A smart cleaning triggering method and system based on photovoltaic module output impedance matching

Abstract

The application provides a kind of intelligent cleaning trigger method and system based on photovoltaic module output impedance matching, the method includes the following steps: (1) multi-dimensional electrical and environmental parameter high-frequency synchronous acquisition and pretreatment, (2) working condition adaptive standard impedance modeling and dynamic impedance ratio accurate calculation, (3) multi-interference layer peeling and dust accumulation exclusive feature vector purification, (4) adaptive hierarchical intelligent cleaning trigger, (5) cleaning effect closed loop verification and algorithm self-iterative optimization.The application constructs the complete technical link of "multi-parameter synchronous acquisition-working condition adaptive impedance modeling-dynamic impedance ratio calculation-multi-interference layer peeling-dust accumulation feature purification-hierarchical closed loop trigger-algorithm self-iteration", completely peels off non-dust accumulation interference factors, realizes the accurate identification of pure dust accumulation state, replaces traditional extensive determination mode, effectively solves the problems of existing technology cleaning misjudgment, omission, excessive cleaning and cleaning lag.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic intelligent operation and maintenance technology, specifically an intelligent cleaning triggering method and system based on photovoltaic module output impedance matching. Background Technology

[0002] The installed capacity of the photovoltaic (PV) power generation industry has continued to grow rapidly, becoming a core component of the clean energy system. PV modules operate in open-air environments for extended periods, making their surfaces highly susceptible to the accumulation of various pollutants such as dust, smog particles, bird droppings, and fallen leaves. This contaminant buildup directly obstructs the light-receiving surface of the solar cells, reducing photoelectric conversion efficiency. Furthermore, uneven dust accumulation can trigger localized hot spots, accelerating the aging and degradation of the solar cells, encapsulation films, and glass covers, significantly shortening the lifespan of PV modules and ultimately resulting in a loss of continuous power generation revenue for the PV power plant. Therefore, sophisticated and intelligent PV module cleaning and maintenance technologies are crucial for ensuring the efficient, stable, and long-term operation of PV power plants.

[0003] Currently, the intelligent cleaning triggering technology for photovoltaic modules has formed a large-scale application system in the industry. The mainstream technical solutions are mainly divided into two categories: fixed timed cleaning and power attenuation threshold triggered cleaning. There are also a small number of machine vision-assisted cleaning solutions. The relevant technologies have been widely used in various scenarios such as centralized ground photovoltaic power stations and distributed rooftop photovoltaic power stations. Existing public patents, journal articles and industry standards can fully prove the current status of technology application.

[0004] Current industry-standard solutions for timed cleaning, power attenuation threshold cleaning, and machine vision-assisted cleaning all have significant technical shortcomings in actual photovoltaic operation and maintenance scenarios. The core common pain point is the inability to accurately distinguish the core causes of module power attenuation and the inability to effectively isolate various interference factors, leading to a series of problems such as cleaning misjudgment, missed judgment, over-cleaning, and cleaning lag. The specific technical shortcomings are detailed below:

[0005] First, fixed-time cleaning technology suffers from severe indiscriminate cleaning, poor adaptability to various scenarios, and low maintenance economics. This technology relies entirely on preset fixed cycles for cleaning, completely disconnected from the real-time dust accumulation status of components, regional environmental conditions, and individual component performance. In areas with clean air and minimal dust, the dust accumulation rate on components is extremely low. Fixed-cycle cleaning results in excessive cleaning, not only wasting the cleaning robot's energy and exacerbating mechanical wear, but also causing frequent friction against the component glass covers, resulting in scratches on the coating layer and glass wear, accelerating component aging and degradation. In areas with frequent dust, sandstorms, and smog, the dust accumulation rate on components is much faster than the preset cleaning cycle, leading to cleaning delays. Long-term dust accumulation not only continuously reduces power generation efficiency but also easily induces hot spot failures and localized component damage, significantly increasing power plant maintenance costs. This technology cannot achieve "on-demand cleaning" and is difficult to adapt to differentiated maintenance needs.

[0006] Secondly, traditional power degradation threshold cleaning technologies suffer from core logical flaws, poor interference resistance, and an extremely high rate of misjudgment during cleaning. During outdoor operation of photovoltaic modules, power degradation is the result of multiple coupled factors, not solely caused by dust accumulation. The core interference factors fall into three categories: first, natural aging degradation from long-term module operation, representing irreversible long-term baseline performance loss; second, instantaneous power degradation caused by weather changes such as light fluctuations, abnormal temperatures, rain, and cloud cover, representing normal operating condition fluctuations; and third, power anomalies caused by equipment failures such as poor module wiring connections, component damage, and partial short circuits. Existing power threshold technologies rely solely on the magnitude of power degradation, failing to identify these non-dust-accumulation interference factors. This easily leads to misjudging power declines caused by aging, weather fluctuations, and equipment failures as dust accumulation, triggering numerous ineffective cleaning operations. Furthermore, when modules experience only slight dust accumulation and power degradation does not reach the preset trigger threshold, but continuous dust accumulation will rapidly deteriorate power generation performance, this technology cannot predict this in advance, resulting in a cleaning lag and hindering preventative maintenance.

[0007] Third, machine vision-assisted cleaning technology has weak anti-interference capabilities and low quantification accuracy, failing to meet the needs of precise operation and maintenance. Visual recognition technology is greatly affected by complex outdoor environments; varying light intensity, shadows, rain, snow, and fog can all severely interfere with image recognition accuracy, making it impossible to accurately distinguish dust thickness and uniformity. Furthermore, this technology can only identify the macroscopic morphology of contamination on component surfaces and cannot quantify the substantial impact of dust accumulation on the component's photoelectric conversion performance and electrical output characteristics. Its judgment dimension is singular and its accuracy insufficient, making it only a supplementary reference and unable to solve the core problems of misjudgment and missed detection during cleaning.

[0008] In summary, the fundamental common defects of existing technologies are: lack of specific feature identification capabilities for dust accumulation pollution, inability to effectively isolate interference factors such as environmental fluctuations, component aging, and equipment failure, and inability to accurately identify the true dust accumulation status of components. Ultimately, this leads to industry pain points such as low accuracy of cleaning triggering, many invalid operations, high energy consumption for operation and maintenance, large equipment wear and tear, and serious loss of power generation revenue. Summary of the Invention

[0009] To address the shortcomings of the existing technologies, this invention provides an intelligent cleaning triggering method and system based on photovoltaic module output impedance matching. It abandons the traditional coarse judgment logic based on a single dimension of time and power, and based on the output impedance matching characteristics of photovoltaic modules, it explores the unique electrical characteristics of dust accumulation, thereby achieving accurate identification of dust accumulation and on-demand intelligent cleaning triggering under multiple interference conditions.

[0010] The technical solution provided by this invention: an intelligent cleaning triggering method based on photovoltaic module output impedance matching, comprising the following steps:

[0011] (1) The electrical operating parameters, environmental operating parameters, and equipment operation and maintenance parameters of the photovoltaic modules are collected in three dimensions synchronously, and the raw collected data is preprocessed.

[0012] (2) A photovoltaic module operating condition-standard impedance fitting model was built. Based on the standard test parameters of the module's factory STC and combined with the equivalent circuit principle of the photovoltaic cell, the optimal output impedance curve of the module corresponding to different irradiance G and different temperatures T was obtained. A standard impedance database of the whole domain operating conditions was constructed, the real-time dynamic output impedance of the module was calculated, and the core judgment index—dynamic impedance ratio K was constructed. The calculation formula is K=Z / Z0, where Z is the real-time dynamic output impedance of the module and Z0 is the standard matching impedance of the module under the same real-time operating conditions.

[0013] (3) Through the three-level layered interference stripping algorithm, interference data is removed step by step according to the progressive logic of "environmental interference - aging interference - fault interference". Finally, the exclusive features corresponding only to ash pollution are purified. After the three-level interference stripping is completed, based on the pure steady-state time-series impedance data, three core ash-specific feature parameters are extracted to construct a three-dimensional quantitative ash feature vector M = [ΔK, VK, SK], where ΔK is the steady-state average offset of impedance ratio, VK is the time-series cumulative change rate of impedance ratio, and SK is the ash feature matching degree.

[0014] (4) Based on the multidimensional ash accumulation feature vector, an adaptive three-level hierarchical triggering logic is constructed, and a scene adaptive threshold correction mechanism is provided to realize refined operation and maintenance control of "minor warning, moderate cleaning, heavy emergency cleaning, and fault isolation".

[0015] (5) Construct a closed-loop operation and maintenance system of “monitoring-judgment-triggering-verification-iteration”. Quantify the cleaning effectiveness by comparing the impedance characteristics before and after cleaning, and continuously iterate and optimize the algorithm model based on the full amount of operation and maintenance data to achieve adaptive and continuous upgrading of the cleaning strategy.

[0016] Furthermore, the electrical operating parameters mentioned in step (1) include single component open-circuit voltage Uoc, short-circuit current Isc, operating voltage U, operating current I, and real-time output power P; the environmental operating parameters include solar irradiance G, component backplane temperature T, and ambient temperature Ta, which are collected in real time by high-precision irradiance sensors and temperature sensors; the equipment operation and maintenance parameters include component factory nominal parameters, component commissioning years, historical fault records, and data on cleaning times and cleaning methods; the preprocessing includes: using a sliding window filtering algorithm to perform mean denoising on the sampled data and removing abnormal outliers that exceed the normal fluctuation range of the parameters; and achieving millisecond-level time synchronization of electrical parameters, environmental parameters, and operation and maintenance parameters based on a timestamp alignment mechanism.

[0017] Furthermore, in step (2), based on the pre-processed steady-state operating voltage U and operating current I, the current dynamic output impedance Z of the component is calculated in real time using the formula Z=U / I, so as to accurately characterize the current real electrical output characteristics of the component.

[0018] Furthermore, the three-level hierarchical interference stripping algorithm in step (3) includes:

[0019] The first level is environmental transient interference filtering: set the steady-state quantitative judgment threshold, filter the steady-state data range with irradiance fluctuation amplitude ≤5% and component temperature fluctuation ≤3℃ within 10 consecutive minutes, directly remove abnormal impedance data caused by transient operating condition fluctuations such as rain, cloud cover, and gust cooling, and only retain the time series data under steady-state operating conditions, completely eliminating the interference of temporary environmental fluctuations on the judgment results.

[0020] The second level is the component aging baseline offset correction: based on the actual years of operation of the component, the preset component aging impedance drift model is called, and according to the general photovoltaic component attenuation standard, the impedance baseline offset of the component corresponding to the years of use is accurately calculated. The real-time dynamic impedance ratio is baseline calibrated, the impedance offset caused by natural aging is accurately deducted, and the characteristic coupling interference of aging attenuation and dust accumulation is completely separated.

[0021] The third level is electrical fault feature filtering: set an impedance change quantification judgment threshold. When the change in impedance ratio of a single sampling exceeds 20%, or when there is a large jump in three consecutive sets of sampling data, it is determined that the component has an electrical fault or abnormal line contact. Such abnormal data is directly removed and the fault location is marked.

[0022] Furthermore, in step (3), it is assumed that the effective steady-state duration contains n sets of impedance ratio sampling data, and the impedance ratio data of the i-th frame is K. i The standard impedance ratio is 1 under clean, dust-free conditions. The formula for calculating the steady-state average offset ΔK of the impedance ratio is:

[0023] ;

[0024] Taking the initial average impedance ratio Kstart and the final average impedance ratio Kend during the steady-state monitoring period, and the total steady-state monitoring duration as T, the formula for calculating the time-series cumulative rate of change of impedance ratio VK is as follows:

[0025] ;

[0026] Let the mean impedance ratio corresponding to the time series of the standard pure ash accumulation template be Kstd. The degree of matching between the current measured time series curve and the standard ash accumulation model is calculated using Euclidean distance similarity. Normalization yields a matching score ranging from 0% to 100%. The formula for calculating the ash accumulation feature matching degree SK is as follows:

[0027] .

[0028] Furthermore, the three-level hierarchical triggering in step (4) includes:

[0029] The first level is a mild dust accumulation early warning mechanism: when 0 < ΔK < 10%, 70% ≤ SK < 85%, and VK approaches 0 with no positive continuous growth trend, the component is determined to have mild dust accumulation. No cleaning operation needs to be initiated; the system only generates a dust accumulation early warning log, continuously tracks the dust accumulation trend, predicts subsequent pollution development, achieves preventative monitoring, and avoids gradual dust accumulation.

[0030] The second level is the moderate dust accumulation cleaning trigger mechanism: when the following conditions are met: 10%≤ΔK≤25%, 85%≤SK≤95%, VK shows a stable low-speed positive growth, the impedance steady-state deviation lasts for ≥10 minutes, the power generation of the module shows a slow decline in the same trend, and the three types of interference factors such as environmental fluctuations, module aging, and electrical faults are completely eliminated, it is determined to be effective moderate dust accumulation pollution. A precise cleaning instruction is immediately generated, and the corresponding intelligent cleaning robot is dispatched to start fixed-point cleaning operation.

[0031] The third level is a mechanism to distinguish between severe dust accumulation and faults: When ΔK > 25%, SK > 95%, and steady-state deviation is continuous without oscillation, it is judged as severe dust accumulation pollution, the power generation loss of the module is significant and there is a high risk of hot spots, and an emergency cleaning operation is triggered. If ΔK shows a large instantaneous jump, SK has no dust accumulation characteristic matching degree, and the data has no steady-state continuity, it is judged as an electrical fault of the module, only a fault warning signal is output, the cleaning command is locked, and invalid cleaning operations are completely eliminated, so as to achieve accurate differentiation between pollution and faults.

[0032] Furthermore, the built-in scene adaptive threshold optimization algorithm in step (4) relies on historical operation and maintenance big data to statistically analyze the dust accumulation patterns and environmental conditions in different months and regions, and automatically fine-tunes the dust accumulation judgment thresholds at each level.

[0033] Furthermore, the scenario-adaptive threshold optimization algorithm constructs an environmental-dust accumulation feature statistical sample library on a monthly basis. It collects effective dust accumulation feature data ΔK, VK, SK, meteorological data, and cleaning efficiency data under all steady-state operating conditions for the current month. A sliding window 95th percentile statistical correction mechanism is used to dynamically adjust the judgment thresholds corresponding to light, moderate, and heavy dust accumulation, ensuring that the thresholds always closely match the actual dust accumulation distribution patterns of the current season and region.

[0034] Threshold adaptive correction calculation formula: ,

[0035] In the formula: θnew is the updated adaptive threshold; θold is the original baseline threshold of the system; α is the learning correction coefficient, with a value of 0.05~0.1; P95 is the 95th percentile of the valid ash accumulation feature samples for the current month; μ is the mean of the ash accumulation feature samples for the current month.

[0036] Furthermore, step (5) includes: after the intelligent cleaning robot completes the standardized cleaning operation of a single area, the system delays for 5 minutes to wait for the component's operating condition to be completely stable, continuously collects the steady-state electrical parameters and environmental parameters for the next 30 minutes, calculates the component's dynamic impedance ratio and dust accumulation feature vector for the second time, and accurately quantifies the cleaning effect by comparing the ΔK offset, power recovery rate, and feature matching degree SK before and after cleaning. If the impedance ratio returns to the standard matching range after cleaning, ΔK≤3%, the dust accumulation feature completely disappears, and the power generation is restored to the standard level of the same operating condition, then the cleaning is deemed qualified and the operation and maintenance loop is completed. If there are still obvious dust accumulation features remaining and the impedance offset has not been restored to the normal range, the cleaning is deemed incomplete. The system automatically and adaptively adjusts the cleaning time, cleaning speed, and cleaning intensity, and issues a second cleaning instruction until the cleaning effect meets the standard. At the same time, the dust accumulation feature data, trigger threshold parameters, cleaning operation parameters, cleaning effect data, and environmental operating condition data of each operation are fully stored. Through big data machine learning algorithms, the interference filtering threshold, dust accumulation feature judgment parameters, and scene adaptive correction coefficient are continuously iterated and optimized.

[0037] Another technical solution provided by this invention: an intelligent cleaning triggering system based on photovoltaic module output impedance matching, which implements the above method, includes:

[0038] The data acquisition layer includes a single-component intelligent acquisition terminal, a high-precision string inverter, and an integrated meteorological monitoring station, which collects the electrical operating parameters of photovoltaic modules and on-site environmental conditions synchronously and at high frequency across the entire domain.

[0039] The data preprocessing and storage layer serves as the local edge computing gateway for the power plant and the cloud-based operation and maintenance database. It receives data from the underlying layer and is used for data purification, time-series alignment, classified storage, and data retrieval.

[0040] The algorithm operation and judgment layer is used to retrieve standardized data from the database and sequentially complete the following steps: adaptive standard impedance modeling under working conditions, dynamic impedance ratio calculation, three-level layered interference stripping, three-dimensional ash accumulation feature vector extraction, multi-level ash accumulation level judgment, adaptive threshold iterative optimization, and cleaning command logic discrimination.

[0041] The equipment execution and feedback layer includes a cluster of all-domain walking intelligent cleaning robots, a component fault alarm terminal, and an operation and maintenance backend monitoring terminal. It receives various instructions issued by the algorithm judgment layer and executes them in a differentiated manner. It collects robot operating parameters, cleaning operation status, and equipment fault status in real time and transmits them back to the upper layer.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] 1. Completely eliminates the problems of misjudgment and omission in cleaning, achieving precise and on-demand cleaning. This invention can accurately distinguish the performance degradation of components caused by four different factors, triggering cleaning operations only in actual dust accumulation and pollution scenarios. It completely eliminates ineffective cleaning caused by weather fluctuations, component aging, and equipment failure, while avoiding the problem of cleaning delays caused by slight dust accumulation. The accuracy of cleaning triggering is improved by more than 90% compared to traditional technologies, truly achieving refined operation and maintenance of "sweeping when needed and not sweeping when there is no dirt".

[0044] 2. Significantly reduces photovoltaic (PV) operation and maintenance costs, and minimizes equipment wear and power generation losses. It effectively avoids the waste of robot power, mechanical wear, and damage to the glass coating of modules caused by excessive cleaning, while also eliminating power generation efficiency degradation and hot spot repair costs resulting from delayed cleaning. Actual measurements show that after applying this invention, the annual number of ineffective cleanings in PV power plants is reduced by more than 60%, the module operation and maintenance failure rate is reduced by 40%, and the overall energy consumption for cleaning and operation and maintenance is reduced by 35%, significantly improving the economic efficiency of operation and maintenance.

[0045] 3. Strong anti-interference capability and extremely strong versatility across all scenarios. This invention does not rely on visual acquisition hardware and is not affected by complex weather conditions such as light, shadow, fog, haze, rain, and snow. It can be perfectly adapted to various operation and maintenance scenarios such as windy and sandy areas, coastal salt fog areas, and urban dusty areas. It is compatible with all centralized and distributed photovoltaic power stations, solving the industry shortcomings of poor anti-interference of visual technology and weak adaptability of traditional threshold technology.

[0046] 4. Establish a long-term optimization mechanism to continuously improve the power plant's power generation efficiency. Through a fully closed-loop operation and maintenance system and an algorithm self-iteration system, the system can autonomously adapt to scene and seasonal changes, continuously optimize the cleaning judgment logic, realize dynamic upgrades of operation and maintenance strategies, and ensure the long-term power generation performance of photovoltaic modules. This can stably increase the overall annual power generation of the power plant by 3%-8%, creating continuous revenue for the power plant. Detailed Implementation

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The core innovation of this invention is based on the specific coupling mechanism of dust accumulation and impedance in photovoltaic modules: dust contaminants on the surface of photovoltaic modules alter the light transmittance and light absorption efficiency of the cells, directly affecting the generation and transport of photogenerated carriers. This causes regular and continuous steady-state changes in the series and parallel resistances within the module, ultimately manifesting as a stable unidirectional shift in the module's output dynamic impedance. The impedance change mechanisms, patterns, and temporal characteristics corresponding to three types of interference factors—natural aging, environmental weather fluctuations, and electrical faults—are fundamentally and distinguishably different from those caused by dust accumulation. Specifically, the impedance change caused by dust accumulation exhibits slow, steady-state, and positive monotonic shift characteristics; the impedance change caused by weather fluctuations is instantaneous and irregularly oscillating; the impedance change caused by module aging is a long-term baseline drift without short-term abrupt changes; and the impedance change caused by electrical faults is an instantaneous, large jump and non-steady-state shift. Based on this differentiated mechanism, this invention constructs a complete technical chain of "multi-parameter synchronous acquisition - working condition adaptive impedance modeling - dynamic impedance ratio calculation - multi-interference layer-by-layer stripping - dust accumulation feature purification - hierarchical closed-loop triggering - algorithm self-iteration," thoroughly eliminating non-dust accumulation interference factors and achieving accurate identification of pure dust accumulation states. This replaces the traditional coarse judgment mode and effectively solves the core pain points of existing technologies, such as cleaning misjudgment, missed judgment, over-cleaning, and cleaning lag. The overall technical solution of this invention includes five progressive core steps, each configured with dedicated algorithm logic, quantitative parameter standards, and optimization mechanisms. The detailed implementation process is as follows:

[0049] Step 1: High-frequency synchronous acquisition and preprocessing of multi-dimensional electrical and environmental parameters

[0050] The core objective of this step is to construct a high-precision, highly synchronous, and highly reliable multi-dimensional data source system, breaking through the limitations of traditional technologies that only collect single power parameters. This provides comprehensive and accurate data support for subsequent impedance calculations, feature extraction, and interference filtering. This step achieves three-dimensional synchronous acquisition of electrical operating parameters, environmental condition parameters, and equipment operation and maintenance parameters. The specific implementation process is as follows: Intelligent acquisition terminals and high-precision inverter sampling modules are deployed at the photovoltaic string and individual module ends, with a high-frequency sampling frequency of 100ms set to collect core module operating parameters in real time around the clock. These parameters include single module open-circuit voltage Uoc, short-circuit current Isc, operating voltage U, operating current I, and real-time output power P. Simultaneously, high-precision irradiance sensors and temperature sensors are used to collect solar irradiance G, module backsheet temperature T, and ambient temperature Ta in real time. At the same time, the system is connected to the power plant operation and maintenance management platform to retrieve static operation and maintenance ledger parameters in real time, including module factory nominal parameters, module commissioning years, historical fault records, and data on cleaning times and methods. To avoid problems such as acquisition noise, instantaneous operating condition interference, and data timing misalignment, this invention adds a dedicated data preprocessing mechanism: a sliding window filtering algorithm is used to perform mean denoising on 10 consecutive sets of sampled data within 1 second, eliminating abnormal outliers that exceed the normal fluctuation range of parameters; based on a timestamp alignment mechanism, millisecond-level timing synchronization of electrical parameters, environmental parameters, and operation and maintenance parameters is achieved, ensuring the consistency, validity, and accuracy of all input data.

[0051] The three parameters Uoc, Isc, and P are mainly used for validating the legality of operating conditions, calibrating the model, verifying the results, and assisting in fault determination. They are key supporting data to ensure the accuracy of the algorithm.

[0052] The core technological advantages and innovations of this step are reflected in two aspects: First, it breaks through the limitations of traditional single power data judgment and constructs a multi-dimensional three-dimensional data system to provide sufficient feature support for differentiating ash accumulation from various interference factors; Second, through filtering and noise reduction and time-series alignment preprocessing, it reduces data acquisition errors and algorithm judgment errors from the source, laying the foundation for subsequent high-precision impedance calculation and accurate purification of ash accumulation features.

[0053] Step 2: Adaptive Standard Impedance Modeling and Accurate Calculation of Dynamic Impedance Ratio under Operating Conditions

[0054] This step is the core quantitative calculation step of this invention. Its core purpose is to construct a standardized impedance evaluation index unaffected by fluctuations in environmental operating conditions, completely eliminating the interference of environmental factors such as irradiance and temperature on impedance determination, and achieving unified quantitative and accurate evaluation of the dust accumulation state of the module under different complex operating conditions. The specific implementation process is as follows: First, a photovoltaic module operating condition-standard impedance fitting model is built. Based on the module's factory STC standard test parameters and combined with the photovoltaic cell equivalent circuit principle, the optimal output impedance curves of the module corresponding to different irradiance G and different temperatures T are fitted to obtain the model. A full-domain operating condition standard impedance database is constructed, which can accurately match the corresponding module standard matching impedance Z0 according to the real-time operating conditions, completely avoiding the shortcomings of traditional fixed impedance parameters that cannot adapt to dynamic operating conditions. Second, the real-time dynamic output impedance Z of the module is calculated. Based on the pre-processed steady-state operating voltage U and operating current I, the current dynamic output impedance of the module is calculated in real time using the formula Z=U / I, accurately characterizing the module's current true electrical output characteristics. Finally, the core judgment index of this invention—the dynamic impedance ratio K—is constructed, calculated using the formula K=Z / Z0, where Z is the real-time dynamic output impedance of the module, and Z0 is the module standard matching impedance under the same real-time operating conditions. This indicator can normalize and eliminate environmental interference, retaining only the impedance shift characteristics caused by changes in the component's own performance, providing a core quantitative basis for accurate dust accumulation determination.

[0055] Based on the nominal Uoc, Isc, and P values ​​of the components as specified by the manufacturer, the photovoltaic equivalent circuit model parameters are matched, and the standard impedance Z0 database under all operating conditions is corrected to ensure accurate fitting of the standard impedance curves under different irradiance and temperature conditions, thus avoiding modeling deviations from the source.

[0056] Based on extensive simulation experiments and field measurement data calibration, this invention clarifies the impedance ratio differentiation characteristics corresponding to four core operating conditions, achieving preliminary state stratification: First, the dust accumulation and contamination condition, where the module surface transmittance decreases, photocurrent decreases, internal series resistance continuously increases, dynamic impedance rises steadily, impedance ratio K is consistently greater than 1, and there are no drastic fluctuations over a long period, with a slow, monotonically increasing trend; second, the natural aging condition, where cell degradation and encapsulation material aging only cause a slow shift in the impedance baseline, exhibiting monthly or annual long-term drift characteristics, with no significant fluctuations or abrupt changes in the K value in a short period; third, the environmental weather fluctuation condition, where cloud cover, sudden changes in illumination, and temperature fluctuations cause instantaneous, irregular oscillations in the K value, with the fluctuation lasting no more than 3-5 minutes, and no steady-state shift characteristics; fourth, the electrical fault condition, where module loose connections, short circuits, and hot spot faults cause instantaneous, sudden increases or decreases in the K value, resulting in significant abnormal jumps and lacking steady-state continuity. This differentiation characteristic system provides core theoretical support for subsequent interference filtering and dust accumulation characteristic purification.

[0057] Step 3: Multi-interference layer stripping and purification of dust-specific feature vectors

[0058] To completely solve the core defects of existing technologies, such as the inability to remove multiple types of interference factors and low feature recognition, this invention designs a three-level hierarchical interference removal algorithm. It removes interference data step by step according to the progressive logic of "environmental interference - aging interference - fault interference", and finally purifies the exclusive features corresponding only to dust accumulation and pollution, so as to achieve accurate identification of the dust accumulation status of components. Each level of filtering logic operates independently and progresses step by step, without feature cross-interference.

[0059] By utilizing the steady-state variation characteristics of real-time Uoc and Isc, we can distinguish between dust accumulation and electrical faults: dust accumulation only slightly reduces the photocurrent and does not change the steady-state characteristics of the open-circuit voltage; short circuit, loose connection, and hot spot faults will cause Uoc and Isc to jump significantly, which helps to enhance the fault interference filtering effect and supplements the impedance change judgment logic.

[0060] The first level is environmental transient interference filtering: A steady-state quantification threshold is set to filter steady-state data ranges where irradiance fluctuations are ≤5% and component temperature fluctuations are ≤3℃ within a continuous 10-minute period. Abnormal impedance data caused by transient operating condition fluctuations such as rain, cloud cover, and gusty cooling are directly eliminated, retaining only time-series data under steady-state conditions, thus completely eliminating the interference of temporary environmental fluctuations on the judgment results. The second level is component aging baseline offset correction: Based on the actual service life of the components, a preset component aging impedance drift model is called. According to industry-standard photovoltaic component attenuation, the impedance baseline offset of the component corresponding to its service life is accurately calculated. The real-time dynamic impedance ratio is baseline-calibrated, accurately deducting the impedance offset caused by natural aging, and completely separating the characteristic coupling interference of aging attenuation and dust accumulation. The third level is electrical fault feature filtering: An impedance mutation quantification threshold is set. When the impedance ratio change in a single sample exceeds 20%, or when three consecutive sets of sampled data show a large jump, it is determined that the component has an electrical fault or abnormal line contact. Such abnormal data is directly eliminated and the fault location is marked, accurately distinguishing between equipment malfunctions and dust accumulation steady-state offsets.

[0061] After completing the three-level interference stripping, based on the clean steady-state time-series impedance data, three core dust accumulation-specific feature parameters are extracted to construct a three-dimensional quantitative dust accumulation feature vector M = [ΔK, VK, SK]. Here, ΔK is the steady-state average offset of the impedance ratio, used to characterize the overall severity of dust accumulation on the module; VK is the time-series cumulative change rate of the impedance ratio, used to characterize the speed of dust accumulation on the module; and SK is the dust accumulation feature matching degree, used to characterize the similarity between the current impedance time-series characteristics and the standard dust accumulation impedance model. Through this multi-dimensional feature vector, the thickness of dust accumulation, coverage uniformity, and degree of contamination accumulation on the module can be accurately quantified, achieving a technological upgrade from traditional "fuzzy judgment based on power appearance" to "precise quantitative judgment based on the essence of electrical impedance," completely solving the industry pain points of low feature recognition and high misjudgment rate in traditional technologies.

[0062] Impedance ratio steady-state average offset ΔK (characterizing the severity of dust accumulation)

[0063] Suppose that the effective steady-state duration contains n sets of impedance ratio sampling data, the impedance ratio data of the i-th frame is Ki, and the standard impedance ratio in a clean, dust-free state is 1, then the formula for calculating the steady-state average offset is:

[0064]

[0065] This parameter is used to characterize the overall deviation of the current component impedance characteristics relative to the clean standard state. The larger the deviation, the more severe the degree of dust accumulation and contamination of the component.

[0066] Impedance ratio time-series cumulative change rate VK (characterizing the rate of dust accumulation)

[0067] Taking the starting average impedance ratio Kstart and the ending average impedance ratio Kend of the steady-state monitoring period, and the total steady-state monitoring time T in minutes, the formula for calculating the time series change rate is:

[0068]

[0069] This parameter characterizes the cumulative growth rate of ash pollution over time. Only slow, steady positive trends represent true ash accumulation characteristics, while instantaneous jumps and violent fluctuations are considered interference data.

[0070] Ash accumulation feature matching degree SK (characterizing the similarity between true and false ash accumulations)

[0071] Let the mean impedance ratio corresponding to the time series of the standard pure ash accumulation template be Kstd. The degree of matching between the current measured time series curve and the standard ash accumulation model is calculated using Euclidean distance similarity. Normalization yields a matching score ranging from 0% to 100%. The calculation formula is as follows:

[0072]

[0073] The higher the SK value, the higher the consistency between the current impedance time-series change pattern and the steady-state monotonic offset characteristics of real uniform ash accumulation, which can effectively distinguish between four types of states: ash accumulation, weather fluctuations, aging drift, and electrical faults.

[0074] Step 4: Adaptive Hierarchical Intelligent Cleaning Trigger Logic

[0075] This invention abandons the traditional, crude triggering mode based on a single fixed threshold. Based on a multi-dimensional dust accumulation feature vector, it constructs an adaptive three-level hierarchical triggering logic, coupled with a scene-adaptive threshold correction mechanism. This enables refined operation and maintenance control, including "minor warnings, moderate cleaning, heavy emergency cleaning, and fault isolation," and can comprehensively adapt to complex operation and maintenance scenarios in different regions and seasons. The core innovative advantage of this logic lies in its dynamic threshold adaptive adjustment. It can automatically correct the dust accumulation feature judgment threshold based on regional differences such as wind, sand, salt spray, and dust, as well as seasonal differences such as rainy season, windy season, and dry season, completely solving the shortcomings of poor adaptability and weak versatility of traditional fixed thresholds.

[0076] Based on massive amounts of measured data from wind, sand, dust, and salt spray conditions, this invention establishes a quantitative grading threshold system that precisely corresponds to feature vectors. The parameter ranges for each level are clearly defined and have distinct boundaries, with no ambiguous judgment intervals. The specific three-level judgment and triggering mechanisms are detailed below:

[0077] The first level is a mild dust accumulation early warning mechanism: when 0 < ΔK < 10%, 70% ≤ SK < 85%, and VK approaches 0 with no positive continuous growth trend, the module is judged to have mild dust accumulation. Under this condition, the dust accumulation is thin and the coverage is highly uniform, affecting the module's photoelectric conversion efficiency by less than 3%, with no risk of hot spot formation. No cleaning operation is required; the system only generates a dust accumulation early warning log, continuously tracks the dust accumulation trend, predicts subsequent pollution development, achieves preventative monitoring, and avoids gradual dust accumulation.

[0078] The second level is the moderate dust accumulation cleaning trigger mechanism (moderate threshold): when 10%≤ΔK≤25%, 85%≤SK≤95%, VK shows a stable low-speed positive growth, the impedance steady-state deviation lasts for ≥10 minutes, the component power generation shows a slow decline in the same trend, and the three types of interference factors such as environmental fluctuations, component aging, and electrical faults are completely eliminated, it is determined to be effective moderate dust accumulation pollution, and a precise cleaning instruction is immediately generated to dispatch the corresponding area intelligent cleaning robot to start fixed-point cleaning operation.

[0079] The third level is a mechanism to distinguish between severe dust accumulation and faults: When ΔK > 25%, SK > 95%, and steady-state deviation is continuous without oscillation, it is judged as severe dust accumulation pollution, indicating significant power generation loss of the module and extremely high hot spot risk. Emergency cleaning is triggered, and a dedicated differentiated severe cleaning operation mode is adopted. The specific execution rules are as follows: priority is given to regular cleaning tasks, and the corresponding area cleaning robot is dispatched to work first. The robot walking speed is reduced by 30%, and the cleaning brush pressure is increased by 20%. The module surface is re-sweeped twice to completely cover it, thoroughly removing stubborn pollutants such as thick dust accumulation and caking dust, and quickly eliminating power generation loss and hot spot safety hazards caused by severe dust accumulation. If ΔK shows a large instantaneous jump, SK has no dust accumulation characteristic matching degree, and the data has no steady-state continuity, it is judged as an electrical fault of the module. Only a fault warning signal is output, the cleaning command is locked, and invalid cleaning operations are completely eliminated, achieving accurate differentiation between pollution and faults.

[0080] If ΔK also shows a significant deviation greater than 25%, but manifests as a transient jump, SK < 70%, and no steady-state data continuity, it is determined to be an electrical fault in the component, abnormal line contact, or transient extreme operating condition disturbance, not dust accumulation or contamination. Under this condition, the component performance degradation stems from equipment and component damage rather than surface contaminants. Cleaning operations have no repair effect, and mechanical cleaning may exacerbate wear on faulty components and create safety hazards. Therefore, the system only outputs a fault warning signal, generates a fault log to push for manual repair, and forcibly locks the cleaning command, prohibiting any cleaning actions to prevent ineffective and harmful blind cleaning operations. Through the differentiated and refined handling logic of the above two high deviation conditions, the system achieves accurate differentiation between effective cleaning of heavy dust accumulation and electrical fault-prohibited cleaning alarms.

[0081] SK < 70% is the non-dust accumulation interference range and has no dust accumulation characteristics.

[0082] Meanwhile, this step incorporates a scenario-adaptive threshold optimization algorithm. The system can leverage historical operational big data to analyze dust accumulation patterns and environmental conditions across different months and regions, automatically fine-tuning the dust accumulation thresholds at each level. For example, during peak dust storm seasons, the threshold for mild dust accumulation can be appropriately lowered to predict dust accumulation trends in advance and avoid delays in cleaning. Conversely, during periods of clean air, the threshold can be appropriately raised to reduce redundant cleaning operations, maximizing the balance between power plant revenue and operational costs, and significantly improving the algorithm's scenario versatility and operational economy.

[0083] The scene-adaptive threshold optimization algorithm constructs an environmental-dust accumulation feature statistical sample library on a monthly basis. It collects effective dust accumulation feature data (ΔK, VK, SK), meteorological data (irradiance fluctuation, temperature and humidity, frequency of wind and sand) and cleaning efficiency data under all steady-state operating conditions in the current month. It adopts a sliding window 95th percentile statistical correction mechanism to dynamically correct the judgment thresholds corresponding to light, moderate and heavy dust accumulation, so that the thresholds always fit the real dust accumulation distribution pattern of the current season and the current region, avoiding the defects of fixed thresholds being too lenient in the wind and sand season and too strict in the clean season.

[0084] Threshold adaptive correction calculation formula

[0085]

[0086] In the formula: θnew is the updated adaptive threshold; θold is the original baseline threshold of the system; α is the learning correction coefficient, with a value of 0.05-0.1 (small step size for smooth iteration to avoid sudden threshold changes); P 95 is the 95th percentile of the valid ash accumulation feature samples for the current month; μ is the mean of the ash accumulation feature samples for the current month.

[0087] Detailed process of dynamic adjustment of stratified thresholds

[0088] (1) Adaptive correction for seasons with high incidence of sandstorms, dust storms, and haze: When the frequency of sandstorms in a month is greater than 30% and the average dust accumulation rate VK increases significantly, the sample P_95 is too large. The algorithm automatically and slightly lowers the warning threshold for mild dust accumulation and the trigger threshold for moderate cleaning. For example, the warning threshold for mild dust accumulation ΔK is slightly adjusted from 10% to 8% to capture the slow accumulation trend of dust in advance, avoid the cleaning delay and power generation loss caused by the gradual thickening of dust accumulation, and achieve preventive cleaning.

[0089] (2) Adaptive correction for seasons with little rain, cleanliness, and good air quality: When there is no significant dust in a month, the overall offset of the dust accumulation sample is low, and the number of invalid cleanings increases, the algorithm automatically slightly increases the judgment thresholds at each level, and slightly adjusts the trigger threshold of moderate cleaning ΔK from 10% to 12%, filtering out very low levels of slight natural dust accumulation, reducing meaningless redundant cleaning operations, reducing robot maintenance energy consumption and component mechanical wear, and balancing maintenance costs and power generation revenue.

[0090] (3) SK feature matching degree threshold linkage correction: Simultaneously adjust the SK matching degree threshold according to the seasonal pollution type. In the season of uniform floating dust pollution, the SK range is appropriately tightened to improve the recognition accuracy of weak dust accumulation; in the season of local fallen leaves and stains non-uniform pollution, the SK range is appropriately widened to avoid missing the local pollution dust accumulation.

[0091] Iterative constraint mechanism (ensuring system stability and preventing jumps)

[0092] To prevent significant threshold fluctuations from disrupting the decision-making logic, the algorithm incorporates threshold correction limits: a single threshold adjustment cannot exceed 2% of the original baseline threshold, and the annual threshold fluctuation range is controlled within ±4%, ensuring smooth algorithm iteration and continuous, stable decision-making logic. Simultaneously, the system records each threshold correction, creating a traceable threshold iteration log to provide data support for subsequent algorithm optimization.

[0093] Step 5: Closed-loop verification of cleaning effect and iterative optimization of algorithm

[0094] This invention constructs an industry-first "monitoring-judgment-triggering-verification-iteration" fully closed-loop operation and maintenance system, which is different from the traditional open-loop control mode of "triggering to terminate". It quantifies the cleaning effectiveness by comparing the impedance characteristics before and after cleaning, and continuously iterates and optimizes the algorithm model based on full operation and maintenance data to achieve adaptive and continuous upgrading of the cleaning strategy.

[0095] The algorithm model for continuous iterative optimization of full-volume operation and maintenance data is a complete adaptive determination model for photovoltaic ash accumulation impedance characteristics, a combination of a statistical adaptive threshold iteration model and an ash accumulation feature discrimination model, consisting of two core models:

[0096] 1. Core Underlying Structure: Dust Accumulation Multidimensional Feature Discrimination Model (Fixed Mechanism, Continuously Optimized Parameters)

[0097] The core logic of the entire cleaning judgment is as follows: taking the three-dimensional feature vector of ΔK (offset amplitude), VK (cumulative rate), and SK (feature matching degree) as input, and the three-level dust accumulation classification threshold as the judgment boundary, the output is the judgment result of: mild warning / moderate cleaning / heavy emergency cleaning / fault isolation.

[0098] 2. Upper-level iterative optimization: Sliding percentile adaptive threshold correction model (self-iterative core)

[0099] The adaptive threshold optimization algorithm is the core model of "continuous iteration":

[0100] Model type: Time series statistical adaptive regression model (lightweight machine learning model)

[0101] Core mechanism: Monthly sliding window 95th percentile correction

[0102] Iteration objects: all judgment thresholds for three-level dust accumulation (ΔK, SK, steady-state duration, VK growth rate threshold)

[0103] The specific execution process is as follows: After the intelligent cleaning robot completes the standardized cleaning operation of a single area, the system delays for 5 minutes to wait for the component's operating condition to fully stabilize. It then continuously collects steady-state electrical and environmental parameters for the next 30 minutes and performs a secondary calculation of the component's dynamic impedance ratio and dust accumulation feature vector. By comparing the ΔK offset, power recovery rate, and feature matching degree SK before and after cleaning, the cleaning effect is accurately quantified: If the impedance ratio returns to the standard matching range after cleaning, ΔK ≤ 3%, the dust accumulation feature completely disappears, and the power generation returns to the standard level under the same operating conditions, the cleaning is deemed qualified, and the maintenance loop is completed; if there are still obvious dust accumulation features remaining and the impedance offset has not returned to the normal range, the cleaning is deemed incomplete. The system automatically and adaptively adjusts the cleaning duration, cleaning speed, and cleaning intensity, and issues a secondary cleaning command until the cleaning effect meets the standard. Meanwhile, the system builds a dedicated operation and maintenance database, storing all data on dust accumulation characteristics, trigger threshold parameters, cleaning operation parameters, cleaning effect data, and environmental conditions for each operation. Through big data machine learning algorithms, it continuously iterates and optimizes interference filtering thresholds, dust accumulation characteristic judgment parameters, and scene adaptive correction coefficients, enabling the algorithm to autonomously adapt to long-term environmental changes in the power station and the slow degradation of component performance, continuously improving the accuracy of cleaning triggers and achieving intelligent, refined, and adaptive upgrades to photovoltaic cleaning operation and maintenance.

[0104] Example

[0105] This embodiment uses a 10MW centralized ground-mounted photovoltaic power station in a typical wind-blown sandy desert region in northern my country as a specific application scenario. The power station uses monocrystalline silicon PERC modules, which have been in operation for 3 years, belonging to the mid-term stable operation stage. This completely eliminates the performance uncertainties of new modules and the interference of significant degradation of old modules, ensuring high reliability of the test data. This region is characterized by year-round drought, frequent sandstorms, and severe dust pollution. Dust accumulation on the surface of photovoltaic modules is characterized by rapid accumulation, poor coverage uniformity, and significant seasonal differences in pollution, making it a typical operating condition where photovoltaic cleaning problems such as misjudgment, omission, and over-cleaning are most prominent. The original operation and maintenance system of this power station has long used a traditional power attenuation threshold-triggered cleaning strategy, which generally suffers from problems such as fluctuating irradiance on cloudy days, seasonal aging, and mis-cleaning caused by minor faults, as well as power generation loss and hot spot hazards caused by the slow accumulation of light dust. The level of operation and maintenance refinement is low, and the power generation gain is limited. This invention requires no modification to the existing hardware equipment of the power plant. It fully relies on existing component acquisition terminals, string inverters, environmental monitoring stations, intelligent cleaning robot clusters, and back-end operation and maintenance platforms to achieve end-to-end implementation, exhibiting strong compatibility and high engineering feasibility. This embodiment will comprehensively describe the specific implementation process from ten dimensions: system construction, parameter calibration, data acquisition, algorithm calculation, interference removal, feature determination, cleaning triggering, closed-loop verification, algorithm iteration, and comparative experiments.

[0106] S1. Overall system architecture setup and hardware / software parameter initialization and calibration

[0107] First, system initialization and precise parameter calibration were completed: All STC standard parameters for monocrystalline silicon photovoltaic modules across the entire station were entered, with standard operating conditions including an irradiance of 1000W / ㎡, a cell temperature of 25℃, a nominal open-circuit voltage Uoc=41.5V, a short-circuit current Isc=13.8A, and a peak power Pm=550W. Based on the five-parameter equivalent circuit model of the photovoltaic cell, standard impedance mapping surfaces corresponding to full irradiance and full temperature conditions were fitted and generated. A global standard impedance database was constructed for an irradiance range of 200–1200W / ㎡ and a temperature range of 0–60℃, achieving millisecond-level precise table lookup matching of the standard matching impedance Z0 under any real-time operating condition. Subsequently, static operation and maintenance parameters of the power station were entered, including the module's 3-year service life, an annual standard degradation rate of 0.7%, historical fault location records, and annual cleaning records, completing the initialization of the aging baseline model parameters. Finally, the system operating parameters were configured uniformly: electrical and environmental parameter sampling frequency of 100ms, sliding filter window of 10 frames, steady state judgment time of 10min, impedance sudden change fault threshold of 20%, and cleanup qualified impedance offset threshold of 3%, thus completing all pre-calibration work.

[0108] S2. Multi-dimensional data synchronous acquisition and refined preprocessing

[0109] The system initiates a 24 / 7 synchronous data acquisition mechanism, with distributed intelligent acquisition terminals collecting real-time operating voltage U, operating current I, open-circuit voltage Uoc, short-circuit current Isc, and real-time output power P of photovoltaic modules block by block. An integrated weather station simultaneously collects real-time data on horizontal solar irradiance G, module backsheet temperature T, ambient temperature Ta, wind speed, and air cleanliness. All data, carrying high-precision timestamps, is uploaded to the edge gateway. The gateway performs a three-level preprocessing operation: the first level is outlier removal, eliminating abnormal data exceeding the normal operating range of the modules to avoid instantaneous interference errors from the acquisition chip; the second level is moving average filtering, using the average of 10 sets of sampled data within 1 second for noise reduction to eliminate high-frequency micro-fluctuations; the third level is time-series alignment, aligning electrical parameters, environmental parameters, and maintenance parameters at the millisecond level to ensure a one-to-one correspondence between operating conditions and electrical parameters at the same moment. The preprocessed data is stored in real-time in a time-series database, providing a clean, synchronous, and high-precision data source for subsequent steady-state screening, impedance calculation, and feature extraction.

[0110] S3. Real-time calculation of adaptive dynamic impedance ratio and steady-state range selection under operating conditions.

[0111] The algorithm system retrieves preprocessed steady-state data in real time and performs dynamic impedance normalization calculations. First, based on the current real-time irradiance G and module temperature T, it calls the operating condition-impedance matching surface to accurately retrieve the standard matching impedance Z0 under the current operating condition, assuming no dust accumulation and no faults. Then, it calculates the real-time dynamic output impedance Z=U / I based on the module's real-time operating voltage and current. Finally, it calculates the normalized dynamic impedance ratio using the core formula K=Z / Z0. To completely avoid misjudgments caused by instantaneous operating condition fluctuations, the system employs a 10-minute sliding window steady-state screening mechanism, retaining only steady-state data sequences with irradiance fluctuations ≤5% and temperature fluctuations ≤3℃ within a continuous 10-minute period. Non-steady-state operating condition data such as cloud cover, gusts of wind cooling, and sudden changes in light intensity are removed, ensuring that the impedance ratio data used for judgment only reflects the module's own continuous performance changes, rather than temporary environmental fluctuations. During the actual measurement process in this embodiment, the power station irradiance was stably maintained at 820–850 W / m² from 10:00 to 10:10 am, and the module temperature was 32–34°C. The operating conditions were stable, and the effective steady-state time series data sequence was obtained through screening, which was used for subsequent ash accumulation feature extraction.

[0112] S4. Three-level layered interference stripping and precise purification of accumulated dust feature vectors

[0113] This embodiment strictly implements a three-level progressive interference stripping algorithm, eliminating and correcting non-dust-accumulated interference one by one to purify pure dust-accumulated characteristics. The first level is environmental interference stripping: through steady-state interval screening, instantaneous weather fluctuations have been completely eliminated, and the current data shows no environmental oscillation characteristics. The second level is aging baseline correction: based on the 3-year operation period of the modules, the aging drift model is retrieved to calculate the annual impedance baseline offset of this batch of modules. The real-time impedance ratio is baseline calibrated, deducting the static offset caused by natural aging, and completely separating aging degradation and dust accumulation characteristics. The third level is fault interference filtering: the system monitors the impedance ratio change amplitude frame by frame. All sampled data in this period show single-frame fluctuations of less than 5%, no jumps exceeding 20%, and no intermittent abnormal fluctuations, completely inconsistent with fault characteristics such as module short circuits, loose connections, and hot spots, indicating no electrical fault interference. After completing the three-level interference stripping, the current time-series data is free of environmental, aging, and fault-related interferences; the data change characteristics are solely caused by surface dust accumulation.

[0114] Based on pure steady-state time-series data, the system automatically calculates the three-dimensional ash accumulation feature vector M = [ΔK, VK, SK]. In this embodiment, the measured steady-state average offset of impedance ratio ΔK = 18%, the cumulative change rate of impedance time series VK shows a steady and slow upward trend, and the ash accumulation feature matching degree SK matches the standard ash accumulation steady-state offset model with a matching degree of up to 96%. There is no oscillation, no jump, and no baseline drift. The features are highly consistent with the characteristics of uniform ash accumulation pollution, completely different from the electrical characteristics of weather fluctuations, aging attenuation, and equipment failure, and accurately determine that the current component has moderate uniform ash accumulation pollution.

[0115] S5. Adaptive hierarchical triggering logic judgment and cleaning command issuance.

[0116] This embodiment employs an adaptive optimization threshold system for windy and sandy areas, executing a three-level refined judgment logic: when ΔK < 10%, it is a light dust accumulation warning zone, only monitoring without cleaning; when 10% ≤ ΔK ≤ 25%, it is a moderate dust accumulation effective cleaning zone, triggering standard cleaning operations; when ΔK > 25%, it is a heavy dust accumulation emergency cleaning zone, initiating an expedited cleaning strategy. In this actual test sample, ΔK = 18%, stably within the moderate dust accumulation zone, and the continuous steady-state duration meets the 10-minute threshold, with power gradually and regularly decreasing, feature matching meeting the standard, and no interference factors. The system determines this as genuine and effective dust pollution, necessitating cleaning. A precise cleaning command is then generated, accurately locating the string and area containing the contaminated components, and scheduling the corresponding track-mounted intelligent cleaning robot to initiate standardized cleaning operations. The default cleaning parameters are configured as standard travel speed, standard brush pressure, full-area full-coverage cleaning mode, and a preset cleaning duration of 15 minutes.

[0117] To highlight the advantages of this invention, this embodiment simultaneously sets up a comparative discrimination mechanism: During the same period, some power components in the power plant experience a brief power drop due to cloud cover. Traditional power threshold algorithms would directly identify this as dust accumulation and trigger cleaning. However, this invention, through impedance feature identification, determines that this is an instantaneous, irregular oscillation without steady-state offset and with extremely low dust accumulation feature matching, directly filtering out interference and not triggering cleaning, thus eliminating misjudgments due to weather interference at the instance level. Furthermore, some components exhibit slight aging power attenuation. This invention, through baseline correction to deduct aging offset, does not classify this as dust accumulation, completely resolving the problem of misjudgment due to aging.

[0118] S6. Steady-state closed-loop verification and quantitative evaluation of cleaning effect after cleaning is completed.

[0119] After the robot completes a 15-minute standardized full-area cleaning operation, the system delays for 5 minutes to allow the airflow on the component surface to stabilize and the component temperature and electrical conditions to return to a steady state. Then, it initiates 30 minutes of continuous monitoring and data acquisition, performing a second calculation of the dynamic impedance ratio and dust accumulation feature vector after cleaning to quantify the cleaning and repair effect. Post-cleaning test data shows that the component's dynamic impedance ratio K has returned to the standard range, and the impedance offset ΔK has decreased from 18% before cleaning to 2.1%, meeting the ≤3% cleaning qualification threshold. The original dust accumulation steady-state offset characteristics have completely disappeared, and the feature matching degree SK has decreased to the invalid range. The component's power generation has increased by 6.3% compared to before cleaning, fully restoring to the standard power generation level under the same operating conditions. The system determines that this cleaning operation was thorough and the repair met the standards, completing a single "monitoring-judgment-triggering-verification" complete maintenance loop, eliminating the need for a second cleaning.

[0120] S7, Big Data Sample Accumulation and Algorithm Adaptive Iterative Upgrade

[0121] The system stores the complete maintenance sample data, including: dust accumulation feature vector parameters before cleaning, environmental condition parameters, seasonal and monthly labels, trigger threshold parameters, robot cleaning operation parameters, impedance comparison data before and after cleaning, power recovery gain data, and cleaning qualification judgment results. Based on massive historical samples from windy and dusty seasons, the system adaptively fine-tunes the regional judgment threshold through machine learning: for periods of high wind and dust intensity, it slightly lowers the warning threshold for mild dust accumulation to proactively capture slow dust accumulation trends and avoid cleaning delays; for dry and dusty weather, it optimizes the steady-state duration judgment parameters to improve the sensitivity of weak dust accumulation feature recognition. Through continuous sample accumulation and iterative optimization, the algorithm can adapt to local environmental patterns over the long term, achieving an adaptive upgrade characteristic that becomes more accurate with use, unlike traditional fixed-threshold algorithms that cannot self-optimize.

[0122] S8. Comparison test and quantitative analysis of the effects of this embodiment with traditional technology

[0123] To objectively verify the advanced nature of this invention, this embodiment sets up a control group test with modules from the same power plant, at the same time, and in the same batch. The method of this invention and the traditional power attenuation threshold cleaning method are continuously compared and tested for 30 days. The quantitative results are as follows: The traditional power threshold solution triggers cleaning 41 times in 30 days, of which only 16 times are effective dust cleaning. 25 times are false cleaning caused by weather fluctuations, module aging, and slight electrical disturbances, with a false judgment rate as high as 38.0%. The proportion of ineffective cleaning is extremely high, resulting in a large amount of energy consumption and equipment wear. The impedance matching intelligent triggering solution of this invention triggers cleaning 17 times in 30 days, all of which are real dust cleaning, with 0 false triggers. The cleaning false judgment rate is reduced to 0%. Only 2 times are there slight dust accumulations that do not reach the cleaning threshold, but only warnings are issued and cleaning is not performed, achieving precise on-demand operation and maintenance. In terms of economic benefits, the solution of this invention reduces monthly robot operation and maintenance energy consumption by 36%, eliminates additional mechanical wear on components, eliminates ineffective cleaning losses, and increases the overall power generation of the power station by 6.2% per month. Long-term operation can effectively delay component aging and reduce the incidence of hot spot failures. The precision, intelligence and economy of operation and maintenance are comprehensively superior to existing traditional technologies.

[0124] This invention features a unique core principle for dust accumulation identification based on output impedance matching, overturning the traditional single-dimensional judgment method. Existing technologies rely solely on external phenomena such as time and power generation to determine the timing of cleaning, failing to distinguish the multiple causes of power degradation. This invention, starting from the electrical core characteristics of photovoltaic modules, uncovers the specific steady-state impedance shift patterns of dust accumulation, accurately distinguishing the differentiated impedance characteristics of four states: dust contamination, natural aging, weather fluctuations, and electrical faults. This fundamentally solves the core problem of misjudgment during cleaning from a technical principle perspective.

[0125] This invention constructs an adaptive dynamic impedance ratio quantification evaluation system to achieve precise quantification of dust accumulation. Overcoming the coarse-grained nature of traditional fixed threshold judgments, it precisely quantifies dust accumulation thickness, contamination level, and accumulation rate by matching the normalized ratio of standard impedance and dynamic output impedance under real-time operating conditions, combined with temporal variation characteristics. This replaces the fuzzy judgment method based solely on power attenuation values, significantly improving the accuracy of dust accumulation status identification.

[0126] This invention designs a three-level, multi-dimensional interference stripping and feature purification mechanism to achieve pure dust accumulation feature extraction. It targets and layers the filtering of three core interference factors: instantaneous weather fluctuations, natural component aging, and equipment electrical faults. It completely strips away non-dust-accumulated performance degradation features and purifies specific dust-accumulated electrical features, eliminating false cleaning triggers caused by non-contamination factors at the data level. This solves the fundamental deficiency of existing technologies in being unable to identify the causes of power degradation.

[0127] This invention establishes a fully closed-loop operation and maintenance system with self-iterative algorithms, enabling adaptive upgrades of operation and maintenance strategies. Unlike the open-loop control mode of existing technologies, this invention verifies the effect by comparing impedance characteristics before and after cleaning. At the same time, it continuously iterates and optimizes the algorithm thresholds and judgment logic based on historical operation and maintenance big data, which can adaptively adapt to the operation and maintenance needs of different scenarios, different seasons, and different component aging states, and continuously optimize the cleaning strategy.

[0128] Embodiments of the present invention also provide an intelligent cleaning triggering system based on photovoltaic module output impedance matching to implement the above method. The system consists of a four-layer architecture: a bottom data acquisition layer, a data preprocessing and storage layer, an algorithm calculation and judgment layer, and a device execution and feedback layer.

[0129] 1. Bottom Data Acquisition Layer (System Sensing Input End)

[0130] This layer is the system's sole data source and the foundational sensing unit for accurate algorithm judgments. Its hardware primarily includes a single-module intelligent data acquisition terminal, a high-precision string inverter, and an integrated meteorological monitoring station. Its core function is to collect photovoltaic module electrical operating parameters and on-site environmental condition parameters synchronously and across the entire domain, at high frequency. Specifically, this includes multi-dimensional data such as module operating voltage, operating current, open-circuit voltage, short-circuit current, real-time output power, solar irradiance, module backsheet temperature, ambient temperature, and wind speed. This layer lacks data processing capabilities; it only handles real-time sampling and timestamped uploading of raw data. It can accurately capture subtle electrical changes in the modules and dynamic changes in environmental conditions, providing the upper-level algorithms with a raw, comprehensive, and complete data source.

[0131] Interaction with other layers: One-way uplink output, continuously pushing raw operating conditions and electrical data with high-precision timestamps to the data preprocessing and storage layer in real time, without downlink command receiving function, providing basic data support for the entire system.

[0132] Specifically, the data collected by the single-module intelligent acquisition terminal includes: electrical parameters of a single photovoltaic module (the most core and finest-grained data). Specific data collected include: single-module operating voltage U, operating current I, module open-circuit voltage Uoc, short-circuit current Isc, and single-module real-time output power P. Its function is to achieve precise sensing of a single module. Unlike traditional string-wide sampling, it can capture subtle impedance shifts caused by minute dust accumulation on a single module. This is the core hardware foundation for the invention to accurately identify light and uniform dust accumulation.

[0133] The high-precision string inverter collects the following data: string-level reference electrical data and operating condition verification data. Specific data collected includes: total string output power, total string current, total voltage, auxiliary verification of single-component data validity, and abnormal switching identification. Its function is to verify data and filter anomalies, assist in judging whether the terminal data of a single component is distorted, whether there is a sampling anomaly, and whether the overall string-level operating condition is normal. It also provides a string-level reference for three-level interference stripping and fault diagnosis.

[0134] The integrated meteorological monitoring station collects environmental operating condition parameters (used to correct the standard impedance Z0). Specific data collected include: solar irradiance G, component backplane temperature T, ambient temperature Ta, ambient wind speed, and auxiliary meteorological parameters such as air cleanliness. Its function is to provide all environmental inputs for the adaptive standard impedance modeling of this invention, and to match the standard optimal impedance Z0 under the current operating condition in real time. It completely eliminates the interference of weather, temperature, and light changes on the judgment, and is the key environmental data source for realizing operating condition normalization and anti-interference judgment.

[0135] 2. Data Preprocessing and Storage Layer (System Data Hub Layer)

[0136] This layer of hardware consists of a local edge computing gateway for the power plant and a cloud-based time-series operation and maintenance database. It receives data from the underlying layer and undertakes four core functions: data purification, time-series alignment, categorized storage, and data retrieval. First, it removes outliers and uses moving average filtering to reduce noise in the uploaded raw data, eliminating collected noise and abnormal outliers. Second, it performs millisecond-level time-series alignment of electrical parameters, environmental parameters, and operation and maintenance parameters, resolving the issue of time-series misalignment in sampling from multiple devices. Finally, it stores the pre-processed, clean, structured data in real time, while also recording operation and maintenance data such as component static parameters, aging logs, historical cleaning records, and fault logs, constructing an integrated database of operating conditions, electrical systems, and operation and maintenance.

[0137] Interaction with other layers: Two-way data interaction. Uplink: Real-time push of stable, clean, and aligned standardized data to the upper-level algorithm operation and judgment layer; Downlink: Real-time reception of raw sampling data from the lower-level acquisition layer, and simultaneously reception and permanent archiving of judgment results, threshold iteration data, and operation and maintenance log data issued by the algorithm layer, achieving full-process data traceability.

[0138] The local edge computing gateway (field core data processing unit) is deployed at the photovoltaic power station site. It is a core relay and preprocessing hardware unit that connects the underlying acquisition hardware with the upper-layer algorithms and cloud databases. It undertakes four core functions: real-time processing of field data, time synchronization, command relay, and local computing power support. It is a key device to ensure low latency and high stability of the system. Its specific functions are as follows:

[0139] 1. Real-time data preprocessing and purification: Receives raw electrical and environmental parameters uploaded by single-component intelligent acquisition terminals, high-precision inverters, and meteorological monitoring stations in real time. Performs outlier removal, high-frequency noise reduction, and moving average filtering preprocessing operations to automatically filter out invalid and abnormal data caused by instantaneous equipment interference and sudden changes in outdoor operating conditions, retaining only valid and accurate sampled data to avoid algorithm misjudgments caused by noisy data from the source.

[0140] 2. Multi-source data time sequence alignment and synchronization: To address the sampling time sequence deviation problem of the three types of acquisition devices, a high-precision timestamp mechanism is used to perform millisecond-level time sequence alignment of electrical parameters, environmental operating condition parameters, and operation and maintenance static parameters. This ensures that all parameters are the corresponding operating condition data at the same time when the algorithm is calculated, and completely solves the technical problems of data misalignment and operating condition mismatch among multiple devices.

[0141] 3. Local low-latency algorithm operation support: It undertakes the on-site edge computing work of the core algorithm of this invention, and completes core logic operations such as dynamic impedance calculation, impedance ratio normalization processing, three-level interference stripping, dust accumulation feature vector extraction, graded threshold determination, and cleaning command generation in real time. It does not rely on remote computing power in the cloud, which greatly reduces the computing latency and ensures that cleaning commands are issued in real time and responded to in a timely manner.

[0142] 4. Two-way command and data relay: Receives full data collection from the underlying hardware and pushes pre-processed standardized data and algorithm judgment results upwards; forwards threshold iteration parameters and operation and maintenance configuration commands issued by the cloud downwards and transmits on-site operation status data upwards. It also has a local caching function when the network is down, automatically caching on-site data and calculation results when the network is abnormal, and synchronizing them in batches after the network is restored to ensure uninterrupted system operation.

[0143] The cloud-based time-series operation and maintenance database (global data storage and iteration support unit) is the core carrier for system global data storage, sample accumulation, and parameter iteration. It is specifically adapted to the time-series and continuous operation data characteristics of photovoltaics. Unlike ordinary databases, it has core capabilities such as efficient storage of time-series data, full-process data traceability, model iteration sample support, and global parameter archiving. Its specific functions are as follows:

[0144] 1. Permanent archiving of full-volume structured data: Long-term storage of all steady-state operating data of the power plant, electrical characteristic parameters, ash accumulation feature vectors (ΔK, VK, SK), environmental meteorological data, robot cleaning operation parameters, comparison data of cleaning effects before and after cleaning, fault logs, component aging baseline data, threshold iteration parameters and other full-dimensional operation and maintenance data, to build a complete power plant operation and maintenance big data sample library.

[0145] 2. Provide core sample support for the adaptive iterative algorithm: Provide all historical sample data for the self-iterative optimization of the lightweight dust accumulation feature statistical adaptive discrimination model of this invention. Summarize seasonal working condition samples such as wind and sand, dust, temperature and humidity on a monthly basis, and statistically analyze core parameters such as 95th percentile and sample mean. Support the calculation formula for dynamic correction of threshold, realize the adaptive fine adjustment of the judgment threshold with the season and scenario, and provide the data foundation for long-term optimization of the system.

[0146] 3. Provides real-time retrieval service for static baseline parameters: Permanently stores static core parameters such as component STC standard parameters, annual aging attenuation parameters, full-domain operating condition standard impedance surface database, and initial benchmark judgment threshold, providing continuous and stable parameter retrieval support for real-time impedance matching, aging baseline correction, and operating condition normalization calculation, ensuring that the algorithm judgment benchmark is consistent and accurate.

[0147] 4. Operation and maintenance closed-loop traceability and effect quantification statistics: Archive the full-process data of each cleaning judgment, instruction issuance, operation execution and effect verification to realize the traceability of the operation and maintenance process and the quantification of cleaning effect. At the same time, it provides data support for monthly and annual operation and maintenance data analysis, power generation gain statistics and invalid operation review, and improves the system's closed-loop operation and maintenance system.

[0148] The local edge computing gateway is responsible for real-time processing, real-time computation, and real-time command interaction on-site, ensuring system response speed and on-site operational stability, and undertaking short-term, dynamic, and real-time business. The cloud-based time-series operation and maintenance database is responsible for long-term data accumulation, model iteration, parameter archiving, and data traceability, ensuring long-term system optimization and data verifiability, and undertaking long-term, static, and iterative business. The two interact bidirectionally in real-time, forming a collaborative mode of "real-time on-site computation + cloud-based iterative optimization," supporting the complete closed-loop operation and maintenance logic of this invention, which includes perception, processing, decision-making, execution, feedback, and iteration.

[0149] 3. Algorithm Calculation and Judgment Layer (Core Layer of System Intelligent Decision-Making)

[0150] This layer is the core brain of the system, consisting of an edge algorithm program equipped with the invention's proprietary lightweight dust accumulation feature statistical adaptive discrimination model. It is a core innovative layer that distinguishes it from traditional fixed-threshold cleaning technologies. Its core functions include retrieving standardized data from the database, sequentially performing adaptive standard impedance modeling, dynamic impedance ratio calculation, three-level layered interference stripping, three-dimensional dust accumulation feature vector extraction, multi-level dust accumulation level determination, adaptive threshold iterative optimization, and cleaning command logic discrimination. Ultimately, it outputs precise decision commands for four categories: early warning, regular cleaning, emergency deep cleaning, and fault-based no-sweep, achieving intelligent transformation from raw data to operational decisions.

[0151] The algorithm operation and judgment layer is equipped with an impedance matching dust accumulation identification algorithm program, which is the edge-end intelligent decision program of this invention. Through working condition adaptive impedance modeling and multi-dimensional electrical feature analysis, it removes all non-dust accumulation interference, accurately quantifies the degree of dust accumulation and pollution of components, and outputs differentiated cleaning / alarm instructions. At the same time, it has the ability to adapt and iteratively optimize the scene, which solves the core defects of traditional cleaning technology such as misjudgment, missed judgment, and poor adaptability in principle.

[0152] Interaction with other layers: Global bidirectional core interaction. Downlink: Real-time retrieval of structured steady-state data and historical operation and maintenance sample data from the preprocessed storage layer to support algorithm computation and model iteration; Uplink: Accurately disseminating various final decision instructions to the device execution and feedback layer; Simultaneously, transmitting each judgment result, threshold iteration parameters, and feature operation data back to the data storage layer for archiving, forming a data closed loop for algorithm self-iteration.

[0153] 4. Equipment Execution and Feedback Layer (System Implementation Terminal Layer)

[0154] This layer of hardware consists of a cluster of intelligent robotic sweepers capable of traversing the entire area, component fault alarm terminals, and operation and maintenance backend monitoring terminals. It serves as the final execution unit for system decisions. Its core function is to accurately receive and execute various instructions from the algorithm-based judgment layer: upon receiving a light dust accumulation warning instruction, it completes ledger recording and displays the warning in the backend; upon receiving a moderate regular cleaning instruction, it initiates standard mode fixed-point full-coverage cleaning; upon receiving a heavy dust accumulation instruction, it executes expedited, low-speed, and pressurized deep cleaning modes; upon receiving an electrical fault judgment instruction, it forcibly locks the cleaning action, triggers fault alarms, and pushes manual repair work orders. Simultaneously, it collects robot operating parameters, cleaning operation status, and equipment fault status in real time and transmits this data back to the upper layer.

[0155] Interaction with other layers: Two-way closed-loop interaction. Downlink: Accurately receives various decision instructions from the algorithm judgment layer and executes them in a differentiated manner; Uplink: Feeds back the cleaning operation completion status, equipment operation data, and operation effect information to the data storage layer and algorithm layer in real time, providing measured feedback data for subsequent closed-loop verification of cleaning effect and adaptive iterative optimization of the algorithm.

[0156] The overall four-layer architecture workflow logic is as follows: The four-layer architecture involves data input, decision-making and calculation, and execution from top to bottom, and feedback of work results, accumulation of sample data, and support for model iteration from bottom to top, forming a complete closed loop of perception-processing-decision-execution-feedback-iteration. Each layer has a clear division of labor and fixed interaction logic, with no data chaos and no instruction conflicts, ensuring the long-term stable, accurate, and adaptive operation of the system.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A smart cleaning triggering method based on photovoltaic module output impedance matching, characterized in that, Includes the following steps: (1) The electrical operating parameters, environmental operating parameters, and equipment operation and maintenance parameters of the photovoltaic modules are collected in three dimensions synchronously, and the raw collected data is preprocessed. (2) A photovoltaic module operating condition-standard impedance fitting model was built. Based on the standard test parameters of the module's factory STC and combined with the equivalent circuit principle of the photovoltaic cell, the optimal output impedance curve of the module corresponding to different irradiance G and different temperatures T was obtained. A standard impedance database of the whole domain operating conditions was constructed, the real-time dynamic output impedance of the module was calculated, and the core judgment index—dynamic impedance ratio K was constructed. The calculation formula is K=Z / Z0, where Z is the real-time dynamic output impedance of the module and Z0 is the standard matching impedance of the module under the same real-time operating conditions. (3) Through the three-level layered interference stripping algorithm, interference data is removed step by step according to the progressive logic of "environmental interference - aging interference - fault interference". Finally, the exclusive features corresponding only to ash pollution are purified. After the three-level interference stripping is completed, based on the pure steady-state time-series impedance data, three core ash-specific feature parameters are extracted to construct a three-dimensional quantitative ash feature vector M = [ΔK, VK, SK], where ΔK is the steady-state average offset of impedance ratio, VK is the time-series cumulative change rate of impedance ratio, and SK is the ash feature matching degree. (4) Based on the multidimensional ash accumulation feature vector, an adaptive three-level hierarchical triggering logic is constructed, and a scene adaptive threshold correction mechanism is provided to realize refined operation and maintenance control of "minor warning, moderate cleaning, heavy emergency cleaning, and fault isolation". (5) Construct a closed-loop operation and maintenance system of "monitoring-judgment-triggering-verification-iteration" to quantify the cleaning effectiveness by comparing the impedance characteristics before and after cleaning, and continuously iterate and optimize the algorithm model based on the full amount of operation and maintenance data to achieve adaptive and continuous upgrading of the cleaning strategy.

2. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, characterized in that, The electrical operating parameters mentioned in step (1) include single component open-circuit voltage Uoc, short-circuit current Isc, operating voltage U, operating current I, and real-time output power P; the environmental operating parameters include solar irradiance G, component backplane temperature T, and ambient temperature Ta, which are collected in real time by high-precision irradiance sensors and temperature sensors; the equipment operation and maintenance parameters include component factory nominal parameters, component commissioning years, historical fault records, and data on cleaning times and cleaning methods; the preprocessing includes: using a sliding window filtering algorithm to perform mean denoising on the sampled data and removing abnormal outliers that exceed the normal fluctuation range of parameters; and achieving millisecond-level time synchronization of electrical parameters, environmental parameters, and operation and maintenance parameters based on a timestamp alignment mechanism.

3. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, characterized in that, In step (2), based on the pre-processed steady-state operating voltage U and operating current I, the current dynamic output impedance Z of the component is calculated in real time using the formula Z=U / I, so as to accurately characterize the current real electrical output characteristics of the component.

4. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, characterized in that, The three-level hierarchical interference stripping algorithm in step (3) includes: The first level is environmental transient interference filtering: set the steady-state quantitative judgment threshold, filter the steady-state data range with irradiance fluctuation amplitude ≤5% and component temperature fluctuation ≤3℃ within 10 consecutive minutes, directly remove abnormal impedance data caused by transient operating condition fluctuations such as rain, cloud cover, and gust cooling, and only retain the time series data under steady-state operating conditions, completely eliminating the interference of temporary environmental fluctuations on the judgment results. The second level is the component aging baseline offset correction: based on the actual years of operation of the component, the preset component aging impedance drift model is called, and according to the general photovoltaic component attenuation standard, the impedance baseline offset of the component corresponding to the years of use is accurately calculated. The real-time dynamic impedance ratio is baseline calibrated, the impedance offset caused by natural aging is accurately deducted, and the characteristic coupling interference of aging attenuation and dust accumulation is completely separated. The third level is electrical fault feature filtering: set an impedance change quantification judgment threshold. When the change in impedance ratio of a single sampling exceeds 20%, or when there is a large jump in three consecutive sets of sampling data, it is determined that the component has an electrical fault or abnormal line contact. Such abnormal data is directly removed and the fault location is marked.

5. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, characterized in that, In step (3), it is assumed that the effective steady-state duration contains n sets of impedance ratio sampling data, and the impedance ratio data of the i-th frame is K. i The standard impedance ratio is 1 under clean, dust-free conditions. The formula for calculating the steady-state average offset ΔK of the impedance ratio is: ； Taking the initial average impedance ratio Kstart and the final average impedance ratio Kend during the steady-state monitoring period, and the total steady-state monitoring duration as T, the formula for calculating the time-series cumulative rate of change of impedance ratio VK is as follows: ； Let the mean impedance ratio corresponding to the time series of the standard pure ash accumulation template be Kstd. The degree of matching between the current measured time series curve and the standard ash accumulation model is calculated using Euclidean distance similarity. Normalization yields a matching score ranging from 0% to 100%. The formula for calculating the ash accumulation feature matching degree SK is as follows: 。 6. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, wherein the three-level hierarchical triggering in step (4) includes: The first level is a mild dust accumulation early warning mechanism: when 0 < ΔK < 10%, 70% ≤ SK < 85%, and VK approaches 0 with no positive continuous growth trend, the component is determined to have mild dust accumulation. No cleaning operation needs to be initiated; the system only generates a dust accumulation early warning log, continuously tracks the dust accumulation trend, predicts subsequent pollution development, achieves preventative monitoring, and avoids gradual dust accumulation. The second level is the moderate dust accumulation cleaning trigger mechanism: when the following conditions are met: 10%≤ΔK≤25%, 85%≤SK≤95%, VK shows a stable low-speed positive growth, the impedance steady-state deviation lasts for ≥10 minutes, the power generation of the module shows a slow decline in the same trend, and the three types of interference factors such as environmental fluctuations, module aging, and electrical faults are completely eliminated, it is determined to be effective moderate dust accumulation pollution. A precise cleaning instruction is immediately generated, and the corresponding intelligent cleaning robot is dispatched to start fixed-point cleaning operation. The third level is a mechanism to distinguish between severe dust accumulation and faults: When ΔK > 25%, SK > 95%, and steady-state deviation is continuous without oscillation, it is judged as severe dust accumulation pollution, the power generation loss of the module is significant and there is a high risk of hot spots, and an emergency cleaning operation is triggered. If ΔK shows a large instantaneous jump, SK has no dust accumulation characteristic matching degree, and the data has no steady-state continuity, it is judged as an electrical fault of the module, only a fault warning signal is output, the cleaning command is locked, and invalid cleaning operations are completely eliminated, so as to achieve accurate differentiation between pollution and faults.

7. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, wherein step (4) incorporates a scene adaptive threshold optimization algorithm, which relies on historical operation and maintenance big data to statistically analyze the dust accumulation patterns and environmental conditions of different months and regions, and automatically fine-tunes the dust accumulation judgment thresholds at each level.

8. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 7, wherein the scene adaptive threshold optimization algorithm constructs an environmental-dust accumulation feature statistical sample library on a monthly basis, collects effective dust accumulation feature data ΔK, VK, SK, meteorological data, and cleaning efficiency data under all steady-state operating conditions in the current month, and adopts a sliding window 95th percentile statistical correction mechanism to dynamically correct the judgment thresholds corresponding to light, moderate, and heavy dust accumulation, so that the thresholds always conform to the actual dust accumulation distribution pattern of the current season and the current region. Threshold adaptive correction calculation formula: , In the formula: θnew is the updated adaptive threshold; θold is the original baseline threshold of the system; α is the learning correction coefficient, with a value of 0.05~0.1; P 95 is the 95th percentile of the valid ash accumulation feature samples for the current month; μ is the mean of the ash accumulation feature samples for the current month.

9. The intelligent cleaning triggering method based on photovoltaic module output impedance matching according to claim 1, wherein step (5) includes: After the intelligent cleaning robot completes a standardized cleaning operation in a single area, the system delays for 5 minutes to allow the component's operating condition to fully stabilize. It then continuously collects steady-state electrical and environmental parameters for the next 30 minutes, and performs a secondary calculation of the component's dynamic impedance ratio and dust accumulation feature vector. By comparing the ΔK offset, power recovery rate, and feature matching degree SK before and after cleaning, the system accurately quantifies the cleaning effect. If the impedance ratio returns to the standard matching range, ΔK ≤ 3%, the dust accumulation feature completely disappears, and the power generation returns to the standard level under the same operating conditions after cleaning, the cleaning is deemed qualified, and the maintenance loop is completed. If there are still obvious dust accumulation features remaining and the impedance offset has not returned to the normal range, the cleaning is deemed incomplete. The system automatically and adaptively adjusts the cleaning duration, cleaning speed, and cleaning intensity, and issues a secondary cleaning command until the cleaning effect meets the standard. At the same time, the system stores all dust accumulation feature data, trigger threshold parameters, cleaning operation parameters, cleaning effect data, and environmental condition data for each operation. Through big data machine learning algorithms, the system continuously iterates and optimizes the interference filtering threshold, dust accumulation feature judgment parameters, and scene adaptive correction coefficients.

10. A smart cleaning triggering system based on photovoltaic module output impedance matching, implementing the method described in any one of claims 1-9, characterized in that, include: The data acquisition layer includes a single-component intelligent acquisition terminal, a high-precision string inverter, and an integrated meteorological monitoring station, which collects the electrical operating parameters of photovoltaic modules and on-site environmental conditions synchronously and at high frequency across the entire area. The data preprocessing and storage layer serves as the local edge computing gateway for the power plant and the cloud-based operation and maintenance database. It receives data from the underlying layer and is used for data purification, time-series alignment, classified storage, and data retrieval. The algorithm operation and judgment layer is used to retrieve standardized data from the database and sequentially complete the following steps: adaptive standard impedance modeling under working conditions, dynamic impedance ratio calculation, three-level layered interference stripping, three-dimensional ash accumulation feature vector extraction, multi-level ash accumulation level judgment, adaptive threshold iterative optimization, and cleaning command logic discrimination. The equipment execution and feedback layer includes a cluster of all-domain walking intelligent cleaning robots, a component fault alarm terminal, and an operation and maintenance backend monitoring terminal. It is used to receive various instructions issued by the algorithm calculation and judgment layer and execute them in a differentiated manner, and to collect robot operating parameters, cleaning operation status, and equipment fault status in real time and transmit them back to the upper layer.

Images