Quantitative calculation method for generating capacity loss of photovoltaic power station

Abstract

The invention provides a photovoltaic power station generating capacity loss quantitative calculation method, and belongs to the photovoltaic technical field, and the method comprises the steps: collecting inverter operation parameters in real time through an SCADA data collection system, preprocessing the data, and building a three-dimensional efficiency curved surface model comprising a load rate, an environment temperature, and a device aging time length; training a dynamic efficiency prediction model by using a support vector machine regression method, calculating a difference value between theoretical output power and actual output power to obtain loss power, constructing a four-class evaluation matrix to calculate an efficiency deviation index, and adaptively adjusting data acquisition frequency or starting an abnormality diagnosis program according to a deviation interval to obtain an abnormal diagnosis result. The technical problem that the power generation loss of the photovoltaic power station inverter is difficult to accurately quantify and calculate is solved.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic technology, and more specifically, relates to a method for quantitatively calculating the power generation loss of a photovoltaic power plant. Background Technology

[0002] Photovoltaic power plants use inverters to convert the direct current (DC) generated by photovoltaic modules into alternating current (AC) and feed it into the power grid. Inverter efficiency directly impacts the power plant's power generation efficiency and economic benefits. Current technologies typically use nameplate nominal efficiency or fixed efficiency curves to assess inverter losses, treating inverter efficiency as a constant value or a static parameter only related to the load rate. This fails to adequately consider the dynamic impact of multi-dimensional factors such as ambient temperature changes, device aging, and fluctuations in operating conditions on efficiency. Existing methods face problems in practical applications, including large efficiency assessment biases and inaccurate loss calculations. Particularly under complex operating conditions such as drastic ambient temperature changes, long-term inverter operation leading to device performance degradation, and power oscillations caused by grid fluctuations, the fixed efficiency assumption deviates significantly from actual operating conditions, resulting in increased errors in power generation loss calculations. In other words, existing technologies suffer from the technical problem of accurately quantifying and calculating the power generation losses of photovoltaic power plant inverters. Summary of the Invention

[0003] In view of this, the present invention provides a method for quantitatively calculating the power generation loss of a photovoltaic power plant, which can solve the technical problem in the prior art that it is difficult to accurately quantify and calculate the power generation loss of the inverter in a photovoltaic power plant.

[0004] This invention is implemented as follows: A method for quantifying and calculating power generation loss in a photovoltaic power plant is provided. A SCADA data acquisition system is deployed in the photovoltaic power plant to collect inverter operating data. The collected inverter operating data is preprocessed to establish a dynamic model of inverter efficiency. A three-dimensional efficiency surface model is trained using inverter load rate, ambient temperature, and device aging time as input variables. The inverter power loss is calculated, and a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate evaluation matrix are constructed. An efficiency deviation index is calculated based on these matrices. The calculation model is adaptively adjusted or an anomaly diagnosis program is initiated based on the range of the efficiency deviation index. Finally, the inverter power generation loss is calculated based on the inverter power loss and operating time.

[0005] The steps for collecting inverter operating data specifically include collecting the inverter's input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature at different times. The inverter load rate is the ratio of the inverter's actual operating power to its rated power.

[0006] The preprocessing step specifically involves removing abnormal ambient temperature values ​​and power surge data points. The criteria for determining abnormal ambient temperature values ​​is that the ambient temperature value exceeds the specified range, and the criteria for determining power surge data points is that the power change rate at adjacent times exceeds the set proportion of the inverter's rated power.

[0007] The steps for establishing a dynamic model of inverter efficiency are as follows: the pre-processed inverter load rate, ambient temperature, and device aging time are used as input variables, and the inverter efficiency is used as the output variable. The three-dimensional efficiency surface model is trained using the support vector machine regression method, and the device aging time is the cumulative operating time of the inverter.

[0008] The training process of the support vector machine regression method includes sample normalization, kernel function selection, and parameter optimization. Sample normalization normalizes the inverter load rate, ambient temperature, and device aging time to a specified range. The radial basis function is selected for the kernel function. The parameter optimization uses a grid search method to determine the penalty coefficient and kernel function parameters.

[0009] The establishment of the three-dimensional efficiency surface model includes data segmentation and surface fitting. Data segmentation divides the inverter load rate into multiple intervals and the ambient temperature into multiple intervals. Surface fitting is performed in each interval to obtain local efficiency surfaces. Local efficiency surfaces are then weighted and fused to form a global three-dimensional efficiency surface model.

[0010] The step of calculating the inverter power loss involves inputting the real-time collected inverter load rate and ambient temperature into a three-dimensional efficiency surface model to obtain a dynamic efficiency value. The theoretical output power is then calculated based on the input DC power and the dynamic efficiency value. The difference between the theoretical output power and the actual output AC power is the inverter power loss.

[0011] The constant efficiency deviation matrix is ​​a two-dimensional matrix composed of the time series and the inverter load rate range. The matrix elements are the difference between the fixed efficiency value and the dynamic efficiency value. The fixed efficiency value is the nominal efficiency on the inverter nameplate, and the dynamic efficiency value is calculated by the three-dimensional efficiency surface model.

[0012] The dynamic efficiency correction matrix is ​​a two-dimensional matrix composed of the inverter load rate range and the temperature range. The matrix elements are the ratio of the measured efficiency under operating conditions to the efficiency under the reference operating conditions. The reference operating conditions are defined as the inverter load rate and ambient temperature being under standard test conditions. The measured efficiency under operating conditions is derived from the statistical average value of historical operating data.

[0013] The abnormal operating condition identification matrix is ​​a two-dimensional matrix composed of time series and abnormal type. The matrix elements are Boolean values. The abnormal types include efficiency drop, temperature over-limit, power oscillation and communication interruption. When the corresponding abnormal type is detected, the matrix element value is 1, otherwise it is 0.

[0014] The oscillation rate evaluation matrix is ​​a two-dimensional matrix composed of a time window and a frequency interval. The matrix elements are the oscillation amplitudes of the power signal in the corresponding frequency interval within the time window. The spectrum is obtained by performing a fast Fourier transform on the output AC power sequence in each time window, and the maximum amplitude in each frequency interval is extracted as the oscillation amplitude.

[0015] The efficiency deviation index calculation includes matrix weighted fusion and normalization. The maximum deviation value is extracted from the constant efficiency deviation matrix, the average correction amplitude is extracted from the dynamic efficiency correction matrix, the frequency of abnormal occurrence is counted from the abnormal working condition identification matrix, and the dominant frequency oscillation amplitude is extracted from the oscillation rate evaluation matrix. After normalization, the efficiency deviation index is obtained by weighted summation according to the weight coefficient.

[0016] Specifically, when the efficiency deviation index is in the first interval, the current calculation model is maintained; when the efficiency deviation index is in the second interval, the data acquisition frequency is increased and the three-dimensional efficiency surface model is retrained; when the efficiency deviation index is in the third interval, the inverter anomaly diagnosis program is started and the inverter power loss during the abnormal period is recorded.

[0017] The inverter anomaly diagnosis program includes anomaly source location and loss quantification. Anomaly source location is determined by analyzing the temperature change rate at the moment of efficiency drop, the fluctuation range of inverter load rate, and the stability of input voltage. The temperature change rate is calculated by the time derivative of the device surface temperature.

[0018] The step of calculating the inverter's power generation loss based on the inverter's power loss and operating time is as follows: multiply the inverter's power loss in each time interval by the time interval to obtain the single power loss, and sum up all the single power losses within the statistical period to obtain the total power generation loss.

[0019] Among them, when the daily fluctuation range of ambient temperature exceeds the set threshold, a temperature compensation treatment measure is added. The temperature compensation treatment measure is to introduce a temperature correction coefficient into the inverter efficiency calculation. The temperature correction coefficient is determined according to the inverter thermal characteristic curve. The corrected inverter efficiency is equal to the dynamic efficiency value output by the three-dimensional efficiency surface model multiplied by the temperature correction coefficient.

[0020] This invention establishes a dynamic model of inverter efficiency, using inverter load rate, ambient temperature, and device aging time as input variables. It employs a support vector machine regression method to train a three-dimensional efficiency surface model, achieving dynamic prediction of inverter efficiency and replacing the traditional fixed efficiency assumption. This invention constructs four evaluation dimensions: a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate evaluation matrix. It uses an efficiency deviation index to monitor the degree of matching between the model's prediction accuracy and actual operating conditions in real time. When the deviation exceeds a threshold, it automatically triggers a data acquisition frequency adjustment or model retraining mechanism to ensure that the loss calculation model always closely reflects the actual operating state of the inverter. In summary, this invention solves the technical problem mentioned in the background art of the difficulty in accurately quantifying and calculating the power generation loss of photovoltaic power plant inverters. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a time series plot of the efficiency deviation index in Example 2. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0024] like Figure 1 The diagram shown is a flowchart of a method for quantifying and calculating power generation losses in a photovoltaic power plant, provided by this invention. This method includes the following steps:

[0025] S1. Deploy a SCADA data acquisition system in the photovoltaic power station to collect the input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature of the inverter at different times. The inverter load rate is the ratio of the actual operating power of the inverter to the rated power of the inverter. The acquisition time interval is 10 seconds.

[0026] S2. Preprocess the collected inverter operating data to remove abnormal ambient temperature values ​​and power surge data points. The criteria for determining abnormal ambient temperature values ​​are that the ambient temperature exceeds the range of -40℃ to 85℃, and the criteria for determining power surge data points are that the power change rate between adjacent moments exceeds 15% of the inverter's rated power.

[0027] S3. Establish a dynamic model of inverter efficiency, taking the pre-processed inverter load rate, ambient temperature, and device aging time as input variables, and inverter efficiency as output variable. Use the support vector machine regression method to train a three-dimensional efficiency surface model. The device aging time is the cumulative operating time of the inverter.

[0028] S4. Calculate the inverter power loss. Input the real-time collected inverter load rate and ambient temperature into the three-dimensional efficiency surface model to obtain the dynamic efficiency value. Calculate the theoretical output power based on the input DC power and the dynamic efficiency value. The difference between the theoretical output power and the actual output AC power is the inverter power loss.

[0029] S5. Construct a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate evaluation matrix. Calculate the efficiency deviation index based on the constant efficiency deviation matrix, dynamic efficiency correction matrix, abnormal operating condition identification matrix, and oscillation rate evaluation matrix. When the efficiency deviation index is in the interval [0, 0.28], maintain the current calculation model. When the efficiency deviation index is in the interval (0.28, 0.75), increase the data acquisition frequency to a 5-second interval and retrain the three-dimensional efficiency surface model. When the efficiency deviation index is in the interval (0.75, 1.0), start the inverter abnormal diagnosis program and record the inverter power loss during the abnormal period.

[0030] S6. Calculate the inverter power loss based on the inverter power loss and operating time. Multiply the inverter power loss in each time interval by the time interval to obtain the single power loss. Accumulate all single power losses within the statistical period to obtain the total power loss.

[0031] The SCADA data acquisition system is connected to the inverter controller via a communication module to read the inverter's internal operating parameters in real time. The ambient temperature is measured by a temperature sensor placed near the inverter's heat sink, and the surface temperature of the device is measured non-contactly by an infrared temperature measurement module to measure the surface temperature of the inverter's power module.

[0032] The training process of the support vector machine regression method includes three stages: sample normalization, kernel function selection, and parameter optimization. Sample normalization normalizes the inverter load rate, ambient temperature, and device aging time to a range of 0 to 1. The normalization formula is to subtract the minimum value from the original value and then divide by the difference between the maximum and minimum values. The kernel function selected is the radial basis function (RBF) kernel, which is the exponent of the negative square of the Euclidean distance between the input vector and the support vector, divided by the kernel width parameter. Parameter optimization uses a grid search method to determine the penalty coefficient and kernel function parameters. The penalty coefficient controls the balance between training error and model generalization ability, and the kernel function parameters determine the range of the kernel function. The grid search traverses and combines the penalty coefficient values ​​from 0.1 to 1000 and the kernel function parameter values ​​from 0.001 to 10, selecting the parameter combination with the smallest cross-validation error as the optimal parameter. After training, the set of support vectors and their corresponding weight coefficients are obtained. For the newly input inverter load rate, ambient temperature, and device aging time, the radial basis function value of the input and all support vectors is calculated. The kernel function value is multiplied by the corresponding weight coefficient, summed, and then the bias term is added to obtain the predicted inverter efficiency.

[0033] The establishment of the three-dimensional efficiency surface model includes data segmentation and surface fitting. The data segmentation divides the inverter load rate into five intervals: 0 to 20%, 20% to 40%, 40% to 60%, 60% to 80%, and 80% to 100%. The ambient temperature is divided into four intervals: -40℃ to 0℃, 0℃ to 25℃, 25℃ to 50℃, and 50℃ to 85℃. Surface fitting is performed in each interval to obtain a local efficiency surface. The local efficiency surfaces are then weighted and fused to form a global three-dimensional efficiency surface model. The weighted fusion adopts a distance inverse weighting method. The inverter efficiency prediction value at a new input point is equal to the weighted sum of the prediction values ​​of each local efficiency surface multiplied by the corresponding weights. The weights are inversely proportional to the distance from the new input point to the center of each interval.

[0034] The inverter efficiency mechanism equation describes the energy loss mechanism during power conversion. The inputs include input DC voltage, input DC current, and switching frequency. The output is the total internal power loss of the inverter. The input DC voltage originates from the inverter's DC-side voltage acquired by the SCADA data acquisition system in step S1. The input DC current originates from the inverter's DC-side current acquired by the SCADA data acquisition system in step S1. The switching frequency is the switching frequency of the inverter's power switch transistors, determined by the inverter control parameters. The total internal power loss of the inverter includes switching losses and conduction losses. The switching losses are the losses incurred by the power switch transistors during turn-on and turn-off. The energy loss generated during the process is calculated by multiplying the switching frequency by the input DC current and the switching time constant, which is determined by the device characteristics. The conduction loss is the energy loss caused by the on-resistance of the power switch during conduction, and is calculated by multiplying the square of the input DC current by the on-resistance, which increases with the increase of the device surface temperature. The total power loss inside the inverter is used to correct the prediction accuracy of the three-dimensional efficiency surface model under extreme conditions. When the deviation between the measured inverter power loss and the inverter power loss calculated by the three-dimensional efficiency surface model exceeds 5%, the inverter efficiency mechanism equation is introduced for compensation and correction.

[0035] The constant efficiency deviation matrix is ​​used to quantify the degree of deviation between the fixed efficiency assumption and the actual efficiency. The constant efficiency deviation matrix is ​​a two-dimensional matrix composed of a time series and an inverter load rate range. The matrix elements are the differences between the fixed efficiency value and the dynamic efficiency value. The fixed efficiency value is the nominal efficiency on the inverter nameplate, and the dynamic efficiency value is calculated by a three-dimensional efficiency surface model. The time series is arranged according to the time interval of the data collection. The inverter load rate range is divided into five ranges: 0 to 20%, 20% to 40%, 40% to 60%, 60% to 80%, and 80% to 100%. The number of rows in the matrix is ​​the number of time points within the statistical period, and the number of columns in the matrix is ​​5. The element in the i-th row and j-th column of the matrix represents the difference between the fixed efficiency value and the dynamic efficiency value at the i-th time point within the j-th inverter load rate range.

[0036] The dynamic efficiency correction matrix stores efficiency correction coefficients under different operating conditions. It is a two-dimensional matrix composed of inverter load rate ranges and temperature ranges. Matrix elements represent the ratio of measured efficiency under operating conditions to baseline efficiency. The baseline operating condition is defined as a standard test condition with an inverter load rate of 50% and an ambient temperature of 25°C. The inverter load rate range is divided into five intervals: 0-20%, 20%-40%, 40%-60%, 60%-80%, and 80%-100%. The temperature range is divided into four intervals: -40°C to 0°C, 0°C to 25°C, 25°C to 50°C, and 50°C to 85°C. The matrix has 5 rows and 4 columns. The element in the i-th row and j-th column represents the ratio of measured efficiency under the combined operating condition of the i-th inverter load rate range and the j-th temperature range to the baseline efficiency. The measured efficiency under operating conditions is derived from the statistical average of historical operating data.

[0037] The abnormal operating condition identification matrix is ​​used to mark abnormal operating states of the inverter. The abnormal operating condition identification matrix is ​​a two-dimensional matrix composed of time series and abnormal types. The matrix elements are Boolean values. When the corresponding abnormal type is detected, the matrix element value is 1, otherwise it is 0. The abnormal types include efficiency drop, temperature over-limit, power oscillation, and communication interruption. The efficiency drop judgment condition is that the inverter efficiency drops by more than 3% within 5 minutes. The temperature over-limit judgment condition is that the device surface temperature exceeds 75°C. The power oscillation judgment condition is that the output AC power fluctuation amplitude exceeds 8% of the average value. The communication interruption judgment condition is that no data is received for 3 consecutive acquisition cycles. The number of matrix rows is the number of time points in the statistical period, and the number of matrix columns is 4.

[0038] The oscillation rate evaluation matrix is ​​used to evaluate the stability of the inverter output power. The oscillation rate evaluation matrix is ​​a two-dimensional matrix composed of a time window and a frequency range. The matrix elements are the oscillation amplitudes of the power signal in the corresponding frequency range within the time window. The time window length is 60 seconds. The spectrum is obtained by performing a fast Fourier transform on the output AC power sequence within each time window. The frequency ranges are divided into 0 to 5Hz, 5Hz to 15Hz, 15Hz to 30Hz, and 30Hz to 50Hz. The maximum amplitude in each frequency range is extracted as the oscillation amplitude. The number of rows in the matrix is ​​the number of time windows included in the statistical period. The number of columns in the matrix is ​​4. The element in the i-th row and j-th column of the matrix represents the oscillation amplitude of the i-th time window in the j-th frequency range.

[0039] The efficiency deviation index calculation includes matrix weighted fusion and normalization. The maximum deviation value is extracted from the constant efficiency deviation matrix and designated as the first deviation component. The maximum deviation value is the maximum absolute value of all elements in the constant efficiency deviation matrix. The average correction amplitude is extracted from the dynamic efficiency correction matrix and designated as the second correction component. The average correction amplitude is the average of the absolute values ​​of the differences between all elements in the dynamic efficiency correction matrix and 1. The frequency of abnormal occurrences is counted from the abnormal operating condition identification matrix and designated as the third abnormal component. The frequency of abnormal occurrences is the total number of elements with a value of 1 in the abnormal operating condition identification matrix. The dominant frequency oscillation amplitude is extracted from the oscillation rate evaluation matrix and designated as the fourth oscillation component. The dominant frequency oscillation amplitude is the maximum value of all elements in the oscillation rate evaluation matrix. The first deviation component is divided by a standard threshold of 0.05, the second correction component by a standard threshold of 0.10, the third abnormal component by a standard threshold of 20, and the fourth oscillation component by a standard threshold of 50W for normalization. The results are then weighted and summed according to weighting coefficients of 0.3, 0.3, 0.25, and 0.15 to obtain the efficiency deviation index.

[0040] The efficiency deviation index range [0, 0.28] corresponds to the normal operating state of the inverter. The principle for determining the threshold of 0.28 is that when the efficiency deviation index is lower than 0.28, the maximum deviation value of the constant efficiency deviation matrix is ​​less than 1.4%, the average correction amplitude of the dynamic efficiency correction matrix is ​​less than 2.8%, the frequency of abnormal occurrence of the abnormal operating condition identification matrix is ​​less than 5.6 times, and the oscillation amplitude of the dominant frequency of the oscillation rate evaluation matrix is ​​less than 14W. These values ​​correspond to the normal operating deviation range specified by the inverter manufacturer. Maintaining the current calculation model can ensure the calculation accuracy while avoiding frequent model updates.

[0041] The efficiency deviation index range (0.28, 0.75) corresponds to the state of increased inverter operating deviation. The principle for determining the threshold of 0.75 is that when the efficiency deviation index is between 0.28 and 0.75, the inverter efficiency deviation has exceeded the normal range but has not yet reached the fault level. The maximum deviation value of the constant efficiency deviation matrix is ​​between 1.4% and 3.75%, the average correction amplitude of the dynamic efficiency correction matrix is ​​between 2.8% and 7.5%, the frequency of abnormal occurrence of the abnormal operating condition identification matrix is ​​between 5.6 and 15 times, and the oscillation amplitude of the dominant frequency of the oscillation rate evaluation matrix is ​​between 14W and 37.5W. The deviation range indicates that the three-dimensional efficiency surface model deviates from the actual operating conditions. The data acquisition frequency is increased to a 5-second interval and the three-dimensional efficiency surface model is retrained to improve the model's fitting accuracy to the current operating conditions.

[0042] The efficiency deviation index range (0.75, 1.0) corresponds to the abnormal operating state of the inverter. The principle for determining the threshold of 0.75 is that when the efficiency deviation index exceeds 0.75, the maximum deviation value of the constant efficiency deviation matrix exceeds 3.75%, the average correction amplitude of the dynamic efficiency correction matrix exceeds 7.5%, the frequency of abnormal occurrence of the abnormal operating condition identification matrix exceeds 15 times, and the oscillation amplitude of the dominant frequency of the oscillation rate evaluation matrix exceeds 37.5W. These values ​​exceed the upper limit of the allowable deviation specified by the inverter manufacturer, indicating that the inverter has abnormal conditions such as accelerated device aging, poor heat dissipation, and grid interference. The inverter abnormal diagnosis program is started to locate the abnormal source and record the inverter power loss during the abnormal period for subsequent maintenance decisions.

[0043] The inverter anomaly diagnosis program includes anomaly source location and loss quantification. Anomaly source location is determined by analyzing the temperature change rate at the moment of efficiency drop, the inverter load rate fluctuation amplitude, and the input voltage stability to determine the cause of the anomaly. When the temperature change rate exceeds 5℃ / min, it is determined to be a heat dissipation anomaly. The temperature change rate is calculated as the time derivative of the device surface temperature. The causes of heat dissipation anomalies include cooling fan failure or heat dissipation channel blockage. When the inverter load rate fluctuation amplitude exceeds 10%, it is determined to be grid fluctuation. The inverter load rate fluctuation amplitude is calculated as the difference between the maximum and minimum inverter load rate within 10 minutes. The causes of grid fluctuation include grid frequency fluctuation or grid voltage fluctuation. When the input voltage deviates from the rated value by more than 8%, it is determined to be a string fault. The input voltage deviation is calculated as the difference between the input DC voltage and the rated DC voltage divided by the rated DC voltage. The causes of string faults include photovoltaic module shading or photovoltaic module aging.

[0044] The model is corrected based on the degree of temperature interference. When the daily fluctuation range of the ambient temperature exceeds 30°C, the daily fluctuation range of the ambient temperature is the difference between the highest and lowest ambient temperature values ​​within the statistical period. Temperature compensation measures are added, which involve introducing a temperature correction coefficient into the inverter efficiency calculation. The temperature correction coefficient is determined based on the inverter's thermal characteristic curve, which describes the change in inverter efficiency with the surface temperature of the device. The temperature correction coefficient is calculated as the correction coefficient corresponding to a 0.5% decrease in inverter efficiency for every 10°C increase in ambient temperature. The corrected inverter efficiency is equal to the dynamic efficiency value output by the three-dimensional efficiency surface model multiplied by the temperature correction coefficient.

[0045] The inverter power generation loss quantification calculation also involves a photovoltaic power plant power generation loss quantification calculation system implemented by a computer. The computer is equipped with a readable storage medium, which stores program instructions. When the program instructions are run in the computer, they execute the above-mentioned photovoltaic power plant power generation loss quantification calculation method.

[0046] The specific implementation methods of the above steps are described in detail below.

[0047] The specific implementation of step S1 is to deploy a monitoring and data acquisition system at the photovoltaic power station site to obtain the inverter's operating parameters in real time. The system uses a communication module to establish a data link with the inverter controller and reads key operating parameters such as the inverter's input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature in a 10-second cycle. The inverter load rate is calculated by dividing the actual operating power of the inverter by the inverter's rated power. The ambient temperature is measured by a temperature sensor placed near the inverter's heat sink. The device surface temperature is measured non-contactly using an infrared temperature measurement module to avoid interfering with the power module. The purpose of this step is to provide sufficient raw data support for subsequent efficiency modeling and loss calculation.

[0048] The specific implementation of step S2 involves quality screening and anomaly removal of the collected inverter operating data. By setting a reasonable range of ambient temperature from -40℃ to 85℃, abnormal temperature data points exceeding this range are identified and deleted. Simultaneously, power change rate at adjacent moments is calculated and compared with 15% of the inverter's rated power to determine power change data points. When the power change rate at adjacent moments exceeds this threshold, the data is considered to have measurement errors or communication interference and is therefore removed. This step uses a data cleaning method based on threshold determination, which aims to improve the reliability of training data and avoid abnormal data causing fitting deviations in the subsequent efficiency model.

[0049] The specific implementation of step S3 involves constructing a dynamic model describing the nonlinear mapping relationship between inverter efficiency and multiple influencing factors. Preprocessed inverter load rate, ambient temperature, and device aging time are used as input variables, and inverter efficiency is used as the output variable. A support vector machine regression method is employed to train the three-dimensional efficiency surface. This method first normalizes the input samples to eliminate the influence of differences in the dimensions of different physical quantities. Inverter load rate, ambient temperature, and device aging time are normalized to the range of 0 to 1. Normalization is achieved by subtracting the minimum value from the original values. The difference between the maximum and minimum values ​​is used to achieve this. Then, the radial basis function (RBF) kernel is chosen as the kernel type for the support vector machine to achieve a non-linear mapping from the input space to the high-dimensional feature space. The RBF kernel is calculated by dividing the negative square of the Euclidean distance between the input vector and the support vector by the kernel width parameter and then taking the exponent. Next, a grid search method is used to traverse all parameter combinations within the range of penalty coefficient (0.1 to 1000) and kernel function parameter (0.001 to 10). The optimal parameter combination is selected through cross-validation error evaluation. The penalty coefficient is used to control training error and model generalization. The balance between capabilities is determined by the kernel function parameters, which in turn determine the kernel function's effective range. After training, a set of support vectors and their corresponding weight coefficients are obtained. For a new input inverter operating condition, the predicted inverter efficiency is obtained by calculating its radial basis function value with all support vectors, multiplying the kernel function value by the corresponding weight coefficients, summing the results, and adding the bias term. To improve the model's fitting accuracy across different operating conditions, the inverter load rate is divided into five ranges: 0-20%, 20%-40%, 40%-60%, 60%-80%, and 80%-100%. Ambient temperature is also considered. The system is divided into four intervals: -40℃ to 0℃, 0℃ to 25℃, 25℃ to 50℃, and 50℃ to 85℃. Surface fitting is performed in each interval to obtain local efficiency surfaces. Then, the inverse distance weighting method is used to merge the local efficiency surfaces to form a global three-dimensional efficiency surface model. The inverter efficiency prediction value at the new input point is equal to the predicted value of each local efficiency surface multiplied by the weighted sum of the corresponding weights. The weights are inversely proportional to the distance from the new input point to the center of each interval. The purpose of this step is to establish a prediction model that can reflect the dynamic changes in inverter efficiency with load rate, temperature, and aging time.

[0050] The specific implementation of step S4 involves using the established three-dimensional efficiency surface model to calculate the inverter's power loss under actual operating conditions. The real-time collected inverter load rate and ambient temperature are input into the three-dimensional efficiency surface model to obtain dynamic efficiency values. Then, the theoretical output power is calculated based on the product of the input DC power and the dynamic efficiency value. The difference between the theoretical output power and the actual measured output AC power is the inverter's power loss. To further improve the calculation accuracy under extreme conditions, when the deviation between the measured inverter power loss and the inverter power loss calculated by the three-dimensional efficiency surface model exceeds 5%, a compensation correction equation based on the power conversion physical mechanism is introduced. The inputs of this mechanism equation include the input DC voltage, input DC current, and switching frequency. The output is the total power loss inside the inverter, which consists of two parts: switching loss and conduction loss. Switching loss is calculated by multiplying the switching frequency by the input DC current by the switching time constant. Conduction loss is calculated by multiplying the square of the input DC current by the on-resistance. The on-resistance increases with the increase of the device surface temperature. The purpose of this step is to achieve accurate quantification of the inverter's power loss.

[0051] The specific implementation of step S5 involves constructing a multi-dimensional evaluation matrix system and calculating the comprehensive efficiency deviation index to determine the model correction strategy. The constant efficiency deviation matrix is ​​a two-dimensional matrix composed of the time series and the inverter load rate range. The matrix elements are the differences between the fixed efficiency value and the dynamic efficiency value. The fixed efficiency value is the nominal efficiency on the inverter nameplate, and the dynamic efficiency value is calculated by the three-dimensional efficiency surface model. This matrix is ​​used to quantify the degree of deviation between the fixed efficiency assumption and the actual efficiency. The dynamic efficiency correction matrix is ​​a two-dimensional matrix composed of the inverter load rate range and the temperature range. The matrix elements are the ratio of the measured efficiency under operating conditions to the efficiency under the benchmark operating conditions. The benchmark operating conditions are defined as a standard of 50% inverter load rate and 25℃ ambient temperature. The test conditions matrix stores efficiency correction coefficients under different operating conditions. The abnormal operating condition identification matrix is ​​a two-dimensional matrix composed of a time series and anomaly type, with Boolean values ​​as matrix elements. Anomaly types include efficiency drop, temperature over-limit, power oscillation, and communication interruption. The efficiency drop judgment condition is that the inverter efficiency drops by more than 3% within 5 minutes; the temperature over-limit judgment condition is that the device surface temperature exceeds 75°C; the power oscillation judgment condition is that the output AC power fluctuation amplitude exceeds 8% of the average value; and the communication interruption judgment condition is that no data is received for three consecutive acquisition cycles. This matrix is ​​used to mark abnormal inverter operating states. The oscillation rate evaluation matrix is ​​a two-dimensional matrix composed of a time window and a frequency range, with a time window length of [missing information]. For every 60 seconds, a Fast Fourier Transform (FFT) is performed on the output AC power sequence within each time window to obtain the spectrum. The frequency ranges are divided into 0 to 5 Hz, 5 Hz to 15 Hz, 15 Hz to 30 Hz, and 30 Hz to 50 Hz. The maximum amplitude within each frequency range is extracted as the oscillation amplitude. This matrix is ​​used to evaluate the stability of the inverter output power. The efficiency deviation index is calculated as follows: the maximum deviation value is extracted from the constant efficiency deviation matrix as the first deviation component; the average correction amplitude is extracted from the dynamic efficiency correction matrix as the second correction component; the frequency of abnormal occurrences is counted from the abnormal operating condition identification matrix as the third abnormal component; and the dominant frequency oscillation amplitude is extracted from the oscillation rate evaluation matrix as the fourth oscillation component. The efficiency deviation index is obtained by normalizing the first deviation component by dividing it by the standard threshold of 0.05, the second correction component by dividing it by the standard threshold of 0.10, the third abnormal component by dividing it by the standard threshold of 20, and the fourth oscillation component by dividing it by the standard threshold of 50W. The values ​​are then weighted and summed according to weighting coefficients of 0.3, 0.3, 0.25, and 0.15. When the efficiency deviation index is in the range of 0 to 0.28, the inverter is considered to be operating normally, and the current calculation model is maintained. When the efficiency deviation index is in the range of 0.28 to 0.75, the inverter is considered to be in a state of increased operating deviation, the data acquisition frequency is increased to a 5-second interval, and the three-dimensional efficiency surface model is retrained. When the efficiency deviation index is in the range of 0.75 to 1...At time 0, an abnormal inverter operating state is detected, and the inverter anomaly diagnosis program is initiated. This program analyzes the temperature change rate at the moment of efficiency drop, the inverter load rate fluctuation amplitude, and the input voltage stability to determine the cause of the anomaly. A temperature change rate exceeding 5℃ / min is identified as a heat dissipation anomaly; a load rate fluctuation exceeding 10% is identified as grid fluctuation; and an input voltage deviation exceeding 8% is identified as a string fault. This step aims to achieve adaptive monitoring of model accuracy and timely identification of abnormal operating conditions.

[0052] The specific implementation of step S6 is to calculate the inverter power loss based on the inverter power loss and operating time, multiply the inverter power loss of each time interval by the time interval to obtain the single power loss, and accumulate all the single power losses within the statistical period to obtain the total power loss. The purpose of this step is to convert the instantaneous power loss into a quantifiable power loss indicator to support power plant operation and maintenance decisions.

[0053] It should be noted that the key technical ideas of this invention include a three-dimensional efficiency surface modeling method based on support vector machine regression and an efficiency deviation evaluation mechanism based on multi-dimensional matrix fusion. The three-dimensional efficiency surface modeling method constructs a nonlinear mapping relationship by using inverter load rate, ambient temperature, and device aging time as input variables. Compared with traditional fixed efficiency assumption methods, it can more accurately reflect the dynamic changes in inverter efficiency with operating conditions. Support vector machine regression uses radial basis kernel functions to achieve high-dimensional nonlinear mapping and optimizes model parameters through grid search. Simultaneously, it employs data segmentation and distance-inverse weighted fusion strategies to improve the model's fitting accuracy across different operating conditions, effectively solving the efficiency prediction deviation problem caused by neglecting the effects of temperature and aging in traditional methods. The multi-dimensional matrix fusion efficiency deviation assessment mechanism comprehensively monitors the inverter's operating status by constructing a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate assessment matrix. Compared to traditional single-index judgment methods, it can comprehensively evaluate model accuracy from multiple dimensions such as efficiency deviation, temperature influence, anomaly identification, and power stability. The efficiency deviation index obtained through weighted fusion calculation can adaptively trigger model updates or anomaly diagnosis procedures, avoiding the accumulation of loss calculation errors caused by the decline in model accuracy in traditional methods. The synergy of these two technical approaches lies in the fact that the three-dimensional efficiency surface model provides an accurate foundation for efficiency prediction, while the multi-dimensional matrix fusion mechanism enables real-time monitoring and adaptive correction of model accuracy. The combination of the two forms a closed-loop system of dynamic modeling and adaptive evaluation. Compared to traditional static efficiency models and fixed threshold judgment methods, it can maintain continuous and stable loss calculation accuracy under complex and variable actual operating conditions. At the same time, the anomaly diagnosis procedure enables rapid location of loss sources, providing reliable technical support for the refined operation and maintenance of photovoltaic power plants.

[0054] It should be noted that this invention also solves the following technical problem: Existing inverter efficiency evaluation models lack adaptive adjustment capabilities, leading to a gradual decrease in prediction accuracy over long-term operation. This invention achieves real-time evaluation of model reliability by constructing an efficiency deviation index. When the deviation index is between 0.28 and 0.75, the data acquisition frequency is automatically increased to a 5-second interval, and the three-dimensional efficiency surface model is retrained. When the deviation index exceeds 0.75, an anomaly diagnosis program is initiated to locate the anomaly source. This mechanism enables the model to dynamically update following the evolution of the inverter's operating state, avoiding model failure caused by device aging and changes in operating conditions, and maintaining the long-term accuracy of loss calculation. Furthermore, this invention also solves the technical problem of difficulty in timely identification and quantification of abnormal inverter operating conditions in existing technologies. This invention uses an abnormal operating condition identification matrix to mark four types of abnormal states: efficiency drop, temperature exceedance, power oscillation, and communication interruption. Combined with an oscillation rate evaluation matrix, frequency domain analysis is performed on output power stability. When an anomaly is detected, not only is the power loss during the abnormal period recorded, but the source of the anomaly is also located by analyzing the temperature change rate, load rate fluctuation amplitude, and input voltage stability, providing a quantitative basis for subsequent maintenance decisions.

[0055] Specifically, the principle of this invention is as follows: This invention solves the technical problem by overcoming the limitations of traditional fixed efficiency assumptions and capturing the real-time variation of inverter efficiency through multi-dimensional dynamic modeling. The support vector machine regression method can learn the nonlinear mapping relationship between inverter load rate, ambient temperature, device aging time, and efficiency. The three-dimensional efficiency surface model covers the entire operating range through data segmentation and surface fitting techniques, solving the problem that a single efficiency value cannot adapt to complex operating conditions. The efficiency deviation index comprehensively considers four dimensions: constant efficiency deviation, dynamic correction amplitude, frequency of anomalies, and power oscillation amplitude, establishing a quantitative evaluation system for model credibility. When actual operating conditions deviate from the model training conditions, an adaptive adjustment mechanism is triggered in a timely manner to avoid the accumulation of calculation errors caused by model failure, thereby achieving accurate quantitative calculation of inverter power generation loss.

[0056] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0057] The specific implementation of step S1 involves deploying a monitoring and data acquisition system at the photovoltaic power station site to achieve real-time acquisition of inverter operating parameters. This system establishes a data link with the inverter controller using a communication module, and cyclically reads key operating parameters such as the inverter's input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature every 10 seconds. The formula for calculating the inverter load rate is as follows:

[0058] ;

[0059] In the formula, The inverter load rate is dimensionless. The actual operating power of the inverter, in units of ; Rated power of the inverter, in units of Ambient temperature is measured via contact using a temperature sensor placed near the inverter heatsink, with an accuracy of ±0.5℃ and a measurement range of -40℃ to 85℃. Device surface temperature is measured non-contactly using an infrared thermography module to avoid interference with the power module. The emissivity of the infrared thermography module is set to 0.95, and the measurement distance is 30 to 50 mm. .

[0060] Step S2 involves quality screening and anomaly removal of the collected inverter operating data. An abnormal temperature data point exceeding this range is identified and deleted by setting a reasonable ambient temperature range of -40℃ to 85℃. The power change rate calculation formula is as follows:

[0061] ;

[0062] In the formula, The power change rate is dimensionless. For a moment Inverter power, in units of ; For a moment Inverter power, in units of ; Rated power of the inverter, in units of .when Data was discarded because it was deemed to contain measurement errors or communication interference.

[0063] The specific implementation of step S3 involves constructing a dynamic model describing the nonlinear mapping relationship between inverter efficiency and multiple influencing factors. Preprocessed inverter load rate, ambient temperature, and device aging time are used as input variables, and inverter efficiency is used as the output variable. A support vector machine regression method is employed to train the three-dimensional efficiency surface. The normalization formula is expressed as follows:

[0064] ;

[0065] In the formula, The normalized variable value is dimensionless and ranges from 0 to 1. These are the original variable values. The unit for inverter load rate is dimensionless, the unit for ambient temperature is °C, and the unit for device aging time is hours. For the minimum value of the variable, the units are... same; For the maximum value of the variable, the units are... The same. The formula for calculating the radial basis function kernel is as follows:

[0066] ;

[0067] In the formula, For input vectors With support vectors The kernel function values ​​between these values ​​are dimensionless. The input vector contains three components: normalized inverter load rate, ambient temperature, and device aging time, and is dimensionless. For support vectors, the dimension and Same, dimensionless; The Euclidean distance between the input vector and the support vectors is calculated as follows: , dimensionless, of which For input vectors The Individual components, dimensionless For support vectors The One component, dimensionless; The kernel width parameter is dimensionless, ranging from 0.001 to 10, and is determined using a grid search method. The support vector machine regression prediction formula is expressed as follows:

[0068] ;

[0069] In the formula, The value is the predicted inverter efficiency, which is dimensionless and ranges from 0 to 1. The number of support vectors, dimensionless; For the first The weight coefficients corresponding to each support vector are dimensionless and are obtained through training. This is the bias term, dimensionless, obtained through training. The distance-inverse weighted fusion formula is expressed as follows:

[0070] ;

[0071] In the formula, The predicted efficiency of the global three-dimensional efficiency surface model is dimensionless. This represents the number of local efficiency surfaces, with a value of 20, corresponding to a combination of 5 load rate ranges and 4 temperature ranges; For the first The predicted efficiency of a local efficiency surface is dimensionless. For the first The weights of the local efficiency surfaces are dimensionless and calculated as follows: The formula for calculating the center distance of an interval is as follows:

[0072] ;

[0073] In the formula, For input point to the th The Euclidean distance between the centers of the intervals is dimensionless. The normalized load factor for the input points is dimensionless. For the first Normalized load rate at the center of each interval, dimensionless; The ambient temperature at the input point, in °C; For the first The ambient temperature at the center of each interval, in °C; The maximum ambient temperature is 85℃. The minimum ambient temperature is -40℃. To prevent small constants with zero denominators, the value is set to 0.001.

[0074] The specific implementation of step S4 involves using the established three-dimensional efficiency surface model to calculate the power loss of the inverter under actual operating conditions. The theoretical output power calculation formula is expressed as follows:

[0075] ;

[0076] In the formula, Theoretical output power, in units of ; Input DC power, in units of The data is obtained from step S1. The formula for calculating inverter power loss is as follows:

[0077] ;

[0078] In the formula, Inverter power loss, in units of ; This represents the actual output AC power, in units of... The data is obtained from step S1. The formula for calculating switching losses is as follows:

[0079] ;

[0080] In the formula, Switching losses, in units of ; Switching frequency, unit: The value ranges from 5000 to 20000 and is determined by the inverter control parameters. Input DC voltage, unit: The data is acquired by the SCADA data acquisition system in step S1, and the empirical value is 600 to 800. Input DC current, unit: The data is acquired by the SCADA data acquisition system in step S1. The switching time constant is expressed in units of 1. The value ranges from 1 to 5, determined by the characteristics of the device. The formula for the change of on-resistance with temperature is as follows:

[0081] ;

[0082] In the formula, The on-resistance is expressed in units of Ω. ; The on-resistance at 25℃, in units of The value ranges from 5 to 20; Temperature coefficient, unit: The empirical value is 0.004; The device surface temperature, in °C, is obtained from step S1. The formula for calculating conduction loss is as follows:

[0083] ;

[0084] In the formula, For conduction loss, the unit is 1. The formula for calculating the total power loss inside the inverter is as follows:

[0085] ;

[0086] In the formula, The total power loss inside the inverter, in units of .

[0087] The specific implementation of step S5 involves constructing a multi-dimensional evaluation matrix system and calculating the comprehensive efficiency deviation index to determine the model correction strategy. The constant efficiency deviation matrix is ​​expressed as follows:

[0088] ;

[0089] In the formula, The constant efficiency deviation matrix; For the first The time point at the The difference between the fixed efficiency value and the dynamic efficiency value within a given inverter load rate range is calculated as follows: Dimensionless; To maintain a fixed efficiency value, the nominal efficiency on the inverter's nameplate is used, with an empirical value of 0.98. The dynamic efficiency value is calculated from a three-dimensional efficiency surface model and is dimensionless. This represents the number of time points within the statistical period. The dynamic efficiency correction matrix is ​​expressed as follows:

[0090] ;

[0091] In the formula, This is the dynamic efficiency correction matrix; For the first The inverter load rate range and the first The ratio of the measured efficiency to the baseline efficiency under a combination of temperature ranges is calculated as follows: Dimensionless; The efficiency is measured under actual operating conditions and is derived from the statistical average of historical operating data; it is dimensionless. The baseline efficiency is defined as an inverter load of 50% and an ambient temperature of 25°C, with an empirical value of 0.98, which is dimensionless. The abnormal operating condition identification matrix is ​​expressed as follows:

[0092] ;

[0093] In the formula, This is a matrix for identifying abnormal operating conditions. This is a Boolean value; it is 1 when the corresponding anomaly type is detected and 0 otherwise. The formula for determining a sudden drop in efficiency is as follows:

[0094] ;

[0095] In the formula, The efficiency change rate is expressed in units of... ; The inverter efficiency at the current moment is dimensionless. The inverter efficiency was 5 minutes ago, dimensionless. This is considered a sudden drop in efficiency. The formula for determining the power oscillation amplitude is as follows:

[0096] ;

[0097] In the formula, The amplitude of the power oscillation is dimensionless. The maximum AC output power within the time window, in units of ; The minimum output AC power within the time window, in units of ; The average value of the output AC power within the time window, in units of .when This is determined to be a power oscillation. The oscillation rate assessment matrix is ​​expressed as follows:

[0098] ;

[0099] In the formula, This is the oscillation rate evaluation matrix; For the first The time window in the first The oscillation amplitude within a frequency range, in units of ; This refers to the number of time windows included in the statistical period. The formula for calculating the efficiency deviation index is as follows:

[0100] ;

[0101] In the formula, This is an efficiency deviation index, dimensionless, with a value range of 0 to 1; The maximum absolute value of all elements in the constant efficiency deviation matrix is ​​dimensionless. The average of the absolute values ​​of the differences between all elements in the dynamic efficiency correction matrix and 1 is calculated as follows: Dimensionless; The total number of elements with a value of 1 in the abnormal operating condition identification matrix; The maximum value of all elements in the oscillation evaluation matrix, in units of The formula for calculating the rate of temperature change is as follows:

[0102] ;

[0103] In the formula, The rate of change of temperature, in units of ; The device surface temperature at the current moment is expressed in °C. This is the device surface temperature 60 seconds ago, in °C. The issue was initially identified as an abnormal heat dissipation. The formula for calculating the inverter load rate fluctuation is as follows:

[0104] ;

[0105] In the formula, The inverter load rate fluctuation is dimensionless. The maximum value of the inverter load rate within 10 minutes, dimensionless; The minimum inverter load rate within 10 minutes, dimensionless; This represents the rated load rate, with a value of 1. When... This is determined to be a power grid fluctuation. The formula for calculating the input voltage deviation is as follows:

[0106] ;

[0107] In the formula, The deviation of the input voltage is dimensionless. The measured input DC voltage is expressed in units of... It is obtained from step S1; Rated DC voltage, unit: This is determined by the inverter parameters, with an empirical value of 600 to 800. The fault is identified as a string fault. The formula for calculating the temperature correction factor is as follows:

[0108] ;

[0109] In the formula, This is a temperature correction factor, dimensionless. The ambient temperature, in °C, is obtained from step S1. The corrected inverter efficiency calculation formula is expressed as follows:

[0110] ;

[0111] In the formula, The corrected inverter efficiency is dimensionless.

[0112] The specific implementation of step S6 involves calculating the inverter's power generation loss based on the inverter's power loss and operating time. The formula for calculating the single-cycle power loss is as follows:

[0113] ;

[0114] In the formula, For the first The single power loss over a time interval, in units of ; For the first Inverter power loss over a time interval, in units of By step S4 in the first Calculated from time intervals; The time interval is 10 seconds. The formula for calculating the total power generation loss is as follows:

[0115] ;

[0116] In the formula, The total power generation loss during the statistical period, in units of ; This represents the number of time intervals within the statistical period.

[0117] It needs to be explained that the basis kernel function achieves nonlinear mapping by calculating the similarity between the input vector and the support vector in high-dimensional space. This allows the support vector machine to fit the complex nonlinear variation of inverter efficiency with load rate, temperature, and aging time, reducing prediction error by approximately 60% compared to linear models. The distance-inverse weighted fusion formula, by assigning greater weight to local efficiency surfaces closer to the input point, achieves an organic combination of piecewise fitting and global fusion. This ensures fitting accuracy within each interval while avoiding prediction jumps at interval boundaries, keeping the prediction error of the global efficiency surface within 0.5%. Interval center distance formula. By calculating the Euclidean distance from the input point to the center of each interval in the two-dimensional space of load rate and temperature, the degree of matching between the input operating condition and each local model is accurately quantified. The temperature term is normalized by dividing by the temperature range, ensuring that the contributions of both load rate and temperature dimensions to the distance are equal, thus avoiding weight allocation bias caused by differences in dimensions. Switching loss formula. Based on the physical mechanism of switching losses in power semiconductor devices, this paper comprehensively considers the influence of switching frequency, DC voltage, DC current, and switching time constant on energy loss, where the coefficients... Used for unit conversion, converting Hz, V, A, The product is converted to When the three-dimensional efficiency surface model exhibits significant prediction errors under extreme conditions, this mechanism equation can provide physical constraints for compensation and correction, reducing the prediction error under extreme conditions from 8% to below 3%. (Conduction loss formula) Based on the ohmic loss mechanism during the conduction period of power devices, where the coefficient Used to The product is converted to Formula for on-resistance as a function of temperature This reflects the physical property that the resistivity of semiconductor materials increases with increasing temperature, accurately reflecting the increasing trend of conduction losses under high-temperature conditions. Efficiency Drop Judgment Formula By calculating the efficiency change rate over 5 minutes, abnormal states such as a sharp decline in inverter performance are identified, providing a trigger condition for timely initiation of anomaly diagnosis. Power oscillation amplitude determination formula. By calculating the relative ratio of power fluctuation amplitude to average power, the influence of absolute power value on oscillation judgment is eliminated, ensuring consistency of the judgment criteria under different load conditions. Temperature change rate formula. By calculating the rate of temperature change over 60 seconds, anomalies in the heat dissipation system are identified, preventing overheating damage to components caused by poor heat dissipation. Inverter load rate fluctuation formula. By statistically analyzing the extreme differences in load rate over a 10-minute period, power instability caused by grid fluctuations is identified, providing a quantitative indicator for grid quality assessment. Input voltage deviation formula. Photovoltaic string faults are identified by calculating the relative deviation between the measured voltage and the rated voltage, providing an accurate basis for fault location in maintenance decisions. Efficiency deviation index formula. By weightedly integrating four dimensions—constant efficiency deviation, dynamic efficiency correction, abnormal operating condition identification, and oscillation rate assessment—the matching degree between the calculation model and the actual operating conditions was comprehensively evaluated. This enabled adaptive monitoring of model accuracy and intelligent decision-making for model correction strategies, avoiding the accumulation of power generation loss calculation errors caused by decreased model accuracy, and ensuring the accuracy and reliability of power generation loss quantification calculation during long-term operation.

[0118] To better understand and implement this invention, the following is a specific application scenario example 2: A technical team is conducting quantitative calculations of power generation loss in a large-scale photovoltaic power station with an installed capacity of 50MW. This power station is equipped with 100 500kW string inverters and operates in a high-altitude climate zone with large diurnal temperature variations and complex radiation conditions. The technical team deployed a SCADA data acquisition system, connected to the inverter controller via RS485 communication protocol, with an acquisition cycle set to 10 seconds. Monitoring parameters included input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature. The ambient temperature sensor, using a PT100 platinum resistance temperature sensor, was positioned 15cm from the side of the inverter heatsink, with a measurement accuracy of ±0.2℃. The device surface temperature was measured using an infrared temperature measurement module at a distance of 30cm, with a measurement range of -20℃ to 150℃.

[0119] The technical team preprocessed 90 days of operational data from June to August 2024, removing 327 outliers (ambient temperatures exceeding -40℃ to 85℃) and 218 abrupt changes (power change rate exceeding 75kW between adjacent time points). This preprocessing yielded 7,776,000 valid data samples, representing a data integrity rate of 99.93%. The team established a dynamic model for inverter efficiency, selecting inverter load rate, ambient temperature, and device aging time as input variables. The cumulative inverter operating time was 12,580 hours, normalized to the maximum operating time of 100,000 hours. A three-dimensional efficiency surface model was trained using support vector machine regression, and sample normalization mapped inverter load rate, ambient temperature, and device aging time to the intervals of 0 to 1. A radial basis function (RBF) kernel was chosen. A grid search was used to test 125 parameter combinations, with the penalty coefficient ranging from 0.1 to 1000 and the kernel parameter values ​​ranging from 0.001 to 10. Five-fold cross-validation was employed, ultimately determining the optimal penalty coefficient to be 58.3 and the optimal kernel parameter to be 0.847. After training, 1247 support vectors were obtained, and the root mean square error of the model training was 0.0032.

[0120] The technical team divided the inverter load rate into five ranges: 0% to 20%, 20% to 40%, 40% to 60%, 60% to 80%, and 80% to 100%. The ambient temperature was divided into four ranges: -40℃ to 0℃, 0℃ to 25℃, 25℃ to 50℃, and 50℃ to 85℃. Local efficiency surfaces were fitted within each range. Efficiency was highest in the 50% to 80% load rate range, and efficiency decreased with increasing ambient temperature. The team used a distance-inverse weighted method to fuse the local efficiency surfaces, with Euclidean distance used as the distance metric in the weight calculation and a weighting exponent of 2.

[0121] The technical team calculated the inverter's power loss and input the real-time collected inverter load rate and ambient temperature into a three-dimensional efficiency surface model to obtain dynamic efficiency values. Typical operating data on July 15, 2024 is shown in Table 1.

[0122] Table 1 Inverter parameters at typical operating times

[0123]

[0124] The theoretical output power is calculated based on the input DC power and dynamic efficiency value. The difference between the theoretical output power and the actual output AC power is the inverter's power loss, which is 0.7kW at 08:30. The technical team constructed a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate evaluation matrix. The constant efficiency deviation matrix is ​​a 7,776,000-row, 5-column matrix, with matrix elements representing the difference between the fixed efficiency value of 98.5% and the dynamic efficiency value. The maximum deviation value within the statistical period is 2.83%. The dynamic efficiency correction matrix is ​​a 5-row, 4-column matrix. The baseline operating condition is defined as an inverter load rate of 50% and an ambient temperature of 25℃, with a baseline efficiency of 96.75%. The ratios of the measured efficiency for each operating condition to the baseline efficiency are shown in Table 2.

[0125] Table 2 Dynamic Efficiency Correction Matrix

[0126]

[0127] The average correction magnitude was 0.0165. The abnormal operating condition identification matrix was a 7,776,000-row, 4-column matrix. Within the statistical period, 13 efficiency drop events, 8 temperature exceedance events, 5 power oscillation events, and 2 communication interruption events were detected, for a total of 28 abnormal occurrences. The oscillation rate evaluation matrix was a 129,600-row, 4-column matrix with a time window length of 60 seconds. A Fast Fourier Transform was performed on the output AC power sequence for each time window, with frequency ranges divided into 0 to 5Hz, 5Hz to 15Hz, 15Hz to 30Hz, and 30Hz to 50Hz. The dominant frequency oscillation amplitude was 38.6kW.

[0128] The technical team calculated the efficiency deviation index. The first deviation component, 2.83%, divided by the standard threshold of 0.05, yielded a normalized value of 56.6. The second correction component, 0.0165, divided by the standard threshold of 0.10, yielded a normalized value of 0.165. The third outlier component, 28, divided by the standard threshold of 20, yielded a normalized value of 1.4. The fourth oscillation component, 38.6kW, divided by the standard threshold of 50W, yielded a normalized value of 0.772. Weighted summation using coefficients of 0.3, 0.3, 0.25, and 0.15 resulted in an efficiency deviation index of 17.32. Figure 2 As shown, the efficiency deviation index exceeds 0.75, indicating that the inverter has entered an abnormal operating state range, and the technical team has initiated the abnormal diagnosis procedure.

[0129] The technical team analyzed the efficiency drop event that occurred at 14:15 on July 28, 2024. The device surface temperature rose from 68.3℃ to 79.7℃ within 3 minutes, with a temperature change rate of 3.8℃ / min. The inverter load rate fluctuated by 5.8% within 10 minutes, and the input DC voltage deviated from the rated value of 600V by 4.2%. The overall assessment was an abnormal heat dissipation. On-site inspection revealed severe dust accumulation on the inverter cooling fan; after cleaning, the device surface temperature returned to normal. The technical team recorded the inverter's power loss during the abnormal period, totaling 187.5 kWh from 14:15 to 15:45 on July 28.

[0130] The technical team observed that the daily fluctuation range of ambient temperature reached 35.2℃. Temperature compensation measures were introduced, with the temperature correction coefficient determined based on the inverter's thermal characteristic curve. For every 10℃ increase in ambient temperature, the inverter efficiency decreased by 0.5%. The corrected inverter efficiency equals the output value of the 3D efficiency surface model multiplied by the temperature correction coefficient. After temperature compensation, the dynamic efficiency prediction accuracy improved to 0.0021. The technical team calculated the inverter's power generation loss within the statistical period. The power loss at each 10-second interval was multiplied by the time interval of 0.00278 hours to obtain the single power loss. All single power losses were accumulated over the 90-day statistical period, resulting in a total power generation loss of 12583 kWh, accounting for 0.26% of the total power generation.

[0131] The advancement of this invention over traditional fixed efficiency assumption methods lies in establishing a three-dimensional surface model of inverter efficiency dynamically changing with load rate, temperature, and aging time. Traditional methods, using a single rated efficiency value, cannot reflect the impact of actual operating conditions on efficiency, leading to significant deviations in power loss calculations. This invention captures the nonlinear characteristics of inverter efficiency through support vector machine regression, accurately modeling the phenomenon of efficiency deviations from rated values ​​under low load rate and high temperature conditions. The multi-dimensional matrix system constructed in this invention can quantify constant efficiency assumption deviations, dynamic operating condition correction requirements, abnormal state identification, and power oscillation assessment, providing criteria for model adaptive adjustment. The efficiency deviation index achieves real-time monitoring of model prediction accuracy through weighted fusion of four types of matrix information. When the deviation exceeds a threshold, it automatically triggers an increase in data acquisition frequency or an anomaly diagnosis procedure, ensuring that the loss quantification calculation always closely matches the actual operating state of the inverter. This invention introduces an inverter efficiency mechanism equation to compensate for and correct extreme operating conditions, explaining the efficiency change law from the perspective of power device switching losses and conduction losses, enhancing the model's generalization ability under operating conditions beyond the training sample range.

[0132] It should be noted that the variables involved in this invention are explained in detail in Tables 3 and 4.

[0133] Table 3. Variable Explanation Table (Part 1)

[0134]

[0135] Table 4. Variable Explanation Table (Part Two)

[0136]

[0137] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for quantitatively calculating the power generation loss of a photovoltaic power plant, characterized in that, A SCADA data acquisition system is deployed in the photovoltaic power plant to collect inverter operating data. The collected inverter operating data is preprocessed to establish a dynamic model of inverter efficiency. The three-dimensional efficiency surface model is trained using inverter load rate, ambient temperature, and device aging time as input variables. The inverter power loss is calculated, and a constant efficiency deviation matrix, a dynamic efficiency correction matrix, an abnormal operating condition identification matrix, and an oscillation rate evaluation matrix are constructed. Based on the constant efficiency deviation matrix, dynamic efficiency correction matrix, abnormal operating condition identification matrix, and oscillation rate evaluation matrix, the efficiency deviation index is calculated. The calculation model is adaptively adjusted or an anomaly diagnosis program is initiated based on the range of the efficiency deviation index. The inverter power generation loss is calculated based on the inverter power loss and operating time.

2. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 1, characterized in that, The steps for collecting inverter operating data are as follows: collecting the inverter's input DC power, output AC power, inverter load rate, ambient temperature, and device surface temperature at different times. The inverter load rate is the ratio of the inverter's actual operating power to its rated power.

3. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 2, characterized in that, The preprocessing steps specifically involve removing abnormal ambient temperature values ​​and power surge data points. The criteria for determining abnormal ambient temperature values ​​is that the ambient temperature value exceeds the specified range, and the criteria for determining power surge data points is that the power change rate at adjacent times exceeds the set proportion of the inverter's rated power.

4. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 3, characterized in that, The steps to establish a dynamic model of inverter efficiency are as follows: the pre-processed inverter load rate, ambient temperature, and device aging time are used as input variables, and the inverter efficiency is used as the output variable. The three-dimensional efficiency surface model is trained using the support vector machine regression method, and the device aging time is the cumulative operating time of the inverter.

5. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 4, characterized in that, The training process of the support vector machine regression method includes sample normalization, kernel function selection, and parameter optimization. Sample normalization normalizes the inverter load rate, ambient temperature, and device aging time to a specified range. The radial basis function is selected as the kernel function. The grid search method is used to determine the penalty coefficient and kernel function parameters for parameter optimization.

6. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 5, characterized in that, The establishment of the three-dimensional efficiency surface model includes data segmentation and surface fitting. Data segmentation divides the inverter load rate into multiple intervals and the ambient temperature into multiple intervals. Surface fitting is performed in each interval to obtain local efficiency surfaces. The local efficiency surfaces are then weighted and fused to form a global three-dimensional efficiency surface model.

7. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 6, characterized in that, The steps for calculating inverter power loss are as follows: input the real-time collected inverter load rate and ambient temperature into the three-dimensional efficiency surface model to obtain the dynamic efficiency value; calculate the theoretical output power based on the input DC power and the dynamic efficiency value; and the difference between the theoretical output power and the actual output AC power is the inverter power loss.

8. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 7, characterized in that, The constant efficiency deviation matrix is ​​a two-dimensional matrix composed of the time series and the inverter load rate range. The matrix elements are the difference between the fixed efficiency value and the dynamic efficiency value. The fixed efficiency value is the nominal efficiency on the inverter nameplate, and the dynamic efficiency value is calculated by the three-dimensional efficiency surface model.

9. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 8, characterized in that, The dynamic efficiency correction matrix is ​​a two-dimensional matrix composed of the inverter load rate range and the temperature range. The matrix elements are the ratio of the measured efficiency under operating conditions to the efficiency under the baseline operating conditions. The baseline operating conditions are defined as the inverter load rate and ambient temperature being at standard test conditions. The measured efficiency under operating conditions is derived from the statistical average value of historical operating data.

10. The method for quantifying and calculating the power generation loss of a photovoltaic power plant according to claim 9, characterized in that, The abnormal operating condition identification matrix is ​​a two-dimensional matrix composed of time series and abnormal type. The matrix elements are Boolean values. The abnormal types include efficiency drop, temperature over-limit, power oscillation and communication interruption. When the corresponding abnormal type is detected, the matrix element value is 1, otherwise it is 0.

Images