A photovoltaic module hierarchical detection optimization method, device, equipment and storage medium

Abstract

This invention provides a method, apparatus, device, and storage medium for optimizing the graded testing of photovoltaic modules. The method includes: during the testing process according to the testing sequence of multiple testing parameters, determining the quality level of the photovoltaic module based on the currently detected values ​​of the testing parameters; determining the cumulative testing cost of the completed testing parameters; determining the misjudgment cost of the completed testing parameters based on the quality level of the photovoltaic module; determining the expected cost of the completed testing parameters and the expected cost of continuing to test the next testing parameter based on the cumulative testing cost and the misjudgment cost; determining a target decision result by comparing the expected cost of the completed testing parameters with the expected cost of continuing to test the next testing parameter; and controlling the testing equipment to perform corresponding testing actions based on the target decision result. This significantly reduces testing costs and time while controlling the risk of misjudgment and improving overall efficiency.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic module testing technology, and particularly relates to a photovoltaic module grading and testing optimization method, a photovoltaic module grading and testing optimization device, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In the current photovoltaic module quality inspection and grading system, there is a common bottleneck in balancing inspection efficiency and cost control. The industry's common practice mainly relies on performing full-process, fixed-item inspections on all modules to be inspected, or simplifying the inspection process to control costs.

[0003] While end-to-end inspection ensures the comprehensiveness and accuracy of evaluation results, it leads to high inspection costs and long time cycles. Especially for components with obvious defect characteristics, subsequent inspections are a waste of resources and result in unnecessary over-inspection. Simplified inspection processes often use classification methods based on a single threshold or traditional static models, lacking a mechanism to adjust subsequent inspection strategies according to individual component differences and previous inspection results. Once the number of inspection items is reduced, the risk of misclassification increases significantly, making it difficult to guarantee grading accuracy. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide a photovoltaic module grading and testing optimization method, a photovoltaic module grading and testing optimization device, an electronic device, and a computer-readable storage medium to overcome or at least partially solve the above problems.

[0005] To address the above problems, a first aspect of the present invention provides a photovoltaic module grading and optimization method, the method comprising: During the process of testing photovoltaic modules using testing equipment according to the testing sequence of multiple testing parameters, the quality level of the photovoltaic module is determined based on the detection value of the currently detected testing parameter. Determine the cumulative testing cost of the tested parameters that have been tested; Based on the quality level corresponding to the photovoltaic module, determine the misjudgment cost of the completed testing parameters; Based on the cumulative detection cost and the misjudgment cost, determine the expected cost of the detection parameters that have been detected, and obtain the expected cost of continuing to detect the next detection parameter; The target decision outcome is determined by comparing the expected cost of the completed detection parameters with the expected cost of continuing to detect the next detection parameter; the target decision outcome includes stopping detection or continuing to detect the next detection parameter. Based on the target decision result, the detection equipment is controlled to perform corresponding detection actions.

[0006] Optionally, determining the quality level of the photovoltaic module based on the detection value of the currently detected detection parameter includes: Obtain the value range of the currently detected detection parameter; Determine the distribution position of the detected value of the currently detected parameter within the value range; The quality level of the photovoltaic module is determined based on its distribution location.

[0007] Optionally, the value range includes a first sub-range and a second sub-range; the distribution position includes being located in the first sub-range and being located in the second sub-range; the first sub-range is the range between the corresponding upper limit threshold and lower limit threshold; the second sub-range is the range greater than the upper limit threshold or less than the lower limit threshold.

[0008] Optionally, determining the quality level of the photovoltaic module based on its distribution location includes: If the detected value of the currently detected parameter is located in the first sub-interval, the quality level of the photovoltaic module is determined based on the preset level classification rules. If the detected value of the currently detected parameter is located in the second sub-interval, the initial quality level of the photovoltaic module is output based on the pre-trained grade classification model, and the initial quality level is corrected to obtain the corrected quality level.

[0009] Optionally, the step of correcting the initial quality level to obtain a corrected quality level includes: The reference quality level corresponding to the detection value located in the second sub-interval is determined according to the classification rules. The initial quality level and the reference quality level are weighted and fused to obtain the corrected quality level.

[0010] Optionally, obtaining the expected cost of continuing to detect the next detection parameter includes: Based on the detection parameters that have been detected, predict all possible detection values ​​and corresponding probabilities for the next detection parameter; For all possible test values, determine all possible quality levels of the photovoltaic module for continuing to test the next test parameter; Based on all possible quality levels of the photovoltaic module for the next detection parameter, determine all possible misjudgment costs for the next detection parameter. Obtain the cumulative detection cost for continuing to detect the next detection parameter; Based on all possible misjudgment costs of continuing to detect the next detection parameter and the cumulative detection cost of continuing to detect the next detection parameter, the expected cost of continuing to detect the next detection parameter with the minimum expected cost is determined.

[0011] Optionally, determining the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module includes: Obtain the misjudgment cost rules; the misjudgment cost rules include the costs incurred when the true quality level of the photovoltaic module is misjudged as another quality level; According to the misjudgment cost rule, the first cost of misjudging the true quality level of the photovoltaic module as the quality level corresponding to the photovoltaic module is determined based on the completed testing parameters. Based on the first cost, determine the misjudgment cost of the detection parameters that have been detected.

[0012] Optionally, determining the cost of all possible misjudgments for continuing to detect the next detection parameter based on all possible quality levels of the photovoltaic module includes: Based on the misjudgment cost rule, the second cost of continuing to test the next test parameter is determined as the misjudgment of the true quality level of the photovoltaic module as all possible quality levels of the photovoltaic module. Based on the second cost, determine all possible misjudgment costs for continuing to detect the next detection parameter.

[0013] Optionally, determining the target decision result by comparing the expected cost of the already detected parameters with the expected cost of continuing to detect the next parameter includes: If the expected cost of the already detected parameter is less than the expected cost of continuing to detect the next parameter, then the decision is to stop detection. If the expected cost of continuing to detect the next detection parameter is less than the expected cost of the already detected detection parameter, then the decision is to continue to detect the next detection parameter.

[0014] Optionally, the testing equipment tests the photovoltaic module according to the testing order of multiple testing parameters in the testing queue; the method further includes: Based on the real-time load and availability of each testing device, the photovoltaic modules are assigned to testing queues with different testing parameters in sequence; When the photovoltaic module enters the testing queue or the photovoltaic module completes the testing of a testing parameter, the photovoltaic module is assigned to the optimal testing queue based on the current equipment load rate, queue waiting length, and equipment availability status of each testing device.

[0015] Optionally, the method further includes: Monitor the operating status of each testing device; When a certain detection device becomes unavailable, all detection sequences using that detection device are marked as unavailable. The photovoltaic modules under test in the affected test queues will be rescheduled to other available test queues.

[0016] According to a second aspect of the present invention, a photovoltaic module grading and optimization device is provided, the device comprising: The quality grade determination module is used to determine the quality grade of the photovoltaic module based on the detection value of the currently detected detection parameter during the process of testing the photovoltaic module by the testing equipment according to the detection sequence of multiple detection parameters. The detection cost determination module is used to determine the cumulative detection cost of the detection parameters that have been detected. The misjudgment cost determination module is used to determine the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module. The expected cost determination module is used to determine the expected cost of the detection parameters that have been detected and obtained based on the cumulative detection cost and the misjudgment cost, and to obtain the expected cost of continuing to detect the next detection parameter. The decision result determination module is used to determine the target decision result by comparing the expected cost of the completed detection parameter with the expected cost of continuing to detect the next detection parameter; the target decision result includes stopping detection or continuing to detect the next detection parameter. The detection action execution module is used to control the detection device to perform corresponding detection actions based on the target decision result.

[0017] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the photovoltaic module grading detection optimization method as described in any of the preceding embodiments.

[0018] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a program is stored, wherein the computer program, when executed by a processor, implements the steps of the photovoltaic module grading detection optimization method as described in any of the preceding embodiments.

[0019] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method, apparatus, device, and storage medium for optimizing the graded testing of photovoltaic modules. The method includes: during the testing of photovoltaic modules using a testing device according to a testing sequence of multiple testing parameters, determining the quality level of the photovoltaic module based on the currently detected values ​​of the testing parameters; determining the cumulative testing cost of the completed testing parameters; determining the misjudgment cost of the completed testing parameters based on the quality level of the photovoltaic module; determining the expected cost of the completed testing parameters and the expected cost of continuing to test the next testing parameter based on the cumulative testing cost and the misjudgment cost; determining a target decision result by comparing the expected cost of the completed testing parameters with the expected cost of continuing to test the next testing parameter; the target decision result includes stopping testing or continuing to test the next testing parameter; and controlling the testing device to perform corresponding testing actions based on the target decision result. During the testing process, the expected costs (including cumulative testing cost and potential misjudgment cost) of the two decisions—stopping testing and continuing to test the next parameter—are evaluated in real time, effectively solving the problem of over-testing. While ensuring the reliability of the classification results, it significantly reduces the testing resources consumed by each module, thereby significantly reducing the overall testing cost. By comparing the total expected cost under different decision-making paths, the economic benefits and risk control were optimized. While reducing the number of testing items, the overall misjudgment risk was still controlled at an acceptable low level, thus ensuring the accuracy of the classification.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0021] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of the steps of a photovoltaic module grading and testing optimization method provided in an embodiment of the present invention; Figure 2 This is a flowchart of another photovoltaic module grading and testing optimization method provided in an embodiment of the present invention; Figure 3 This is a logic block diagram of a photovoltaic module grading and testing optimization method provided in an embodiment of the present invention; Figure 4 This is a structural block diagram of a photovoltaic module grading and testing optimization device provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Figure 1 This is a flowchart illustrating the steps of a photovoltaic module grading and testing optimization method provided in an embodiment of the present invention. See also... Figure 1 The method specifically includes the following steps: Step 101: During the process of testing the photovoltaic module in the order of multiple testing parameters using the testing equipment, the quality level of the photovoltaic module is determined based on the detection value of the currently detected testing parameter. In this invention embodiment, the testing equipment is not a single device, but an automated testing system integrating multiple specialized testing instruments, a central control unit, and a data communication network. This testing system comprises specialized testing instrument units configured for different testing parameters, typically including: an EL (electroluminescence) detector to capture the luminescence image of the photovoltaic module under power to identify internal defects such as microcracks, broken grids, and fragments; its core output is the EL image and quantitative parameters extracted after image analysis software (such as maximum microcrack length, total defect area, etc.); a power tester (solar simulator) to measure the maximum output power of the module under standard test conditions and calculate its attenuation rate relative to the nominal value; and an insulation withstand voltage and wet leakage current tester to evaluate the electrical safety performance of the module, detecting whether its insulation resistance and wet leakage current meet safety standards.

[0024] The testing sequence for multiple parameters refers to the order in which various quality tests (such as EL microcrack detection, power degradation testing, and wet leakage current testing) are performed on a photovoltaic module. The system pre-stores several feasible and validated testing sequence schemes, forming a sequence strategy library. For example: Sequence A: First EL testing, then power testing, and finally wet leakage current testing. Sequence B: First power testing, then EL testing, and finally wet leakage current testing. More sequences can be defined based on the actual equipment layout and process characteristics.

[0025] A test value specifically refers to the specific, quantified result data obtained after performing a single measurement on a specific test parameter for a particular photovoltaic module according to a predetermined test method. The test value carries the real-time status information of the module in a specific quality dimension.

[0026] Testing parameters refer to the quantitative data or characteristic values ​​obtained after performing specific physical or electrical testing on photovoltaic modules, reflecting a specific performance, state, or defect of the module. Testing parameters originate from a series of standard or non-standard tests performed on photovoltaic modules and can be mainly divided into several categories: Image and optical testing parameters: sourced from electroluminescence (EL) imaging, infrared thermal imaging (IR), etc. Typical parameters / features include crack / hidden crack features, hot spot features, broken grid / poor solder joint features, and overall image features such as dimensionality-reduced feature vectors extracted through principal component analysis (PCA) that comprehensively reflect image texture, contrast, and brightness. Electrical performance testing parameters: sourced from current-voltage (IV) characteristic tests, insulation resistance tests, wet leakage current tests, etc. Typical parameters / features include direct measurements, feature vectors extracted from IV curves using time-series models such as LSTM that reflect the nonlinear shape of the curve, and can be used to identify anomalies such as increased series resistance, decreased parallel resistance, bypass diode failure, and local shading.

[0027] Quality grades are discretized, tiered category labels for photovoltaic modules. These grades characterize the continuous quality status of modules, from excellent to poor, from usable to unusable.

[0028] In this embodiment of the invention, a definite interval and a fuzzy interval are predefined for each detection parameter. When a detection value (such as the length of an EL microcrack) is obtained, its interval is first determined. If it falls into the definite interval (e.g., microcrack length > 3mm), a preset hard rule (e.g., "greater than 3mm is Class D") is directly applied, immediately assigning the component a definite grade (Class D) and 100% confidence. If it falls into the fuzzy interval (e.g., microcrack length between 1-3mm), a machine learning classification model (e.g., SVM) trained based on historical data is invoked. This model outputs a preliminary grade probability distribution (e.g., [A:0, B:0.1, C:0.3, D:0.6]) based on the current single or multiple measured parameter values. Subsequently, expert prior knowledge (e.g., "microcracks between 1-3mm tend to be Class C / D") is integrated to correct this probability distribution, ultimately generating a graded confidence level, i.e., the probability vector of the component belonging to each grade. This probability vector is the complete mathematical expression of the determined "quality level". It is not a single label, but a probability distribution that includes a measure of uncertainty.

[0029] After the initial detection, an initial quality level probability distribution is formed based on the detected value of the first parameter (such as EL). If the decision is made to continue detection, after obtaining the detected value of the second parameter (such as power attenuation), the current probability distribution is used as a priori, combined with the new detected value, to generate an updated graded confidence level that incorporates information from both parameters. This process iterates, and with each new detected value, the system redetermines (updates) the quality level probability distribution corresponding to the photovoltaic module. This ensures that the assessment of the module's condition is continuously refined as the amount of detected information increases, providing real-time and optimal status basis for each "stop / inspect" decision.

[0030] Step 102: Determine the cumulative testing cost of the completed testing parameters; Cumulative testing cost refers to the total cost of all testing actually paid up to the current decision point in order to obtain the existing testing information of the component. Its calculation is based on a pre-defined, static testing cost list, which explicitly specifies the fixed cost required to perform each testing parameter measurement. Whenever a testing parameter is measured (i.e., a test value is obtained), the corresponding testing cost is looked up from the cost list and added to the component's cost counter. For example, if the component has completed EL testing and power testing, its cumulative testing cost is the sum of the costs of both.

[0031] In this embodiment of the invention, an independent cost counter is maintained for each photovoltaic module under test, with an initial value of zero. Whenever a measurement of a specific parameter (such as EL microcrack detection) is completed, the fixed detection cost of that parameter is accumulated into the counter according to a predefined detection cost schedule (e.g., EL detection cost = 50 yuan / time, power detection cost = 30 yuan / time). Therefore, the accumulated detection cost accurately represents the total economic resources consumed to obtain information about the current state of the module up to the current decision point.

[0032] When assessing the expected total cost of ceasing detection, this cost equals the current cumulative detection cost plus the expected false positive cost calculated based on the current confidence level. When assessing the expected total cost of continuing detection, it equals the current cumulative detection cost plus the fixed detection cost for the next parameter, plus the weighted average of the minimum expected total cost predicted for all possible future states after continuing detection. Cumulative cost is the cornerstone for weighing the economic cost (marginal cost) of each detection step against the marginal benefit reflected in reducing the expected false positive cost on the same economic scale, thereby achieving the optimal dynamic allocation of detection resources.

[0033] Step 103: Determine the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module; The misclassification cost is a predefined cost matrix that measures the potential economic loss or risk caused by incorrect classification results. It is an M×M square matrix (M being the number of quality levels, such as A / B / C / D), where each element C_{ij} represents the cost of misclassifying a component with a true quality level of j as a quality level of i. The diagonal elements C_{jj} are typically 0, indicating that the cost of correct classification is zero. The values ​​of the off-diagonal elements are set according to business logic, with the basic principle being that the loss from misclassifying a high-quality product as a low-quality product is usually small, while the loss from misclassifying a low-quality product as a high-quality product is very high. For example, misclassifying a D-class component with serious safety hazards as a A-class component and allowing it to enter the power plant could lead to high subsequent maintenance costs and compensation, such as fires and power generation losses; therefore, the value of C_{A,D} would be set extremely high. Conversely, misclassifying an A-class component as a B-class component (C_{B,A}) might only result in a slight reduction in its selling price, with a lower cost setting.

[0034] During the decision-making process, the expected misjudgment cost is calculated using the current state's tiered confidence level (i.e., the probability distribution P(j) of a component belonging to each true level) and the misjudgment cost matrix. The calculation formula is: Σ_{j}[P(j)*C_{k,j}]. This expected value directly maps the uncertainty of the classification result to economic risk, enabling a quantitative comparison and optimal trade-off between paying more for detection to achieve more accurate classification and bearing the potential misjudgment loss caused by the uncertainty of the current classification.

[0035] In this embodiment of the invention, the misjudgment cost of the completed testing parameters is determined based on the quality level corresponding to the photovoltaic module. Here, the quality level is not a fixed label, but a calculated level of confidence, i.e., a probability distribution vector [P(A), P(B), P(C), P(D)]. This classification uncertainty is quantified into a specific, calculable economic risk value, the misjudgment cost.

[0036] The specific calculation relies on a pre-defined, static misclassification cost matrix. This matrix defines the business loss caused by misclassifying a component's true level as any other level. The expected misclassification cost is calculated based on the predicted level to be output at the current decision point (usually the highest level k in the confidence level). Expected misclassification cost = Σ(for all true levels j)[P(j)*C(k,j)], where P(j) is the probability that the component's true level is j (i.e., the tiered confidence level), and C(k,j) is the cost of misclassifying the true level j as k.

[0037] Step 104: Based on the cumulative detection cost and the misjudgment cost, determine the expected cost of the detection parameters that have been detected and obtain the expected cost of continuing to detect the next detection parameter; Expected cost is the mathematical expectation of all economic costs anticipated from the current decision point until the final completion of component classification and the assumption of all consequences. Expected cost consists of two core components: cumulative detection cost, which is the sum of all known costs actually paid to obtain all current detection information; and expected misclassification cost, which is the probability-weighted average of economic losses caused by potential future classification errors based on current information (classification confidence level).

[0038] In this embodiment of the invention, the expected cost of the detected parameters that have been detected is determined. This cost is the minimum expected total cost corresponding to choosing to stop detection and output the classification result in the current state, which is the sum of the cumulative detection cost and the current misjudgment cost. Among them, the cumulative detection cost is the sunk cost that has been incurred, while the current expected misjudgment cost is the average risk faced by immediate decision-making based on the current classification confidence level and using the misjudgment cost matrix.

[0039] Secondly, the expected cost of continuing to detect the next parameter is obtained; this cost is the expected total cost corresponding to choosing to continue the action. Based on the currently available information, all possible detection results for the next parameter (e.g., falling into a definite Class D interval, a specific value in an ambiguous interval, etc.) and their probabilities are predicted. For each possible future result, a new state (updated confidence level and cumulative cost) is simulated and calculated, and the minimum expected total cost of that new state is retrieved from the pre-calculated optimal value function table. The expected cost of continuing detection equals the fixed detection cost of the next parameter plus the weighted average of all possible future states according to their probabilities.

[0040] Step 105: By comparing the expected cost of the completed detection parameters with the expected cost of continuing to detect the next detection parameter, a target decision result is determined; the target decision result includes stopping detection or continuing to detect the next detection parameter. In this embodiment of the invention, based on the calculated expected cost, if the expected cost of the completed detection parameter is less than or equal to the expected cost of continuing to detect the next detection parameter, the target decision is to stop detection. If the expected cost of the completed detection parameter is greater than the expected cost of continuing to detect the next detection parameter, the target decision is to continue detecting the next detection parameter. When the expected cost of the completed detection parameter is smaller, it means that the cost of the next detection to obtain information is already higher than the marginal benefit it can bring (the reduction in expected misjudgment cost), so "stopping" is the more economically superior choice. Conversely, it indicates that the next detection, in the long run, can reduce the overall risk and cost, so it should continue.

[0041] Through this continuous comparison and decision-making, instead of pre-planning a complete path for each component, a high-performance detection path with the lowest expected total cost is generated in real time for each component, realizing the transformation from a fixed process to an economically optimal adaptive process.

[0042] Step 106: Based on the target decision result, control the detection device to perform the corresponding detection action.

[0043] In this embodiment of the invention, two different control actions are executed based on the target decision result: If the decision is to stop detection, an instruction is immediately sent to the material handling unit (such as a robotic arm or conveyor belt controller) to remove the current component from the detection line and guide it to the corresponding classification exit (such as Class A or Class B area) based on the highest confidence level in the current classification. Simultaneously, the component's final classification, detection path, cumulative cost, and confidence level are recorded and archived, and the detection task ends. If the decision is to continue detecting the next detection parameter, the next parameter to be tested is first determined according to a preset or dynamically scheduled detection sequence (e.g., after EL detection, the next step is power testing). Then, the component is precisely transported to the detection station corresponding to the next parameter, and the target detection equipment is triggered to start and execute a standardized detection procedure (e.g., powering on the component and taking an EL image).

[0044] By seamlessly integrating intelligent decision-making based on probability and economic models into the execution layer, not only is the detection cost of individual components minimized, but the equipment utilization rate of the entire production line is also optimized through dynamic scheduling (such as assigning different starting detection sequences to different components). Ultimately, the economic benefits of intelligent decision-making are transformed into tangible productivity improvements and operating cost savings.

[0045] Figure 2 This is a flowchart illustrating the steps of a photovoltaic module grading and testing optimization method provided in an embodiment of the present invention. See also... Figure 2 The method specifically includes the following steps: Step 201: During the process of testing the photovoltaic module by the testing equipment according to the testing order of multiple testing parameters, the quality level corresponding to the photovoltaic module is determined according to the detection value of the currently detected testing parameter; the testing equipment tests the photovoltaic module according to the testing order of multiple testing parameters in the testing queue. In this embodiment of the invention, after obtaining a detection value (such as the length of an EL microcrack), the range to which it belongs is first determined. If it falls into a defined range (e.g., microcrack length > 3mm), a preset hard rule (e.g., "greater than 3mm is Class D") is directly applied, immediately assigning the component a defined grade (Class D) and 100% confidence. If it falls into a vague range (e.g., microcrack length between 1-3mm), a machine learning classification model (e.g., SVM) trained based on historical data is invoked. This model outputs a preliminary grade probability distribution based on the current single or multiple measured parameter values ​​(e.g., [A:0, B:0.1, C:0.3, D:0.6]). Subsequently, expert prior knowledge (e.g., "microcracks of 1-3mm are more likely to be Class C / D") is integrated to correct this probability distribution, ultimately generating a graded confidence level, i.e., the probability vector of the component belonging to each grade. This probability vector is the complete mathematical expression of the determined "quality grade," which is not a single label but a probability distribution containing a measure of uncertainty.

[0046] In some embodiments, step 201 may include the following sub-steps: Sub-step S11: Obtain the value range of the currently detected detection parameter; Sub-step S12: Determine the distribution position of the detected value of the currently detected parameter within the value range; Sub-step S13: Determine the quality level corresponding to the photovoltaic module based on the distribution location.

[0047] In this embodiment of the invention, the core logic of determining the quality level based on the current detection value lies in quickly locating and preliminarily interpreting a single detection value based on a preset numerical range and rules, thereby laying the foundation for subsequent comprehensive evaluation and decision-making.

[0048] The system retrieves the static value range of the detection parameter from a pre-stored knowledge base. This range is typically divided into sub-ranges with different properties by a preset threshold. It determines which preset sub-range the currently obtained detection value (e.g., 2.5mm) falls into. Different grade determination processes are triggered based on the distribution location. If the range the detection value falls into is explicitly associated with a grade by a preset rule, that rule is directly applied, outputting the determined grade and 100% confidence. If the range the detection value falls into is a preset ambiguous zone (e.g., "≤3mm"), the system does not provide a definitive conclusion. Instead, it typically uses this detection value as input to call a pre-trained data-driven model (e.g., a classifier) ​​to output a preliminary probability distribution (i.e., a preliminary quality grade confidence). Therefore, the final output is either a definitive classification or a probability distribution, ensuring maximum decision-making efficiency in scenarios with explicit rule coverage, while in complex scenarios, it relies on the data model for refined probability inference.

[0049] In some embodiments, the value range includes a first sub-range and a second sub-range; the distribution position includes being located in the first sub-range and being located in the second sub-range; the first sub-range is an interval between the corresponding upper limit threshold and lower limit threshold; the second sub-range is an interval greater than the upper limit threshold or less than the lower limit threshold.

[0050] In some embodiments, "first sub-interval" and "second sub-interval" specifically refer to regions with clear logical meanings that are divided according to preset dual thresholds.

[0051] Specifically, the first sub-interval is defined as the numerical range between a preset "lower threshold" and a preset "upper threshold." For example, for a performance indicator, its acceptable range can be set, such as "between 90 and 110," and this intermediate range is usually associated with a clear acceptable state. The second sub-interval is defined as the union of numerical ranges below the lower threshold (e.g., "less than 90") and above the upper threshold (e.g., "greater than 110"). These two outer intervals are usually designed as fuzzy intervals, where the indicator is in a deteriorated or abnormal state, but this single parameter is still insufficient to accurately distinguish subtle quality differences within the component (e.g., all are acceptable products, but may belong to categories A and B), so it is necessary to combine other information or models for judgment.

[0052] This partitioning method transforms the continuous parameter space into decision regions with clear semantics: two outer "further analysis / fuzzy" regions and a central "definite" region. The essence of determining the distribution location is to perform a rapid threshold comparison and classify the detected value into one of these regions. This embodies the hybrid decision-making concept of this invention, which efficiently combines simple and clear rule-based judgment with probabilistic inference for complex situations.

[0053] In some embodiments, step S13 may include the following sub-steps: Sub-step S131: If the detection value of the currently detected detection parameter is located in the first sub-interval, then the quality level corresponding to the photovoltaic module is determined based on the preset level classification rules. In sub-step S132, if the detection value of the currently detected detection parameter is located in the second sub-interval, the initial quality level of the photovoltaic module is output based on the pre-trained grade classification model, and the initial quality level is corrected to obtain the corrected quality level.

[0054] In this embodiment of the invention, based on the nature of the interval into which the detected value falls, two essentially different level determination logics are executed separately, reflecting the hybrid architecture of "combining rules and models" in this embodiment of the invention, and the targeted strategies adopted for problems with different degrees of determinism.

[0055] When the detected value falls into the first sub-interval (i.e., the interval between the upper and lower thresholds), this interval is usually preset as a defined interval. At this point, the preset classification rule bound to this interval is directly activated. For example, the rule might directly specify that a leakage current greater than the safety threshold constitutes a Class D non-compliance. Based on this, a defined quality level and its 100% confidence level are output.

[0056] When a detected value falls into the second sub-interval (i.e., the interval greater than the upper threshold or less than the lower threshold), this interval typically corresponds to a fuzzy interval, where a single parameter cannot provide a definitive conclusion. In this case, all currently acquired detected values ​​(possibly more than one) are used as features and input into a pre-trained hierarchical classification model (such as SVM or a neural network). Based on complex and non-linear patterns in historical data, this model outputs an initial quality level probability distribution (e.g., A: 0.7, B: 0.25, C: 0.05, D: 0). Subsequently, this result is not directly adopted but is corrected using expert knowledge. For example, if expert experience indicates that "within this parameter range, the probability of class C should be higher," the system will fuse the model output with the encoded expert prior probabilities using linear weighting or Bayesian methods to obtain the corrected quality level (probability distribution). By fusing prior knowledge to improve robustness, and through the synergy of these two paths, both decision-making efficiency and reliability in explicit scenarios are ensured, while also possessing the ability to perform intelligent and refined evaluations in fuzzy scenarios.

[0057] In some embodiments, step S132 may include the following sub-steps: Sub-step S1321: Determine the reference quality level corresponding to the detection value located in the second sub-interval according to the grade classification rules; Sub-step S1322: The initial quality level and the reference quality level are weighted and fused to obtain the corrected quality level.

[0058] In this embodiment of the invention, the correction of the initial quality level is specified as a quantitative process that integrates data-driven results with expert rule guidance. Even for detection values ​​currently falling into an ambiguous range, expert knowledge may contain rule guidance regarding their associated defined ranges, which can serve as a reference for correction. Rules are not directly applied to the current ambiguous value; instead, a "reference quality level" is derived from the classification rules related to the parameter and its adjacent defined ranges, based on the type and value of the current detection parameter. For example, for an EL microcrack length of 2mm (located in the ambiguous range of 1-3mm), although the rule does not directly determine this, the system will refer to the rule "greater than 3mm is Class D," combined with expert experience (such as "the closer to 3mm, the higher the risk"), to generate a guiding probability distribution, such as biased towards Class C / D.

[0059] The initial quality level probability distribution output is linearly fused with the reference quality level probability distribution derived from rules and experience in the previous step using preset weighting coefficients. The resulting quality level probability distribution is the corrected quality level probability distribution. The judgment made after the fusion of the two has both data-level accuracy and consistency with domain knowledge, significantly improving the reliability and interpretability of decision-making in ambiguous areas.

[0060] Step 202: Determine the cumulative testing cost of the completed testing parameters; In this embodiment of the invention, an independent cost counter is maintained for each photovoltaic module under test, with an initial value of zero. Whenever a measurement of a specific parameter (such as EL microcrack detection) is completed, the fixed detection cost of that parameter is accumulated into the counter according to a predefined detection cost schedule (e.g., EL detection cost = 50 yuan / time, power detection cost = 30 yuan / time). Therefore, the accumulated detection cost accurately represents the total economic resources consumed to obtain information about the current state of the module up to the current decision point.

[0061] Step 203: Determine the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module; In this embodiment of the invention, the misjudgment cost of the completed testing parameters is determined based on the quality level corresponding to the photovoltaic module. Here, the quality level is not a fixed label, but a calculated level of confidence, i.e., a probability distribution vector [P(A), P(B), P(C), P(D)]. This classification uncertainty is quantified into a specific, calculable economic risk value, the misjudgment cost.

[0062] The specific calculation relies on a pre-defined, static misclassification cost matrix. This matrix defines the business loss caused by misclassifying a component's true level as any other level. The expected misclassification cost is calculated based on the predicted level to be output at the current decision point (usually the highest level k in the confidence level). Expected misclassification cost = Σ(for all true levels j)[P(j)*C(k,j)], where P(j) is the probability that the component's true level is j (i.e., the tiered confidence level), and C(k,j) is the cost of misclassifying the true level j as k.

[0063] In some embodiments, step 203 may include the following sub-steps: Sub-step S21: Obtain the misjudgment cost rule; the misjudgment cost rule includes the cost to be incurred when the true quality level of the photovoltaic module is misjudged as another quality level; Sub-step S22: According to the misjudgment cost rule, determine the first cost of misjudging the true quality level of the photovoltaic module as the quality level corresponding to the photovoltaic module based on the completed detection parameters; Sub-step S23: Based on the first cost, determine the misjudgment cost of the detection parameters that have been detected.

[0064] The misclassification cost rule is a set of predefined criteria for quantifying and mapping the business consequences of incorrect quality grading of photovoltaic modules. It is not a simple formula or a single numerical value, but rather typically manifested as a misclassification cost matrix. The core of this rule is that the severity of losses caused by classification errors in different directions varies significantly and asymmetrically.

[0065] The cost of missing high-risk components is extremely high: For example, if a component with serious safety hazards or severely degraded performance (true class D) is mistakenly classified as high-performance (class A or class B), the defective component will enter the power plant, which may cause fire, serious loss of power generation, high maintenance costs, or even compensation for safety accidents. Therefore, its cost C(A,D) or C(B,D) will be set very high.

[0066] The cost of overclassifying low-risk components is relatively low: For example, misclassifying a high-performance component (true grade A) as a lower grade (grade B or C) mainly leads to downgraded sales of the product or additional re-inspection costs. Although it causes a loss of value, it usually does not lead to catastrophic consequences. Therefore, the cost C(B,A) or C(C,A) is set low.

[0067] The cost of a correct determination is zero: the elements on the diagonal of the matrix, namely C(A,A), C(B,B), etc., are all defined as 0.

[0068] In this embodiment of the invention, the core objective of determining the cost of misclassification is to quantify the business consequences of misclassification and integrate them into the economic considerations of the current decision. It is not based on a single true situation, but rather on the current perception of uncertainty regarding the level of truth (i.e., graded confidence levels) to calculate a statistically significant average risk.

[0069] Cost data is obtained from a predefined, static misjudgment cost matrix (i.e., the specific representation of misjudgment cost rules). This matrix specifies the economic losses corresponding to all possible combinations of misjudgments (such as misjudging a true category D as category A, or category B as category C). For example, "the cost of omitting high-risk components (such as misjudging category D as category A) is extremely high, while the cost of overjudging low-risk components (such as misjudging category A as category C) is relatively low."

[0070] Based on the current state, determine the output judgment level (usually the highest level k in the current confidence level) if stopping immediately. Then, according to the misjudgment cost matrix, query and list the cost C_{kj} corresponding to misjudging each possible true level j as k; this set is the "first cost". Combine the current classification confidence level (i.e., the probability P(j) that the photovoltaic module's true level is j) with the first cost, and calculate the misjudgment cost = Σ_j[P(j)*C_{kj}]. This result is essentially a probability-weighted average, representing the average expected economic loss due to misjudgment if stopping now and making a judgment of level k under the current uncertainty of information.

[0071] The resulting uncertainty (confidence level) is multiplied by the severity of the business-level consequences (cost matrix) to output a risk value. This risk value is then added to the detection costs already incurred to form the total cost of halting the action, which is used to compare with the expected total cost of continuing detection. This ensures that every "stop / detect" decision is the economically optimal choice made after clearly weighing the costs of information acquisition against the benefits of risk aversion.

[0072] Step 204: Based on the cumulative detection cost and the misjudgment cost, determine the expected cost of the detection parameters that have been detected and obtain the expected cost of continuing to detect the next detection parameter; In this embodiment of the invention, the expected cost of the detected parameters that have been detected is determined. This cost is the minimum expected total cost corresponding to choosing to stop detection and output the classification result in the current state, which is the sum of the cumulative detection cost and the current misjudgment cost. Among them, the cumulative detection cost is the sunk cost that has been incurred, while the current expected misjudgment cost is the average risk faced by immediate decision-making based on the current classification confidence level and using the misjudgment cost matrix.

[0073] Secondly, the expected cost of continuing to detect the next parameter is obtained; this cost is the expected total cost corresponding to choosing to continue the action. Based on the currently available information, all possible detection results for the next parameter (e.g., falling into a definite Class D interval, a specific value in an ambiguous interval, etc.) and their probabilities are predicted. For each possible future result, a new state (updated confidence level and cumulative cost) is simulated and calculated, and the minimum expected total cost of that new state is retrieved from the pre-calculated optimal value function table. The expected cost of continuing detection equals the fixed detection cost of the next parameter plus the weighted average of all possible future states according to their probabilities.

[0074] In some embodiments, step 204 may include the following sub-steps: Sub-step S31: Based on the detection parameters that have been detected, predict all possible detection values ​​and corresponding probabilities for continuing to detect the next detection parameter; Sub-step S32: For all possible detection values, determine all possible quality levels of the photovoltaic module to continue detecting the next detection parameter; Sub-step S33: Based on all possible quality levels of the photovoltaic module for the next detection parameter, determine all possible misjudgment costs for the next detection parameter. In this embodiment of the invention, based on all currently obtained detection information (measured parameter values ​​and their correlation patterns in historical data), a statistical model or historical distribution is used to predict all possible detection results and their probabilities for the next parameter to be tested. For example, if power attenuation is to be detected next, based on the current EL detection results (such as the presence of microcracks), it may be predicted that there is a 70% probability of attenuation rate <3% (good), a 20% probability of attenuation rate between 3% and 5% (slight degradation), and a 10% probability of attenuation rate >5% (significant degradation).

[0075] For each possible future detection value predicted, the logic for determining the quality level is simulated and invoked. That is, the predicted value is added to the currently known information, and the probability distribution of all possible quality levels of the photovoltaic module under this new information is calculated (i.e., the updated confidence level). This is equivalent to generating multiple possible future states. Finally, for each of these future states (corresponding to a predicted detection value and its resulting quality level probability distribution), the logic for determining the cost of misjudgment is invoked again to calculate the expected cost of misjudgment if the optimal decision is made in that future state.

[0076] For all possible future information acquisition outcomes, we assessed how each outcome would alter our perception of the component's state and ultimately quantified the risk costs remaining under each new perceived state. By weighting these costs according to their probability of occurrence and adding the fixed cost of the next detection, we obtained the overall expected cost of continuing detection, enabling a fair and forward-looking comparison with the cost of stopping.

[0077] In some embodiments, step S33 may include the following sub-steps: Sub-step S331: Based on the misjudgment cost rule, determine the second cost of continuing to detect the next detection parameter. The true quality level of the photovoltaic module is misjudged as all possible quality levels of the photovoltaic module. Sub-step S332: Based on the second cost, determine all possible misjudgment costs for continuing to detect the next detection parameter.

[0078] In this embodiment of the invention, for a given, predicted future detection result (e.g., a predicted power degradation rate of 4%), the probability distribution of all possible quality levels of the photovoltaic module under this result has been determined (e.g., the updated confidence level is [A:0.1, B:0.6, C:0.3, D:0]). First, the output level (assuming the highest probability, class B) is determined if a judgment is made under this condition. Then, according to the misjudgment cost rule (i.e., the misjudgment cost matrix), the costs C(B,A), C(B,B), C(B,C), and C(B,D) incurred for misjudging each possible true level (A, B, C, D) as this output level (class B) are queried and listed. This cost set is the second cost, representing a list of losses for all misjudgment directions in the future scenario.

[0079] The obtained quality level probability distribution (as the probability weight of the true level) in the future state is combined with the second cost list obtained from the query, and the misjudgment cost is calculated according to the formula: Misjudgment Cost = P(A)*C(B,A) + P(B)*C(B,B) + P(C)*C(B,C) + P(D)*C(B,D). The calculation result is the expected misjudgment loss that would occur if detection were stopped and a judgment was made under specific predicted future information.

[0080] Sub-step S34: Obtain the cumulative detection cost for continuing to detect the next detection parameter; Sub-step S35: Based on all possible misjudgment costs of continuing to detect the next detection parameter and the cumulative detection cost of continuing to detect the next detection parameter, determine the expected cost of continuing to detect the next detection parameter with the minimum expected cost.

[0081] In this embodiment of the invention, obtaining the cumulative detection cost of continuing to detect the next detection parameter clarifies the direct economic expenditure that inevitably arises from the decision to continue detecting the next detection parameter. The cumulative detection cost here does not refer to the total historical cost of the component up to the present, but specifically refers to the estimated total cost at the moment the next parameter detection is completed, assuming the continuation action is performed. Estimated cumulative detection cost = current cumulative detection cost + fixed detection cost of the next parameter. For example, if the current cumulative cost is 50 yuan and the cost of the next power detection is 30 yuan, then this value is 80 yuan.

[0082] From all possible future paths, find the path with the minimum expected total cost, and define this minimum as the expected cost of continuing to test the next parameter. Integrate all predicted possible future outcomes, their probabilities, and the corresponding expected misjudgment cost calculated for each future outcome. For each predicted future, the total expected cost of its complete path = estimated cumulative testing cost + expected misjudgment cost for that future. However, this is only the cost of stopping immediately after testing the next parameter. According to the optimality principle of dynamic programming, after testing the next parameter, one can still choose to continue or stop. Therefore, from the pre-calculated optimal value function table, query the optimal residual expected cost (V_next) corresponding to each future state. This value represents the minimum total cost from that state to the end by adopting the optimal strategy. Finally, the overall expected cost of continuing is: E_continue = (fixed testing cost of the next parameter) + Σ[probability(future i) * V_next(future i)]. This represents the overall minimum average cost achievable by adopting the optimal response strategy for all possible future evolutions after paying the fixed cost of the next test.

[0083] Step 205: By comparing the expected cost of the completed detection parameters with the expected cost of continuing to detect the next detection parameter, a target decision result is determined; the target decision result includes stopping detection or continuing to detect the next detection parameter. In this embodiment of the invention, based on the calculated expected cost, if the expected cost of the completed detection parameter is less than or equal to the expected cost of continuing to detect the next detection parameter, the target decision is to stop detection. If the expected cost of the completed detection parameter is greater than the expected cost of continuing to detect the next detection parameter, the target decision is to continue detecting the next detection parameter. When the expected cost of the completed detection parameter is smaller, it means that the cost of the next detection to obtain information is already higher than the marginal benefit it can bring (the reduction in expected misjudgment cost), so "stopping" is the more economically superior choice. Conversely, it indicates that the next detection, in the long run, can reduce the overall risk and cost, so it should continue.

[0084] In some embodiments, step 205 may include the following sub-steps: In sub-step S41, if the expected cost of the already detected parameter is less than the expected cost of continuing to detect the next parameter, the decision result is to stop detection; if the expected cost of continuing to detect the next parameter is less than the expected cost of the already detected parameter, the decision result is to continue to detect the next parameter.

[0085] In this embodiment of the invention, if the expected cost of the completed detection parameter is less than the expected cost of continuing to detect the next detection parameter, the decision is to stop detection; if the expected cost of continuing to detect the next detection parameter is less than the expected cost of the completed detection parameter, the decision is to continue detecting the next detection parameter. Comparing the expected cost of the completed detection parameter with the expected cost of continuing to detect the next detection parameter is essentially comparing the risk cost of stopping and bearing the current information, and investing more to obtain more information, thereby potentially reducing future risks. When the expected cost of the completed detection parameter is smaller, it means that the expected marginal benefit of continuing detection (i.e., the expected reduction in the expected cost of misjudgment) is already lower than the cost of the next detection, so stopping is an economically better choice. Conversely, it indicates that continuing is worthwhile.

[0086] This ensures that the entire decision-making process satisfies the principle that, regardless of the initial state and previous decisions, the current decision must constitute the optimal strategy for the remaining decisions. At each step, the path leading to the lowest expected total cost is automatically selected, thus dynamically generating a unique, cost-effective testing sequence for each photovoltaic module. This achieves a transformation from fixed-process testing to economically optimal adaptive testing.

[0087] Step 206: Based on the target decision result, control the detection device to perform the corresponding detection action.

[0088] In this embodiment of the invention, two different control actions are executed based on the target decision result: If the decision is to stop detection, an instruction is immediately sent to the material handling unit (such as a robotic arm or conveyor belt controller) to remove the current component from the detection line and guide it to the corresponding classification exit (such as Class A or Class B area) based on the highest confidence level in the current classification. Simultaneously, the component's final classification, detection path, cumulative cost, and confidence level are recorded and archived, and the detection task ends. If the decision is to continue detecting the next detection parameter, the next parameter to be tested is first determined according to a preset or dynamically scheduled detection sequence (e.g., after EL detection, the next step is power testing). Then, the component is precisely transported to the detection station corresponding to the next parameter, and the target detection equipment is triggered to start and execute a standardized detection procedure (e.g., powering on the component and taking an EL image).

[0089] Step 207: Based on the real-time load and availability of each testing device, the photovoltaic modules are assigned to testing queues with different testing parameters in sequence; In this embodiment of the invention, based on the real-time operation status of the entire production line, an initial or subsequent detection path (i.e., the order of detection parameters) is dynamically selected for each newly entered component or component that has completed a certain step, thereby achieving efficient utilization of equipment resources and maximizing overall throughput.

[0090] Specifically, a "testing sequence strategy library" is maintained, where each strategy corresponds to a specific parameter testing sequence (e.g., "EL first, then power" sequence A, "Power first, then EL" sequence B). Simultaneously, the real-time load (e.g., current queue length, equipment utilization) and availability status (e.g., online, busy, calibrating, faulty) of all testing stations (equipment) are monitored. When a queue needs to be allocated to a component, the scheduling algorithm makes decisions based on minimizing the estimated completion time or balancing the load across stations. For example, if the queue for EL testing is too long, while the queue for power testing is idle, the scheduler will tend to allocate the new component to the "Power first, then EL" (sequence B) queue, allowing the component to utilize idle resources to complete the power test first, thereby alleviating congestion at the EL station. By dynamically adjusting the order of testing tasks, load balancing is achieved, preventing individual devices from becoming bottlenecks due to fixed processes. This solution not only saves costs for each photovoltaic module but also saves time for the entire production line, achieving a dual improvement in economic benefits and production efficiency.

[0091] Step 208: When the photovoltaic module enters the testing queue or the photovoltaic module completes the testing of a testing parameter, the photovoltaic module is assigned to the optimal testing queue according to the current equipment load rate, queue waiting length and equipment availability status of each testing device.

[0092] In this embodiment of the invention, when the component first enters the detection process, and after the component completes each detection parameter (at which point it is at a decision point, needing to decide not only "stop / detect" but also "which path to take next"), its core logic is based on continuous optimization using global real-time state awareness.

[0093] A snapshot of a testing device is captured, including: the current load rate of each testing station (device) (reflecting the device's busyness), the waiting length of each testing queue (directly reflecting queuing time), and the availability status of all devices (operating, faulty, under maintenance, etc.). Based on the shortest queue, shortest estimated completion time, or load balancing rules, an optimal testing queue is calculated and selected for the current component. Here, "optimal" typically refers to minimizing the component's estimated total testing time or best balancing the overall line load. For example, a component has just completed EL testing and is originally scheduled to enter the power testing phase of the "EL → Power" queue. However, the scheduler finds that the power testing equipment queue is too long, while the wet leakage current testing equipment is idle. It may decide to reassign the component to a new path (such as "EL completed → Wet leakage current test →...") queue, thus bypassing the congestion point.

[0094] This optimized architecture of global scheduling and path optimization enables each component to select the best detection order based on real-time device load and stop detection in advance when the destination is clear enough, thereby achieving the combined optimality of overall efficiency and individual economy in complex production environments.

[0095] In some embodiments, the method further includes: Monitor the operating status of each testing device; when a testing device becomes unavailable, mark all testing sequences using that testing device as unavailable; reschedule the photovoltaic modules under test in the affected testing queue to other available testing queues.

[0096] In this embodiment of the invention, when a critical resource fails unexpectedly, the photovoltaic modules under test in the affected testing queue are rescheduled to other available testing queues to ensure that the entire testing system can continue to operate.

[0097] Specifically, the operational status of each testing device is continuously monitored (e.g., through device heartbeat signals, fault codes, or operator manual marking). When a testing device is detected to be "unavailable" (e.g., EL camera malfunction, power tester calibration timeout), a response process is immediately triggered. In the internal testing sequence strategy library, all testing sequences containing the testing steps corresponding to the malfunctioning device (e.g., all sequences that use EL testing as the first or intermediate station) are marked as unavailable.

[0098] This not only prevents new components from entering these disabled routes, but more importantly, it involves emergency rescheduling of photovoltaic modules already on these routes that have not yet completed the corresponding testing steps for the faulty equipment. These modules are removed from their current queues, and, based on the status of all available equipment, rescheduled to other available testing queues. For example, a module waiting for EL testing in the "EL→Power" queue will be reassigned to an available "Power→Wet Leakage" queue after an EL device failure, starting from its currently executable first step.

[0099] This mechanism ensures production continuity, preventing a complete blockage due to a single point of failure; it maintains the effectiveness of decision-making, ensuring that all components are always on an effective and executable detection path. This allows optimization to work not only under ideal conditions but also to operate stably under real-world fluctuations.

[0100] Reference Figure 3 The diagram shows a logic block diagram of a photovoltaic module graded detection optimization method provided by an embodiment of the present invention. Figure 3This invention describes the key decision-making process of a photovoltaic module grading detection optimization method according to an embodiment of the present invention. First, the detected value is acquired and its corresponding interval (first sub-interval or second sub-interval) is determined. Different classification strategies are selected based on the interval position: if it is in the first sub-interval, a preset rule is applied to directly determine the grade; if it is in the second sub-interval, a pre-trained model is invoked to determine the grade. Subsequently, the expected costs of the two decision paths—complete detection and continuing to detect the next parameter—are calculated and compared. If the expected cost of completing detection is lower, detection stops and the result is output; otherwise, the detection of the next parameter continues, thereby achieving a dynamic balance and optimization between detection cost and classification accuracy.

[0101] This invention, during the detection process, evaluates in real-time the expected costs (including cumulative detection costs and potential misjudgment costs) of two decisions: stopping and continuing to detect the next parameter. This effectively solves the over-detection problem and significantly reduces the detection resources consumed by each component while ensuring the reliability of the classification results, thereby significantly reducing the overall detection cost. By comparing the total expected costs under different decision paths, the invention optimizes economic benefits and risk control. While reducing the number of detection items, it still keeps the overall misjudgment risk at an acceptable low level, thus ensuring the accuracy of the classification.

[0102] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0103] Figure 4 This is a structural block diagram of a photovoltaic module grading and optimization device provided in an embodiment of the present invention. (Refer to...) Figure 4 The device specifically includes the following modules: The quality grade determination module 301 is used to determine the quality grade of the photovoltaic module based on the detection value of the currently detected detection parameter during the process of testing the photovoltaic module by the testing equipment according to the detection sequence of multiple detection parameters. The detection cost determination module 302 is used to determine the cumulative detection cost of the detection parameters that have been detected. The misjudgment cost determination module 303 is used to determine the misjudgment cost of the completed detection parameters based on the quality level corresponding to the photovoltaic module. The expected cost determination module 304 is used to determine the expected cost of the detection parameters that have been detected based on the cumulative detection cost and the misjudgment cost, and to obtain the expected cost of continuing to detect the next detection parameter. The decision result determination module 305 is used to determine the target decision result by comparing the expected cost of the completed detection parameter with the expected cost of continuing to detect the next detection parameter; the target decision result includes stopping detection or continuing to detect the next detection parameter. The detection action execution module 306 is used to control the detection device to perform corresponding detection actions based on the target decision result.

[0104] In some embodiments, the quality level determination module 301 includes: The value range acquisition submodule is used to acquire the value range of the currently detected detection parameter; The distribution location determination submodule is used to determine the distribution location of the detected values ​​of the currently detected parameters within the value range; The component grade determination submodule is used to determine the quality grade corresponding to the photovoltaic module based on the distribution location.

[0105] In some embodiments, the value range includes a first sub-range and a second sub-range; the distribution position includes being located in the first sub-range and being located in the second sub-range; the first sub-range is an interval between the corresponding upper limit threshold and lower limit threshold; the second sub-range is an interval greater than the upper limit threshold or less than the lower limit threshold.

[0106] In some embodiments, the component level determination submodule includes: The first detection value judgment unit is used to determine the quality level of the photovoltaic module based on a preset level classification rule if the detection value of the currently detected detection parameter is located in the first sub-interval. The second detection value judgment unit is used to output the initial quality level of the photovoltaic module based on the pre-trained grade classification model if the detection value of the currently detected detection parameter is located in the second sub-interval, and to correct the initial quality level to obtain the corrected quality level.

[0107] In some embodiments, the second detection value determination unit includes: The reference quality level determination subunit is used to determine the reference quality level corresponding to the detection value located in the second sub-interval according to the grade classification rules. The quality level correction subunit is used to perform weighted fusion of the initial quality level and the reference quality level to obtain the corrected quality level.

[0108] In some embodiments, the expected cost determination module 304 includes: The detection value prediction submodule is used to predict all possible detection values ​​and their corresponding probabilities for the next detection parameter based on the detection parameters that have been detected. The quality level prediction submodule is used to determine all possible quality levels of the photovoltaic module for continuing to detect the next detection parameter, based on all possible detection values. The parameter cost determination submodule is used to determine all possible misjudgment costs for continuing to detect the next detection parameter based on all possible quality levels of the photovoltaic module. The cumulative detection cost acquisition submodule is used to obtain the cumulative detection cost for continuing to detect the next detection parameter; The parameter expected cost determination submodule is used to determine the expected cost of the next detection parameter by minimizing the expected cost based on all possible misjudgment costs of the next detection parameter and the cumulative detection cost of the next detection parameter.

[0109] In some embodiments, the misjudgment cost determination module 303 includes: The cost rule acquisition submodule is used to acquire misjudgment cost rules; the misjudgment cost rules include the costs that need to be incurred when the true quality level of the photovoltaic module is misjudged as another quality level; The first cost determination submodule is used to determine the first cost of misjudging the true quality level of the photovoltaic module as the quality level corresponding to the photovoltaic module, based on the misjudgment cost rule; The measured parameter cost determination submodule is used to determine the misjudgment cost of the measured parameters based on the first cost.

[0110] In some embodiments, the parameter cost determination submodule includes: The second cost determination unit is used to determine, according to the misjudgment cost rule, the second cost of continuing to detect the next detection parameter of the photovoltaic module, where the true quality level of the photovoltaic module is misjudged as all possible quality levels of the photovoltaic module; The next parameter cost determination unit is used to determine, based on the second cost, all possible misjudgment costs of continuing to detect the next detection parameter.

[0111] In some embodiments, the decision result determination module 305 includes: The first expected cost judgment submodule is used to decide whether to stop detection if the expected cost of the already detected detection parameter is less than the expected cost of continuing to detect the next detection parameter. The second expected cost judgment submodule is used to decide whether to continue detecting the next detection parameter if the expected cost of continuing to detect the next detection parameter is less than the expected cost of the detection parameter that has been detected.

[0112] In some embodiments, the testing device tests the photovoltaic module according to the testing order of multiple testing parameters in the testing queue; the device further includes: The photovoltaic module allocation module is used to allocate the photovoltaic modules to detection queues with different detection parameters according to the real-time load and availability status of each detection device. The detection queue allocation module is used to allocate the photovoltaic module to the optimal detection queue based on the current equipment load rate, queue waiting length, and equipment availability status of each detection device when the photovoltaic module enters the detection queue or after the photovoltaic module completes the detection of a detection parameter.

[0113] In some embodiments, the apparatus further includes: The operation status monitoring module is used to monitor the operation status of each testing device; The detection sequence marking module is used to mark all detection sequences using the detection device as unavailable when the status of a certain detection device becomes unavailable. The photovoltaic module scheduling module is used to reschedule photovoltaic modules under test in the affected testing queue to other available testing queues.

[0114] As the apparatus embodiment is basically similar to the method embodiment, it is described in a relatively simple manner. For relevant details, please refer to the description of the method embodiment.

[0115] This invention also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, it implements the various processes of the above-described photovoltaic module graded detection optimization method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0116] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described photovoltaic module graded detection optimization method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0117] 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 optimizing the graded testing of photovoltaic modules, characterized in that, The method includes: During the process of testing photovoltaic modules using testing equipment according to the testing sequence of multiple testing parameters, the quality level of the photovoltaic module is determined based on the detection value of the currently detected testing parameter. Determine the cumulative testing cost of the tested parameters that have been tested; Based on the quality level corresponding to the photovoltaic module, determine the misjudgment cost of the completed testing parameters; Based on the cumulative detection cost and the misjudgment cost, determine the expected cost of the detection parameters that have been detected, and obtain the expected cost of continuing to detect the next detection parameter; The target decision outcome is determined by comparing the expected cost of the completed detection parameters with the expected cost of continuing to detect the next detection parameter; the target decision outcome includes stopping detection or continuing to detect the next detection parameter. Based on the target decision result, the detection equipment is controlled to perform corresponding detection actions.

2. The photovoltaic module grading and testing optimization method according to claim 1, characterized in that, The step of determining the quality level of the photovoltaic module based on the detection value of the currently detected detection parameter includes: Obtain the value range of the currently detected detection parameter; Determine the distribution position of the detected value of the currently detected parameter within the value range; The quality level of the photovoltaic module is determined based on its distribution location.

3. The photovoltaic module grading and testing optimization method according to claim 2, characterized in that, The value range includes a first sub-range and a second sub-range; the distribution position includes the first sub-range and the second sub-range; the first sub-range is the range between the corresponding upper limit threshold and lower limit threshold; the second sub-range is the range greater than the upper limit threshold or less than the lower limit threshold.

4. The photovoltaic module grading and testing optimization method according to claim 3, characterized in that, Determining the quality level of the photovoltaic module based on its distribution location includes: If the detected value of the currently detected parameter is located in the first sub-interval, the quality level of the photovoltaic module is determined based on the preset level classification rules. If the detected value of the currently detected parameter is located in the second sub-interval, the initial quality level of the photovoltaic module is output based on the pre-trained grade classification model, and the initial quality level is corrected to obtain the corrected quality level.

5. The photovoltaic module grading and testing optimization method according to claim 4, characterized in that, The step of correcting the initial quality level to obtain the corrected quality level includes: The reference quality level corresponding to the detection value located in the second sub-interval is determined according to the classification rules. The initial quality level and the reference quality level are weighted and fused to obtain the corrected quality level.

6. The photovoltaic module grading and testing optimization method according to claim 1, characterized in that, The expected cost of obtaining the parameters for continuing to detect the next step includes: Based on the detection parameters that have been detected, predict all possible detection values ​​and corresponding probabilities for the next detection parameter; For all possible test values, determine all possible quality levels of the photovoltaic module for continuing to test the next test parameter; Based on all possible quality levels of the photovoltaic module for the next detection parameter, determine all possible misjudgment costs for the next detection parameter. Obtain the cumulative detection cost for continuing to detect the next detection parameter; Based on all possible misjudgment costs of continuing to detect the next detection parameter and the cumulative detection cost of continuing to detect the next detection parameter, the expected cost of continuing to detect the next detection parameter with the minimum expected cost is determined.

7. The photovoltaic module grading and testing optimization method according to claim 1, characterized in that, The step of determining the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module includes: Obtain the misjudgment cost rules; the misjudgment cost rules include the costs incurred when the true quality level of the photovoltaic module is misjudged as another quality level; According to the misjudgment cost rule, the first cost of misjudging the true quality level of the photovoltaic module as the quality level corresponding to the photovoltaic module is determined based on the completed testing parameters. Based on the first cost, determine the misjudgment cost of the detection parameters that have been detected.

8. The photovoltaic module grading and testing optimization method according to claim 6, characterized in that, The step of determining all possible misjudgment costs for continuing to detect the next detection parameter based on all possible quality levels of the photovoltaic module includes: Based on the misjudgment cost rule, the second cost of continuing to test the next test parameter is determined as the misjudgment of the true quality level of the photovoltaic module as all possible quality levels of the photovoltaic module. Based on the second cost, determine all possible misjudgment costs for continuing to detect the next detection parameter.

9. The photovoltaic module grading and testing optimization method according to claim 1, characterized in that, The process of determining the target decision result by comparing the expected cost of the already detected parameters with the expected cost of continuing to detect the next parameter includes: If the expected cost of the already detected parameter is less than the expected cost of continuing to detect the next parameter, then the decision is to stop detection. If the expected cost of continuing to detect the next detection parameter is less than the expected cost of the already detected detection parameter, then the decision is to continue to detect the next detection parameter.

10. The photovoltaic module grading and testing optimization method according to claim 1, characterized in that, The testing equipment tests the photovoltaic module according to the testing order of multiple testing parameters in the testing queue; the method further includes: Based on the real-time load and availability of each testing device, the photovoltaic modules are assigned to testing queues with different testing parameters in sequence; When the photovoltaic module enters the testing queue or the photovoltaic module completes the testing of a testing parameter, the photovoltaic module is assigned to the optimal testing queue based on the current equipment load rate, queue waiting length, and equipment availability status of each testing device.

11. The photovoltaic module grading and testing optimization method according to claim 10, characterized in that, The method further includes: Monitor the operating status of each testing device; When a certain detection device becomes unavailable, all detection sequences using that detection device are marked as unavailable. The photovoltaic modules under test in the affected test queues will be rescheduled to other available test queues.

12. A photovoltaic module grading and optimization device, characterized in that, The device includes: The quality grade determination module is used to determine the quality grade of the photovoltaic module based on the detection value of the currently detected detection parameter during the process of testing the photovoltaic module by the testing equipment according to the detection sequence of multiple detection parameters. The detection cost determination module is used to determine the cumulative detection cost of the detection parameters that have been detected. The misjudgment cost determination module is used to determine the misjudgment cost of the completed testing parameters based on the quality level corresponding to the photovoltaic module. The expected cost determination module is used to determine the expected cost of the detection parameters that have been detected and obtained based on the cumulative detection cost and the misjudgment cost, and to obtain the expected cost of continuing to detect the next detection parameter. The decision result determination module is used to determine the target decision result by comparing the expected cost of the completed detection parameter with the expected cost of continuing to detect the next detection parameter; the target decision result includes stopping detection or continuing to detect the next detection parameter. The detection action execution module is used to control the detection device to perform corresponding detection actions based on the target decision result.

13. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the photovoltaic module grading and testing optimization method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the steps of the photovoltaic module grading detection optimization method as described in any one of claims 1-11.

Images