Artificial intelligence-based photovoltaic power station abnormal information generation method and device

Abstract

Embodiments of the present disclosure disclose an artificial intelligence-based photovoltaic power station abnormal information generation method and device. A specific embodiment of the method includes: obtaining a historical power generation data set; generating a first sub-feature group and a second sub-feature group according to a preset correlation threshold and the correlation between sub-features and power generation; obtaining a pre-constructed model pool and determining the input features corresponding to each model; training the model pool using the historical power generation data set to obtain a pre-trained model pool; sorting each pre-trained model to obtain a pre-trained model sequence; training the pre-trained model sequence to obtain a power generation prediction model; inputting the features of the time point to be predicted into the power generation prediction model to obtain a predicted power generation; and when the difference between the actual power generation and the predicted power generation is greater than a preset difference threshold, generating abnormal information and pushing the abnormal information to a maintenance terminal, thereby improving the accuracy of generating abnormal information and reducing false positives.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, specifically to a method and apparatus for generating abnormal information of photovoltaic power plants based on artificial intelligence. Background Technology

[0002] Photovoltaic power plants convert solar energy into electrical energy, representing a green energy development project. Due to their large footprint and wide distribution, photovoltaic arrays are prone to various anomalies. Failure to detect these anomalies promptly can disrupt the operation of the power plant. Currently, anomaly information is typically generated by comparing predicted power generation with actual power generation, and maintenance is then performed based on this information.

[0003] However, the inventors discovered that when generating abnormal information using the above method, the following technical problems often occur:

[0004] First, it frequently generates incorrect abnormal information, leading to frequent false alarms from the maintenance terminal, which greatly increases the workload of maintenance and wastes human and material resources.

[0005] Secondly, in reality, due to the wide distribution of photovoltaic arrays, there is a lack of methods to re-inspect abnormalities before manual visual inspection and maintenance, resulting in a significant waste of manpower and resources due to false alarms.

[0006] Third, the abnormal information contains little information, and maintenance personnel need to manually find the cause of the abnormality, which is time-consuming and labor-intensive.

[0007] Fourth, relying solely on abnormal information during extreme weather events can easily damage photovoltaic modules. Summary of the Invention

[0008] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0009] Some embodiments of this disclosure propose a method and apparatus for generating abnormal information of photovoltaic power plants based on artificial intelligence, in order to solve one or more of the technical problems mentioned in the background section above.

[0010] In a first aspect, some embodiments of this disclosure provide a method for generating abnormal information of a photovoltaic power station based on artificial intelligence, comprising: acquiring a preprocessed historical power generation dataset of a target photovoltaic power station, wherein the historical power generation data in the historical power generation dataset includes environmental features, photovoltaic panel features, and power generation, and the environmental features and photovoltaic panel features each include multiple sub-features and each includes a sub-feature set composed of multiple sub-features; grouping multiple sub-features according to a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, to obtain a first sub-feature group and a second sub-feature group, wherein the correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold; acquiring a pre-constructed model pool, and for each model in the model pool, performing the following feature selection steps: determining a target number of sub-features according to the second sub-feature group to obtain a target sub-feature group; The union of each sub-feature in the first sub-feature group and the target sub-feature group is determined as the input feature corresponding to the model. Based on the input feature corresponding to each model, each model in the model pool is trained using the first subset of historical power generation data from the historical power generation dataset to obtain the trained model pool, which is then designated as the pre-trained model pool. For each pre-trained model in the pre-trained model pool, they are sorted according to the number of input features and the model type to obtain the pre-trained model sequence. For each pre-trained model in the pre-trained model sequence, they are trained using the second subset of historical power generation data from the historical power generation dataset to obtain the power generation prediction model. The environmental features and photovoltaic panel features at the time point to be predicted are input into the power generation prediction model to obtain the predicted power generation at the time point to be predicted. When the difference between the actual power generation and the predicted power generation at the time point to be predicted exceeds a preset difference threshold, an anomaly message is generated and pushed to the maintenance terminal.

[0011] Secondly, some embodiments of this disclosure provide an artificial intelligence-based photovoltaic power plant anomaly information generation device, including: an acquisition unit configured to acquire a preprocessed historical power generation dataset of a target photovoltaic power plant, wherein the historical power generation data in the historical power generation dataset includes environmental features, photovoltaic panel features, and power generation, and the environmental features and photovoltaic panel features include multiple sub-features; a grouping unit configured to group the multiple sub-features according to a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, to obtain a first sub-feature group and a second sub-feature group, wherein the correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold; and a model pool acquisition unit configured to acquire a pre-built model pool, and for each model in the model pool, perform the following feature selection steps: determine a target number of sub-features according to the second sub-feature group to obtain a target sub-feature group; and group each sub-feature in the first sub-feature group and the target sub-feature group... The union of features is used to determine the input features corresponding to the model; the pre-training unit is configured to train each model in the model pool using the first subset of historical power generation data from the historical power generation dataset based on the input features corresponding to each model, to obtain the trained model pool, and to determine the trained model pool as the pre-trained model pool; the sorting unit is configured to sort each pre-trained model in the pre-trained model pool according to the number of input features and the model type, to obtain a pre-trained model sequence; the training unit is configured to train each pre-trained model in the pre-trained model sequence using the second subset of historical power generation data from the historical power generation dataset, to obtain a power generation prediction model; the prediction unit is configured to input the environmental features and photovoltaic panel features of the time point to be predicted into the power generation prediction model to obtain the predicted power generation at the time point to be predicted; the generation unit is configured to generate abnormal information and push the abnormal information to the maintenance terminal when the difference between the actual power generation and the predicted power generation at the time point to be predicted is greater than a preset difference threshold.

[0012] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0013] The embodiments disclosed above have the following beneficial effects: improved accuracy in generating abnormal information and reduced false alarms. Specifically, the inventors have found that the reason for frequently generating erroneous abnormal information is the low accuracy in predicting power generation. Based on this, some embodiments of this disclosure, by fusing models in a model pool, group sub-features according to the correlation of power generation. Furthermore, sub-features with high correlation are shared among different models, while sub-features with low correlation are differentiated across different models. Since sub-features with high correlation are shared among different models, the prediction accuracy of each model is maintained at a high level. The differentiation of sub-features with low correlation across different models helps to explore the correlation between different sub-features and power generation, while increasing the overall generalization ability of the model, resulting in better fusion effects and thus improving the accuracy of power generation prediction. In addition, training is performed based on the ranking of each pre-trained model. The output of the pre-trained model ranked first is added to the input of the pre-trained model ranked later. That is, super features are introduced to explore the correlation between super features and the actual power generation, thereby improving the accuracy of the prediction model ranked later in predicting power generation, which in turn improves the accuracy of generating abnormal information and reduces false alarms. Attached Figure Description

[0014] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0015] Figure 1 This is a flowchart of some embodiments of the method for generating abnormal information of photovoltaic power plants based on artificial intelligence according to the present disclosure;

[0016] Figure 2 This is a schematic diagram of the structure of some embodiments of the artificial intelligence-based photovoltaic power plant anomaly information generation device according to the present disclosure;

[0017] Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0018] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0019] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0023] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] refer to Figure 1 The diagram illustrates a flow 100 of some embodiments of the AI-based photovoltaic power plant anomaly information generation method according to this disclosure. The AI-based photovoltaic power plant anomaly information generation method includes the following steps:

[0025] Step 101: Obtain the preprocessed historical power generation dataset of the target photovoltaic power station. The historical power generation dataset includes environmental features, photovoltaic panel features, and power generation. The environmental features and photovoltaic panel features each include multiple sub-features, and each includes a set of sub-features.

[0026] In some embodiments, the executing entity of this AI-based photovoltaic power plant anomaly information generation method can be any electronic device, such as the server of the target photovoltaic power plant. The server can be equipped with various software to implement various functions. Based on this, the executing entity can first obtain a pre-processed historical power generation dataset of the target photovoltaic power plant. For example, it can be obtained from a database. The historical power generation dataset includes environmental features, photovoltaic panel features, and power generation. The environmental features and photovoltaic panel features each include multiple sub-features. For example, the environmental features may include at least one of the following: location information (latitude and longitude), altitude information, solar radiation, wind direction and speed, etc. The photovoltaic panel features may include at least one of the following: installation parameters (tilt angle, azimuth angle), panel surface temperature, etc. The target photovoltaic power plant can be any photovoltaic power plant, such as the photovoltaic power plant currently being predicted. The environmental features and photovoltaic panel features each include a sub-feature set composed of multiple sub-features; for example, if the environmental features include 10 sub-features and the photovoltaic panel features include 13 sub-features, then the sub-feature set includes 13 sub-features.

[0027] Step 102: Based on the preset correlation threshold and the correlation between each sub-feature in the sub-feature set and the power generation, the multiple sub-features are grouped to obtain the first sub-feature group and the second sub-feature group. The correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the correlation threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the correlation threshold.

[0028] In some embodiments, the above execution can calculate the correlation between each sub-feature and the power generation. Specifically, the Pearson correlation coefficient or the Spearman correlation coefficient can be calculated and used as the correlation. Based on this, sub-features with a correlation threshold greater than or equal to a correlation threshold (e.g., 0.5) can be assigned to a first sub-feature group, and sub-features with a correlation threshold less than the threshold can be assigned to a second sub-feature group.

[0029] Step 103: Obtain the pre-built model pool, and for each model in the model pool, perform the following feature selection steps: Based on the second sub-feature group, determine the target number of sub-features to obtain the target sub-feature group; determine the union of each sub-feature in the first sub-feature group and the target sub-feature group as the input feature corresponding to the model.

[0030] In some embodiments, the above execution can obtain a pre-built model pool, which includes multiple types of models, such as tree-structured models, such as LightGBM, XGBoost, etc. Alternatively, it can be artificial neural networks, such as LSTM (Long Short Term Memory) networks, RNNs (Recurrent Neural Networks), etc.

[0031] In some embodiments, for each model in the model pool, the input features of that model can be determined first. Sub-features with high relevance can be shared among different models. All sub-features in the first sub-feature group are included in the input features of each model. Since these sub-features have a high correlation with power generation, this ensures that the prediction accuracy of each model remains at a high level. Additionally, for each model in the model pool, a target number of sub-features are determined based on the second sub-feature group, resulting in a target sub-feature group. For example, a target number of sub-features can be randomly selected from the second sub-feature group without replacement to obtain the target sub-feature group. This allows for differentiation between different models for sub-features with low relevance. Each sub-feature in the second sub-feature group is only included in the input features of one model. This facilitates the discovery of the correlation between different sub-features and power generation, while increasing the overall generalization ability of the model, achieving better fusion results, and thus improving the accuracy of power generation prediction. Then, the union of the sub-features in the first sub-feature group and the target sub-feature group is determined as the input feature of the corresponding model. During training or prediction, the feature values ​​can be concatenated to obtain the input feature values ​​of each model. The number of objectives can be positively correlated with the model depth. Alternatively, the number of objectives can be a pre-specified value.

[0032] Step 104: Based on the input features corresponding to each model, train each model in the model pool using the first subset of historical power generation data from the historical power generation dataset to obtain the trained model pool, and determine the trained model pool as the pre-trained model pool.

[0033] In some embodiments, based on the input features of each model determined in step 103, specific input feature values ​​can be determined using a first subset of historical power generation data. These input feature values ​​are then input into the corresponding models to obtain the predicted values ​​for each model. For each model, the parameters are adjusted based on the difference between its predicted value and the power generation in the corresponding historical power generation data. This trains each model, resulting in a trained model pool. Depending on the needs, each model can be trained independently or jointly. The first subset of historical power generation data is a subset of the historical power generation dataset. Based on this, machine learning algorithms, such as backpropagation and stochastic gradient descent, can be used to train each model until all data in the first subset of historical power generation data is used, thus obtaining the trained model pool, which is then designated as the pre-trained model pool.

[0034] Step 105: Sort each pre-trained model in the pre-trained model pool according to the number of input features and the model type to obtain the pre-trained model sequence.

[0035] In some embodiments, a pre-trained model sequence can be generated through the following steps:

[0036] Step 1: Group each pre-trained model according to its model type to obtain multiple pre-trained model groups;

[0037] Step 2: For each pre-trained model in each of the multiple pre-trained model groups, sort them in ascending order of the number of input features to obtain the pre-trained model subsequence corresponding to each pre-trained model group.

[0038] Step 3: Based on the pre-trained model sub-sequences corresponding to the multiple pre-trained model groups, a pre-trained model sequence is obtained. These sub-sequences can be sorted according to pre-defined rules. For example, a pre-defined sorting rule could be: tree structure models are placed before artificial neural networks. Within each group, models are sorted according to the number of input features, with prediction accuracy gradually increasing as the number of features increases. As another example, a pre-defined sorting rule could be: sorting the pre-trained model sub-sequences in ascending order of average model depth. The average model depth can be the average depth of each pre-trained model within the sub-sequence. For example, the depth of a pre-trained model could be the maximum number of layers in a tree structure model (counting leaf nodes from the root node) or the number of layers in an artificial neural network.

[0039] Step 106: For each pre-trained model in the pre-trained model sequence, train it using the second historical power generation data subset of the historical power generation dataset to obtain the power generation prediction model.

[0040] In some embodiments, the pre-trained model sequence includes a first pre-trained model, a second pre-trained model, and a third pre-trained model. Each of these models comprises at least two different model types, enabling cascading and fusion of multiple models to complement each other and improve the overall accuracy of power generation prediction. Specifically, the first prediction model may be LightGBM, the second pre-trained model may be XGBoost, and the third pre-trained model may be LSTM.

[0041] Based on this, for each pre-trained model in the pre-trained model sequence, a second subset of historical power generation data from the historical power generation dataset is used for training to obtain a power generation prediction model, including the following steps:

[0042] Step one involves selecting target historical power generation data from the second historical power generation data subset, and determining the first input feature value of the first pre-trained model, the second input feature value of the second pre-trained model, and the third input feature value of the third pre-trained model based on the target historical power generation data and the input features of each pre-trained model. The target historical power generation data can be any historical power generation data. It can be understood that the first, second, and third input feature values ​​are all data, with each dimension corresponding to a sub-feature. The historical power generation dataset can be divided into a second and first historical power generation data subsets according to a certain proportion; the first and second historical data subsets have no overlap. The input features of each pre-trained model are the same as the input features of the model corresponding to that pre-trained model.

[0043] Step 2: Input the first input feature value into the first pre-trained model to obtain the first predicted power generation value;

[0044] Step 3: Replace the feature value of the dimension with the lowest correlation to power generation among all dimensions of the second input feature value with the predicted power generation value of the first dimension, thus obtaining the updated second input feature value. First, determine the correlation between the sub-features corresponding to each dimension of the second input feature value and power generation; the dimension corresponding to the sub-feature with the lowest correlation is the dimension with the lowest correlation to power generation.

[0045] Step 4: Input the updated second input feature value into the second pre-trained model to obtain the second power generation prediction value;

[0046] Step 5: Replace the feature value of the dimension with the lowest correlation to power generation among all dimensions of the third input feature value with the second power generation prediction value to obtain the updated third input feature value; wherein, firstly, determine the correlation between the sub-features corresponding to each dimension of the third input feature value and power generation, wherein the dimension corresponding to the sub-feature with the lowest correlation is the dimension with the lowest correlation to power generation.

[0047] Step 6: Input the updated third input feature value into the third pre-trained model to obtain the third predicted power generation value;

[0048] Step 7: The weighted sum of the first power generation forecast, the second power generation forecast, and the third power generation forecast is determined as the predicted power generation; where the weights can be pre-specified.

[0049] Step 8: Determine the difference between the predicted power generation and the power generation in the target historical power generation data, and adjust the parameters of the first pre-trained model, the second pre-trained model and the third pre-trained model according to the difference until the training stop condition is met, so as to obtain the power generation prediction model.

[0050] The difference between the predicted power generation and the target historical power generation data can be determined using a preset loss function, such as MSE (mean squared error). This difference is then propagated layer by layer from the last layer forward, updating the parameters of each layer. This process is iterated multiple times until a training stopping condition is met, resulting in the power generation prediction model. The training stopping condition can be: the number of iterations reaches a preset number.

[0051] Step 107: Input the environmental characteristics and photovoltaic panel characteristics of the time point to be predicted into the power generation prediction model to obtain the predicted power generation of the time point to be predicted.

[0052] In some embodiments, the aforementioned executing entity may use the trained power generation prediction model to predict the predicted power generation at the time point to be predicted.

[0053] Step 108: When the difference between the actual power generation and the predicted power generation at the time point to be predicted is greater than the preset difference threshold, anomaly information is generated and pushed to the maintenance terminal.

[0054] The method provided in some embodiments of this disclosure, based on the fusion of models in a model pool, groups sub-features according to the magnitude of their relevance to power generation. Sub-features with high relevance are shared among different models, while sub-features with low relevance are differentiated across models. Sharing highly relevant sub-features among different models ensures that the prediction accuracy of each model remains at a high level. Differentiating less relevant sub-features among different models helps to uncover the correlation between different sub-features and power generation, while also increasing the overall generalization ability of the model, resulting in better fusion effects and improved accuracy in predicting power generation. Furthermore, training is performed based on a ranking of the pre-trained models. The output of the top-ranked pre-trained model is added to the input of the next-ranked pre-trained model, i.e., super-features are introduced to uncover the correlation between super-features and actual power generation, thereby improving the accuracy of the next-ranked prediction model in predicting power generation. This, in turn, improves the accuracy of generating abnormal information and reduces false alarms.

[0055] While the above embodiments can produce the described technical effects, in reality, due to the widespread distribution of photovoltaic arrays, the lack of re-inspection for any abnormalities before manual visual inspection and maintenance leads to a significant waste of manpower and resources due to false alarms. To address this technical problem, some embodiments of this disclosure provide methods that further include the following steps:

[0056] Maintenance personnel can operate the maintenance terminal, which then sends a re-inspection request to the aforementioned entity. For example, maintenance personnel can click the "Request Re-inspection" button.

[0057] Based on this, in response to a re-inspection request for abnormal information sent by the maintenance terminal, the aforementioned execution entity can obtain the power generation data of multiple associated photovoltaic power stations of the target photovoltaic power station. The distance between each of the multiple associated photovoltaic power stations and the target photovoltaic power station meets a distance condition; the distance condition can be that the distance is less than a distance threshold. Then, based on the power generation data of the multiple associated photovoltaic power stations, the expected power generation data of the target photovoltaic power station is determined; for example, the power generation data of the multiple associated photovoltaic power stations can be converted into the expected power generation data of the target photovoltaic power station using a conversion ratio. The conversion ratio can be specified. A re-inspection result is generated based on the expected power generation data and the predicted power generation. If the difference between the expected power generation data and the predicted power generation is less than a first difference threshold, a re-inspection result indicating that the predicted power generation has been verified is generated, and the re-inspection result is pushed to the maintenance terminal.

[0058] While the above embodiments can produce the described technical effects, the abnormal information contains limited details, requiring maintenance personnel to manually investigate the cause of the abnormality, which is time-consuming and labor-intensive. Therefore, some embodiments of this disclosure further include the following steps:

[0059] Step 1: In response to receiving a re-inspection request for abnormal information sent to the maintenance terminal, determine the target drone that meets the re-inspection shooting conditions based on the shooting parameter list of each drone in the drone station. The re-inspection shooting conditions are that the shooting resolution is greater than or equal to the first resolution threshold.

[0060] Step 2: Control the target drone to capture re-inspection images of the target photovoltaic power station. The re-inspection images include infrared images and visible light images.

[0061] The drone can be equipped with infrared and visible light imaging devices, enabling it to capture infrared and visible light images in real time and transmit them to the aforementioned execution entity. Thus, the execution entity can receive infrared and visible light images captured of the target photovoltaic power station. The infrared and visible light images can be captured simultaneously or with a time difference of less than a preset threshold. In practice, visible light and infrared images have different characteristics. Visible light images are easily affected by lighting conditions but provide more detail, while infrared images are unaffected by lighting conditions but lack detail and texture information. Therefore, using both infrared and visible light images simultaneously can be complementary, resulting in better anomaly detection. The infrared and visible light images can be obtained by taking aerial views of the target photovoltaic power station, showing the individual photovoltaic modules within the station.

[0062] Step 3: Generate anomaly detection results based on infrared and visible light images.

[0063] Specifically, firstly, an infrared image is input into the first feature extraction layer of the anomaly detection model to generate an infrared image feature map, and a visible light image is input into the second feature extraction layer of the anomaly detection model to generate a visible light image feature map. The anomaly detection model includes a first feature extraction layer, a second feature extraction layer, a feature fusion network, a first prediction layer, a second prediction layer, a third prediction layer, and a detection result output layer.

[0064] Secondly, the infrared image feature map and the visible light image feature map are input into the feature fusion network to generate a fused feature map. The aforementioned execution entity can first adjust the infrared image feature map and the visible light image feature map to make their sizes consistent through pixel fusion. Based on this, for example, the average value of two corresponding pixels can be calculated to generate the fused feature map.

[0065] Then, the infrared image feature map is input into the first prediction layer to generate the first anomaly detection result, the visible light image feature map is input into the second prediction layer to generate the second anomaly detection result, and the fused feature map is input into the third prediction layer to generate the third anomaly detection result.

[0066] The structure and parameters of the first, second, and third prediction layers can be set according to actual needs. The first, second, and third anomaly detection results can include the location information and anomaly category information of the anomalous components in the image. The anomaly category information can be used to characterize the anomaly category, which includes, but is not limited to, occlusion, fragmentation, and hot spots. Depending on the needs, the anomaly category information can be a numerical identifier. The location information of the anomalous components in the image can be the two-dimensional coordinates of the anomalous components displayed in the image.

[0067] Finally, the first anomaly detection result, the second anomaly detection result, and the third anomaly detection result are input into the detection result output layer to generate the anomaly detection result corresponding to the target photovoltaic power station.

[0068] The anomaly detection model includes a first feature extraction layer, a second feature extraction layer, a feature fusion network, a first prediction layer, a second prediction layer, a third prediction layer, and a detection result output layer. The first and second feature extraction layers can be, for example, multi-layer convolutional neural networks, which can effectively extract image features. The structures and parameters of the first and second feature extraction layers differ, allowing for targeted extraction of infrared and visible light images, thus fully extracting the characteristics of different image types and improving anomaly detection accuracy. Furthermore, the feature fusion network can fuse feature maps through concatenation or pixel fusion. The first, second, and third prediction layers can, for example, include a softmax function. It is understood that, depending on the needs, the anomaly detection model can also include structures such as downsampling layers, fully connected layers, and intermediate hidden layers.

[0069] This enables the automatic generation of anomaly detection results, including the location information of abnormal components and the anomaly category information, eliminating the need for manual investigation of the cause of the anomaly and helping to improve the efficiency of anomaly elimination.

[0070] Furthermore, the above method may also include the following steps: determining the actual location information of the abnormal component based on the location information of the abnormal component in the image and a pre-generated electronic map of photovoltaic components in the target photovoltaic power station, wherein the electronic map of photovoltaic components is constructed based on the actual location information of multiple photovoltaic components in the target photovoltaic power station.

[0071] In some embodiments, an electronic map of photovoltaic modules can be established based on the actual location information and distribution relationship of each photovoltaic module in the target photovoltaic power station. Each photovoltaic module in the electronic map is mapped one-to-one with its actual location. Thus, the execution entity can extract a visible light image block from the corresponding area of ​​the visible light image based on the abnormal module's location information, and identify the module number painted on the abnormal module. Then, based on the module number, the corresponding photovoltaic module is marked in the electronic map. Furthermore, the actual location information of the abnormal module is determined through the electronic map. Based on this, an anomaly troubleshooting task information is generated according to the abnormal module's actual location information, and this task information is pushed to the maintenance terminal. The anomaly troubleshooting task information includes the abnormal module's actual location information, enabling the maintenance terminal to display the anomaly troubleshooting task information in the electronic map of the photovoltaic modules in the target photovoltaic power station, and to generate route information based on the actual location information and the maintenance terminal's current location information.

[0072] Among these implementation methods, the troubleshooting task information is displayed on the electronic map of the photovoltaic modules in the target photovoltaic power station through the maintenance terminal. Maintenance personnel can intuitively see the location of the abnormal photovoltaic modules on the electronic map and quickly reach the faulty photovoltaic modules according to the route information, which helps to improve the efficiency of troubleshooting and ensure the stable operation of the photovoltaic power station.

[0073] While some of the above embodiments can produce the described technical effects, the fourth technical problem described in the background section still exists: "In the event of extreme weather, relying solely on abnormal information can easily damage photovoltaic modules." Therefore, in some embodiments, each photovoltaic module in the target photovoltaic power station is mounted on a smart support, which controls the angle between each photovoltaic module and the horizontal plane; and the method further includes: acquiring meteorological information, including current weather information and forecast weather information; and in response to determining that the current weather information or forecast weather information meets the warning conditions, sending control information to the smart support to adjust the angle between each photovoltaic module and the horizontal plane to the target angle.

[0074] The intelligent support system includes a communication component, a controller, and an electric telescopic pole. The photovoltaic modules are electrically connected to and power the electric telescopic pole, which adjusts the angle between the support's movable plate and a horizontal plate fixed to the horizontal plane. The communication unit can communicate with the aforementioned execution entity to send and receive data. Upon receiving an angle adjustment command from the execution entity, the angle between each photovoltaic module and the horizontal plane can be adjusted to the target angle. The photovoltaic modules are fixed to the movable plate of the support system, thus moving synchronously with the movable plate.

[0075] The aforementioned implementing entity can acquire meteorological information and input it into a pre-trained angle prediction network, which can be a trained MLP (Multilayer Perceptron). This yields the target angle and adjustment time corresponding to the meteorological information. In practice, stronger winds necessitate keeping photovoltaic modules as horizontal as possible to minimize damage. This solves the problem of manual angle adjustment required in existing technologies, reducing manpower and material costs, and allowing for precise angle and adjustment time.

[0076] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an artificial intelligence-based photovoltaic power plant anomaly information generation device. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0077] like Figure 2As shown, an AI-based photovoltaic power plant anomaly information generation device 200 in some embodiments includes: an acquisition unit 201 configured to acquire a preprocessed historical power generation dataset of a target photovoltaic power plant, wherein the historical power generation data in the historical power generation dataset includes environmental features, photovoltaic panel features, and power generation, and the environmental features and photovoltaic panel features each include multiple sub-features and each includes a sub-feature set composed of multiple sub-features; a grouping unit 202 configured to group multiple sub-features according to a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, to obtain a first sub-feature group and a second sub-feature group, wherein the correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold; and a model pool acquisition unit 203 configured to acquire a pre-built model pool, and for each model in the model pool, perform the following feature selection steps: determine a target number of sub-features according to the second sub-feature group to obtain a target sub-feature group; and combine the first sub-feature group and the target sub-feature group. The union of each sub-feature in the dataset is determined as the input feature corresponding to the model; the pre-training unit 204 is configured to train each model in the model pool according to the input feature corresponding to each model, using the first subset of historical power generation data in the historical power generation dataset, to obtain the trained model pool, and to determine the trained model pool as the pre-trained model pool; the sorting unit 205 is configured to sort each pre-trained model in the pre-trained model pool according to the number of input features and the model type, to obtain the pre-trained model sequence; the training unit 206 is configured to train each pre-trained model in the pre-trained model sequence using the second subset of historical power generation data in the historical power generation dataset, to obtain the power generation prediction model; the prediction unit 207 is configured to input the environmental features and photovoltaic panel features of the time point to be predicted into the power generation prediction model to obtain the predicted power generation of the time point to be predicted; the generation unit 208 is configured to generate abnormal information and push the abnormal information to the maintenance terminal when the difference between the actual power generation and the predicted power generation of the time point to be predicted is greater than a preset difference threshold.

[0078] It is understandable that the units described in the device 200 are related to the reference. Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the device 200 and the units contained therein, and will not be repeated here.

[0079] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0080] like Figure 3 As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0081] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0082] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0083] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0084] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0085] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: acquire a preprocessed historical power generation dataset of the target photovoltaic power plant, wherein the historical power generation data in the historical power generation dataset includes environmental features, photovoltaic panel features, and power generation, and the environmental features and photovoltaic panel features each include multiple sub-features, and each of the included sub-features constitutes a sub-feature set; group the multiple sub-features according to a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, to obtain a first sub-feature group and a second sub-feature group, wherein the correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold; acquire a pre-constructed model pool, and for each model in the model pool, perform the following feature selection steps: based on the second sub-feature group, determine a target number of sub-features, to obtain a target sub-feature... Feature groups; the union of each sub-feature in the first sub-feature group and the target sub-feature group is determined as the input feature corresponding to the model; based on the input feature corresponding to each model, each model in the model pool is trained using the first subset of historical power generation data from the historical power generation dataset to obtain the trained model pool, and the trained model pool is determined as the pre-trained model pool; for each pre-trained model in the pre-trained model pool, they are sorted according to the number of input features and model type to obtain the pre-trained model sequence; for each pre-trained model in the pre-trained model sequence, they are trained using the second subset of historical power generation data from the historical power generation dataset to obtain the power generation prediction model; the environmental features and photovoltaic panel features at the time point to be predicted are input into the power generation prediction model to obtain the predicted power generation at the time point to be predicted; when the difference between the actual power generation and the predicted power generation at the time point to be predicted is greater than a preset difference threshold, anomaly information is generated and pushed to the maintenance terminal.

[0086] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0087] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0088] The units described in some embodiments of this disclosure can be implemented in software or in hardware.

[0089] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0090] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for generating abnormal information in photovoltaic power plants based on artificial intelligence, comprising: Obtain the preprocessed historical power generation dataset of the target photovoltaic power station. The historical power generation data in the historical power generation dataset includes environmental features, photovoltaic panel features, and power generation. The environmental features and photovoltaic panel features each include multiple sub-features and each includes a sub-feature set composed of multiple sub-features. Based on a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, the multiple sub-features are grouped to obtain a first sub-feature group and a second sub-feature group. The correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold. Obtain a pre-built model pool, and for each model in the model pool, perform the following feature selection steps: determine a target number of sub-features based on the second sub-feature group to obtain a target sub-feature group; determine the union of each sub-feature in the first sub-feature group and the target sub-feature group as the input feature corresponding to the model; Based on the input features corresponding to each model, the models in the model pool are trained using the first subset of historical power generation data in the historical power generation dataset to obtain the trained model pool, and the trained model pool is determined as the pre-trained model pool. For each pre-trained model in the pre-trained model pool, sort them according to the number of input features and the model type to obtain a pre-trained model sequence; For each pre-trained model in the pre-trained model sequence, a power generation prediction model is obtained by training it using the second historical power generation data subset of the historical power generation dataset. The environmental characteristics and photovoltaic panel characteristics at the time point to be predicted are input into the power generation prediction model to obtain the predicted power generation at the time point to be predicted. When the difference between the actual power generation at the predicted time point and the predicted power generation exceeds a preset difference threshold, an anomaly information is generated and pushed to the maintenance terminal. The pre-trained model sequence includes a first pre-trained model, a second pre-trained model, and a third pre-trained model. The step of training each pre-trained model in the pre-trained model sequence using a second subset of historical power generation data from the historical power generation dataset to obtain a power generation prediction model includes: Select target historical power generation data from the second historical power generation data subset, and determine the first input feature value of the first pre-trained model, the second input feature value of the second pre-trained model, and the third input feature value of the third pre-trained model based on the target historical power generation data and the input features of each pre-trained model. The first input feature value is input into the first pre-trained model to obtain the first power generation prediction value; Replace the feature value of the dimension with the lowest correlation to power generation in each dimension of the second input feature value with the first power generation prediction value to obtain the updated second input feature value. The updated second input feature value is input into the second pre-trained model to obtain the second power generation prediction value; The feature value of the dimension with the lowest correlation to power generation among the dimensions of the third input feature value is replaced with the second power generation prediction value to obtain the updated third input feature value. The updated third input feature value is input into the third pre-trained model to obtain the third predicted power generation value; The predicted power generation is determined by weighting the first predicted power generation value, the second predicted power generation value, and the third predicted power generation value. The difference between the predicted power generation and the power generation in the target historical power generation data is determined, and the parameters of the first pre-trained model, the second pre-trained model and the third pre-trained model are adjusted according to the difference until the training stop condition is met, so as to obtain the power generation prediction model.

2. The method according to claim 1, characterized in that, The pre-trained models in the pre-trained model pool are sorted according to the number of input features and model type to obtain a pre-trained model sequence, including: For each of the pre-trained models, they are grouped according to model type to obtain multiple pre-trained model groups; For each pre-trained model in each of the multiple pre-trained model groups, sort them in ascending order of the number of input features to obtain the pre-trained model subsequence corresponding to each pre-trained model group; The pre-trained model sequence is obtained based on the pre-trained model sub-sequences corresponding to the multiple pre-trained model groups.

3. The method according to claim 2, characterized in that, The method further includes: In response to receiving a re-inspection request for the abnormal information sent by the maintenance terminal, the power generation data of multiple associated photovoltaic power stations of the target photovoltaic power station are obtained, and the distance between each of the multiple associated photovoltaic power stations and the target photovoltaic power station meets the distance condition. Based on the power generation data of the multiple associated photovoltaic power plants, the expected power generation data of the target photovoltaic power plant is determined; Based on the expected power generation data and the predicted power generation, a re-examination result is generated; The re-inspection results are pushed to the maintenance terminal.

4. The method according to any one of claims 1-3, characterized in that, Each photovoltaic module in the target photovoltaic power station is mounted on an intelligent support, which is used to control the angle between each photovoltaic module and the horizontal plane. as well as The method further includes: Obtain meteorological information, including current weather information and forecast weather information; In response to determining that the current weather information or the predicted weather information meets the warning conditions, control information is sent to the smart bracket so that the smart bracket adjusts the angle between each photovoltaic module and the horizontal plane to the target angle.

5. An artificial intelligence-based photovoltaic power plant anomaly information generation device, comprising: The acquisition unit is configured to acquire a preprocessed historical power generation dataset of the target photovoltaic power station. The historical power generation dataset includes environmental features, photovoltaic panel features, and power generation. The environmental features and photovoltaic panel features each include multiple sub-features and each includes a sub-feature set composed of multiple sub-features. The grouping unit is configured to group the multiple sub-features according to a preset relevance threshold and the correlation between each sub-feature in the sub-feature set and the power generation, to obtain a first sub-feature group and a second sub-feature group, wherein the correlation between each sub-feature in the first sub-feature group and the power generation is greater than or equal to the relevance threshold, and the correlation between each sub-feature in the second sub-feature group and the power generation is less than the relevance threshold. The model pool acquisition unit is configured to acquire a pre-built model pool, and for each model in the model pool, perform the following feature selection steps: determine a target number of sub-features based on the second sub-feature group to obtain a target sub-feature group; and determine the union of each sub-feature in the first sub-feature group and the target sub-feature group as the input feature corresponding to the model. The pre-training unit is configured to train each model in the model pool using a first subset of historical power generation data from the historical power generation dataset based on the input features corresponding to each model, thereby obtaining a trained model pool, and to determine the trained model pool as the pre-trained model pool. The sorting unit is configured to sort the pre-trained models in the pre-trained model pool according to the number of input features and the model type to obtain a pre-trained model sequence. The training unit is configured to train each pre-trained model in the pre-trained model sequence using a second subset of historical power generation data from the historical power generation dataset to obtain a power generation prediction model. The prediction unit is configured to input the environmental characteristics of the time point to be predicted and the photovoltaic panel characteristics of the time point to be predicted into the power generation prediction model to obtain the predicted power generation of the time point to be predicted. The generation unit is configured to generate abnormal information and push the abnormal information to the maintenance terminal when the difference between the actual power generation at the time point to be predicted and the predicted power generation is greater than a preset difference threshold. The pre-trained model sequence includes a first pre-trained model, a second pre-trained model, and a third pre-trained model; and The training unit is further configured as follows: Select target historical power generation data from the second historical power generation data subset, and determine the first input feature value of the first pre-trained model, the second input feature value of the second pre-trained model, and the third input feature value of the third pre-trained model based on the target historical power generation data and the input features of each pre-trained model. The first input feature value is input into the first pre-trained model to obtain the first power generation prediction value; Replace the feature value of the dimension with the lowest correlation to power generation in each dimension of the second input feature value with the first power generation prediction value to obtain the updated second input feature value. The updated second input feature value is input into the second pre-trained model to obtain the second power generation prediction value; The feature value of the dimension with the lowest correlation to power generation among the dimensions of the third input feature value is replaced with the second power generation prediction value to obtain the updated third input feature value. The updated third input feature value is input into the third pre-trained model to obtain the third predicted power generation value; The predicted power generation is determined by weighting the first predicted power generation value, the second predicted power generation value, and the third predicted power generation value. The difference between the predicted power generation and the power generation in the target historical power generation data is determined, and the parameters of the first pre-trained model, the second pre-trained model and the third pre-trained model are adjusted according to the difference until the training stop condition is met, so as to obtain the power generation prediction model.

6. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-4.

Images