A photovoltaic module consistency evaluation method and device based on irradiance, temperature and power joint response surface

Abstract

The application provides a photovoltaic module consistency evaluation method and device based on irradiance, temperature and power joint response surface, and relates to the technical field of photovoltaic module performance evaluation. The method collects outdoor operation data of K brand photovoltaic module groups in the same operating cycle in the same demonstration site, divides MxN grid units in the irradiance-temperature two-dimensional plane, calculates the power concentration trend statistics of the grid units meeting the reliability conditions to construct the annual average response surface matrix; after Min-Max global normalization, the brand surface distance is quantified by Frobenius norm, the average value of all brand pair distances is calculated and mapped as a consistency index CI, and the consistency level is evaluated and the interchangeability judgment result is output. The application breaks through the limitation of single component performance characterization under laboratory standard test conditions, realizes the quantitative evaluation of multi-brand component response consistency based on outdoor demonstration data, and can guide the selection of power station components, brand replacement and operation and maintenance management.

Description

Technical Field

[0001] This application relates to the field of photovoltaic module performance evaluation technology, and in particular to a method and apparatus for evaluating the consistency of photovoltaic modules based on the joint response surface of irradiance, temperature and power. Background Technology

[0002] With the rapid development of photovoltaic (PV) module technology and the continuous expansion of power plant installed capacity, PV power plants often employ a mixed arrangement of PV modules from the same brand or even multiple different brands in actual operation. Even with the same nominal power output, PV modules from different brands may exhibit significant differences in power response characteristics in actual outdoor operating environments due to variations in cell materials, encapsulation processes, and encapsulation materials. This inconsistency in power response between brands directly affects the overall power generation efficiency and operation and maintenance strategies of the power plant, and in severe cases may lead to increased module mismatch losses, increased hot spot risk, and difficulties in operation and maintenance replacement. Therefore, a scientific and quantitative evaluation of the consistency between PV module brands is of significant engineering importance for power plant selection, module replacement decisions, and operation and maintenance optimization.

[0003] Currently, the performance characterization and consistency evaluation of photovoltaic (PV) modules mainly rely on the IEC 61853-1 standard published by the International Electrotechnical Commission (IEC). This standard proposes a method for performance testing and energy rating of individual PV modules under controlled laboratory conditions. Its core is to divide the two environmental parameters, irradiance and temperature, into multiple discrete levels (7 levels for irradiance and 4 levels for temperature, totaling 23 test conditions), forming a performance matrix. The module's performance under different environmental combinations is characterized by measuring the module's power output point-by-point under each test condition. Furthermore, the single-variable binning method commonly used in existing outdoor performance characterization typically only uses irradiance as the single binning variable for binning statistics, failing to introduce temperature at the binning dimension, making it difficult to characterize the coupled influence of irradiance and temperature on module power output.

[0004] However, the above-mentioned existing technical solutions have the following limitations: First, the existing evaluation system only evaluates components of a single brand or model, without involving horizontal comparisons between components of multiple brands, and has not established a consistent quantitative evaluation mechanism among multiple brands, thus failing to provide a basis for component replacement decisions in power plant operation and maintenance.

[0005] Second, the tests conducted according to the IEC 61853-1 standard are performed under controlled conditions using a laboratory solar simulator. The irradiance and temperature at each test point are preset fixed values, rather than the natural variations in actual outdoor environments. Laboratory conditions cannot reflect the power response characteristics of the modules under complex outdoor operating conditions such as real climate, seasonal changes, and diurnal temperature variations.

[0006] In summary, the aforementioned technical methods, whether for laboratory performance characterization of a single component or outdoor calibration considering only a single environmental variable, are insufficient to effectively evaluate the power response consistency of multi-brand components in actual power plant operation. They are also unable to meet the engineering requirements for quantitative consistency determination in component selection, brand replacement, and operation and maintenance management of power plants. Summary of the Invention

[0007] The technical problem this application aims to solve is: how to effectively evaluate the power response consistency of multi-brand photovoltaic modules under actual outdoor operating conditions of power plants, and provide a quantitative basis for power plant module selection, brand replacement and operation and maintenance management.

[0008] To solve the above-mentioned technical problems, the embodiments of this application are implemented through the following aspects.

[0009] Firstly, this application provides a photovoltaic module consistency evaluation method based on the joint response surface of irradiance, temperature, and power, comprising: acquiring outdoor operation data of K brands of photovoltaic module strings to be evaluated within the same experimental site and the same operating period, and performing data cleaning, wherein the outdoor operation data includes irradiance, ambient temperature, and string DC power, and K≥2; dividing the irradiance-temperature two-dimensional plane into M×N grid cells, and for each brand, assigning each cleaned data point to the corresponding grid cell according to its irradiance value and temperature value, wherein M is the number of intervals in the irradiance dimension, and N is the number of intervals in the temperature dimension; for each brand, using a preset time period as... The granularity method assigns the cleaned data to the corresponding grid cells according to the irradiance and temperature values. Reliability checks are performed on the power data of each grid cell within each period. For grid cells that meet preset reliability conditions, the central tendency statistic of the power data within the cell is used as the representative power value of that grid cell. Based on this, an M×N periodic response surface matrix is ​​constructed for each brand, and an M×N annual average response surface matrix is ​​calculated based on each periodic response surface matrix. The annual average response surface matrix of each brand is normalized, and the surface distance between any two brand normalized matrices is calculated. The average of the pairing distances of all brands is then taken, and the consistency index between brands is determined based on this average value.

[0010] Furthermore, the preset reliability condition includes: the number of valid data points within a grid cell is not less than the minimum sample number threshold.

[0011] Furthermore, the central tendency statistic includes: median, truncated mean, or arithmetic mean.

[0012] Furthermore, the truncated mean is calculated as follows: the power data in the grid cells are sorted from smallest to largest, the data with the highest and lowest α proportions after sorting are removed, and the arithmetic mean of the remaining data is taken, with α ranging from 3% to 10%.

[0013] Furthermore, the step of constructing an M×N periodic response surface matrix based on the power representative value of each grid cell with a preset time period as the granularity, and calculating an M×N annual average response surface matrix based on each of the periodic response surface matrices, includes: using the natural month as the time granularity, calculating the central tendency statistic of the power data within the grid cell that meets the preset reliability conditions as the power representative value of the grid cell; constructing an M×N monthly response surface matrix for each brand based on the power representative value of each grid cell; taking the arithmetic mean of the 12 monthly response surface matrices of each brand element by element according to the corresponding grid cell to obtain the M×N annual average response surface matrix of each brand; when the number of valid data points in a certain month is less than the preset minimum sample size threshold, the grid cell corresponding to that month is automatically excluded when taking the element-by-element average, and the average is taken with the actual number of valid months as the denominator.

[0014] Furthermore, the normalization process employs Min-Max normalization, performed according to the following equation: in, Let m be the element in the m-th row and n-th column of the annual average response surface matrix of brand k. and Let be the global minimum and maximum values ​​of all valid elements in the annual average response surface matrix of all K brands, respectively. These are the normalized matrix elements.

[0015] Furthermore, in one implementation, the Frobenius norm is used to calculate the surface distance between any two brand normalized matrices. For matrices with missing grids, the summation is performed only on the corresponding cells where both brands are valid grids, and the result is divided by the number of valid cells and the square root is taken. ; The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in a valid grid. The number of effective units, For the annual average response surface matrix element of brand i, the element in the m-th row and n-th column is... The element in the m-th row and n-th column of the annual average response surface matrix of brand j.

[0016] In another implementation, the Frobenius norm is used to calculate the surface distance between any two brand normalized matrices. For matrices with missing grids, the summation is performed only on the corresponding cells where both brands have valid grids, and a data quality weighting is introduced. ; The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in a valid grid. For the annual average response surface matrix element of brand i, the element in the m-th row and n-th column is... The element in the m-th row and n-th column of the annual average response surface matrix of brand j, with weights The variation coefficients of the two brands within that grid cell are jointly determined: ; and denoted as the coefficients of variation for brand i and brand j within grid cell (m,n), respectively.

[0017] Furthermore, the consistency index among brands is determined based on the average value, including: calculating the average of the pairing surface distances for all brands. Substituting into the following equation, we obtain the consistency index CI: ; Where CI∈(0,1], and the consistency level is determined based on the CI value: When CI≥0.90, the rating is excellent, and the irradiance-temperature-power response characteristics of photovoltaic modules from various brands are highly consistent. When 0.80≤CI<0.90, the rating is good, and the response characteristics of photovoltaic modules from different brands are relatively similar; When 0.70≤CI<0.80, the grade is general, and the response characteristics of photovoltaic modules from different brands vary significantly. When CI < 0.70, the rating is poor, and the response characteristics of photovoltaic modules from different brands vary significantly.

[0018] Secondly, this application provides a photovoltaic module consistency evaluation device based on the joint response surface of irradiance, temperature, and power, comprising: a data acquisition module for acquiring outdoor operation data of K brands of photovoltaic module strings to be evaluated within the same experimental site and the same operating period, wherein the outdoor operation data includes irradiance, ambient temperature, and string DC power, where K≥2; and a data processing module for cleaning the outdoor operation data, dividing the irradiance-temperature two-dimensional plane into M×N grid cells, and assigning each cleaned data point to the corresponding grid cell for each brand according to its irradiance and temperature values. The grid cell is used to calculate the central tendency statistics of the power data within the grid cell that meets the preset reliability conditions, and use them as the power representative value of the grid cell. An M×N periodic response surface matrix is ​​constructed based on the power representative value of each grid cell with a preset time period as the granularity. An M×N annual average response surface matrix is ​​then calculated based on the periodic response surface matrix. The consistency analysis module is used to normalize the annual average response surface matrix of each brand, calculate the surface distance between any two brand normalized matrices and take the average of the pairing distances of all brands, and determine the consistency index between each brand based on the average value.

[0019] The technical advantages of this application are as follows: By dividing the irradiance-temperature two-dimensional plane into M×N grid cells, and constructing the annual average response surface matrix of each brand using central tendency statistics, the surface distance between brands is quantified using the Frobenius norm after Min-Max global normalization and mapped to the consistency index CI. This overcomes the limitation of characterizing the performance of a single component only under standard laboratory test conditions, and can reflect the differences in power response characteristics of different brands of components in the actual outdoor full-condition range, realizing a quantitative evaluation of horizontal consistency among multi-brand components. The Min-Max global normalization uses global extreme values ​​across all brands for unified normalization, eliminating the absolute power difference caused by different nominal power and string configurations, so that the evaluation results focus only on the consistency of the response surface shape, avoiding the loss of amplitude information caused by independent normalization of each brand. Furthermore, the data quality weighting mechanism based on the coefficient of variation makes high-reliability grid cells dominate the surface distance calculation, suppressing the interference of low-quality data such as local occlusion and sensor drift, and improving the accuracy and robustness of the evaluation results. The consistency index corresponds to different engineering decision-making scenarios at different levels, which can guide the design of mixed-use schemes for photovoltaic power plants with different brands of modules, and provide quantitative judgment basis for power plant module selection, brand replacement and operation and maintenance management. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the photovoltaic module consistency evaluation method based on the joint response surface of irradiance, temperature, and power, provided in the embodiments of this application.

[0021] Figure 2A schematic diagram of the structure of the photovoltaic module consistency evaluation device based on the joint response surface of irradiance, temperature and power provided in the embodiments of this application.

[0022] Figure 3 This is a schematic diagram comparing the 5×5 two-dimensional annual average response surface of photovoltaic modules of brand A.

[0023] Figure 4 This is a schematic diagram comparing the 5×5 two-dimensional annual average response surface of the photovoltaic module of brand B.

[0024] Figure 5 This is a schematic diagram comparing the 5×5 two-dimensional annual average response surface of the C brand photovoltaic module.

[0025] Figure 6 This is a schematic diagram comparing the 5×5 two-dimensional annual average response surface of the D brand photovoltaic module. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application are described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0028] The following detailed description, in conjunction with the accompanying drawings, through specific embodiments and application scenarios, provides a detailed explanation of the photovoltaic module consistency evaluation method and apparatus based on the joint response surface of irradiance, temperature, and power provided in this application.

[0029] like Figure 1 As shown, the photovoltaic module consistency evaluation method based on the joint response surface of irradiance, temperature and power provided in this embodiment specifically includes the following steps.

[0030] S1: Obtain outdoor operation data of K brand photovoltaic module strings to be evaluated within the same experimental site and the same operating cycle, and perform data cleaning.

[0031] Outdoor operation data includes irradiance, ambient temperature, and string DC power, where K≥2.

[0032] In this application, K brand photovoltaic module strings installed in the same demonstration site were selected. Each string had the same orientation, tilt angle, and installation method. Minute-level outdoor operation data were collected synchronously within the same operating cycle. The operating cycle was no less than 12 calendar months to cover the complete range of seasonal irradiance and temperature variations.

[0033] Data cleaning includes the following processes: (1) Remove nighttime irradiance values ​​below the irradiance threshold G. night Data points, for example, G night 20W / m can be taken 2 When the irradiance is below this threshold, the component's output power is close to zero and has no evaluative significance.

[0034] (2) Power outage values ​​are removed. Power outage values ​​include negative power values ​​and power values ​​that exceed the preset power limit. Negative power values ​​are usually caused by sensor zero-point drift, and power values ​​that exceed the preset power limit (10000W in this embodiment) are usually caused by sensor failure or data acquisition error. Both are removed.

[0035] S2: Divide the irradiance-temperature two-dimensional plane into M×N grid cells, and for each brand, assign each piece of data after cleaning to the corresponding grid cell according to its irradiance value and temperature value.

[0036] Where M is the number of intervals in the irradiance dimension and N is the number of intervals in the temperature dimension; In this application, G night Starting point, measured maximum irradiance G max The irradiance is divided into M intervals, with the lowest measured temperature T as the endpoint; min Starting point, highest measured temperature T max The system is divided into N temperature ranges, with M ranging from 3 to 8 and N ranging from 3 to 8.

[0037] When data is binned, for each brand, each cleaned data point is assigned to an irradiance range based on its irradiance value and a temperature range based on its temperature value, thus being assigned to the corresponding grid cell. Data from different brands is binned independently, with each brand forming its own set of grid data.

[0038] S3: For each brand, the cleaned data is assigned to the corresponding grid cells according to the irradiance and temperature values ​​at a preset time period. The reliability of the power data of each grid cell in each period is checked. For grid cells that meet the preset reliability conditions, the central tendency statistic of the power data in the cell is calculated as the power representative value of the grid cell. Based on this, an M×N periodic response surface matrix for each brand is constructed, and an M×N annual average response surface matrix is ​​calculated based on the periodic response surface matrix.

[0039] In this embodiment, the time granularity is the calendar month. For each brand's monthly data, the cleaned data is assigned to the corresponding M×N grid cells according to the same grid division method in S2. Then, the power data of each grid cell within that month is subjected to reliability testing. The preset reliability conditions include: the number of valid data points in the grid cell is not less than the minimum sample size threshold n. min n min Use a range of 20 to 50. When the number of data points is insufficient, the estimated variance of the statistic is large and not representative; therefore, the grid cell is marked as missing.

[0040] For grid cells that pass the above reliability test, the central tendency statistic of their internal power data is calculated as the monthly representative power value for that grid cell. The central tendency statistic can be the median, truncated mean, or arithmetic mean.

[0041] In one implementation, a truncated mean is used. The calculation method involves sorting the power data within the grid cells from smallest to largest, removing the data with the highest and lowest proportions (α) after sorting, and then taking the arithmetic mean of the remaining data, where α ranges from 3% to 10%. The truncated mean balances the statistical efficiency of the mean with the robustness of the median, allowing for the utilization of more data points after removing extreme outliers.

[0042] In another implementation, the median is used as a central tendency statistic. The median is naturally robust to outliers and is suitable for grid cells with large data distribution skew.

[0043] In another implementation, when the data distribution is approximately symmetrical, the arithmetic mean is used as a statistic of central tendency.

[0044] Repeat the above process month by month to obtain the M×N monthly response surface matrix for each brand for 12 months. The element in the m-th row and n-th column of the matrix is ​​the monthly power representative value of the corresponding brand in the grid cell of the m-th irradiance interval and the n-th temperature interval.

[0045] The annual average response surface matrix of each brand is obtained by taking the arithmetic mean of each element of the corresponding grid cell in the 12-month response surface matrix. Where k = 1, 2, ..., K, and the matrix elements are denoted as . , m=1,2,...,M, n=1,2,...,N.

[0046] When a grid cell in a given month has fewer valid data points than the minimum sample size threshold, that cell for that month is automatically excluded when averaging element by element, and the average is calculated using the actual number of valid months as the denominator. Let brand k have T at grid cell (m,n). k If there are valid monthly data points, then the annual average power representative value of this grid cell is: in, The valid monthly data for brand k at grid (m,n) for month t, where t represents the valid month for that grid cell.

[0047] This processing method avoids the problem of missing monthly data being mistakenly treated as zero values, which would lower the annual average and ensure the accuracy of the annual average response surface matrix.

[0048] S4: Normalize the annual average response surface matrix of each brand, calculate the surface distance between any two brand normalized matrices and take the average of the pairing distances of all brands, and determine the consistency index between each brand based on the average value.

[0049] Since the nominal power of photovoltaic modules from different brands may vary, the absolute power values ​​in the annualized response surface matrix exhibit systematic differences. Directly calculating the matrix distance cannot reflect the consistency of the response characteristics. Therefore, normalization processing is required to eliminate absolute power differences and retain only the shape information of the response surface.

[0050] In this embodiment, the normalization process uses Min-Max normalization, performed according to the following equation: in, Let m be the element in the m-th row and n-th column of the annual average response surface matrix of brand k. and Let be the global minimum and maximum values ​​of all valid elements in the annual average response surface matrix of all K brands, respectively. The elements are normalized matrix elements, and their values ​​range from [0,1].

[0051] It should be noted that, and It represents the global extremum across all brands and all valid grid cells, rather than the extremum for each brand individually. Using global extremum ensures that different brands are normalized to the same dimensional scale, making the normalization results comparable.

[0052] After normalization, the surface distance between any two brand normalization matrices is calculated. This embodiment provides two parallel methods for calculating surface distance.

[0053] Method 1: Equal-weighted Frobenius norm The Frobenius norm is used to calculate the surface distance between any two brand normalized matrices. For matrices with missing meshes, the sum is calculated only for the corresponding cells where both brands have valid meshes, and then divided by the number of valid cells, with the square root taken. ; The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in a valid grid. The number of effective units, For the annual average response surface matrix element of brand i, the element in the m-th row and n-th column is... The element in the m-th row and n-th column of the annual average response surface matrix of brand j. Divide by Instead of a fixed M×N, this eliminates the problem of incomparable distances between different brands due to varying numbers of missing grids.

[0054] Method 2: CV-weighted Frobenius norm The Frobenius norm is used to calculate the surface distance, and data quality weighting is introduced: ; The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in the valid grid, with weights... The variation coefficients of the two brands within that grid cell are jointly determined: ; and Here, CV represents the coefficient of variation (CV) for brand i and brand j within grid cell (m,n). A smaller CV indicates more concentrated data and more reliable power representation within the grid cell, with a weight closer to 1; a larger CV indicates a smaller weight, but it is always greater than zero. The denominator is divided by the sum of weights rather than the number of effective cells to ensure that the weighted distance value remains within the [0,1] scale range, making it comparable to the results of the equal-weighting method. The CV is the ratio of the standard deviation to the arithmetic mean of the power data within the grid cell, i.e., CV = σ / μ, where σ is the sample standard deviation of all effective power data points within the grid cell, and μ is the arithmetic mean of all effective power data points within the grid cell. CV is dimensionless, its value is independent of the absolute dimension of power, and it allows for comparison of data dispersion across brands and grid cells. When constructing the annual average response surface matrix, the annual average CV value of each grid cell is obtained by taking the arithmetic mean of the monthly CV values ​​of each effective month for that cell, with months without effective data automatically skipped.

[0055] The core idea of ​​the CV weighting method is that different grid cells have different data quality. Grid cells with high data quality should contribute more to the surface distance, while the contribution of grid cells with low data quality should be suppressed, thereby improving the reliability and robustness of the surface distance.

[0056] After obtaining the surface distances for all brand pairings, calculate the average of all C(K,2) surface distance values. : ; Will Substituting into the following equation, we obtain the consistency index CI: ; Where CI∈(0,1). When the normalized response surfaces of all brands are exactly the same, =0, CI=1, best consistency; the greater the difference in response surfaces between brands. The larger the value, the closer CI is to 0.

[0057] Based on CI values, consistency levels are classified as follows: When CI ≥ 0.90, the rating is excellent. The irradiance-temperature-power response characteristics of photovoltaic modules from different brands are highly consistent, allowing for free interchangeability within the same string or the same Maximum Power Point Tracking (MPPT) loop. MPPT is the inverter's maximum power point tracking function, which adjusts the module's operating voltage in real time to ensure the module always operates near its maximum power point. When brand modules with highly consistent response characteristics are connected to the same MPPT loop, the maximum power point operating voltage and current of each module are essentially the same, preventing mutual interference.

[0058] When 0.80 ≤ CI < 0.90, the rating is good. The response characteristics of photovoltaic modules from different brands are relatively similar, and they can be used interchangeably within the same MPPT circuit, but should not be mixed within the same string. Within the same MPPT circuit, different strings are connected in series, and there is a parallel voltage matching relationship between the strings. Small differences can be absorbed by the MPPT adjustment range. However, within the same string, the modules are connected in series, and the current is consistent. Differences in response characteristics can lead to the "weakest link" effect.

[0059] When 0.70 ≤ CI < 0.80, the rating is considered average. The response characteristics of photovoltaic modules from different brands vary significantly, and each brand should be configured with its own independent MPPT circuit. They must not be mixed within the same MPPT circuit. The maximum power point operating voltages of modules from different brands differ considerably; mixing them under the same MPPT will cause some modules to operate outside their maximum power point, resulting in power generation losses.

[0060] When CI < 0.70, the rating is poor, indicating significant differences in the response characteristics of photovoltaic modules from different brands. These modules should not be connected to the same inverter and should be configured with separate inverters and MPPT systems. If the differences are too large, even if allocated to different MPPT channels, the DC bus voltage range and MPPT voltage window of the same inverter may not be compatible with the operating voltage requirements of modules from different brands simultaneously.

[0061] The above-mentioned threshold levels are determined based on the tolerance of different hybrid connection methods in photovoltaic power plant engineering design to the differences in component response characteristics. In specific applications, they can be adjusted according to the inverter model, MPPT number, and power plant design requirements.

[0062] The aforementioned photovoltaic module consistency evaluation method based on the joint response surface of irradiance, temperature, and power divides the irradiance-temperature two-dimensional plane into grid cells, constructs an M×N response surface matrix for each brand using central tendency statistics, and after normalization, quantifies the surface distance between brands using the matrix norm and maps it to a consistency index. This method overcomes the limitations of traditional methods that only compare nameplate parameters under standard test conditions, and can comprehensively reflect the differences in response characteristics of different brands of modules across the entire actual operating range. Min-Max global normalization eliminates the absolute power differences caused by different nominal capacity and string configurations, allowing the evaluation results to focus only on the consistency of the response surface shape. The data quality weighting mechanism based on the coefficient of variation ensures that the high-reliability grid dominates the distance calculation, suppressing the interference of abnormal data such as local shading and sensor drift. The consistency index directly corresponds to engineering decision-making scenarios such as string-level mixed connection, MPPT loop-level mixed use, and inverter-level independent configuration in four levels, which can directly guide the design of mixed use schemes for photovoltaic power plants using different brands of modules. Moreover, all calculations are automatically completed based on existing monitoring data of the power plant, without the need for additional testing equipment, resulting in low implementation costs and strong scalability.

[0063] The following examples, based on a specific application scenario, further illustrate the above evaluation method.

[0064] Photovoltaic module strings of K=4 brands (brand A, brand B, brand C, and brand D) were installed at a photovoltaic power station demonstration site. Each string had the same orientation, tilt angle, and installation method. The operation cycle was 12 calendar months. Minute-level outdoor operation data was collected simultaneously, including irradiance, ambient temperature, and string DC power.

[0065] Step 1: Data cleaning. Remove data with irradiance below 20W / m². 2 Nighttime data; negative power values ​​and power values ​​greater than 10000W were removed.

[0066] Step 2: Divide the irradiance-temperature two-dimensional plane into 25 grid cells with M=5 and N=5.

[0067] The irradiance ranges are as follows: [20,200), [200,400), [400,600), [600,800), [800,1400)W / m 2； The temperature ranges are as follows: [-30,0), [0,10), [10,20), [20,30), [30,50)℃.

[0068] For each brand, each piece of data after cleaning is assigned to the corresponding grid cell according to its irradiance and temperature values.

[0069] Step 3: For each brand's monthly data, perform a reliability test on each grid cell, with a minimum sample size threshold n. min 20 is selected. Grid cells with insufficient data points are marked as missing and do not participate in surface construction. For grid cells that pass the test, the median of their internal power data is calculated as the monthly representative power value for that grid cell. A 5×5 monthly response surface matrix is ​​constructed for each brand at the granularity of a calendar month, resulting in 12 monthly matrices. The 5×5 annual average response surface matrix for each brand is obtained by taking the arithmetic mean of each of the 12 monthly matrices element by element, automatically excluding missing months, and taking the mean with the actual number of valid months as the denominator.

[0070] For example, Table 1 shows the 5×5 response surface matrix of Brand A in November 2022, in units of W. The 5×5 response surface matrices for other months of Brand A and other brands are not shown here to avoid redundancy.

[0071] Table 1. 5×5 Response Surface Matrix of Brand A in November 2022 For example, Table 2 is the 5×5 annual average response surface matrix of brand A, Table 3 is the 5×5 annual average response surface matrix of brand B, Table 4 is the 5×5 annual average response surface matrix of brand C, and Table 5 is the 5×5 annual average response surface matrix of brand D. The unit is W.

[0072] Table 2. Brand A's 5×5 Annual Average Response Surface Matrix Table 3. Brand B's 5×5 Annual Average Response Surface Matrix Table 4. Brand C's 5×5 Annual Average Response Surface Matrix Table 5. 5×5 Annual Average Response Surface Matrix of Brand D Figures 3-6 The images show the 5×5 two-dimensional annual average response surfaces constructed by brands A, B, C, and D based on their respective response surface matrices. The horizontal axis represents the irradiance zones G1 to G5 (corresponding to 20–1400 W / m²). 2 The vertical axis represents temperature zones T1 to T5 (corresponding to -30 to 50℃), and the value in each cell is the annual median power (unit: W) within that operating range. The color gradation transitions from blue (low power) through white to red (high power).

[0073] Depend on Figures 3-6 It can be seen that: (1) Irradiance is the dominant factor affecting the output power of the module. Within the same temperature range, the power value increases significantly from G1 to G5 with the irradiance zone, and the color changes from blue to red, indicating that the module output is most sensitive to the response of solar irradiance intensity.

[0074] (2) Temperature has a negative impact on the overall power of the module. In the medium-high irradiance range (G3~G5), the power value within the same irradiance zone decreases in general from T1 to T5 with the temperature zone, which reflects the negative temperature coefficient characteristic of photovoltaic modules; while in the low irradiance range (G1~G2), the negative effect of temperature on power is weakened due to the nonlinear operating characteristics of the module under low irradiance.

[0075] (3) The response surface morphology of the four brands is highly consistent. The power distribution patterns of each brand are similar in the same operating range, and the color gradient direction is consistent, indicating that the output response characteristics of different brands of components under complex outdoor operating conditions are comparable, providing a data basis for subsequent evaluation of the interchangeability between brands based on the consistency of response surfaces.

[0076] Step 4: Find the global minimum value Pmin and global maximum value Pmax of all valid elements in the annual average response surface matrix of all brands, where Pmin = 408.00W and Pmax = 6033.33W. Perform Min-Max global normalization based on Pmin and Pmax.

[0077] For example, Table 6 is the normalized 5×5 annual average response surface matrix of brand A, Table 7 is the normalized 5×5 annual average response surface matrix of brand B, Table 8 is the normalized 5×5 annual average response surface matrix of brand C, and Table 9 is the normalized 5×5 annual average response surface matrix of brand D.

[0078] Table 6. Normalized 5×5 Annual Average Response Surface Matrix of Brand A Table 7. Normalized 5×5 Annual Average Response Surface Matrix of Brand B Table 8. Normalized 5×5 Annual Average Response Surface Matrix for Brand C Table 9. Normalized 5×5 Annual Average Response Surface Matrix of Brand D Step 5: For any two brands, calculate the surface distance using the CV-weighted Frobenius norm. For example, Table 10 is the annual average CV value matrix for brand A, Table 11 is the annual average CV value matrix for brand B, Table 12 is the annual average CV value matrix for brand C, and Table 13 is the annual average CV value matrix for brand D; Table 14 shows the surface distance calculation results for the 6 brand pairings.

[0079] Table 10. Average Annual CV Value Matrix for Brand A Table 11 Average Annual CV Value Matrix for Brand B Table 12 Average Annual CV Value Matrix for Brand C Table 13 Annual CV Value Matrix of Brand D The surface distances for the six brand pairings are calculated using the following formula, as shown in Table 14.

[0080] ; .

[0081] Table 14 shows the results of surface distance calculation for the six brand pairings.

[0082] Step 6: Calculate the average distance of all C(4,2) = 6 brand pairing surfaces. Substitute Obtain the Consistency Index (CI).

[0083] =(0.030023 + 0.114843 + 0.015312 + 0.144627 + 0.016099 + 0.129888) / 6=0.075132; =1 / (1+0.075132)=0.9301.

[0084] Step 7: Assess the consistency level based on the CI value and output the interchangeability judgment result: CI=0.9301, CI≥0.90, the consistency level is "Excellent". The irradiance-temperature-power response characteristics of photovoltaic modules from different brands are highly consistent, and they can be freely interchanged and mixed within the same string or the same maximum power point tracking (MPPT) loop.

[0085] Based on the photovoltaic module consistency evaluation method based on the joint response surface of irradiance, temperature, and power described above, this application discloses a photovoltaic module consistency evaluation device based on the joint response surface of irradiance, temperature, and power. Figure 2 As shown, the device mainly includes: The data acquisition module 210 is used to acquire outdoor operation data of K brands of photovoltaic module strings to be evaluated within the same experimental site and the same operating cycle. The outdoor operation data includes irradiance, ambient temperature, and string DC power, where K ≥ 2. Specifically, the data acquisition module 210 synchronously acquires outdoor operation data of each brand of string from the monitoring system of each brand within the experimental site at a preset acquisition frequency (e.g., minute-level), with an operating cycle of no less than 12 calendar months to cover the complete range of irradiance and temperature variations throughout the four seasons. The data acquisition module 210 is also used to perform data cleaning on the acquired raw data, including removing nighttime data points with irradiance below a preset nighttime threshold and removing power anomalies, including negative power values ​​and power values ​​exceeding a preset power upper limit.

[0086] The data processing module 220 is used to perform gridding, reliability testing, and response surface matrix construction on the cleaned data. Specifically, it includes: dividing the data into M×N grid cells on an irradiance-temperature two-dimensional plane; assigning each cleaned data point to its corresponding grid cell based on its irradiance and temperature values ​​for each brand; performing reliability testing on the power data of each grid cell within each preset time period; calculating the central tendency statistic of the power data within the grid cell that meets the preset reliability conditions as the representative power value of that grid cell; constructing the M×N periodic response surface matrix for each brand based on this; and calculating the M×N annual average response surface matrix based on the periodic response surface matrix. The preset reliability conditions include that the number of valid data points within the grid cell is not less than the minimum sample size threshold; the central tendency statistic includes the median, truncated mean, or arithmetic mean; and the annual average response surface matrix is ​​obtained by taking the arithmetic mean of each element of the periodic response surface matrix, automatically excluding missing periods and taking the mean with the actual number of valid periods as the denominator.

[0087] The consistency analysis module 230 is used to normalize the annual average response surface matrix of each brand, calculate the surface distance between any two brand normalized matrices, and take the average of the paired distances of all brands. Based on this average, the consistency index between each brand is determined. Specifically, the consistency analysis module 230 uses Min-Max global normalization, uniformly normalizing each brand's matrix using the global minimum and global maximum values ​​of all valid elements in the annual average response surface matrix of all brands; it uses the Frobenius norm to calculate the surface distance between any two brand normalized matrices; and it calculates the arithmetic mean of the paired surface distances of all brands. Substitute Obtain the consistency index (CI); assess the consistency level based on the CI value, and output the interchangeability judgment results of photovoltaic modules of different brands in power plant scenarios.

[0088] Optionally, the device further includes: The storage module is used to store outdoor operation data, grid division parameters, response surface matrices for each period, annual average response surface matrix, and consistency evaluation results for each brand.

[0089] The output module is used to output the consistency index (CI), consistency level assessment results, and interchangeability determination report. The report includes the surface distance matrix between each brand, the normalized response surface matrix, and recommended mixed solutions.

[0090] Finally, it should be noted that: The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for evaluating the consistency of photovoltaic modules based on the joint response surface of irradiance, temperature, and power, characterized in that, include: Obtain outdoor operation data of K brand photovoltaic module strings to be evaluated within the same experimental site and the same operating cycle, and perform data cleaning. The outdoor operation data includes irradiance, ambient temperature and string DC power, where K≥2. On the irradiance-temperature two-dimensional plane, divide it into M×N grid cells. For each brand, each piece of data after cleaning is assigned to the corresponding grid cell according to its irradiance value and temperature value. Here, M is the number of intervals in the irradiance dimension and N is the number of intervals in the temperature dimension. For each brand, the cleaned data is categorized into corresponding grid cells based on irradiance and temperature values ​​at a preset time period. The reliability of the power data in each grid cell within each period is verified. For grid cells that meet the preset reliability conditions, the central tendency statistic of the power data within the cell is calculated as the representative power value of that grid cell. Based on this, an M×N periodic response surface matrix is ​​constructed for each brand, and an M×N annual average response surface matrix is ​​calculated based on each periodic response surface matrix. The annual average response surface matrix of each brand is normalized, the surface distance between any two brand normalized matrices is calculated, and the average of the pairing distances of all brands is taken. The consistency index between each brand is determined based on the average value.

2. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The preset reliability conditions include: the number of valid data points within a grid cell is not less than a preset minimum sample number threshold.

3. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The central tendency statistics include: median, truncated mean, or arithmetic mean.

4. The photovoltaic module consistency evaluation method according to claim 3, characterized in that, The truncated mean is calculated as follows: sort the power data in the grid cells from smallest to largest, remove the data with the highest and lowest α proportions after sorting, and take the arithmetic mean of the remaining data, where α is 3% to 10%.

5. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The central tendency statistic of the power data within the grid cell that meets the preset reliability conditions is used as the power representative value of the grid cell. Based on this, an M×N periodic response surface matrix for each brand is constructed, and an M×N annual average response surface matrix is ​​calculated based on each of the periodic response surface matrices, including: Using the natural month as the time granularity, the central tendency statistics of the power data within the grid cell that meets the preset reliability conditions are used as the representative power value of the grid cell. Construct an M×N monthly response surface matrix for each brand based on the power representative value of each grid cell; The M×N annual average response surface matrix of each brand is obtained by taking the arithmetic mean of each element of the corresponding grid cell in the 12-month response surface matrix. When the number of valid data points for a given month is less than the preset minimum sample size threshold, the grid cell corresponding to that month is automatically excluded when averaging element by element, and the average is taken with the actual number of valid months as the denominator.

6. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The normalization process uses Min-Max normalization, performed according to the following equation: in, Let m be the element in the m-th row and n-th column of the annual average response surface matrix of brand k. and Let be the global minimum and maximum values ​​of all valid elements in the annual average response surface matrix of all K brands, respectively. These are the normalized matrix elements.

7. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The Frobenius norm is used to calculate the surface distance between any two brand normalized matrices. For matrices with missing grids, only the corresponding cells of both brands that are valid grids are summed, and the sum is divided by the number of valid cells and the square root is taken. ； The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in a valid grid. The number of effective units, For the annual average response surface matrix element of brand i, the element in the m-th row and n-th column is... The element in the m-th row and n-th column of the annual average response surface matrix of brand j.

8. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The Frobenius norm is used to calculate the surface distance between any two brand normalized matrices. For matrices with missing grids, the summation is performed only on the corresponding cells where both brands have valid grids, and a data quality weighting is introduced. ； The summation range is (m,n)∈ , Let brand i and brand j be the set of cells in a valid grid. For the annual average response surface matrix element of brand i, the element in the m-th row and n-th column is... The element in the m-th row and n-th column of the annual average response surface matrix of brand j, with weights The variation coefficients of the two brands within that grid cell are jointly determined: ； and denoted as the coefficients of variation for brand i and brand j within grid cell (m,n), respectively.

9. The photovoltaic module consistency evaluation method according to claim 1, characterized in that, The consistency index among brands is determined based on the average value, including: Calculate the average of the surface distances for all brand pairings. Substituting into the following equation, we obtain the consistency index CI: ； Where CI∈(0,1], and the consistency level is determined based on the CI value: When CI≥0.90, the rating is excellent, and the irradiance-temperature-power response characteristics of photovoltaic modules from various brands are highly consistent. When 0.80≤CI<0.90, the rating is good, and the response characteristics of photovoltaic modules from different brands are relatively similar; When 0.70≤CI<0.80, the grade is general, and the response characteristics of photovoltaic modules from different brands vary significantly. When CI < 0.70, the rating is poor, and the response characteristics of photovoltaic modules from different brands vary significantly.

10. A photovoltaic module consistency evaluation device based on the joint response surface of irradiance, temperature, and power, characterized in that, include: The data acquisition module is used to acquire outdoor operation data of K brand photovoltaic module strings to be evaluated within the same operating period at the same test site. The outdoor operation data includes irradiance, ambient temperature and string DC power, where K≥2. The data processing module is used to clean the outdoor operation data, divide the irradiance-temperature two-dimensional plane into M×N grid cells, and assign each cleaned data to the corresponding grid cell according to its irradiance and temperature values ​​for each brand. For grid cells that meet the preset reliability conditions, the central tendency statistics of the power data in the cell are calculated as the power representative value of the grid cell. An M×N periodic response surface matrix is ​​constructed based on the power representative value of each grid cell with a preset time period as the granularity, and an M×N annual average response surface matrix is ​​calculated based on each periodic response surface matrix. The consistency analysis module is used to normalize the annual average response surface matrix of each brand, calculate the surface distance between any two brand normalized matrices and take the average of the pairing distances of all brands, and determine the consistency index between each brand based on the average value.

Images