Photovoltaic module surface detection method and system

Abstract

According to the photovoltaic module surface detection method and system provided by the invention, the first module surface state description information and the surface anomaly description information are loaded to the remote module surface recognition model, and the surface information recognition result is generated based on the remote module surface recognition model; and determining a photovoltaic module surface detection result corresponding to the specified photovoltaic module object according to a surface detection step in the surface information identification result, which can be understood that the photovoltaic module surface detection result corresponding to the specified photovoltaic module object is determined according to the first module surface state description information and the surface anomaly description information. Therefore, the surface detection result of the photovoltaic module can be accurately and reliably determined, the condition that the surface of the photovoltaic module is damaged can be determined in time, the damage of photovoltaic module equipment is avoided, and the photovoltaic module is protected in time.

Description

Technical Field

[0001] The present application relates to the field of surface recognition technology, and in particular to a photovoltaic module surface detection method and system. Background Art

[0002] Photovoltaic modules have a wide coverage and are also vulnerable parts. Therefore, there will be protective materials on the surface of photovoltaic modules to protect the photovoltaic modules in a targeted manner. However, the surface of photovoltaic modules is easily damaged. Without protection, photovoltaic modules are easily damaged and cause failures. Therefore, a technical solution is urgently needed to perform real-time detection of the surface of photovoltaic modules to solve the above technical problems. Summary of the invention

[0003] In view of this, the present application provides a photovoltaic module surface detection method and system.

[0004] In a first aspect, a photovoltaic module surface detection method is provided, which is applied to a remote control system, and the method comprises: Searching for first component surface state description information corresponding to a specified photovoltaic component object in a photovoltaic component surface description record; Determine surface anomaly difference data between the first component surface state description information and the surface state description information of a sample component of the first component surface state description information, and determine surface anomaly description information corresponding to the surface anomaly difference data; The first component surface state description information and the surface abnormality description information are loaded into a remote component surface recognition model that meets the model deployment conditions, so as to identify component surface description features from the first component surface state description information based on the remote component surface recognition model, identify surface change features from the surface abnormality description information, cluster the component surface description features and the surface change features, and obtain a clustering vector matrix; wherein the component surface description features represent the surface condition description information of the photovoltaic component in the first component surface state description information, and the surface change features represent the data after the surface of the photovoltaic component in the surface abnormality description information has a difference; Loading the clustering vector matrix and the component surface description features into the remote component surface recognition model to generate a surface information recognition result based on the remote component surface recognition model; The photovoltaic component surface detection result corresponding to the designated photovoltaic component object is determined according to the surface detection step in the surface information recognition result.

[0005] In the present application, the generating of the surface anomaly description information corresponding to the surface anomaly difference data includes: If it is determined that the surface abnormality difference data includes a recognition angle and a recognition method, the recognition angle is matched to the recognition angle feature of the sample feature set, and the recognition method is matched to the recognition method feature of the sample feature set; The surface anomaly description information is generated according to the recognition angle feature and the recognition method feature.

[0006] In the present application, the remote component surface recognition model includes a first fully connected layer, a second fully connected layer, and a third fully connected layer; The method of identifying component surface description features from the first component surface state description information based on the remote component surface recognition model, identifying surface change features from the surface abnormality description information, clustering the component surface description features and the surface change features to obtain a clustering vector matrix includes: The first fully connected layer based on the remote component surface recognition model identifies component surface description features from the first component surface state description information, and loads the component surface description features to the third fully connected layer; The second fully connected layer based on the remote component surface recognition model identifies surface change features from the surface anomaly description information, and loads the surface change features to the third fully connected layer; The component surface description features and the surface change features are clustered based on the third fully connected layer to obtain a clustering vector matrix, and the clustering vector matrix is ​​loaded into the first fully connected layer.

[0007] In the present application, the component surface description features and the surface change features are clustered based on the third fully connected layer to obtain a clustering vector matrix, including: Performing vector expansion on the surface change feature to obtain the surface change feature after vector expansion, and performing vector data update processing on the surface change feature after vector expansion using a data update layer to obtain the surface change feature after vector data update processing; Clustering is performed on the component surface description features and the surface change features after the vector data update processing to obtain a clustering vector matrix.

[0008] In the present application, after determining the photovoltaic component surface detection result corresponding to the specified photovoltaic component object according to the surface detection step in the surface information recognition result, it also includes: Determining a remote identification operation that matches the photovoltaic module surface detection result; Determining identification step training data corresponding to the specified photovoltaic component object according to the remote identification operation; The training data of the statistical identification step covers the training item coefficients of the preset training data; The photovoltaic component surface detection result is run based on the identification step training data and the training item coefficients.

[0009] In this application, the remote component surface recognition model is obtained based on the following steps: Searching for first reference surface abnormality description information corresponding to a specified photovoltaic component object in the photovoltaic component surface description record; Determine the difference description content between the first reference surface abnormality description information and the surface abnormality description information example of the first reference surface abnormality description information, and generate the difference feature description content corresponding to the difference description content; The first reference surface abnormality description information and the distinguishing feature description content are loaded into the original remote component surface recognition model, so as to identify an abnormal feature vector from the first reference surface abnormality description information based on the original remote component surface recognition model, identify a component surface example from the distinguishing feature description content, cluster the abnormal feature vector and the component surface example, and obtain a cluster reference vector; wherein the abnormal feature vector represents the surface condition description information of the photovoltaic component in the first reference surface abnormality description information, and the component surface example represents the data after the surface of the photovoltaic component in the distinguishing feature description content has a difference; Loading the cluster reference vector and the abnormal feature vector into the original remote component surface recognition model to generate predicted reference surface abnormality description information based on the original remote component surface recognition model; Determine a prediction cost coefficient according to the prediction reference surface anomaly description information and the prediction operation annotated surface anomaly description information carrying annotations; The model weight of the original remote component surface recognition model is updated according to the predicted cost coefficient to obtain a remote component surface recognition model.

[0010] In the present application, the updating of the model weight of the original remote component surface recognition model according to the predicted cost coefficient to obtain the remote component surface recognition model includes: updating the weight data of the original remote component surface recognition model according to the predicted cost coefficient to obtain an updated candidate remote component surface recognition model, and determining whether the updated candidate remote component surface recognition model meets the model deployment condition; If the determination is no, the updated candidate remote component surface recognition model is determined as the original remote component surface recognition model, and the operation of loading the first reference surface anomaly description information and the distinguishing feature description content into the original remote component surface recognition model is returned to execute; If the determination is yes, the updated candidate remote component surface recognition model is determined as the remote component surface recognition model.

[0011] In a second aspect, a photovoltaic module surface detection system is provided, comprising a processor and a memory communicating with each other, wherein the processor is used to read a computer program from the memory and execute the computer program to implement the above method.

[0012] A photovoltaic module surface detection method and system provided in an embodiment of the present application loads first component surface state description information and surface abnormality description information into a remote component surface recognition model, generates a surface information recognition result based on the remote component surface recognition model, and determines the photovoltaic module surface detection result corresponding to the specified photovoltaic module object according to the surface detection step in the surface information recognition result. It can be understood that the photovoltaic module surface detection result corresponding to the specified photovoltaic module object is determined based on the first component surface state description information and the surface abnormality description information, and then the photovoltaic module surface detection result is accurately and reliably determined, so that the surface damage can be determined in time, the damage of the photovoltaic module equipment can be avoided, and the photovoltaic module can be protected in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A flow chart of a photovoltaic module surface detection method provided in an embodiment of the present application.

[0015] Figure 2 A block diagram of a photovoltaic module surface detection device provided in an embodiment of the present application.

[0016] Figure 3 This is an architectural diagram of a photovoltaic module surface detection system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0018] See also Figure 1 , shows a photovoltaic module surface detection method, which may include the technical solutions described in the following steps 201-205.

[0019] Step 201: searching for first component surface state description information corresponding to a specified photovoltaic component object in a photovoltaic component surface description record.

[0020] Exemplarily, the photovoltaic component surface description record can find the first surface abnormality description information corresponding to the specified photovoltaic component object, and load the first surface abnormality description information to the recording device. In order to better identify related data, the first surface abnormality description information in the training operation can be recorded as the first reference surface abnormality description information. Further, the recording device can obtain the first reference surface abnormality description information corresponding to the specified photovoltaic component object.

[0021] Step 202: determining surface anomaly difference data between the first component surface state description information and the sample component surface state description information of the first component surface state description information, and determining surface anomaly description information corresponding to the surface anomaly difference data.

[0022] Exemplarily, the photovoltaic component surface description record can search for multiple groups of first reference surface abnormality description information corresponding to the specified photovoltaic component object, and load several first reference surface abnormality description information into the recording device. Further, the recording device can obtain multiple groups of first reference surface abnormality description information. When the recording device obtains each group of first reference surface abnormality description information, it can first match the surface abnormality description information example of the first reference surface abnormality description information (that is, the first group of first reference surface abnormality description information belonging to the previous round of the first reference surface abnormality description information), and then determine the difference information between the first reference surface abnormality description information and the surface abnormality description information example of the first reference surface abnormality description information. In order to better identify related data, the difference information in the training operation can be understood as the difference description content.

[0023] When using the AI ​​artificial model method to determine the difference description content, an AI artificial model can be trained in advance. The AI ​​artificial model is used to query the difference information of abnormal data between two sets of data that have a relationship. Therefore, the AI ​​artificial model can be used to determine the difference description content between the first reference surface abnormality description information and the surface abnormality description information example of the first reference surface abnormality description information.

[0024] Further, determining the distinguishing feature description content corresponding to the distinguishing description content may include the following contents, but is not limited to the following contents.

[0025] The recognition angle is matched to the recognition angle feature of the sample feature set, and the recognition method is matched to the recognition method feature of the sample feature set, and the distinguishing feature description content is generated according to the recognition angle feature and the recognition method feature.

[0026] Exemplarily, for the first key node in the first reference surface anomaly description information, the recognition angle corresponding to the key node is matched to the recognition angle feature of the sample feature set, and the recognition method corresponding to the key node is matched to the recognition method feature of the sample feature set, and the recognition angle feature and the recognition method feature constitute the key feature expression of the first key node of the distinguishing feature description content, and the key feature expression of the global key node of the distinguishing feature description content can be obtained, and the key feature expression of the global key node can constitute the distinguishing feature description content, that is, determine the distinguishing feature description content.

[0027] The distinguishing feature description content of the sample feature set is described. The distinguishing feature description content can be a surface abnormality description information that can be displayed. The attributes of the distinguishing feature description content represent the identification angles of the distinguishing feature description content. The descriptions of the differences in the distinguishing feature description content are used to characterize different identification methods in the distinguishing description content. Each key node in the distinguishing feature description content is described. The attributes of the key nodes represent the identification angles in the distinguishing description content. The descriptions of the key nodes are used to characterize the identification methods in the distinguishing description content.

[0028] Step 203: Load the first component surface state description information and the surface abnormality description information into a remote component surface recognition model that meets the model deployment conditions, so as to identify component surface description features from the first component surface state description information based on the remote component surface recognition model, identify surface change features from the surface abnormality description information, cluster the component surface description features and the surface change features, and obtain a clustering vector matrix; wherein the component surface description features represent the surface condition description information of the photovoltaic component in the first component surface state description information, and the surface change features represent the data after differences occur on the surface of the photovoltaic component in the surface abnormality description information.

[0029] Step 204, loading the clustering vector matrix and the component surface description features into the remote component surface recognition model to generate a surface information recognition result based on the remote component surface recognition model. Exemplarily, the predicted reference surface abnormality description information may include a surface detection step and a non-surface detection step, the key feature expression of each key node in the surface detection step is a first set value, and the key feature expression of each key node in the non-surface detection step is a second set value.

[0030] Exemplarily, an original remote component surface recognition model (i.e., a candidate remote component surface recognition model to be trained) can be built in advance. The original remote component surface recognition model can be an AI artificial model or another candidate remote component surface recognition model. The original remote component surface recognition model is not described in detail. The only input of the original remote component surface recognition model is the first reference surface anomaly description information and the distinguishing feature description content. The generation of the original remote component surface recognition model is to predict the reference surface anomaly description information.

[0031] Exemplarily, the original remote component surface recognition model may include a first fully connected layer, a second fully connected layer and a third fully connected layer, the first fully connected layer is used to process the first reference surface anomaly description information and the important attributes corresponding to the first reference surface anomaly description information, the second fully connected layer is used to process the distinguishing feature description content and the important attributes corresponding to the distinguishing feature description content, and the third fully connected layer is used to process the important attributes corresponding to the first reference surface anomaly description information and the important attributes corresponding to the distinguishing feature description content.

[0032] For explanation of step 203 and step 204, the first reference surface abnormal description information may be loaded into the first fully connected layer, an abnormal feature vector may be identified from the first reference surface abnormal description information based on the first fully connected layer, and the abnormal feature vector may be loaded into the third fully connected layer. Exemplarily, the first fully connected layer may include no less than one sub-model unit, such as a vector expansion unit, a feature extraction unit, a data update layer, etc., and the description vector generated by one sub-model unit may be used as an abnormal feature vector, or the description vectors generated by several sub-model units may all be used as abnormal feature vectors.

[0033] The distinguishing feature description content can be loaded into the second fully connected layer, and a component surface example can be identified from the distinguishing feature description content based on the second fully connected layer, and the component surface example can be loaded into the third fully connected layer. Exemplarily, the second fully connected layer can include no less than one sub-model unit, such as a vector expansion unit, a feature extraction unit, a data update layer, etc., and the description vector generated by one sub-model unit can be used as a component surface example, or the description vectors generated by several sub-model units can all be used as component surface examples, that is, the number of component surface examples can be one or several.

[0034] After the abnormal feature vector and the component surface example are loaded into the third fully connected layer, the abnormal feature vector and the component surface example are clustered based on the third fully connected layer to obtain a cluster reference vector. After obtaining the cluster reference vector, the third fully connected layer loads the cluster reference vector into the first fully connected layer. For example, the third fully connected layer includes no less than one sub-model unit, and each sub-model unit is described. The input of the sub-model unit is the abnormal feature vector and the component surface example, and the sub-model unit generates a cluster reference vector. The sub-model unit is used to cluster the abnormal feature vector and the component surface example to obtain a cluster reference vector.

[0035] According to the above description, the cluster reference vector and the abnormal feature vector have been loaded into the first fully connected layer, and the first fully connected layer can generate the predicted reference surface abnormal description information based on the cluster reference vector and the abnormal feature vector. Exemplarily, after the cluster reference vector and the abnormal feature vector are loaded into a sub-model unit of the first fully connected layer, the sub-model unit can process the cluster reference vector and the abnormal feature vector to obtain the predicted reference surface abnormal description information.

[0036] It can be understood that the original remote component surface recognition model includes the following contents: the first fully connected layer includes the sub-model unit 11 and the sub-model unit 12, the second fully connected layer includes the sub-model unit 21, and the third fully connected layer includes the sub-model unit 31. The first reference surface abnormal description information can be loaded into the sub-model unit 11, and the sub-model unit 11 processes the first reference surface abnormal description information to obtain a description vector 41, which can be understood as an abnormal feature vector. The sub-model unit 11 loads the description vector 41 into the sub-model unit 31, and loads the description vector 41 into the sub-model unit 12. The distinguishing feature description content can be loaded into the sub-model unit 21, and the sub-model unit 21 processes the distinguishing feature description content to obtain a description vector 42, which can be understood as a component surface example, and the sub-model unit 21 loads the description vector 42 into the sub-model unit 31. After the sub-model unit 31 obtains the description vector 41 and the description vector 42, the description vector 41 and the description vector 42 are clustered to obtain the description vector 43, which can be understood as a cluster reference vector, and the sub-model unit 31 loads the description vector 43 to the sub-model unit 12. After the sub-model unit 12 obtains the description vector 41 and the description vector 43, the description vector 44 can be determined based on the description vector 41 and the description vector 43, such as the description vector 44 is the sum of the description vector 41 and the description vector 43, and the predicted reference surface anomaly description information matching the description vector 44 is determined.

[0037] Further, the original remote component surface recognition model may include the following: the first fully connected layer includes sub-model unit 11, sub-model unit 12 and sub-model unit 13, the second fully connected layer includes sub-model unit 21 and sub-model unit 22, and the third fully connected layer includes sub-model unit 31 and sub-model unit 32. Sub-model unit 11 processes the abnormal description information of the first reference surface to obtain a description vector 51 (abnormal feature vector), and loads the description vector 51 to sub-model unit 31 and sub-model unit 12. Sub-model unit 21 processes the distinguishing feature description content to obtain a description vector 52 (component surface example), and loads the description vector 52 to sub-model unit 31 and sub-model unit 22. Sub-model unit 31 clusters description vector 51 and description vector 52 to obtain description vector 53 (clustering reference vector), and loads description vector 53 to sub-model unit 12. Sub-model unit 12 generates description vector 54 based on description vector 51 and description vector 53, processes description vector 54, obtains description vector 55 (abnormal feature vector), and loads description vector 55 to sub-model unit 32 and sub-model unit 13. Sub-model unit 22 processes description vector 52 to obtain description vector 56 (component surface example), and loads description vector 56 to sub-model unit 32. Sub-model unit 32 clusters description vector 55 and description vector 56 to obtain description vector 57 (clustering reference vector), and loads description vector 57 to sub-model unit 13. Sub-model unit 13 generates description vector 58 (such as the sum of description vector 55 and description vector 57) based on description vector 55 and description vector 57, and determines the predicted reference surface abnormal description information matching description vector 58.

[0038] Therefore, there is no one-to-one limitation on the original remote component surface recognition model, as long as the original remote component surface recognition model can identify the abnormal feature vector from the first reference surface abnormal description information, identify the component surface example from the distinguishing feature description content, cluster the abnormal feature vector and the component surface example, obtain the cluster reference vector, and generate the predicted reference surface abnormal description information based on the cluster reference vector and the abnormal feature vector.

[0039] The predicted reference surface anomaly description information is explained as a binary surface anomaly description information, which can be understood as a binary surface anomaly description information, that is, the key feature expression of the key node is only the first set value or the second set value. In the generation operation of the predicted reference surface anomaly description information, the predicted reference surface anomaly description information can be divided into a surface detection step and a non-surface detection step according to the operating status trend data and the distinction information. The key feature expression of each key node in the surface detection step is the first set value, and the key feature expression of each key node in the non-surface detection step is the second set value.

[0040] To summarize, after the first reference surface anomaly description information and the distinguishing feature description content are loaded into the original remote component surface recognition model, the original remote component surface recognition model generates predicted reference surface anomaly description information, and subsequent processing can be performed based on the predicted reference surface anomaly description information.

[0041] In the present application, clustering of abnormal feature vectors and component surface examples is performed based on the third fully connected layer to obtain a cluster reference vector, which may include the following contents, but is not limited to the following contents.

[0042] The component surface example is vector-expanded to obtain the component surface example after vector expansion, and the component surface example after vector expansion is matched using a data update layer to obtain the component surface example after vector data update processing.

[0043] Then, the abnormal feature vectors and component surface examples after the vector data update processing are clustered to obtain the cluster reference vector.

[0044] Step 205 : determining the photovoltaic component surface detection result corresponding to the designated photovoltaic component object according to the surface detection step in the surface information recognition result.

[0045] The prediction cost coefficient is determined according to the prediction reference surface anomaly description information and the prediction operation annotated surface anomaly description information carrying annotations. Exemplarily, the prediction operation annotated surface anomaly description information may be annotated surface anomaly description information of the first reference surface anomaly description information, and the prediction operation annotated surface anomaly description information may include a surface detection step and a non-surface detection step, and the key feature expressions of each key node in the surface detection step are all first set values, and the key feature expressions of each key node in the non-surface detection step are all second set values.

[0046] Exemplarily, before step 205, the labeled surface anomaly description information of the first reference surface anomaly description information can be trained, and the labeled surface anomaly description information is recorded as the predicted operation labeled surface anomaly description information. Exemplarily, when the first reference surface anomaly description information is labeled, the surface detection step and the non-surface detection step in the first reference surface anomaly description information can be obtained, so the surface detection step and the non-surface detection step in the first reference surface anomaly description information can be labeled, and the predicted operation labeled surface anomaly description information can be generated based on the labeled data. When generating the predicted operation labeled surface anomaly description information, the key feature expression of each key node corresponding to the surface detection step is the first set value, and the key feature expression of each key node corresponding to the non-surface detection step is the second set value. In this way, the predicted operation labeled surface anomaly description information can be generated, and the predicted operation labeled surface anomaly description information is a binary surface anomaly description information, which can be understood as a binary surface anomaly description information. After obtaining the predicted operation labeled surface anomaly description information, the set of the global first set value can be understood as the surface detection step, and the set of the global second set value can be understood as the non-surface detection step.

[0047] Furthermore, after obtaining the predicted reference surface anomaly description information and the predicted operation annotated surface anomaly description information, the predicted cost coefficient can be determined according to the difference between the predicted reference surface anomaly description information and the predicted operation annotated surface anomaly description information. Obviously, when the predicted cost coefficient is lower, it means that the predicted reference surface anomaly description information and the predicted operation annotated surface anomaly description information are more similar, that is, the predicted reference surface anomaly description information generated by the original remote component surface recognition model is more reliable, and when the predicted cost coefficient is larger, it means that the predicted reference surface anomaly description information and the predicted operation annotated surface anomaly description information are farther apart, that is, the predicted reference surface anomaly description information generated by the original remote component surface recognition model is less reliable.

[0048] Step 206: Update the model weight of the original remote component surface recognition model according to the predicted cost coefficient to obtain the remote component surface recognition model.

[0049] Exemplarily, the weight data of the original remote component surface recognition model can be updated according to the prediction cost coefficient to obtain an updated candidate remote component surface recognition model. Exemplarily, based on the weight data of the updated original remote component surface recognition model, an updated candidate remote component surface recognition model is obtained, and this update operation is not limited one by one. Then, it is determined whether the updated candidate remote component surface recognition model meets the model deployment conditions. If it is determined to be no, the updated candidate remote component surface recognition model is determined as the original remote component surface recognition model, and the step of loading the first reference surface anomaly description information and the distinguishing feature description content into the original remote component surface recognition model is returned to execute. If it is determined to be yes, the updated candidate remote component surface recognition model is determined as a remote component surface recognition model that meets the model deployment conditions.

[0050] After obtaining the remote component surface recognition model that meets the model deployment conditions, the photovoltaic component object remote control can be implemented based on the remote component surface recognition model that meets the model deployment conditions. For this detection operation, the following description content can be included.

[0051] Step 501: searching for first component surface state description information corresponding to a specified photovoltaic component object in a photovoltaic component surface description record.

[0052] Exemplarily, the PV component surface description record can search for the first surface abnormality description information corresponding to the specified PV component object, load the first surface abnormality description information into the recording device, and record the first surface abnormality description information in the verification operation as the first component surface status description information.

[0053] Step 502, determining surface anomaly difference data between the first component surface state description information and the sample component surface state description information of the first component surface state description information, and generating surface anomaly description information corresponding to the surface anomaly difference data.

[0054] Exemplarily, when the recording device obtains each group of first component surface state description information, it can first match the sample component surface state description information of the first component surface state description information, and then determine the difference information between the first component surface state description information and the sample component surface state description information of the first component surface state description information, and the difference information in the verification operation can be regarded as surface abnormality difference data.

[0055] The surface anomaly difference data may include an identification angle and an identification method. For each key node in the surface state description information of the first component, a sample node matching the key node is queried from the surface state description information of the sample component, and the identification angle and identification method corresponding to the key node are determined according to the correspondence between the key node and the sample node. Obviously, the surface anomaly difference data includes the identification angle and identification method corresponding to each key node in the surface state description information of the first component.

[0056] Further, determining the surface anomaly description information corresponding to the surface anomaly difference data may include the following contents, but is not limited to the following contents.

[0057] The recognition angle is matched to the recognition angle feature of the sample feature set, and the recognition method is matched to the recognition method feature of the sample feature set, and surface anomaly description information is generated based on the recognition angle feature and the recognition method feature.

[0058] The surface anomaly description information of the sample feature set is explained. The surface anomaly description information can be a visualized surface anomaly description information. The different attributes in the surface anomaly description information represent different recognition angles in the surface anomaly difference data. The different descriptions in the surface anomaly description information represent different recognition methods in the surface anomaly difference data.

[0059] Step 503, the first component surface state description information and the surface abnormality description information are loaded into a remote component surface recognition model that meets the model deployment conditions, so as to identify component surface description features from the first component surface state description information based on the remote component surface recognition model, identify surface change features from the surface abnormality description information, cluster the component surface description features and the surface change features, and obtain a clustering vector matrix. The component surface description features represent the surface condition description information of the photovoltaic component in the first component surface state description information, and the surface change features represent the data after the surface of the photovoltaic component in the surface abnormality description information has differences.

[0060] Step 504, loading the clustering vector matrix and the component surface description features into the remote component surface recognition model to generate a surface information recognition result based on the remote component surface recognition model. Exemplarily, the surface information recognition result may include a surface detection step and a non-surface detection step, the key feature expressions of each key node in the surface detection step are all first set values, and the key feature expressions of each key node in the non-surface detection step are all second set values.

[0061] Exemplarily, the model structure of the remote component surface recognition model is similar to the model structure of the original remote component surface recognition model. Therefore, the remote component surface recognition model may include a first fully connected layer, a second fully connected layer and a third fully connected layer. The first fully connected layer is used to process the first component surface state description information and the important attributes corresponding to the first component surface state description information, the second fully connected layer is used to process the surface abnormality description information and the important attributes corresponding to the surface abnormality description information, and the third fully connected layer is used to process the important attributes corresponding to the first component surface state description information and the important attributes corresponding to the surface abnormality description information.

[0062] Furthermore, the first component surface state description information is loaded into the first fully connected layer, component surface description features are identified from the first component surface state description information based on the first fully connected layer, and the component surface description features are loaded into the third fully connected layer.

[0063] The surface anomaly description information may be loaded into the second fully connected layer, and the surface change feature may be identified from the surface anomaly description information based on the second fully connected layer, and the surface change feature may be loaded into the third fully connected layer. After the component surface description feature and the surface change feature are loaded into the third fully connected layer, the component surface description feature and the surface change feature may be clustered based on the third fully connected layer to obtain a clustering vector matrix. After obtaining the clustering vector matrix, the third fully connected layer may load the clustering vector matrix into the first fully connected layer.

[0064] According to the above description, the clustering vector matrix and the component surface description features have been loaded into the first fully connected layer, and the first fully connected layer generates a surface information recognition result based on the clustering vector matrix and the component surface description features.

[0065] In this embodiment, the component surface description feature can represent the surface condition description information of the photovoltaic component and the surface condition description information of the non-photovoltaic component in the first component surface state description information, that is, the component surface description feature can characterize the difference between the operating process and the non-operating process. The surface change feature can represent the data after the surface of the photovoltaic component in the surface abnormality description information differs and the data after the surface of the non-photovoltaic component differs, that is, the surface change feature can represent the difference between the operating process and the non-operating process. The clustering vector matrix can represent the clustering data of the operating state trend data and the distinguishing information, and the clustering data can represent the difference between the operating process and the non-operating process, that is, the difference between the operating process and the non-operating process is reflected based on the operating state trend data and the distinguishing information.

[0066] In summary, after the first component surface state description information and the surface abnormality description information are loaded into the remote component surface recognition model, the remote component surface recognition model generates a surface information recognition result, and subsequent processing can be performed based on the surface information recognition result.

[0067] Exemplarily, clustering the component surface description features and surface change features based on the third fully connected layer to obtain a clustering vector matrix may include the following content, but is not limited to the following content.

[0068] Performing vector expansion on the surface change feature to obtain the surface change feature after vector expansion, and using a data update layer to match the surface change feature after vector expansion to obtain the surface change feature after vector data update processing; Step 505 , determining the photovoltaic component surface detection result corresponding to the specified photovoltaic component object according to the surface detection step in the surface information recognition result, that is, determining the surface detection step as the photovoltaic component surface detection result corresponding to the specified photovoltaic component object.

[0069] Exemplarily, the surface information recognition result includes a surface detection step and a non-surface detection step, the key feature expression of each key node in the surface detection step is the first set value, and the key feature expression of each key node in the non-surface detection step is the second set value. Therefore, the set consisting of the key nodes of the global first set value is determined as the photovoltaic component surface detection result corresponding to the specified photovoltaic component object (also called the running process set), that is, the photovoltaic component surface detection result is a set consisting of the key nodes of the global first set value.

[0070] In the present application, a remote recognition operation matching the photovoltaic component surface detection result can also be determined, and the recognition step training data corresponding to the specified photovoltaic component object can be determined based on the remote recognition operation. Then, the training item coefficient of the recognition step training data covering the preset training data can be counted. Exemplarily, within a preset time period, a specific time period is used as a segment, and steps 501-505 are performed in each segment to obtain the photovoltaic component surface detection result of the current segment. If it is determined that the recognition step training data corresponding to the photovoltaic component surface detection result of the current segment covers the preset training data, the training item coefficient can be updated. If it is determined that the recognition step training data corresponding to the photovoltaic component surface detection result of the current cycle does not cover the preset training data, the training item coefficient can be kept unchanged. Then, the photovoltaic component surface detection result (i.e., the photovoltaic component surface detection result of the current segment), the recognition step training data (i.e., the recognition step training data of the current segment) and the training item coefficient (i.e., the training item coefficient within the set time sequence) can be run.

[0071] In the embodiment of the present application, the surface state description information and the surface abnormality description information of the first component can be loaded into the remote component surface recognition model, and the surface information recognition result can be generated based on the remote component surface recognition model. The photovoltaic component surface detection result corresponding to the specified photovoltaic component object is determined according to the surface detection step in the surface information recognition result. It can be understood that the photovoltaic component surface detection result corresponding to the specified photovoltaic component object is determined according to the first component surface state description information and the surface abnormality description information, so as to accurately and reliably determine the photovoltaic component surface detection result, and assist in determining the operation process in real time and reliably, prompting the remote control system to quickly query the operation process and process the operation process. Based on loading the surface abnormality description information into the remote component surface recognition model, the difference information (i.e., the data after the surface of the photovoltaic component has a difference) can be loaded into the remote component surface recognition model, so as to use the difference information to control the remote component surface recognition model to make a more reliable judgment, thereby improving the credibility and accuracy of the recognition of the photovoltaic component surface detection result.

[0072] Based on the above, please refer to Figure 2 , a photovoltaic module surface detection device 200 is provided, which is applied to a photovoltaic module object remote control system, and the device comprises: The data search module 210 is used to search for first component surface state description information corresponding to a specified photovoltaic component object in the photovoltaic component surface description record; A data determination module 220 is used to determine surface anomaly difference data between the first component surface state description information and the sample component surface state description information of the first component surface state description information, and determine the surface anomaly description information corresponding to the surface anomaly difference data; A matrix acquisition module 230 is used to load the first component surface state description information and the surface abnormality description information into a remote component surface recognition model that meets the model deployment conditions, so as to identify component surface description features from the first component surface state description information based on the remote component surface recognition model, identify surface change features from the surface abnormality description information, cluster the component surface description features and the surface change features, and obtain a clustering vector matrix; wherein the component surface description features represent the surface condition description information of the photovoltaic component in the first component surface state description information, and the surface change features represent the data after the surface of the photovoltaic component in the surface abnormality description information has a difference; A result recognition module 240, used to load the clustering vector matrix and the component surface description features into the remote component surface recognition model to generate a surface information recognition result based on the remote component surface recognition model; The instruction execution module 250 is used to determine the photovoltaic component surface detection result corresponding to the specified photovoltaic component object according to the surface detection step in the surface information recognition result.

[0073] Based on the above, please refer to Figure 3 , shows a photovoltaic module surface detection system 300, including a processor 310 and a memory 320 that communicate with each other, and the processor 310 is used to read and execute a computer program from the memory 320 to implement the above method.

[0074] Based on the above, a computer-readable storage medium is also provided, on which a computer program stored implements the above method when running.

[0075] In summary, based on the above scheme, the surface state description information and the surface abnormality description information of the first component are loaded into the remote component surface recognition model, and the surface information recognition result is generated based on the remote component surface recognition model. The photovoltaic component surface detection result corresponding to the specified photovoltaic component object is determined according to the surface detection step in the surface information recognition result. It can be understood that the photovoltaic component surface detection result corresponding to the specified photovoltaic component object is determined based on the first component surface state description information and the surface abnormality description information, and then the photovoltaic component surface detection result is accurately and reliably determined, so that the surface damage can be determined in time, the damage of the photovoltaic component equipment can be avoided, and the photovoltaic component can be protected in time.

[0076] It should be understood that the system and its modules shown above can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or a dedicated design hardware. Those skilled in the art will understand that the above methods and systems can be implemented using computer executable instructions and / or included in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The system and its modules of the present application can not only be implemented by hardware circuits such as ultra-large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., but can also be implemented by software such as executed by various types of processors, and can also be implemented by a combination of the above hardware circuits and software (e.g., firmware).

[0077] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects that may be produced may be any one or a combination of the above, or any other beneficial effects that may be obtained.

[0078] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of the present application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements and amendments to the present application. Such modifications, improvements and amendments are suggested in the present application, so such modifications, improvements and amendments still belong to the spirit and scope of the exemplary embodiments of the present application.

[0079] At the same time, the present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to no less than one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0080] In addition, it will be appreciated by those skilled in the art that various aspects of the present application may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present application may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.

[0081] A computer storage medium may include a propagated data signal containing computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination. A computer storage medium may be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, device or apparatus to communicate, propagate or transmit the program for use. The program code on the computer storage medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.

[0082] The computer program codes required for the operation of each part of the present application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages ​​such as C language, Visual Basic, Fortran 2003, Perl, COBOL 2002, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy, or other programming languages, etc. The program code can be run entirely on the user's computer, or run on the user's computer as an independent software package, or run partially on the user's computer and partially on a remote computer, or run entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer in any model form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0083] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some invention embodiments that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0084] Similarly, it should be noted that in order to simplify the description of the disclosure of this application and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this application, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the object of this application are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0085] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers allow adaptive changes. Accordingly, in some embodiments, the numerical coefficients used in the specification and claims are approximate values, which can be changed according to the required features of individual embodiments. In some embodiments, the numerical coefficients should take into account the specified significant digits and adopt the general method of retaining the number of digits. Although the numerical domains and coefficients used to confirm the breadth of the scope in some embodiments of the present application are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

[0086] Each patent, patent application, patent application disclosure, and other materials, such as articles, books, instructions, publications, documents, etc., cited in this application are hereby incorporated by reference in their entirety. Except for application history documents that are inconsistent with or conflicting with the content of this application, documents that limit the broadest scope of the claims of this application (currently or later attached to this application) are also excluded. It should be noted that if the descriptions, definitions, and / or use of terms in the attached materials of this application are inconsistent or conflicting with the content described in this application, the descriptions, definitions, and / or use of terms in this application shall prevail.

[0087] Finally, it should be understood that the embodiments described in this application are only used to illustrate the principles of the embodiments of this application. Other variations may also fall within the scope of this application. Therefore, as an example and not a limitation, the alternative training of the embodiments of this application can be considered to be consistent with the teachings of this application. Accordingly, the embodiments of this application are not limited to the embodiments explicitly introduced and described in this application.

[0088] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A photovoltaic module surface detection method, characterized in that: The method comprises: Searching for first component surface state description information corresponding to a specified photovoltaic component object in a photovoltaic component surface description record; Determine surface anomaly difference data between the first component surface state description information and the surface state description information of a sample component of the first component surface state description information, and determine surface anomaly description information corresponding to the surface anomaly difference data; The first component surface state description information and the surface abnormality description information are loaded into a remote component surface recognition model that meets the model deployment conditions, so as to identify component surface description features from the first component surface state description information based on the remote component surface recognition model, identify surface change features from the surface abnormality description information, cluster the component surface description features and the surface change features, and obtain a clustering vector matrix; wherein the component surface description features represent the surface condition description information of the photovoltaic component in the first component surface state description information, and the surface change features represent the data after the surface of the photovoltaic component in the surface abnormality description information has a difference; Loading the clustering vector matrix and the component surface description features into the remote component surface recognition model to generate a surface information recognition result based on the remote component surface recognition model; The photovoltaic component surface detection result corresponding to the designated photovoltaic component object is determined according to the surface detection step in the surface information recognition result.

2. The method according to claim 1, characterized in that The generating of the surface anomaly description information corresponding to the surface anomaly difference data includes: If it is determined that the surface abnormality difference data includes a recognition angle and a recognition method, the recognition angle is matched to the recognition angle feature of the sample feature set, and the recognition method is matched to the recognition method feature of the sample feature set; The surface anomaly description information is generated according to the recognition angle feature and the recognition method feature.

3. The method according to claim 1, characterized in that The remote component surface recognition model includes a first fully connected layer, a second fully connected layer and a third fully connected layer; The method of identifying component surface description features from the first component surface state description information based on the remote component surface recognition model, identifying surface change features from the surface abnormality description information, clustering the component surface description features and the surface change features to obtain a clustering vector matrix includes: The first fully connected layer based on the remote component surface recognition model identifies component surface description features from the first component surface state description information, and loads the component surface description features to the third fully connected layer; The second fully connected layer based on the remote component surface recognition model identifies surface change features from the surface anomaly description information, and loads the surface change features to the third fully connected layer; The component surface description features and the surface change features are clustered based on the third fully connected layer to obtain a clustering vector matrix, and the clustering vector matrix is ​​loaded into the first fully connected layer.

4. The method according to claim 3, characterized in that Clustering the component surface description features and the surface change features based on the third fully connected layer to obtain a clustering vector matrix includes: Performing vector expansion on the surface change feature to obtain the surface change feature after vector expansion, and performing vector data update processing on the surface change feature after vector expansion using a data update layer to obtain the surface change feature after vector data update processing; Clustering is performed on the component surface description features and the surface change features after the vector data update processing to obtain a clustering vector matrix.

5. The method according to any one of claims 1 to 4, characterized in that: After determining the photovoltaic component surface detection result corresponding to the specified photovoltaic component object according to the surface detection step in the surface information recognition result, the method further includes: Determining a remote identification operation that matches the photovoltaic module surface detection result; Determining identification step training data corresponding to the specified photovoltaic component object according to the remote identification operation; The training data of the statistical identification step covers the training item coefficients of the preset training data; The photovoltaic component surface detection result is run based on the identification step training data and the training item coefficients.

6. The method according to any one of claims 1 to 4, characterized in that: The remote component surface recognition model is obtained based on the following steps: Searching for first reference surface abnormality description information corresponding to a specified photovoltaic component object in the photovoltaic component surface description record; Determine the difference description content between the first reference surface abnormality description information and the surface abnormality description information example of the first reference surface abnormality description information, and generate the difference feature description content corresponding to the difference description content; The first reference surface abnormality description information and the distinguishing feature description content are loaded into the original remote component surface recognition model, so as to identify an abnormal feature vector from the first reference surface abnormality description information based on the original remote component surface recognition model, identify a component surface example from the distinguishing feature description content, cluster the abnormal feature vector and the component surface example, and obtain a cluster reference vector; wherein the abnormal feature vector represents the surface condition description information of the photovoltaic component in the first reference surface abnormality description information, and the component surface example represents the data after the surface of the photovoltaic component in the distinguishing feature description content has a difference; Loading the cluster reference vector and the abnormal feature vector into the original remote component surface recognition model to generate predicted reference surface abnormality description information based on the original remote component surface recognition model; Determine a prediction cost coefficient according to the prediction reference surface anomaly description information and the prediction operation annotated surface anomaly description information carrying annotations; The model weight of the original remote component surface recognition model is updated according to the predicted cost coefficient to obtain a remote component surface recognition model.

7. The method according to claim 6, characterized in that The updating of the model weight of the original remote component surface recognition model according to the predicted cost coefficient to obtain the remote component surface recognition model includes: updating the weight data of the original remote component surface recognition model according to the predicted cost coefficient to obtain an updated candidate remote component surface recognition model, and determining whether the updated candidate remote component surface recognition model meets the model deployment condition; If the determination is no, the updated candidate remote component surface recognition model is determined as the original remote component surface recognition model, and the operation of loading the first reference surface anomaly description information and the distinguishing feature description content into the original remote component surface recognition model is returned to execute; If the determination is yes, the updated candidate remote component surface recognition model is determined as the remote component surface recognition model.

8. A photovoltaic module surface detection system, characterized in that: The invention comprises a processor and a memory communicating with each other, wherein the processor is used to read a computer program from the memory and execute the computer program to implement the method according to any one of claims 1 to 7.

Images