt removal efficiency, and energy consumption. When calculating the fitness value, the path length was weighted as 0.3, the dirt removal efficiency was weighted as 0.4, and the energy consumption was weighted as 0.3.

[0112] The crossover operation uses a sequential crossover method to randomly select a segment from the parent chromosome for exchange, ensuring that the daughter chromosome does not contain repeated grids. The mutation operation uses a reversal mutation method to randomly select two positions on the chromosome and reverse the gene sequence between them.

[0113] After 100 iterations of the genetic algorithm, the 10 individuals with the highest fitness are selected as the optimization results to initialize the pheromone distribution of the ant colony algorithm.

[0114] The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and the multi-pheromone ant colony algorithm based on grid energy efficiency gradient is used to generate the global optimal cleaning path.

[0115] The multi-pheromone system includes cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. The cleaning efficiency pheromone is proportional to the grid's dirt density. A grid with a dirt density of 10 has an initial pheromone value of 1.0, a grid with a dirt density of 5 has an initial pheromone value of 0.5, and so on. The energy consumption pheromone is related to the grid's terrain complexity, with an initial value of 0.2 for flat areas, 0.5 for areas with slight obstacles, and 0.8 for areas with more obstacles. The time cost pheromone is inversely proportional to the distance between grids. The initial value is 0.9 for adjacent grids, 0.7 for grids with a distance of 2, 0.5 for grids with a distance of 3, and so on.

[0116] When choosing the next grid to visit, ants consider the influence of three pheromones: the cleaning efficiency pheromone has a weight of 0.5, the energy consumption pheromone has a weight of 0.3, and the time cost pheromone has a weight of 0.2. The total number of ants is set to 30, and the number of iterations is 200.

[0117] In each iteration, the pheromone volatilization rate is set to 0.1, which means that 90% of the original pheromone is retained. After the ant completes a complete path, it adds pheromone to the grids it passes through, proportional to the path quality. Path quality evaluation metrics include total cleaning efficiency, total energy consumption, and total time cost, with weights of 0.5, 0.3, and 0.2, respectively.

[0118] To prevent the algorithm from falling into a local optimum, the pheromone upper limit is set to 10.0 and the lower limit is set to 0.1. When the pheromone of a path exceeds the upper limit, it is limited to the upper limit; when it is lower than the lower limit, it is set to the lower limit.

[0119] After 200 iterations, the path with the strongest pheromone was selected as the global optimal cleaning path. In actual tests, this path was able to cover all dirty areas, with a total path length of 1250m, a cleaning efficiency of 95%, and energy consumption of 75% of the standard value.

[0120] Finally, an environment-adaptive multi-neighborhood simulated annealing algorithm is used to locally optimize the global optimal cleaning path. The multi-neighborhoods include reversal, exchange, insertion, and region reconstruction. The reversal neighborhood randomly selects two points in the path and reverses the order of grid access between them; the exchange neighborhood randomly selects two grids and swaps their access order; the insertion neighborhood randomly selects a grid and inserts it into another position in the path; and the region reconstruction neighborhood randomly selects a subregion and reorders the access order within that region.

[0121] The environment-aware temperature regulation mechanism dynamically adjusts the annealing temperature based on the current environmental characteristics. The initial temperature is set to 100. In high-contamination areas, the temperature decreases by a factor of 0.95; in medium-contamination areas, the temperature decreases by a factor of 0.9; and in low-contamination areas, the temperature decreases by a factor of 0.85.

[0122] The dynamic acceptance criterion strategy dynamically adjusts the acceptance probability based on the current iteration count and the quality of the solution. In the early stages of an iteration, even if a new solution is slightly worse than the current one, it has a high probability of being accepted. As the iteration progresses, the acceptance probability gradually decreases, and the algorithm becomes more inclined to accept the better solution.

[0123] The simulated annealing algorithm was iterated 300 times with a termination temperature of 0.01. In each iteration, a neighborhood operation was randomly selected to generate a new solution, and the acceptance criterion was used to decide whether to accept the new solution.

[0124] After local optimization, the cleaning path was further optimized, the total path length was reduced to 1180m, the cleaning efficiency was increased to 97%, and the energy consumption was reduced to 70% of the standard value, which was significantly better than the global path optimization.

[0125] Through a three-stage fusion optimization approach combining a genetic algorithm, a multi-pheromone ant colony algorithm, and an environmentally adaptive multi-neighborhood simulated annealing algorithm, this invention achieves efficient and energy-efficient intelligent cleaning path planning. The genetic algorithm's dirt density encoding strategy and environmentally sensitive adaptive mechanism provide a high-quality initial solution for path optimization. The multi-pheromone ant colony algorithm effectively balances cleaning quality and resource consumption through the synergistic effects of three pheromones: cleaning efficiency, energy consumption, and time cost. The environmentally adaptive multi-neighborhood simulated annealing algorithm further achieves localized, refined optimization of the path.

[0126] In an optional embodiment, the genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism including:

[0127] Obtain multi-type dirt features on the photovoltaic panel surface, divide the photovoltaic panel surface into grid cells, and calculate the dirt density of each grid cell;

[0128] The chromosome of the genetic algorithm is constructed as a multi-level coding structure of space sequence coding, resource allocation coding and density association coding, in which the space sequence coding represents the cleaning path, the resource allocation coding represents the cleaning parameters, and the density association coding represents the dirt treatment strategy.

[0129] Constructing a density weight mapping function based on the dirt density, the density weight mapping function simultaneously considering the dirt density of the target grid cell and the dirt density distribution of its adjacent grid cells, and initializing the population according to the density weight mapping function;

[0130] Acquiring environmental parameters and calculating environmental impact factors, dynamically adjusting the crossover probability and mutation probability of the genetic algorithm based on the environmental impact factors and the density weight mapping function, optimizing the cleaning path sequence through the crossover operation, and adjusting the resource allocation strategy through the mutation operation;

[0131] The spatial sequence coding, resource allocation coding and density association coding of the best individual in the optimized population are mapped to the initial pheromone distribution matrix of the ant colony algorithm.

[0132] For example, multiple types of contamination features are acquired from the photovoltaic panel surface. Image data of the photovoltaic panel surface is acquired using an image acquisition device, and image processing techniques are used to identify different types of contamination, such as dust, bird droppings, and leaves. The acquired image data is preprocessed, including denoising and contrast enhancement, to improve the accuracy of contamination identification. For example, for a 1.5 m x 1 m photovoltaic panel, a 2000 x 1500 pixel high-definition image can be acquired. Different types of contamination areas can be identified using HSV color space conversion and threshold segmentation. The photovoltaic panel surface is divided into grid cells, and the contamination density of each grid cell is calculated. The photovoltaic panel surface is evenly divided into m x n grid cells, for example, a 10 x 8 grid. For each grid cell, the ratio of the number of contamination pixels within the grid cell to the total number of pixels in the grid cell is calculated as the contamination density of the grid cell. For example, for a grid cell numbered (3, 4), if it contains 5000 pixels, of which 2000 are contamination pixels, the contamination density of the grid cell is 0.4.

[0133] The genetic algorithm's multi-level chromosome encoding structure consists of three parts: spatial sequence encoding, resource allocation encoding, and density-dependent encoding. The spatial sequence encoding represents the cleaning path, using integer encoding. Its length is the total number of grid cells, and each gene value represents the number of the grid cell visited. For example, for a 10×8 grid, the spatial sequence encoding length is 80, and the encoding "1-25-36-42..." indicates that the cleaning order is grid cells 1, 25, 36, 42, and so on. The resource allocation encoding represents cleaning parameters, including cleaning pressure and detergent dosage, using real numbers. For example, the encoding "0.5-0.3-0.8..." indicates that the corresponding grid cells have a cleaning pressure of 0.5 MPa, a detergent dosage of 0.3 ml, and a cleaning time of 0.8 seconds. The density-dependent encoding represents the dirt treatment strategy, using binary encoding to indicate whether special treatment is required. For example, the encoding "1-0-1..." indicates that the first and third grid cells require special treatment, while the second receives standard treatment.

[0134] A density weight mapping function is constructed based on the dirt density. This function not only considers the dirt density of the target grid cell, but also the dirt density distribution of its adjacent grid cells. For any grid cell, its density weight value is calculated as follows: first calculate the dirt density value of the grid cell, then calculate the average dirt density value of its eight adjacent grid cells, and finally perform a weighted summation of the two density values to obtain the final density weight value. The weight coefficient can be adjusted according to actual conditions. For example, the weight of the dirt density of the target grid cell is 0.7, and the weight of the average dirt density of the adjacent cells is 0.3. For edge grid cells, only existing adjacent cells are considered. For example, for grid cell (5, 6) with a dirt density of 0.6, the average dirt density of its adjacent cells is 0.4, then its density weight value is 0.6×0.7+0.4×0.3=0.54.

[0135] The population is initialized using a density-weighted mapping function. The population size is set to 100. The spatial sequence encoding component of each individual prioritizes grid cells with high density weights to form the initial cleaning path. The resource allocation encoding component sets initial values based on the dirt density of the corresponding grid cell. The higher the dirt density, the more cleaning resources are allocated. The density-related encoding component is set based on the dirt type. For difficult-to-clean dirt types, it is set to 1, indicating that special treatment is required.

[0136] Obtain environmental parameters and calculate environmental impact factors. Environmental parameters include ambient temperature, humidity, light intensity, etc. For example, when the ambient temperature is 30°C, the humidity is 60%, and the light intensity is 800W / m 2The environmental impact factor can be set to 0.85. Higher ambient temperatures make dirt easier to dry, making cleaning more difficult. Higher humidity softens dirt, making cleaning easier. Higher light intensity shortens the cleaning window.

[0137] The genetic algorithm's crossover and mutation probabilities are dynamically adjusted based on environmental factors and a density weight mapping function. The initial crossover probability is set to 0.8, and the mutation probability is set to 0.1. The crossover probability is adjusted based on the environmental factors. A larger environmental factor increases the crossover probability, enhancing the algorithm's global search capabilities. The mutation probability is also adjusted based on the density weight mapping function. Regions with higher density weights have lower mutation probabilities, preserving superior genes. For example, when the environmental factor is 0.85, the adjusted crossover probability is 0.8 × 0.85 = 0.68. When the density weight of a grid cell is 0.54, the mutation probability at that location is 0.1 × (1 - 0.54) = 0.046.

[0138] Optimize the cleaning path sequence through crossover. A partial matching crossover (PMX) method is used to crossover the spatial sequence code to ensure that the crossover code remains a valid path. For example, for the spatial sequence code of parent individual A, "1-3-5-7-9," and the spatial sequence code of parent individual B, "2-4-6-8-10," the crossover points are selected as the second and fourth positions. After crossover, the spatial sequence codes of offspring individual C, "1-4-6-7-9," and offspring individual D, "2-3-5-8-10," are obtained.

[0139] The resource allocation strategy is adjusted through mutation. A Gaussian mutation is performed on the resource allocation code, with the magnitude of the mutation proportional to the dirt density of the corresponding grid cell. For example, if the resource allocation code "0.5-0.3-0.8" has a dirt density of 0.6, the resulting code might be "0.55-0.3-0.8," increasing the cleaning pressure to cope with the higher dirt density.

[0140] The spatial sequence encoding, resource allocation encoding, and density-related encoding of the optimal individual in the optimized population are mapped into the initial pheromone distribution matrix of the ant colony algorithm. For pairs of adjacent grid cells in the spatial sequence encoding, initial pheromone values are set at corresponding positions in the pheromone matrix, with the values proportional to the density weights of the two grid cells. The resource allocation encoding and density-related encoding are converted into decision-making probability adjustment factors for the ants in the ant colony algorithm. For example, for the spatial sequence encoding of the optimal individual "1-25-36-42...", higher initial pheromone values are set at positions (1, 25), (25, 36), and (36, 42) in the pheromone matrix to guide the ants to explore along the optimized path.

[0141] Figure 2 This is a comparison chart of the cleaning path optimization rate versus the number of iterations, as shown in Figure 2 As shown in the figure, the algorithm of the present invention (black solid line) achieved a path optimization rate of more than 20% after only 40 iterations, while the traditional single encoding algorithm (dashed line in the figure) required more than 80 iterations to reach the same level. As the number of iterations increases, the performance gap between the two algorithms becomes more obvious. After 100 iterations, the path optimization rate of the algorithm of the present invention is higher than that of the other two algorithms. What is more noteworthy is that the algorithm of the present invention shows a faster convergence speed and a larger slope of the curve at the beginning of the iteration. This fully proves that the encoding strategy based on dirt density weight can effectively improve the algorithm's search efficiency and the quality of the solution, providing a better solution for photovoltaic panel cleaning path planning.

[0142] like Figure 3 The performance retention rate comparison of different algorithms under different environmental conditions is shown in the figure. The horizontal axis represents the amplitude of environmental condition changes (from 10% to 90%), and the vertical axis represents the algorithm performance retention rate (from 50% to 100%). The environmental adaptive algorithm of the present invention (black solid line and solid dots) shows a significant robustness advantage when the environmental conditions change drastically. When the amplitude of environmental change reaches 90%, the algorithm of the present invention can still maintain about 78% of the performance, while the performance of the traditional fixed parameter algorithm (dashed line and hollow squares) drops to about 58%. The adaptive mechanism of the present invention can effectively respond to environmental changes and provide reliable guarantee for the stable operation of the photovoltaic panel cleaning system in complex and changing environments.

[0143] The present invention combines a multi-level coding structure with an environmental perception adaptive mechanism to achieve intelligent optimization of photovoltaic panel cleaning paths and resource allocation, providing an effective solution for improving photovoltaic system efficiency and reducing maintenance costs.

[0144] In an optional embodiment, the optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and a multi-pheromone ant colony algorithm based on the grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone, including:

[0145] The optimization result of the genetic algorithm is mapped to an initial distribution of multidimensional pheromones, wherein the multidimensional pheromones include cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. Each pheromone component is obtained by a weighted combination of the sequence information corresponding to the optimal individual and the grid dirt density.

[0146] Calculating a grid energy efficiency ratio based on a ratio of the cleaning benefit to the energy consumption of the grid unit, and constructing a grid energy efficiency gradient vector based on the horizontal and vertical partial derivatives of the grid energy efficiency ratio;

[0147] Performing a weighted product operation on the heuristic information of each pheromone component to obtain comprehensive heuristic information, and calculating a state transition probability matrix based on the comprehensive heuristic information, the multi-dimensional pheromone initial distribution, and the grid energy efficiency gradient vector;

[0148] Determining the ant's movement path according to the state transition probability matrix, performing local pheromone updates on the grid cells on the movement path, wherein the local pheromone updates take into account the influence of the pheromone volatility coefficient and the grid energy efficiency ratio, and performing pheromone enhancement on the global optimal path based on the evaluation index of the movement path;

[0149] The variance of each pheromone component is calculated, and a multidimensional convergence judgment is performed based on the relative change rate of the variance. When the multidimensional convergence judgment result is less than a preset convergence threshold, a global optimal cleaning path is output.

[0150] Cleaning path optimization method combining genetic algorithm with multi-pheromone ant colony algorithm

[0151] For example, the optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and the multi-pheromone ant colony algorithm based on the grid energy efficiency gradient is used to generate the global optimal cleaning path. The specific implementation process is as follows:

[0152] During the genetic algorithm optimization phase, the cleaning area is divided into a grid. For example, a 10-meter by 10-meter cleaning area is divided into 100 1-meter by 1-meter grid cells. Each grid cell has a dirt density value, a decimal between 0 and 1, with higher values indicating greater dirt levels. The genetic algorithm optimizes the initial path, resulting in an optimized path sequence, such as [25, 26, 36, 46, 56, 55, 54, 53, 43, 33, 23, 13], which represents the grid cell numbers that the cleaning robot passes through.

[0153] When mapping the optimization results of the genetic algorithm into an initial multidimensional pheromone distribution, three types of pheromones are considered: cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. For adjacent grid cells at the start and end of a path, the initial pheromone value is calculated by weightedly combining the sequence information corresponding to the optimal individual with the grid dirt density. For example, if the start and end grid cells are adjacent in the optimal path and have dirt densities of 0.8 and 0.7, respectively, the initial value of the cleaning efficiency pheromone can be set to 0.5 × (0.8 + 0.7) = 0.75. Similarly, the initial pheromone values for energy consumption and time cost are calculated.

[0154] The grid energy efficiency ratio (EER) is calculated based on the ratio of a grid cell's cleaning revenue to its energy consumption. The cleaning revenue is defined as the grid cell's dirt density multiplied by the cleaning area. Energy consumption includes both movement and cleaning energy. For example, for a grid cell with a dirt density of 0.8, a cleaning area of 1 square meter, 0.2 kWh of cleaning energy, and 0.05 kWh of movement energy, the grid EER is (0.8 × 1) / (0.2 + 0.05) = 3.2, representing the cleaning revenue per unit of energy consumption.

[0155] The grid energy efficiency gradient vector is constructed based on the horizontal and vertical partial derivatives of the grid energy efficiency ratio. Assuming that the energy efficiency ratios of two adjacent points in the horizontal direction of adjacent grid cells are 3.2 and 3.5, respectively, the horizontal partial derivative is 3.5-3.2=0.3; and if the energy efficiency ratios of two adjacent points in the vertical direction of adjacent grid cells are 3.2 and 2.9, respectively, the vertical partial derivative is 2.9-3.2=-0.3. The grid energy efficiency gradient vector thus constructed is (0.3, -0.3), indicating the direction and magnitude of energy efficiency growth.

[0156] The combined heuristic information is obtained by performing a weighted product operation on the heuristic information of each pheromone component. The cleaning efficiency heuristic is proportional to the grid dirt density; the energy consumption heuristic is inversely proportional to the travel distance and cleaning energy consumption; and the time cost heuristic is inversely proportional to the travel time and cleaning time. If the weight coefficients are set to 0.4, 0.3, and 0.3, respectively, the combined heuristic information is equal to the weighted product of the individual heuristics.

[0157] The state transition probability matrix is calculated based on comprehensive heuristic information, the initial multidimensional pheromone distribution, and the grid energy efficiency gradient vector. The probability of an ant moving from its current grid cell to a candidate grid cell is determined by taking into account the weighted product of the three pheromones, the comprehensive heuristic information, and the grid energy efficiency gradient vector. For example, if the pheromone weights are set to 0.35, 0.35, and 0.3, respectively, and the gradient influence factor is 0.2, the state transition probability is proportional to the product of the pheromone weights, the heuristic information, and the gradient angle factor, where the angle factor is the cosine of the angle between the ant's movement direction and the energy efficiency gradient vector.

[0158] The ant's movement path is determined based on the state transition probability matrix, and local pheromone updates are performed on the grid cells along the movement path. Setting the pheromone volatility coefficient to 0.1 and the local update intensity factor to 0.05, the local pheromone update formula is: the original pheromone value multiplied by (1-volatility coefficient) plus the local update intensity factor multiplied by the grid energy efficiency ratio. For example, if the original pheromone value is 0.75 and the grid energy efficiency ratio is 3.2, the updated pheromone value is (1-0.1)×0.75+0.05×3.2=0.835.

[0159] For the global optimal path, pheromone reinforcement is performed. If the global update intensity factor is set to 0.2, the global pheromone update formula is: the original pheromone value multiplied by (1-volatility coefficient) plus the global update intensity factor multiplied by the constant divided by the total path length. For example, if the global optimal path length is 25 meters and the constant is 100, the pheromone increment is 0.2 × 100 / 25 = 0.8.

[0160] The variance of each pheromone component is calculated, and multidimensional convergence is determined based on the relative rate of change of the variance. For example, after two consecutive iterations, the variance of the cleaning efficiency pheromone decreases from 0.05 to 0.03, with a relative rate of change of (0.05-0.03) / 0.05 = 0.4; the variance of the energy consumption pheromone decreases from 0.06 to 0.04, with a relative rate of change of 0.33; and the variance of the time cost pheromone decreases from 0.04 to 0.03, with a relative rate of change of 0.25. The average of these three values, 0.33, is taken as the multidimensional convergence determination result. If the preset convergence threshold is 0.1, the algorithm continues to iterate. When the multidimensional convergence determination result is less than 0.1, the global optimal cleaning path is output.

[0161] like Figure 4 The cleaning path simulation diagram of the multi-pheromone ant colony algorithm based on the grid energy efficiency gradient of the present invention is shown in the figure. Different gray levels are used in the figure to represent the dirt density distribution of each grid unit in the 10×10 meter area. The dark area represents the location with high dirt density and urgent cleaning needs, and the light area represents the location with less dirt. The arrow represents the grid energy efficiency gradient vector, indicating the direction of energy efficiency improvement. The black solid line represents the optimal cleaning path generated by the algorithm of the present invention, which adopts an optimized serpentine pattern to ensure full coverage while minimizing the path length. It can be seen from the simulation results that the algorithm can reasonably plan the path according to the characteristics of the dirt distribution, especially in the heavily dirty area of the central area (3, 3) to (5, 5), the path planning is more compact, reflecting the algorithm's sensitivity to dirt density and the reasonable allocation of cleaning resources. Compared with the traditional algorithm, this path reduces the moving distance by about 9.1%, while increasing the dirt removal rate by 3.5%, effectively solving the multi-objective optimization problem in photovoltaic panel cleaning.

[0162] like Figure 5As shown in the convergence process analysis diagram of the multi-pheromone ant colony algorithm of the present invention, the horizontal axis represents the number of algorithm iterations, the vertical axis represents the variance value of each pheromone, and the three curves of different line types represent the variance change trends of the cleaning efficiency pheromone (solid line), energy consumption pheromone (dashed line) and time cost pheromone (dotted line). The variance values of the three pheromones all show a downward trend as the iteration proceeds, indicating that the ant colony's choice of the optimal path gradually converges to a consensus; the energy consumption pheromone converges the fastest, and its initial variance is the highest (0.25) but the fastest decline rate, indicating that the energy factor plays a leading role in the optimization process; at the 42nd iteration (the multidimensional convergence point marked in the figure), the comprehensive convergence judgment result of the three pheromones is lower than the preset threshold line (0.1), and the algorithm reaches the convergence condition and stops iterating. This multidimensional convergence mechanism effectively avoids the local optimal problem caused by the premature convergence of a single pheromone.

[0163] The present invention combines genetic algorithm and multi-pheromone ant colony algorithm, uses grid energy efficiency gradient to guide path optimization, reduces moving distance and improves dirt removal rate, and effectively solves the multi-objective optimization problem of cleaning tasks.

[0164] In an optional embodiment, the multi-robot collaborative cleaning control step includes:

[0165] Obtaining the dirt density distribution and area of the cleaning area, calculating the workload of the sub-areas, and dividing the cleaning area into multiple sub-areas based on the balance of the workload of the sub-areas and assigning them to multiple cleaning robots;

[0166] The cleaning trajectory in each sub-area is divided into a high-density trajectory segment and a low-density trajectory segment according to the dirt density threshold. A first sliding surface is constructed for the high-density trajectory segment using a weighted combination of velocity error and position error. A second sliding surface is constructed for the low-density trajectory segment using a weighted combination of position error integral term and position error.

[0167] An equivalent control variable is calculated based on the state quantities of the first sliding surface and the second sliding surface, the equivalent control variable being used to compensate for the deterministic dynamic characteristics of the cleaning process; an adaptive switching gain is calculated based on the dirt density, the cleaning resistance, and the state quantities of the sliding surface, the adaptive switching gain being used to suppress uncertain disturbances during the cleaning process; and a piecewise adaptive sliding mode control input is generated by combining the equivalent control variable with a sign function of the adaptive switching gain.

[0168] The proportion of the cleaned area and the distance to the target point of each cleaning robot are calculated to determine the obstacle avoidance priority. When the distance between adjacent cleaning robots is less than the safety distance threshold, the segmented adaptive sliding mode control input is corrected according to the obstacle avoidance priority, and the correction amount is proportional to the difference in the obstacle avoidance priorities of the adjacent robots and the distance gradient.

[0169] For example, the dirt density distribution and area of the cleaning area are obtained. The dirt density distribution can be obtained by capturing images with the robot's onboard camera and calculating them using an image processing algorithm. For example, the cleaning area is divided into a 10×10 grid, with the dirt density value of each grid ranging from 0 to 10, where 0 represents completely clean and 10 represents extremely dirty. The area of the area is obtained by measurement, such as an office floor area of 100 square meters. Based on the obtained dirt density distribution, the workload of each sub-area is calculated. The workload calculation formula is the product of the sub-area area and the average dirt density of the area. For example, if the average dirt density of a 5-square-meter sub-area is 7, the workload of this area is 35 work units. To achieve workload balance, an iterative bisection method is used to divide the cleaning area into multiple sub-areas. The specific steps are as follows: first, the area is divided into two parts along the largest dimension, and the workload difference between the two parts is calculated. If the difference is greater than a preset threshold (e.g., 5% of the total workload), the position of the dividing line is adjusted. This process is repeated until the workload difference is less than the threshold or a preset number of iterations (e.g., 10) is reached. For the case of three cleaning robots, the area is finally divided into three sub-areas with workloads of 105, 108, and 102 work units, respectively, achieving a balanced distribution of workload.

[0170] After the sub-area division is complete, a cleaning trajectory is planned for each cleaning robot. A modified zigzag cleaning path is used to ensure 100% coverage. The cleaning trajectory is divided into high-density and low-density segments based on a preset dirt density threshold (e.g., a density of 6).

[0171] For high-density trajectory segments, a first sliding surface is constructed. This sliding surface comprehensively considers both velocity error and position error, with a velocity error weight of 0.7 and a position error weight of 0.3. This configuration enables the robot to maintain a low and stable speed in high-contamination areas, improving cleaning performance. For example, when the robot is cleaning an area with a dirt density of 8, the target speed is set to 0.1 m / s, the actual speed is 0.15 m / s, and the position error is 0.05 m. The calculated sliding surface state quantity is 0.035, indicating that control input needs to be reduced. For low-density trajectory segments, a second sliding surface is constructed. This sliding surface comprehensively considers both the position error integral term and the position error, with a position error integral term weight of 0.4 and a position error weight of 0.6. This configuration enables the robot to quickly pass through low-contamination areas while maintaining trajectory accuracy. For example, when the robot is cleaning an area with a dirt density of 3, the cumulative position error is 0.1 m, and the current position error is 0.03 m. The calculated sliding surface state quantity is 0.058, indicating that appropriate control input adjustment is required. Based on the sliding surface state, equivalent control variables are calculated. These are primarily used to compensate for the deterministic aspects of the robot's dynamics model, including parameters such as mass, inertia, and friction. For example, for a cleaning robot with a mass of 15 kg and a wheelbase of 0.4 m, when the sliding surface state is 0.035, the calculated equivalent control variables are 0.42 N·m for the left wheel motor torque and 0.38 N·m for the right wheel motor torque.

[0172] The adaptive switching gain is calculated based on the dirt density, cleaning resistance, and the sliding surface state. When the dirt density is high, the switching gain is increased to enhance disturbance rejection; as the cleaning resistance increases, the switching gain is also increased accordingly. For example, when the dirt density is 8, the cleaning resistance is 5 Newtons, and the sliding surface state is 0.035, the adaptive switching gain is calculated to be 0.25.

[0173] The equivalent control variable is combined with the sign function of the adaptive switching gain to generate the piecewise adaptive sliding mode control input. When the absolute value of the sliding surface state variable is less than 0.01, a continuous approximation is used to replace the sign function to avoid control input jitter. When the absolute value of the state variable is greater than 0.01, the standard sign function is used.

[0174] To avoid robot collisions, the percentage of cleaned area and distance to the target point of each cleaning robot are calculated in real time to determine the obstacle avoidance priority. Robots with a higher percentage of cleaned area and closer distance to the target point have a higher obstacle avoidance priority. For example, if robot A has cleaned 75% of the area and is 0.8 meters away from the target point, while robot B has cleaned 60% of the area and is 1.5 meters away from the target point, robot A's obstacle avoidance priority is higher than robot B. When the distance between adjacent cleaning robots falls below the safety distance threshold (e.g., 0.5 meters), the segmented adaptive sliding mode control input is corrected based on the obstacle avoidance priority. The correction amount is calculated as: Correction amount = basic correction coefficient × priority difference × distance gradient.

[0175] This invention achieves efficient coordination of a multi-robot cleaning system through workload-balanced sub-area division and a differentiated segmented sliding mode control strategy. Different control strategies are adopted for areas with different dirt densities, prioritizing cleaning quality in high-dirt areas and improving cleaning efficiency in low-dirt areas. Furthermore, an adaptive switching gain effectively suppresses external disturbances and parameter uncertainties during the cleaning process. An obstacle avoidance priority mechanism, based on the percentage of cleaned area and target distance, effectively prevents robot collisions, improving system safety and reliability.

[0176] According to a second aspect of the embodiments of the present invention,

[0177] Provides an adaptive photovoltaic panel cleaning path planning system based on environmental perception, including:

[0178] The first unit is configured to obtain historical weather data of the area where the photovoltaic panels are located and status monitoring data of the photovoltaic panels, calculate a cleanliness value of the photovoltaic panels, and generate a cleaning trigger instruction when the cleanliness value is lower than a cleaning threshold dynamically calculated based on power generation loss and cleaning cost;

[0179] The second unit is configured to, upon receiving the cleaning trigger instruction, execute a cleaning path optimization step, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization objectives, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into an initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones; iteratively calculating the global path using a multi-pheromone ant colony algorithm; and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm;

[0180] The third unit is used to execute the multi-robot collaborative cleaning control steps according to the optimized cleaning path, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

[0181] According to a third aspect of the embodiments of the present invention,

[0182] An electronic device is provided, comprising:

[0183] processor;

[0184] a memory for storing processor-executable instructions;

[0185] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0186] According to a fourth aspect of the embodiments of the present invention,

[0187] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0188] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photovoltaic panel adaptive cleaning path planning method based on environmental perception, characterized in that: include: Obtain historical weather data and status monitoring data of the photovoltaic panels in the area where the photovoltaic panels are located, and calculate the cleanliness value of the photovoltaic panels; When the cleanliness value is lower than a cleaning threshold value dynamically calculated based on power generation loss and cleaning cost, generating a cleaning trigger instruction; After receiving the cleaning trigger instruction, a cleaning path optimization step is executed, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization goals, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones, iteratively calculating the global path through a multi-pheromone ant colony algorithm, and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm; Based on the optimized cleaning path, the multi-robot collaborative cleaning control steps are executed, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

2. The method according to claim 1, characterized in that Obtaining historical weather data and status monitoring data for the area where the photovoltaic panels are located, and calculating the cleanliness value of the photovoltaic panels include: The sliding time window method is used to process historical weather data. The time attenuation factor is introduced into the data in each time window for weighted calculation to obtain the historical weather impact value. The historical weather impact value and the status monitoring data are multiplied by the corresponding weight coefficients and summed to obtain the photovoltaic panel cleanliness value, wherein the initial value of each weight coefficient is determined by the fuzzy analytic hierarchy process, and the gradient direction of the weight is dynamically adjusted based on the cleanliness value during the calculation process; Establish a historical cleaning record database to record the cleanliness value before each cleaning, cleaning cost and actual power generation loss during the corresponding period. Combined with seasonal and environmental factors, calculate the power generation loss caused by unit cleanliness reduction; A cleaning benefit evaluation function is established based on the unit cleaning cost and the power generation loss caused by the unit cleanliness reduction. The corresponding cleaning benefits are calculated by traversing different cleanliness values, and the cleanliness value that maximizes the cleaning benefit is determined as the cleaning threshold for the current time period. The cleaning benefit evaluation function also considers the cleaning resource scheduling cost and the cleaning operation time window constraints.

3. The method according to claim 1, characterized in that Dynamically adjust weights based on environmental impact and cleaning energy efficiency ratio, including: Establishing an environmental impact function, the environmental impact function is used to characterize the influence of temperature parameters, humidity parameters, wind speed parameters, light intensity parameters, and precipitation probability parameters on the cleaning effect, wherein the temperature parameter is calculated using an exponential function to determine the degree of influence of temperature deviation from the optimal cleaning temperature, and the humidity parameter is calculated using an S-shaped response curve to determine the degree of influence of humidity exceeding a critical value; constructing a coupling coefficient between the temperature parameter and the humidity parameter based on the environmental impact function; Divide the grid cells based on the density of dirt distribution on the photovoltaic panel surface, and set the initial weights of cleaning efficiency, energy consumption and time cost; Calculating an environmental adjustment coefficient based on the environmental impact function and the coupling coefficient, wherein the environmental adjustment coefficient is positively correlated with the environmental sensitivity of the grid unit; calculating an energy efficiency ratio adjustment coefficient based on the cleaning energy efficiency ratio of the grid unit, wherein the energy efficiency ratio adjustment coefficient is a power function relationship between the current cleaning energy efficiency ratio and the maximum cleaning energy efficiency ratio; The product of the environmental adjustment coefficient and the energy efficiency ratio adjustment coefficient is used as the weight adjustment amount, so that the cleaning efficiency weight is negatively correlated with the weight adjustment amount, and the energy consumption weight and the time cost weight are positively correlated with the weight adjustment amount, and the initial weight is dynamically updated.

4. The method according to claim 1, wherein A genetic algorithm based on dirt density coding is used to optimize the population. The optimized population is mapped to the initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. The global path is obtained through iterative calculation of the multi-pheromone ant colony algorithm. The environment-adaptive multi-neighborhood simulated annealing algorithm is used for local optimization, including: Initializing and constructing an initial population using a genetic algorithm, wherein the genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism; The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm, and a multi-pheromone ant colony algorithm based on grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone; The global optimal cleaning path is locally optimized using an environment-adaptive multi-neighborhood simulated annealing algorithm. The multi-neighborhood includes a reverse neighborhood, an exchange neighborhood, an insertion neighborhood, and a regional reconstruction neighborhood. Local path optimization is performed using an environment-aware temperature regulation mechanism and a dynamic acceptance criterion strategy.

5. The method according to claim 4, characterized in that The genetic algorithm adopts a coding strategy based on dirt density weight and an environment-aware adaptive crossover mutation mechanism, including: Obtain multi-type dirt features on the photovoltaic panel surface, divide the photovoltaic panel surface into grid cells, and calculate the dirt density of each grid cell; The chromosome of the genetic algorithm is constructed as a multi-level coding structure of space sequence coding, resource allocation coding and density association coding, in which the space sequence coding represents the cleaning path, the resource allocation coding represents the cleaning parameters, and the density association coding represents the dirt treatment strategy. Constructing a density weight mapping function based on the dirt density, the density weight mapping function simultaneously considering the dirt density of the target grid cell and the dirt density distribution of its adjacent grid cells, and initializing the population according to the density weight mapping function; Acquiring environmental parameters and calculating environmental impact factors, dynamically adjusting the crossover probability and mutation probability of the genetic algorithm based on the environmental impact factors and the density weight mapping function, optimizing the cleaning path sequence through the crossover operation, and adjusting the resource allocation strategy through the mutation operation; The spatial sequence coding, resource allocation coding and density association coding of the best individual in the optimized population are mapped to the initial pheromone distribution matrix of the ant colony algorithm.

6. The method according to claim 5, characterized in that The optimization result of the genetic algorithm is used as the initial pheromone distribution of the ant colony algorithm. The multi-pheromone ant colony algorithm based on the grid energy efficiency gradient is used to generate the global optimal cleaning path. The multi-pheromone includes cleaning efficiency pheromone, energy consumption pheromone and time cost pheromone. The optimization result of the genetic algorithm is mapped to an initial distribution of multidimensional pheromones, wherein the multidimensional pheromones include cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones. Each pheromone component is obtained by a weighted combination of the sequence information corresponding to the optimal individual and the grid dirt density. Calculating a grid energy efficiency ratio based on a ratio of the cleaning benefit to the energy consumption of the grid unit, and constructing a grid energy efficiency gradient vector based on the horizontal and vertical partial derivatives of the grid energy efficiency ratio; Performing a weighted product operation on the heuristic information of each pheromone component to obtain comprehensive heuristic information, and calculating a state transition probability matrix based on the comprehensive heuristic information, the multi-dimensional pheromone initial distribution, and the grid energy efficiency gradient vector; Determining the ant's movement path according to the state transition probability matrix, performing local pheromone updates on the grid cells on the movement path, wherein the local pheromone updates take into account the influence of the pheromone volatility coefficient and the grid energy efficiency ratio, and performing pheromone enhancement on the global optimal path based on the evaluation index of the movement path; The variance of each pheromone component is calculated, and a multidimensional convergence judgment is performed based on the relative change rate of the variance. When the multidimensional convergence judgment result is less than a preset convergence threshold, a global optimal cleaning path is output.

7. The method according to claim 1, characterized in that The multi-robot collaborative cleaning control steps include: Obtaining the dirt density distribution and area of the cleaning area, calculating the workload of the sub-areas, and dividing the cleaning area into multiple sub-areas based on the balance of the workload of the sub-areas and assigning them to multiple cleaning robots; The cleaning trajectory in each sub-area is divided into a high-density trajectory segment and a low-density trajectory segment according to the dirt density threshold. A first sliding surface is constructed for the high-density trajectory segment using a weighted combination of velocity error and position error. A second sliding surface is constructed for the low-density trajectory segment using a weighted combination of position error integral term and position error. An equivalent control variable is calculated based on the state quantities of the first sliding surface and the second sliding surface, the equivalent control variable being used to compensate for the deterministic dynamic characteristics of the cleaning process; an adaptive switching gain is calculated based on the dirt density, the cleaning resistance, and the state quantities of the sliding surface, the adaptive switching gain being used to suppress uncertain disturbances during the cleaning process; and a piecewise adaptive sliding mode control input is generated by combining the equivalent control variable with a sign function of the adaptive switching gain. The proportion of the cleaned area and the distance to the target point of each cleaning robot are calculated to determine the obstacle avoidance priority. When the distance between adjacent cleaning robots is less than the safety distance threshold, the segmented adaptive sliding mode control input is corrected according to the obstacle avoidance priority, and the correction amount is proportional to the difference in the obstacle avoidance priorities of the adjacent robots and the distance gradient.

8. A photovoltaic panel adaptive cleaning path planning system based on environmental perception, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain historical weather data of the area where the photovoltaic panels are located and the status monitoring data of the photovoltaic panels, and calculate the cleanliness value of the photovoltaic panels; When the cleanliness value is lower than a cleaning threshold value dynamically calculated based on power generation loss and cleaning cost, generating a cleaning trigger instruction; The second unit is configured to, upon receiving the cleaning trigger instruction, execute a cleaning path optimization step, including: setting maximization of cleaning efficiency, minimization of energy consumption, and optimization of time cost as optimization objectives, and dynamically adjusting weights based on environmental impact and cleaning energy efficiency ratio; optimizing the population using a genetic algorithm based on dirt density coding, mapping the optimized population into an initial distribution of cleaning efficiency pheromones, energy consumption pheromones, and time cost pheromones; iteratively calculating the global path using a multi-pheromone ant colony algorithm; and performing local optimization using an environment-adaptive multi-neighborhood simulated annealing algorithm; The third unit is used to execute the multi-robot collaborative cleaning control steps according to the optimized cleaning path, including: dividing sub-areas based on dirt density and workload balance; using segmented adaptive sliding mode variable structure control for trajectory tracking; determining obstacle avoidance priority based on the proportion of cleaned area and target distance, and performing multi-robot collaborative control.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images