A method for predicting light attenuation of photovoltaic array and autonomous cleaning decision for unattended border energy sentry station

Abstract

The application discloses a photovoltaic array light transmission attenuation prediction and autonomous cleaning decision method for an unattended border energy sentry station, and relates to the technical field of photovoltaic energy generation, and specifically relates to a photovoltaic array light transmission attenuation prediction and autonomous cleaning decision method for an unattended border energy sentry station. The method comprises the following steps: acquiring multi-source time series data of a photovoltaic array and performing pretreatment; constructing a surface deposition state dynamic model to obtain surface deposition state; calculating a transmission attenuation rate according to an exponential attenuation relationship between the surface deposition state and the irradiation transmittance; inputting meteorological variables, surface deposition state, transmission attenuation rate and time characteristics into a recurrent residual network to obtain a hierarchical state representation; outputting a transmission attenuation rate prediction sequence of a future period through a prediction head and outputting a cleaning trigger probability through a decision head; and generating a cleaning decision result according to the transmission attenuation rate prediction sequence, a seasonal threshold, a minimum cleaning interval and a cumulative loss constraint. The application can realize predictive maintenance of photovoltaic components in a border energy scene where manual access is difficult, reduce the number of invalid cleaning times, and improve energy supply continuity and operation and maintenance economy.

Description

Technical Field

[0001] This invention belongs to the field of new energy intelligent operation and maintenance, photovoltaic power generation status assessment and autonomous maintenance decision-making technology, and in particular relates to a photovoltaic array light transmission attenuation prediction and autonomous surface clearing decision method for unmanned border energy outposts. Background Technology

[0002] Unmanned border energy outposts are typically deployed in environments prone to sandstorms, difficult to reach manually, and subject to unpredictable rainfall. Photovoltaic arrays, as their primary power source, are constantly affected by airborne particulate matter deposition, humidity buildup, and the random erosion from rain or strong winds. Once a persistent layer of contamination forms on the photovoltaic surface, it reduces the transmission capacity of incident radiation, causing a decrease in output power. In severe cases, this can lead to insufficient replenishment of the energy storage system, consequently affecting the continuous operation of the outpost's communication, sensing, and alarm equipment.

[0003] To address the issue of surface contamination in photovoltaic arrays, existing technologies mainly employ the following methods: First, fixed-cycle cleaning, i.e., manual cleaning is arranged at fixed intervals such as 7 days, 14 days, or 30 days; second, threshold alarms are triggered based on the current power deviation, and an alarm is triggered when the power drop exceeds a set threshold; third, the photovoltaic contamination status is modeled separately to estimate the degree of contamination, but this is not linked to subsequent operation and maintenance decisions.

[0004] However, the above methods have the following shortcomings: they cannot reflect the non-stationary characteristics of pollution accumulation caused by border sandstorms and sudden weather events; they cannot distinguish between "short-term recoverable losses" and "continuous losses requiring human intervention"; they cannot make proactive decisions based on the risk of power generation loss in the next few hours; and in unattended scenarios, they are prone to over-cleaning or delayed cleaning, resulting in both maintenance costs and energy risks. Therefore, it is necessary to propose a unified method that can simultaneously complete pollution state estimation, loss prediction, and net surface decision-making. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for predicting the light transmittance attenuation of photovoltaic arrays and making autonomous surface clearing decisions for unmanned border energy outposts, comprising the following steps: Acquire multi-source time-series data of the photovoltaic array and perform preprocessing; Based on the preprocessed data, a dynamic model of surface deposition state quantities is constructed to characterize the evolution of surface contamination, so as to obtain the surface deposition state quantities. The energy transmission attenuation rate is calculated based on the exponential decay relationship between the surface sedimentation state and the irradiance transmittance. Meteorological variables, the surface subsidence state quantity, the energy permeability attenuation rate, and time characteristics are input into the hierarchical recurrent reserve network to obtain a hierarchical state representation; Based on the hierarchical state representation, the prediction head outputs a prediction sequence of energy permeability attenuation rate for future periods, and the decision head outputs the net surface trigger probability. Based on the energy permeability attenuation rate prediction sequence, seasonal threshold, minimum net area interval, and cumulative loss constraint, the net area decision result is generated. When the decision result is to execute the net surface, the surface settlement state quantity is reset and the rolling prediction continues.

[0006] Optionally, a dynamic model of surface settlement states is constructed, specifically including: The surface settlement state is updated at the current moment based on the previous moment's surface settlement state, settlement input increment, rain jump net release factor, and cyclone desorption factor; wherein the settlement input increment is determined based on particulate matter concentration, tidal adhesion enhancement factor, and wind resistance settlement factor.

[0007] Optionally, the energy transmission attenuation rate is calculated based on the exponential decay relationship between the surface sedimentation state and the irradiance transmittance, specifically including: Based on the surface sedimentation state, the irradiance is determined according to the exponential decay relationship; based on the difference between 1 and the irradiance, the energy transmission decay rate is obtained.

[0008] Optionally, the hierarchical reverberation reserve network includes at least three layers of reserve pools, and the recursive weight matrix of each layer is subjected to spectral radius constraint processing.

[0009] Optionally, the recursive weight matrix is ​​subjected to spectral radius constraints, specifically including: Obtain the original recursive weight matrix and its spectral radius; based on the ratio of a preset spectral radius adjustment factor to the spectral radius of the original recursive weight matrix, scale the original recursive weight matrix so that the spectral radius of the constrained recursive weight matrix is ​​equal to the spectral radius adjustment factor.

[0010] Optionally, meteorological variables, the surface subsidence state quantity, the energy permeability attenuation rate, and time characteristics are input into the hierarchical recurrence reserve network to obtain a hierarchical state representation, specifically including: The meteorological variables, the surface subsidence state, the energy permeability attenuation rate, and the time features are combined into a historical input sequence; wherein, the time features include at least hourly features, weekday features, and monthly features; The historical input sequence is input into the first layer of at least three serially connected storage pools, and the first layer state representation is obtained after recursive processing by the first layer storage pool. The first-level state representation is input into the second-level reserve pool, and the second-level state representation is obtained after recursive processing in the second-level reserve pool. The second-level state representation is input into the third-level reserve pool, and the third-level state representation is obtained after recursive processing in the third-level reserve pool. The first-level state representation, the second-level state representation, and the third-level state representation are concatenated to obtain the hierarchical state representation.

[0011] Optionally, based on the hierarchical state representation, a prediction sequence of energy permeability attenuation rate for future time periods is output by the prediction head, and a net surface trigger probability is output by the decision head, specifically including: The hierarchical state representation is input into the future energy permeability attenuation rate prediction head and the autonomous surface clearing decision head, respectively. The future energy permeability attenuation rate prediction head uses a two-layer fully connected network to perform a nonlinear mapping on the hierarchical state representation and outputs the predicted energy permeability attenuation rate sequence for each time point within a preset future period. The autonomous surface clearing decision head uses a two-layer fully connected network followed by a Sigmoid function to perform a nonlinear mapping on the hierarchical state representation and convert it into a probability value between 0 and 1, outputting the surface clearing trigger probability at the current moment.

[0012] Optionally, based on the energy permeability attenuation rate prediction sequence, seasonal threshold, minimum net area interval, and cumulative loss constraint, a net area decision result is generated, specifically including: Based on the predicted energy transmission attenuation rate sequence, the average energy transmission attenuation rate within a preset time period is calculated, and it is determined whether the average energy transmission attenuation rate exceeds the seasonal threshold corresponding to the current season, thereby obtaining a first determination result. Obtain the time interval between the current time and the last clean-up time, determine whether the time interval has reached the minimum clean-up interval, and obtain a second determination result; The predicted values ​​of energy transmission attenuation rate at each time since the last clean surface are obtained, and the cumulative predicted loss is obtained by summing them. The cumulative predicted loss is then determined to see if it reaches the preset cumulative loss threshold, and a third determination result is obtained. When the first determination result, the second determination result, and the third determination result all meet the conditions, the decision result of the execution net surface is generated.

[0013] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0014] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention first acquires multi-source time-series data on temperature, humidity, irradiance, wind speed, rainfall, particulate matter concentration, and output power in the photovoltaic array's operating environment, and performs anomaly identification, missing data repair, and timescale alignment. Based on this, a dynamic model of surface deposition states is constructed. This model comprehensively considers the combined effects of airborne particle deposition, rainfall erosion, and wind-induced desorption, and can accurately characterize the dynamic evolution of photovoltaic surface contamination under conditions of frequent border sandstorms and random rainfall. It overcomes the shortcomings of existing technologies where fixed-period cleaning cannot reflect the non-stationary characteristics of contamination.

[0016] This invention calculates the energy attenuation rate based on the exponential decay relationship between surface sedimentation mass and irradiance transmittance. This method directly transforms physical sedimentation mass into a quantifiable energy attenuation index, establishing a clear mathematical correlation between pollution levels and power generation loss, and providing a measurable basis for subsequent loss prediction and decision-making.

[0017] This invention inputs meteorological variables, surface deposition state, permeability decay rate, and temporal characteristics into a hierarchical recurrent reservoir network to obtain a hierarchical state representation. This network, connected serially through multiple reservoirs, can sense short-term weather fluctuations, medium-term deposition accumulation, and long-term seasonal variations, thereby effectively capturing the multi-timescale dynamic characteristics of pollution evolution and outputting a predicted permeability decay rate sequence for future periods. This solves the problem that existing technologies cannot predict future power generation loss risks.

[0018] This invention, based on a hierarchical state representation, outputs a predicted sequence of energy permeability decay rates for future time periods through a prediction head, while simultaneously outputting the net surface trigger probability at the current moment through a decision head. This design unifies pollution state estimation, loss prediction, and maintenance decision-making within the same framework, avoiding the judgment lag or over-intervention problems caused by the separation of pollution modeling and operation and maintenance decisions in existing technologies.

[0019] This invention generates clean-up decision results based on a predicted energy dissipation rate sequence, a seasonal threshold, a minimum clean-up interval, and a cumulative loss constraint. The seasonal threshold distinguishes between winter and non-winter pollution sensitivities, the minimum clean-up interval avoids maintenance waste from frequent cleaning, and the cumulative loss constraint ensures that clean-up is only triggered when avoidable power generation losses accumulate to a level significant for maintenance. The combined effect of these three factors effectively distinguishes between "short-term recoverable losses" and "continuous losses requiring manual intervention," enabling predictive, low-redundancy autonomous clean-up decision-making in unattended scenarios.

[0020] This invention resets the surface settling state after cleaning and continues rolling prediction for the next time step. This closed-loop mechanism ensures that the system can continuously track changes in the pollution state, dynamically adjust subsequent decisions, and ensure long-term power supply continuity and operational economy.

[0021] In summary, this invention enables predictive maintenance of photovoltaic modules in border energy scenarios where manual access is difficult and infrequent, reducing unnecessary cleaning operations and improving energy supply continuity and operational economy. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the photovoltaic power supply scenario for an unmanned border energy outpost, according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the layered echo reserve network structure according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the rolling prediction closed loop after the surface is triggered according to an embodiment of the present invention. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0025] Example 1 like Figure 1 As shown, this embodiment provides a method for predicting the light transmittance attenuation of photovoltaic arrays and making autonomous surface clearing decisions for unmanned border energy outposts, including the following steps: Data on temperature, humidity, irradiance, wind speed, rainfall, particulate matter concentration, and photovoltaic output power in the photovoltaic array's operating environment are collected. The data is then processed for anomaly identification, missing segment repair, and unified time resolution alignment. A dynamic model of photovoltaic surface settlement state was constructed, with the settlement mass per unit area as the state variable. Based on particulate matter concentration, humidity, wind speed and rainfall, the settlement input increment, rain jump net release factor and wind cyclone desorption factor were calculated to obtain the surface settlement state at each time. Based on the exponential decay relationship between surface sedimentation state quantity and irradiance transmittance, the transmittance decay rate of the photovoltaic array at each time point is calculated. A hierarchical response reserve network is constructed, which combines meteorological variables, surface subsidence state, energy permeability attenuation rate and time characteristics into a historical input sequence, and obtains hierarchical state representation through multi-layer reserve pools; Based on the hierarchical state representation, the future energy penetration attenuation rate prediction head outputs the energy penetration attenuation rate prediction sequence within a preset time period, and the autonomous surface clearing decision head outputs the surface clearing trigger probability at the current moment. Based on the predicted energy permeability attenuation rate over a future preset period, seasonal thresholds, minimum netting intervals, and cumulative loss constraints, a decision on whether to perform netting is generated. When the decision result is to execute the net surface, the surface settlement state is reset and the rolling prediction continues into the next time step.

[0026] The feasible data acquisition and preprocessing process includes: like Figure 2 As shown in the example, in the unmanned border energy outpost targeted in this embodiment, the photovoltaic array is deployed in an environment with frequent sandstorms and where it is difficult for personnel to frequently reach. As the main energy supply unit of the outpost, it provides power support for the downstream energy storage system and communication, sensing, and alarm equipment. The photovoltaic surface is constantly affected by air particle deposition, humidity adhesion, and random rainfall or strong wind erosion. The power attenuation caused by surface contamination will directly affect the continuity of the outpost's power supply.

[0027] The data collected includes: ambient temperature. relative humidity Global horizontal irradiance Wind speed hourly rainfall Output power Air particulate matter concentration Where t is the discrete time step, which is uniformly scaled to hours.

[0028] The preprocessing process includes: ① Abnormal removal: A local outlier detection method is used to detect anomalies in the original observation sequence. Let the sample points be... Its local achievable density is , yes Given the local reachability density of neighboring points, the local outlier factor is defined as: ; in, Indicates sample of Nearest neighbor set; It is a local outlier; yes of Any neighbor sample point in the nearest neighbor set.

[0029] When satisfied Then determine These are considered abnormal samples and are removed. The set threshold for anomaly detection.

[0030] ② Time standardization: Resample the original 5-minute sampled variables by hourly mean: ; in, For the first Within an hour Five-minute sample values.

[0031] ③ Missing item repair: Let the duration of consecutive missing values ​​be L, and the sampling interval be... Define the missing threshold Missing threshold .

[0032] when Linear interpolation is used; when Spline interpolation is used; when If so, the corresponding time period is removed to avoid introducing false dynamic structures.

[0033] A feasible dynamic model of surface settlement states can be constructed, specifically including: The surface settlement state is updated at the current moment based on the previous moment's surface settlement state, settlement input increment, rain jump net release factor, and cyclone desorption factor; wherein the settlement input increment is determined based on particulate matter concentration, tidal adhesion enhancement factor, and wind resistance settlement factor.

[0034] As a specific implementation method, the mass of sedimentation per unit area of ​​the photovoltaic surface is defined as... ;in, Units are , indicating the first Surface settlement state at time t.

[0035] Its state update equation is: ; in: Input increment for settlement; Rain-induced net release factor; It is a cyclone desorption factor.

[0036] ①Settlement input increment: definition: ; in: The basic settlement coefficient; This refers to the particulate matter concentration. It is a moisture-enhancing factor; This is the wind resistance and settling factor.

[0037] ② Hygroscopic Enhancement Factor: Considering that increased humidity enhances particle adhesion, the definition is: ; in: Normalized humidity; This is the humidity enhancement factor.

[0038] ③ Wind resistance and settling factor: Considering that increased wind speed may inhibit stable particle settling, the definition is: ; in, This is the wind speed suppression coefficient.

[0039] ④ Rain Leap Net Release Factor: Let the critical rainfall threshold be ,definition: ; in, This represents the cleanliness factor of rainfall.

[0040] In a preferred embodiment, when the rainfall intensity is significantly greater than the threshold, it can be directly set... This indicates that the surface returns to a near-clean state after this time step.

[0041] ⑤ Cyclone desorption factor: Let the critical wind speed for wind-induced resuspension be... ,definition: ; in, is the rate constant for wind-induced desorption.

[0042] The feasible process of calculating the transmittance decay rate of a photovoltaic array at various times, based on the exponential decay relationship between surface sedimentation state quantity and irradiance transmittance, includes: In the sedimentation state quantity Based on this, a surface light transmittance attenuation model is established: Let the effective incident radiation on the clean surface be... The effective irradiation after passing through the contamination layer is Then, according to the law of exponential transmission: ; in, The total extinction coefficient.

[0043] Define the energy transmission attenuation rate for: ; but This characterizes the degree of relative energy decay caused by surface contamination.

[0044] Therefore, we can conclude that: when When it increases, Monotonically increasing; when effective rainfall or artificial surface cleaning occurs, Therefore .

[0045] Feasible, such as Figure 3 As shown, in order to characterize short-term weather disturbances, medium-term subsidence accumulation and long-term seasonal changes, this embodiment constructs a three-layer cascaded echo reserve network, and the recursive weight matrix of each layer is constrained by the spectral radius.

[0046] Furthermore, the recursive weight matrix is ​​subjected to spectral radius constraint processing, specifically including: obtaining the original recursive weight matrix and its spectral radius; scaling the original recursive weight matrix based on the ratio of a preset spectral radius adjustment factor to the spectral radius of the original recursive weight matrix, so that the spectral radius of the constrained recursive weight matrix is ​​equal to the spectral radius adjustment factor.

[0047] Furthermore, the meteorological variables, the surface subsidence state, the energy permeability attenuation rate, and the time characteristics are input into the hierarchical recurrence reserve network to obtain a hierarchical state representation, specifically including: The meteorological variables, the surface subsidence state, the energy permeability attenuation rate, and the time features are combined into a historical input sequence; wherein, the time features include at least hourly features, weekday features, and monthly features; The historical input sequence is input into the first layer of at least three serially connected storage pools, and the first layer state representation is obtained after recursive processing by the first layer storage pool. The first-level state representation is input into the second-level reserve pool, and the second-level state representation is obtained after recursive processing in the second-level reserve pool. The second-level state representation is input into the third-level reserve pool, and the third-level state representation is obtained after recursive processing in the third-level reserve pool. The first-level state representation, the second-level state representation, and the third-level state representation are concatenated to obtain the hierarchical state representation.

[0048] As a specific implementation method, in order to characterize short-term weather disturbances, medium-term subsidence accumulation and long-term seasonal changes, this embodiment constructs a three-layered cascaded echo reserve network.

[0049] ① Input vector: At any moment Define the input vector: ; in: Characteristic of hours; Characteristics of the week; This is a monthly characteristic.

[0050] History window length set Preferred selection That is, a continuous historical sequence of 168 hours.

[0051] Preferably, the number of neurons in the three layers can be set as follows: .

[0052] The final splicing state is represented as follows: .

[0053] Implementable, based on the hierarchical state representation, the prediction head outputs a prediction sequence of energy permeability attenuation rates for future time periods, and the decision head outputs the net surface trigger probability, specifically including: The hierarchical state representation is input into the future energy permeability attenuation rate prediction head and the autonomous surface clearing decision head, respectively. The future energy permeability attenuation rate prediction head uses a two-layer fully connected network to perform a nonlinear mapping on the hierarchical state representation and outputs the predicted energy permeability attenuation rate sequence for each time point within a preset future period. The autonomous surface clearing decision head uses a two-layer fully connected network followed by a Sigmoid function to perform a nonlinear mapping on the hierarchical state representation and convert it into a probability value between 0 and 1, outputting the surface clearing trigger probability at the current moment.

[0054] As a specific implementation method, defining the future The predicted energy transmission attenuation rate vector for each time step is: ; Two layers of fully connected mapping are used: ; in: , , ReLU activation function Right now ; Its loss function is defined as mean squared error: ; in, This refers to predicting risks in the next 24 hours.

[0055] Define the clean face trigger tag as: ; in: This indicates that a clean face should be triggered at the current moment; This indicates that the cleanup will not be triggered at the current moment.

[0056] The output probability of a clean surface is obtained by using two fully connected layers plus a sigmoid function. ; in, It is a non-linear activation function.

[0057] When satisfied If so, a clean surface trigger suggestion will be output.

[0058] The corresponding classification loss function is binary cross-entropy: .

[0059] Feasible approach: Jointly train future energy permeability attenuation rate prediction with net surface triggering decision, defining the total loss function: ; in: In this embodiment, we select .

[0060] Feasible, based on the energy permeability attenuation rate prediction sequence, seasonal threshold, minimum net area interval, and cumulative loss constraint, generate net area decision results, specifically including: Based on the predicted energy transmission attenuation rate sequence, the average energy transmission attenuation rate within a preset time period is calculated, and it is determined whether the average energy transmission attenuation rate exceeds the seasonal threshold corresponding to the current season, thereby obtaining a first determination result. Obtain the time interval between the current time and the last clean-up time, determine whether the time interval has reached the minimum clean-up interval, and obtain a second determination result; The predicted values ​​of energy transmission attenuation rate at each time since the last clean surface are obtained, and the cumulative predicted loss is obtained by summing them. The cumulative predicted loss is then determined to see if it reaches the preset cumulative loss threshold, and a third determination result is obtained. When the first determination result, the second determination result, and the third determination result all meet the conditions, the decision result of the execution net surface is generated.

[0061] As a specific implementation method, in order to make the output directly usable for the operation and maintenance of unmanned border energy outposts, the following triggering conditions are defined: Condition 1: Risk conditions for the next 24 hours: Assume the predicted average energy permeability attenuation rate for the next 24 hours is: ; when If so, it is believed that there is a significant avoidable loss in the next 24 hours.

[0062] in, is the seasonal threshold, and s represents the seasonal category.

[0063] Preferably: ; Condition 2: Minimum clear surface interval condition: Let the last time of face cleaning be... Then the following is required: , .

[0064] Condition 3: Cumulative Loss Constraint Define the cumulative predicted loss since the last netting as: ; when If so, it is considered that a cumulative loss with operational significance has been formed.

[0065] Therefore, the final net face decision function is defined as: ; in, This indicates that you are performing a self-cleaning procedure.

[0066] If feasible, when the decision result is to execute the net surface, the surface settlement state is reset, and the rolling prediction continues into the next time step. For example... Figure 4 As shown, the state reset after the clean surface is triggered and the rolling prediction process form a closed-loop control, ensuring the autonomy of the system's long-term operation.

[0067] Specifically, after performing the cleanup, the status will be reset to: ; And update .

[0068] In summary, this embodiment first acquires multi-source time-series data such as temperature, humidity, irradiance, wind speed, rainfall, particulate matter concentration, and power generation, and performs anomaly identification, missing data repair, and unified time-scale alignment on the data. Secondly, it establishes a dynamic equilibrium model of surface deposition states to characterize the evolution of photovoltaic surface pollution under the combined effects of air particle deposition, rainfall erosion, and wind-induced desorption. Further, it constructs a transmission attenuation rate model based on the exponential transmission attenuation relationship. Then, it inputs meteorological variables, surface deposition states, transmission attenuation rate, and time characteristics into a layered recurrent reserve network, outputting future multi-period transmission attenuation rate predictions and net surface trigger probabilities. Finally, it combines seasonal thresholds, minimum net surface intervals, and cumulative loss constraints to form an autonomous net surface decision.

[0069] Compared to existing technologies, this embodiment provides a method for predicting the light transmittance decay of photovoltaic arrays and making autonomous surface cleaning decisions for unmanned border energy outposts. This method enables predictive maintenance of photovoltaic modules in border energy scenarios where manual access is infrequent, reducing unnecessary cleaning and improving power supply continuity and operational economy. Furthermore, this method is applicable to unmanned border scenarios, addressing not only power generation efficiency but also directly serving the continuity of power supply and the scarcity of maintenance at outposts. It forms a complete closed loop from contamination to decision-making, unifying surface settlement state estimation, light transmittance decay rate prediction, and surface cleaning decisions. It balances mechanistic reliability with time-series prediction capabilities, employing a settlement-scouring-transmission decay physical model on one hand, and modeling complex nonlinear dynamics through a layered echo reserve network on the other. It is more economical than fixed-period cleaning, allowing for predictive triggering based on future risks, reducing unnecessary maintenance and lowering power generation losses.

[0070] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0071] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0072] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the light transmittance attenuation of photovoltaic arrays and making autonomous netting decisions for unmanned border energy outposts, characterized in that, Includes the following steps: Acquire multi-source time-series data of the photovoltaic array and perform preprocessing; Based on the preprocessed data, a dynamic model of surface deposition state quantities is constructed to characterize the evolution of surface contamination, so as to obtain the surface deposition state quantities. The energy transmission attenuation rate is calculated based on the exponential decay relationship between the surface sedimentation state and the irradiance transmittance. Meteorological variables, the surface subsidence state quantity, the energy permeability attenuation rate, and time characteristics are input into the hierarchical recurrent reserve network to obtain a hierarchical state representation; Based on the hierarchical state representation, the prediction head outputs a prediction sequence of energy permeability attenuation rate for future periods, and the decision head outputs the net surface trigger probability. Based on the energy permeability attenuation rate prediction sequence, seasonal threshold, minimum net area interval, and cumulative loss constraint, the net area decision result is generated. When the decision result is to execute the net surface, the surface settlement state quantity is reset and the rolling prediction continues.

2. The method according to claim 1, characterized in that, Constructing a dynamic model of surface settlement states, specifically including: The surface settlement state is updated at the current moment based on the previous moment's surface settlement state, settlement input increment, rain jump net release factor, and cyclone desorption factor; wherein the settlement input increment is determined based on particulate matter concentration, tidal adhesion enhancement factor, and wind resistance settlement factor.

3. The method according to claim 1, characterized in that, Based on the exponential decay relationship between the surface sedimentation state and the irradiance transmittance, the energy transmission attenuation rate is calculated, specifically including: Based on the surface sedimentation state, the irradiance is determined according to the exponential decay relationship; based on the difference between 1 and the irradiance, the energy transmission decay rate is obtained.

4. The method according to claim 1, characterized in that, The hierarchical recurrent reserve network includes at least three layers of reserve pools, and the recursive weight matrix of each layer is subjected to spectral radius constraint processing.

5. The method according to claim 4, characterized in that, The recursive weight matrix is ​​subjected to spectral radius constraints, specifically including: Obtain the original recursive weight matrix and its spectral radius; based on the ratio of a preset spectral radius adjustment factor to the spectral radius of the original recursive weight matrix, scale the original recursive weight matrix so that the spectral radius of the constrained recursive weight matrix is ​​equal to the spectral radius adjustment factor.

6. The method according to claim 4, characterized in that, Meteorological variables, the surface subsidence state quantity, the energy permeability attenuation rate, and time characteristics are input into the hierarchical recurrent reserve network to obtain a hierarchical state representation, specifically including: The meteorological variables, the surface subsidence state, the energy permeability attenuation rate, and the time features are combined into a historical input sequence; wherein, the time features include at least hourly features, weekday features, and monthly features; The historical input sequence is input into the first layer of at least three serially connected storage pools, and the first layer state representation is obtained after recursive processing by the first layer storage pool. The first-level state representation is input into the second-level reserve pool, and the second-level state representation is obtained after recursive processing in the second-level reserve pool. The second-level state representation is input into the third-level reserve pool, and the third-level state representation is obtained after recursive processing in the third-level reserve pool. The first-level state representation, the second-level state representation, and the third-level state representation are concatenated to obtain the hierarchical state representation.

7. The method according to claim 1, characterized in that, Based on the hierarchical state representation, the prediction head outputs a prediction sequence of energy permeability attenuation rate for future time periods, and the decision head outputs the net surface trigger probability, specifically including: The hierarchical state representation is input into the future energy permeability attenuation rate prediction head and the autonomous surface clearing decision head, respectively. The future energy permeability attenuation rate prediction head uses a two-layer fully connected network to perform a nonlinear mapping on the hierarchical state representation and outputs the predicted energy permeability attenuation rate sequence for each time point within a preset future period. The autonomous surface clearing decision head uses a two-layer fully connected network followed by a Sigmoid function to perform a nonlinear mapping on the hierarchical state representation and convert it into a probability value between 0 and 1, outputting the surface clearing trigger probability at the current moment.

8. The method according to claim 1, characterized in that, Based on the energy permeability attenuation rate prediction sequence, seasonal threshold, minimum net area interval, and cumulative loss constraint, net area decision results are generated, specifically including: Based on the predicted energy transmission attenuation rate sequence, the average energy transmission attenuation rate within a preset time period is calculated, and it is determined whether the average energy transmission attenuation rate exceeds the seasonal threshold corresponding to the current season, thereby obtaining a first determination result. Obtain the time interval between the current time and the last clean-up time, determine whether the time interval has reached the minimum clean-up interval, and obtain a second determination result; The predicted values ​​of energy transmission attenuation rate at each time since the last clean surface are obtained, and the cumulative predicted loss is obtained by summing them. The cumulative predicted loss is then determined to see if it reaches the preset cumulative loss threshold, and a third determination result is obtained. When the first determination result, the second determination result, and the third determination result all meet the conditions, the decision result of the execution net surface is generated.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images