Photovoltaic performance anomaly diagnosis method and system based on continuous weather characteristic coefficient

Abstract

The application discloses a photovoltaic performance anomaly diagnosis method and system based on continuous weather characteristic coefficients, and mainly relates to the technical field of distributed photovoltaic intelligent operation and maintenance. The method comprises the following steps: acquiring actual operation data of a photovoltaic unit, calculating a normalized weather characteristic coefficient representing continuous weather changes based on the rated power of the photovoltaic unit under standard conditions and actual solar radiation and component temperature, constructing a dynamic theoretical power reference surface based on the normalized weather characteristic coefficient and the component temperature, acquiring real-time measured power of the photovoltaic unit, calculating the residual error of the real-time measured power and the corresponding theoretical power on the dynamic theoretical power reference surface, and decoupling and diagnosing multiple types of faults of the photovoltaic unit according to the dynamic characteristics of the normalized weather characteristic coefficient and the change trend of the residual error. The application has the beneficial effects that it overcomes the defects of discrete weather evaluation dimension, non-smooth reference and high dependence on hardware in the existing photovoltaic performance anomaly diagnosis technology.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic intelligent operation and maintenance technology, specifically a photovoltaic performance anomaly diagnosis method and system based on continuous weather characteristic coefficients. Background Technology

[0002] As a crucial component of the new power system, real-time monitoring and efficiency early warning of photovoltaic power generation are essential for ensuring investment returns. Traditional photovoltaic operation and maintenance early warning technologies have evolved from manual inspections to threshold alarms, and then to digital monitoring. However, in current engineering practice, mainstream efficiency assessment methods still have significant shortcomings.

[0003] Currently, the discretization distortion of photovoltaic power generation evaluation indicators is a recognized pain point in the industry. Existing operation and maintenance platforms typically classify weather into a limited number of discrete levels such as "sunny, cloudy, overcast, and rainy" based on meteorological forecasts or irradiance sensors. However, at the physical level, atmospheric transparency, cloud thickness, and the proportion of light scattering are continuously changing functions. This "step-like" classification logic causes the efficiency benchmark to jump instantaneously at the critical point of weather transition (such as thin clouds passing by, overcast turning cloudy). For early warning algorithms based on residual analysis, this jump generates a large amount of pseudo-random noise, resulting in a high false alarm rate. Secondly, the deep coupling between meteorology and equipment loss leads to difficulties in "causal localization." Traditional operation and maintenance indicators such as performance ratio (PR) are effective in long-term evaluations, but in minute-level online early warnings, they cannot distinguish whether the power reduction is due to natural fluctuations caused by environmental obstruction (such as smog or dust storms) or performance degradation caused by inherent equipment defects (such as hot spots on modules or overheating of combiner boxes). Furthermore, for large-scale distributed photovoltaic systems, especially rooftop photovoltaic systems implemented across entire counties, the cost of installing high-precision radiometers is extremely high, and maintenance is difficult. Without reliable local meteorological references, the accuracy of existing algorithms for early warning drops sharply. At the same time, the limited communication bandwidth of power stations in old urban areas or remote regions cannot support the backhaul of full high-frequency raw data, placing extremely high demands on the edge computing capabilities and data compression efficiency of early warning algorithms.

[0004] Therefore, there is an urgent need for a method and system for diagnosing photovoltaic performance anomalies based on continuous weather characteristic coefficients to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a photovoltaic performance anomaly diagnosis method and system based on continuous weather characteristic coefficients, which overcomes the shortcomings of existing photovoltaic performance anomaly diagnosis technologies, such as discrete weather evaluation dimensions, unsmooth benchmarks, and high hardware dependence.

[0006] To achieve the above objectives, the present invention employs the following technical solution: On the one hand, the present invention provides a method for diagnosing photovoltaic performance anomalies based on continuous weather characteristic coefficients, comprising the following steps: Step S1: Obtain the actual operating data of the photovoltaic unit, including but not limited to actual solar irradiance and module temperature. and actual effort; Rated power and actual solar irradiance, module temperature under standard photovoltaic unit conditions. Calculate the normalized weather characteristic coefficients that characterize continuous weather variations. The normalized weather characteristic coefficient The range of values ​​is ,in Characterizing clear-sky conditions with no atmospheric attenuation. This characterizes a completely blocked, no-output working condition. Step S2: Based on normalized weather characteristic coefficients and component temperature Construct a dynamic theoretical power reference surface; Step S3: Obtain the real-time measured power of the photovoltaic unit and calculate the residual between the real-time measured power and the corresponding theoretical power on the dynamic theoretical power reference surface. ; Step S4: Based on the normalized weather characteristic coefficients Dynamic characteristics and residuals Based on the changing trends, multi-type fault decoupling diagnosis is performed on photovoltaic units.

[0007] Preferably, in step S1, the normalized weather characteristic coefficients are solved using the least squares method. The optimization objective is to minimize the sum of the actual output of the photovoltaic unit and the output under clear skies by a coefficient. The sum of squared residuals between them.

[0008] Preferably, under conditions of weak communication or no radiation sensor operation, the normalized weather characteristic coefficients mentioned in step S1 The calculation method includes one of the following: Parent-daughter station collaboration mode: Select a parent station with real-time calculation capabilities within a preset radius of the target daughter station, and calculate the solar altitude angle phase difference based on the latitude and longitude difference between the parent station and the daughter station. And the weather characteristic coefficients calculated by the mother station Perform time shift correction to obtain the substation time. Weather characteristic coefficient ; Sensorless reverse estimation method: The voltage and current signals of the inverter's DC side are collected at a high sampling rate using an edge computing acquisition device. The time-domain fluctuation components are extracted and frequency-domain transformed. The frequency-domain energy distribution characteristics are input into a pre-trained deep learning model to estimate the weather characteristic coefficients in reverse. .

[0009] Preferably, in step S2, the specific method for constructing the dynamic theoretical power reference surface is as follows: Using radial basis function neural networks or Gaussian process regression to normalize weather characteristic coefficients. and component temperature As input, with theoretical power For output, establish a three-dimensional mapping function. This generates a smooth, continuous theoretical power reference surface.

[0010] Preferably, in step S4, the multi-type fault decoupling diagnosis includes ash accumulation loss diagnosis, and its diagnostic conditions are as follows: After a series of sunny days and Within the interval, if the residual It exhibits a slow, linear growth trend over time, especially at noon. When the value reaches its peak and the deviation is the largest, it is determined that there is dust accumulation on the surface of the photovoltaic unit, and a cleaning and maintenance prompt is triggered.

[0011] Preferably, in step S4, the multi-type fault decoupling diagnosis includes local shadow or hot spot diagnosis, and the diagnostic conditions are as follows: exist If the measured power decreases in a stepwise manner and the current dispersion between strings exceeds 10%, provided that the value remains stable and its variance is less than 0.05, it is determined to be hard blocking or hot spot of the module.

[0012] Preferably, in step S4, the multi-type fault decoupling diagnosis includes inverter MPPT tracking anomaly diagnosis, and its diagnostic conditions are as follows: In the case of frequent weather fluctuations and Under the condition that the residual A sharp jump occurred, and the power recovery rate lagged behind. If the value increases rapidly, it is determined that the inverter MPPT optimization has failed, and inverter parameter optimization suggestions are triggered synchronously.

[0013] On the other hand, the present invention also provides a photovoltaic performance anomaly diagnosis system based on continuous weather characteristic coefficients, for implementing the photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients as described above, including: The data acquisition module is used to acquire the actual operating data of the photovoltaic unit; The weather characteristic coefficient calculation module is used to calculate normalized weather characteristic coefficients that characterize continuous weather changes based on the actual operational data. ; The performance benchmark surface construction module is used to construct a performance benchmark surface based on the normalized weather characteristic coefficients. and component temperature Construct a dynamic theoretical power reference surface; The residual calculation module is used to calculate the residual between the measured power and the corresponding theoretical power on the dynamic theoretical power reference surface. ; The fault decoupling diagnosis module is used to determine the normalized weather characteristic coefficients. Dynamic characteristics and residuals The changing trend is used to perform multi-type fault decoupling diagnosis.

[0014] Preferably, the weather characteristic coefficient calculation module includes an edge computing unit, which is configured to extract time-domain fluctuation characteristics by sampling the voltage and current signals on the DC side of the inverter at high frequency in the event of communication interruption or the absence of a high-precision radiometer, and then back-estimate the time-domain fluctuation characteristics using a pre-trained deep learning model. value.

[0015] Preferably, the deep learning model in the edge computing unit is a convolutional neural network (CNN) or a long short-term memory network (LSTM), whose input is the peak-to-peak value, standard deviation, and low-frequency energy proportion characteristics of voltage time-domain fluctuations, and whose output is a reverse estimation. The edge computing unit is also used to reverse-estimate the value; The value is compared with historical operating benchmarks. If it remains low, the power warning threshold is automatically reduced to avoid false alarms.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention maps complex meteorological conditions to... The continuous space eliminates the "step-like" benchmark abrupt changes caused by traditional discrete weather levels. The system can build a smooth "power fingerprint" reference for photovoltaic power plants, accurately filter out pseudo-random noise caused by weather switching thresholds (such as thin clouds passing by), significantly reduce the false alarm rate of operation and maintenance early warning, and improve the stability of the monitoring system. 2. This invention differs from traditional single residual judgment by utilizing... The dynamic characteristics of the value and the spatiotemporal correlation of the fault residual can accurately distinguish between natural weather fluctuations and physical hazards of the equipment itself. Under various complex operating conditions such as continuous sunny weather, continuous stable shading and violent weather fluctuations, the system can clearly identify different faults such as dust accumulation, local shadows / hot spots and inverter MPPT optimization failure, allowing maintenance personnel to shift directly from "blind troubleshooting" to "targeted maintenance", which greatly shortens the fault discovery and handling time. 3. In the face of the current situation that distributed photovoltaics are "large in scale and wide in area" and it is difficult to configure high-precision radiometers, this invention innovatively adopts "parent-child station collaboration" and a sensorless mechanism based on the embedded feature of DC side electrical signal back-inference. Even in weak communication environments with communication interruption or no meteorological station reference, it can still estimate weather characteristics with high accuracy. At the same time, the edge computing architecture effectively reduces the bandwidth requirements for high-frequency data backhaul, which is extremely friendly to power stations in old urban areas and remote areas with limited communication conditions. 4. By accurately identifying slow-onset performance anomalies such as dust accumulation, this invention can assist operators in scientifically calculating the balance point of maintenance revenue within the remaining lifespan, effectively avoiding blind and frequent cleaning and excessive repairs, making the allocation of operation and maintenance resources more economical and reasonable, and ensuring the long-term operational benefits of photovoltaic assets. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This invention relates to the real-time coefficient of a photovoltaic power station in a certain region. Calculation diagram; Figure 3 This invention is based on coefficients A schematic diagram of the dynamic weather coefficient-theoretical power-ambient temperature performance reference surface; Figure 4 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0018] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0019] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0020] Example: like Figure 1 As shown in the figure, this embodiment provides a method for diagnosing photovoltaic performance anomalies based on continuous weather characteristic coefficients, including the following steps: Step S1: Obtain actual operating data of the photovoltaic unit; calculate the normalized weather characteristic coefficients characterizing continuous weather changes based on the rated power, actual solar irradiance, and module temperature under standard conditions of the photovoltaic unit; Step S2: Construct a dynamic theoretical power reference surface based on normalized weather characteristic coefficients and component temperature; Step S3: Obtain the real-time measured power of the photovoltaic unit and calculate the residual between the real-time measured power and the corresponding theoretical power on the dynamic theoretical power reference surface; Step S4: Based on the dynamic characteristics of the normalized weather characteristic coefficients and the changing trend of the residuals, perform multi-type fault decoupling diagnosis on the photovoltaic unit.

[0021] In step S1, the normalized weather characteristic coefficients The calculation includes the following two methods: 1. Continuous Weather Characteristic Coefficient under Normal Operating Conditions Physical modeling and calculation: This embodiment defines weather characteristic coefficients using physical formulas. Unlike traditional absolute values ​​of irradiance, It is a normalized relative quantity that characterizes "weather transparency". The value calculation steps are as follows: (1) Factors affecting the output of photovoltaic units: Unlike wind power, while photovoltaic (PV) power generation also exhibits randomness and intermittency, it also possesses deterministic daily characteristics and seasonality. The actual output of a PV unit... The calculation is as follows: (1); In the formula, Photovoltaic unit output under standard conditions; solar radiation intensity under standard conditions. for ,temperature It is 25 degrees Celsius. Photovoltaic unit Actual solar irradiance at any given time Photovoltaic unit Real-time temperature The temperature coefficient of the photovoltaic unit; As can be seen from equation (1), the main factor affecting actual photovoltaic output is solar irradiance. and temperature And temperature The impact on photovoltaics is minor and can be ignored. (2) Physical meaning of continuous weather characteristic coefficient: Assuming that the various influencing factors do not interact with each other, the actual power output of the photovoltaic unit... The output decomposition model is shown in equation (2): (2); In the formula, For weather characteristic coefficients, To account for the power output in clear skies without considering the effects of Earth's atmosphere, This is the part of the photovoltaic system that generates power due to its highly random nature; (3) Coefficient Real-time calculation: Equation (2) has already given the coefficients. The basic definition, in fact, is the coefficient. This coefficient is used to characterize the influence of weather type on various factors affecting photovoltaic (PV) output. If the influence of the Earth's atmosphere on PV output is small and can be ignored, then the actual PV output is the theoretically calculated clear-sky PV output. The value is 1; conversely, if the Earth's atmosphere has a significant impact on photovoltaic output, resulting in almost zero actual photovoltaic output, the coefficient... The value of this coefficient is 0; therefore, the coefficient... The range of values ​​is ; The coefficients are calculated using the least squares method. The value of , which serves as a decision variable in the optimization problem, is as follows: (3); In the formula, the coefficients For weather characteristic coefficients, A set of time period numbers, Number the time period; We selected actual photovoltaic power output data from a region in Qinghai to verify the coefficient. The calculation method uses a data time granularity of 15 minutes, totaling 1056 time periods, or 11 days. The calculation results are as follows: Figure 2 As shown, the coefficients for days 2, 4, 5, and 6. A value close to 1 indicates that the influence of Earth's atmosphere is relatively weak during these days. The coefficients for days 1, 8, 10, and 11 are... A value around 0.6 indicates that the influence of Earth's atmosphere is stronger on these days. On other days, such as the 3rd and 7th days, the strong influence of Earth's atmosphere leads to a lower coefficient. The value is low; 2. Coefficient under weak communication conditions Calculation method: Distributed photovoltaic systems are geographically widespread and have small individual capacities, which presents problems such as the inability to install expensive irradiance meters and frequent communication link interruptions. According to step (1) under normal operating conditions, the coefficient is calculated. Real-time transmission of active power from the inverter's AC side is required. The aforementioned issues prevent the coefficients from being calculated in real time, even after data collection is complete. ; This embodiment proposes coefficients under weak communication environments. The calculation method is designed to maintain high accuracy even in the absence of direct meteorological observations or in the event of data loss. The value is calculated as follows: (1) Method 1: Calculation coefficient of parent-child station collaboration mode : It will have the ability to calculate coefficients in real time. Distributed photovoltaic sites meeting certain conditions are defined as "mother stations"; however, the lack of real-time calculation coefficients... The distributed photovoltaic sites under certain conditions are defined as daughter stations, and the calculation coefficient for the parent-daughter station collaboration method is as follows. The applicable target is substations that have a parent station within a certain distance range; it mainly utilizes the spatiotemporal correlation between the parent and substations to calculate coefficients. Calculation; When the target substation lacks an irradiance sensor or communication is interrupted, the system uses a parent-substation collaborative method to calculate the coefficient. The main steps are as follows: Mother station selection and clustering: The system automatically searches for mother stations within a 20km radius of the sub-stations based on the Geographic Information System (GIS). The selection criteria can be modified and set, including but not limited to: having a high-precision irradiator, belonging to the same climate microzone as the sub-station, and having a stable real-time communication link. Spatial location difference decoupling: Considering the geometric laws of solar motion, there are latitude and longitude differences between the mother station and the daughter station. The system first calculates the solar altitude angle phase difference between the two stations. ; Time phase shift: If the mother station is east of the target daughter station, the weather characteristic coefficients observed by the mother station... This will serve as a future moment for the sub-site. The system uses reference forecasts; if data from a substation is interrupted, the system utilizes real-time indicators from the parent station. , combined The translation correction is performed using the formula shown in equation (4): (4); In the formula, These are the weather characteristic coefficients observed at the mother station; Cloud movement vector correction: Further, the system retrieves cloud image vector data from regional meteorological satellites and adjusts the vector based on wind speed and direction. The shift of the indicator is fine-tuned a second time, thereby constructing a high-confidence "virtual weather feature value" at the substation. (2) Method 2: Sensor-independent back-calculation coefficients method: In the case of complete isolation and no weather station reference, this embodiment proposes to use the "endogenous characteristics" of the electrical signal on the DC side of the inverter for inverse estimation. The method relies on edge computing acquisition devices deployed at distributed photovoltaic sites; the main steps are: Signal Extraction and Noise Reduction: The edge computing acquisition device acquires the inverter DC bus voltage at a high-frequency sampling rate (greater than or equal to 10kHz). and current By detrending, its fluctuation components are extracted. ; Frequency domain feature transformation (FFT / wavelet transform: using Fast Fourier Transform (FFT) to transform time-domain fluctuations into frequency-domain energy distribution; experimental studies have found that: clear weather conditions ( The power output is stable, with fluctuations mainly concentrated in the low-frequency band, and the spectrum curve is smooth; complex weather / cloudy days ( Due to the random edge effect of cloud shadows and the instability of scattered light, the DC bus voltage will produce a unique "pink noise" characteristic in the low frequency range of 0.1Hz to 5Hz. Value inverse mapping model: A pre-trained deep convolutional neural network (CNN) or long short-term memory network (LSTM) is established. Its input is the feature vector of DC-side voltage fluctuations (including peak-to-peak value, standard deviation, and low-frequency energy proportion), and its output is an estimated... The inverse calculation is mainly based on the following physical mechanism: the dynamic internal resistance of a photovoltaic array under different irradiance intensities and spectral compositions. The difference is directly reflected in the voltage ripple shape generated by MPPT closed-loop control; Self-healing closed loop: The edge computing acquisition device will calculate the reverse... Compare with local historical operating benchmarks; if the reverse calculation is... If the power level remains low, the system will automatically reduce the power warning threshold to avoid issuing erroneous warnings of "abnormal efficiency" due to the misconception that the weather is sunny.

[0022] Step S2 specifically involves constructing a coefficient-based... The dynamic KPT performance benchmark includes: 1. Data cleaning and healthy sample extraction: The system automatically retrieves data from the "golden period" of the power plant's first year of operation, removes periods of grid curtailment, inverter outages, and known power outages, and uses these healthy samples to construct a three-dimensional mapping function. ; 2. Performance reference surface generation: Radial basis function neural networks or Gaussian process regression are used, with coefficients... and component temperature As input, with theoretical power For output, when the coefficient When fixed, the curved surface behaves as... The thermal properties increase and then decrease linearly and slowly, when When fixed, the curved surface behaves as... The irradiation characteristics increase monotonically and rise; This three-dimensional surface is for each group Value and temperature The combination provides a unique, smooth theoretical power reference, such as Figure 3 As shown.

[0023] Steps S3-S4 are specifically based on coefficients The efficiency anomaly decoupling online diagnostic matrix includes: When the measured power When the value falls below the preset confidence limit of the reference surface, the system enters the decoupling diagnostic process. Diagnostic issues include, but are not limited to, dust accumulation loss, localized shading or hot spot diagnosis, and inverter MPPT tracking anomalies; details are as follows: 1. Dust accumulation and loss diagnosis: Anomaly identification: During consecutive sunny days ( For values ​​greater than 0.85, the residuals It exhibits a slow, linear increase over time, and at noon ( The deviation is most obvious when the value reaches its peak; this situation is judged as surface dust accumulation, and the system automatically retrieves local rainfall records. If there has been no rain in the past 15 days, a cleaning suggestion is triggered. 2. Diagnosis based on localized shadows or hot spots: Anomaly feature identification: The value remained stable. (less than 0.05), but A specific step-like decrease occurs, accompanied by a current dispersion between strings exceeding 10%; this situation is identified as hard obstruction (such as new buildings or taller vegetation) or hot spots on the module. 3. Inverter MPPT tracking error: Anomaly identification: In situations with frequent weather fluctuations ( and When it is relatively large, Abrupt changes, and the power recovery rate lags significantly behind. The rate of increase in value; this situation is determined to be the failure of the inverter control algorithm to optimize under complex weather conditions, and the inverter parameter optimization suggestion is triggered synchronously.

[0024] like Figure 4 As shown, this embodiment also provides a photovoltaic performance anomaly diagnosis system based on continuous weather characteristic coefficients, including: The data acquisition module is used to acquire the actual operating data of the photovoltaic unit; The weather characteristic coefficient calculation module is used to calculate normalized weather characteristic coefficients that characterize continuous weather changes based on actual operational data. ; The performance benchmark construction module is used to construct benchmarks based on normalized weather characteristic coefficients. and component temperature Construct a dynamic theoretical power reference surface; The residual calculation module is used to calculate the residual between the measured power and the corresponding theoretical power on the dynamic theoretical power reference surface. ; The fault decoupling diagnosis module is used to diagnose based on normalized weather characteristic coefficients. Dynamic characteristics and residuals Based on the changing trends, multi-type fault decoupling diagnosis is performed; The weather characteristic coefficient calculation module includes an edge computing unit. This unit is configured to extract time-domain fluctuation characteristics by frequently sampling the voltage and current signals on the DC side of the inverter in the event of communication interruption or the absence of a high-precision radiometer, and then back-estimate the time-domain fluctuations using a pre-trained deep learning model. value; The deep learning model in the edge computing unit is either a Convolutional Neural Network (CNN) or a Long Short-Term Memory (LSTM) network. Its input consists of the peak-to-peak value, standard deviation, and low-frequency energy proportion characteristics of voltage time-domain fluctuations, and its output is an inverse estimation. value; Edge computing units are also used to back-estimate The value is compared with the historical operating benchmark. If it remains at a low level, the power warning threshold is automatically reduced to avoid false alarms.

[0025] The above is a detailed description of the preferred embodiments of the present invention, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients, characterized in that, Includes the following steps: Step S1: Obtain the actual operating data of the photovoltaic unit, including but not limited to actual solar irradiance and module temperature. and actual effort; Rated power and actual solar irradiance, module temperature under standard photovoltaic unit conditions. Calculate the normalized weather characteristic coefficients that characterize continuous weather variations. The normalized weather characteristic coefficient The range of values ​​is ,in Characterizing clear-sky conditions with no atmospheric attenuation. This characterizes a completely blocked, no-output working condition. Step S2: Based on normalized weather characteristic coefficients and component temperature Construct a dynamic theoretical power reference surface; Step S3: Obtain the real-time measured power of the photovoltaic unit and calculate the residual between the real-time measured power and the corresponding theoretical power on the dynamic theoretical power reference surface. ; Step S4: Based on the normalized weather characteristic coefficients Dynamic characteristics and residuals Based on the changing trends, multi-type fault decoupling diagnosis is performed on photovoltaic units.

2. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, In step S1, the normalized weather characteristic coefficients are solved using the least squares method. The optimization objective is to minimize the sum of the actual output of the photovoltaic unit and the output under clear skies by a coefficient. The sum of squared residuals between them.

3. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, Under conditions of weak communication or no radiation sensor operation, the normalized weather characteristic coefficients mentioned in step S1 The calculation method includes one of the following: Parent-daughter station collaboration mode: Select a parent station with real-time calculation capabilities within a preset radius of the target daughter station, and calculate the solar altitude angle phase difference based on the latitude and longitude difference between the parent station and the daughter station. And the weather characteristic coefficients calculated by the mother station Perform time shift correction to obtain the substation time. Weather characteristic coefficient ; Sensorless reverse estimation method: The voltage and current signals of the inverter's DC side are collected at a high sampling rate using an edge computing acquisition device. The time-domain fluctuation components are extracted and frequency-domain transformed. The frequency-domain energy distribution characteristics are input into a pre-trained deep learning model to estimate the weather characteristic coefficients in reverse. .

4. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, In step S2, the specific method for constructing the dynamic theoretical power reference surface is as follows: Using radial basis function neural networks or Gaussian process regression to normalize weather characteristic coefficients. and component temperature As input, with theoretical power For output, establish a three-dimensional mapping function. This generates a smooth, continuous theoretical power reference surface.

5. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, In step S4, the multi-type fault decoupling diagnosis includes ash accumulation loss diagnosis, and its diagnostic conditions are as follows: After a series of sunny days and Within the interval, if the residual It exhibits a slow, linear growth trend over time, especially at noon. When the value reaches its peak and the deviation is the largest, it is determined that there is dust accumulation on the surface of the photovoltaic unit, and a cleaning and maintenance prompt is triggered.

6. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, In step S4, the multi-type fault decoupling diagnosis includes the diagnosis of local shadows or hot spots, and the diagnostic conditions are as follows: exist If the measured power decreases in a stepwise manner and the current dispersion between strings exceeds 10%, provided that the value remains stable and its variance is less than 0.05, it is determined to be hard blocking or hot spot of the module.

7. The photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients according to claim 1, characterized in that, In step S4, the multi-type fault decoupling diagnosis includes inverter MPPT tracking anomaly diagnosis, and its diagnostic conditions are as follows: In the case of frequent weather fluctuations and Under the condition that the residual A sharp jump occurred, and the power recovery rate lagged behind. If the value increases rapidly, it is determined that the inverter MPPT optimization has failed, and inverter parameter optimization suggestions are triggered synchronously.

8. A photovoltaic performance anomaly diagnosis system based on continuous weather characteristic coefficients, used to implement the photovoltaic performance anomaly diagnosis method based on continuous weather characteristic coefficients as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the actual operating data of the photovoltaic unit; The weather characteristic coefficient calculation module is used to calculate normalized weather characteristic coefficients that characterize continuous weather changes based on the actual operational data. ; The performance benchmark surface construction module is used to construct a performance benchmark surface based on the normalized weather characteristic coefficients. and component temperature Construct a dynamic theoretical power reference surface; The residual calculation module is used to calculate the residual between the measured power and the corresponding theoretical power on the dynamic theoretical power reference surface. ; The fault decoupling diagnosis module is used to determine the normalized weather characteristic coefficients. Dynamic characteristics and residuals The changing trend is used to perform multi-type fault decoupling diagnosis.

9. The photovoltaic performance anomaly diagnosis system based on continuous weather characteristic coefficients according to claim 8, characterized in that, The weather characteristic coefficient calculation module includes an edge computing unit. This edge computing unit is configured to extract time-domain fluctuation characteristics by frequently sampling the voltage and current signals on the DC side of the inverter in the event of communication interruption or the absence of a high-precision radiometer, and then back-estimate the time-domain fluctuation characteristics using a pre-trained deep learning model. value.

10. The photovoltaic performance anomaly diagnosis system based on continuous weather characteristic coefficients according to claim 8, characterized in that, The deep learning model in the edge computing unit is either a Convolutional Neural Network (CNN) or a Long Short-Term Memory (LSTM) network. Its inputs are the peak-to-peak value, standard deviation, and low-frequency energy proportion characteristics of voltage time-domain fluctuations, and its output is a reverse estimation. The edge computing unit is also used to reverse-estimate the value; The value is compared with the historical operating benchmark. If it remains at a low level, the power warning threshold is automatically reduced to avoid false alarms.

Images