A wind farm equipment health degree evaluation method and system based on multi-source data fusion and adaptive dynamic modeling

Abstract

The application provides a wind farm equipment health degree evaluation method and system based on multi-source data fusion and adaptive dynamic modeling, comprising: collecting multi-source data, preprocessing the data, using a timestamp as an index, fusing data of different sources by using a data fusion technology, intercepting sequence data within a fixed time period p, reducing the dimension of the sequence data by using PCA, and extracting main components; training a dynamic LSTM sequence model, calculating a sliding step q according to a deviation index, setting p = p + q, returning to training; and outputting a prediction result. An adaptive dynamic LSTM model is constructed by using multi-source data and a sliding time window with variable width, real-time discrimination of fan faults is realized by combining a dynamic threshold algorithm, data-driven equipment health degree evaluation and degradation trend prediction are supported, an attention mechanism is added, important moments in the time sequence are captured, the convergence speed is accelerated, the accuracy of the model is improved, and power generation optimization and ecological compatibility decision-making are supported.

Description

Technical Field

[0001] The present application relates to the field of wind farm equipment health assessment, and in particular to a wind farm equipment health assessment method and system based on multi-source data fusion and adaptive dynamic modeling. Background Art

[0002] my country's photovoltaic power plants are numerous and occupy vast areas. They are typically located in deserts, Gobi deserts, wastelands, and hillsides, all of which have poor natural environments, resulting in poor network connectivity for these plants. This necessitates the use of intelligent devices such as drones for inspections. During these inspections, drones retrieve data from the photovoltaic plants to determine if any anomalies are present. During operation, photovoltaic plants generate a vast amount of data, and the data formats vary significantly between devices. Drones have limited computing power and are unable to process this vast array of data in a timely manner. Therefore, an edge computing device within the photovoltaic plant is required to normalize and analyze the data format, transmitting the analysis results to intelligent devices.

[0003] With the large-scale development of wind farms, equipment health assessment has become critical to ensuring efficient operation of wind farms. Traditional methods often rely on a single data source or static models, making them difficult to cope with the complex and changing wind farm environment.

[0004] Therefore, designing a method that combines multi-source data fusion with adaptive dynamic modeling methods for wind farm equipment health assessment, improving the accuracy and real-time performance of the assessment, reducing the risk of errors caused by human subjectivity, and saving manpower and material resources is an important research topic for technical personnel in this field. Summary of the Invention

[0005] In view of this, it is necessary to provide a wind farm equipment health assessment method and system based on multi-source data fusion and adaptive dynamic modeling. An adaptive dynamic LSTM model is constructed through multi-source data and a sliding time window with variable width, and the time series data related to anomalies is analyzed, thereby improving the model performance and prediction accuracy.

[0006] In a first aspect, an embodiment of the present application provides a method for evaluating the health of wind farm equipment based on multi-source data fusion and adaptive dynamic modeling, the method comprising:

[0007] S1: Collects time-stamped operation and monitoring data from various devices, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones;

[0008] S2: Preprocess the data and perform time series data processing;

[0009] S3: Using timestamps as indexes, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment;

[0010] S4: intercepting sequence data within a fixed time period p, using PCA to reduce the dimension of the sequence data, extracting the main components, reducing the feature dimensions, and dividing the sequence data into a training set, a validation set, and a test set according to a preset ratio;

[0011] S5: Construct the extracted main components into a dynamic LSTM sequence model according to the time step and train it; use the dynamic LSTM sequence model to verify the time series data of the validation set, evaluate the model performance, and adjust the hyperparameters; finally evaluate the generalization ability of the model on the test set;

[0012] S6: Evaluate the dynamic LSTM sequence model using the mean square error evaluation index. If the threshold index is not met, calculate the sliding step size q based on the deviation index, set p = p + q, and return to S4;

[0013] S7: Otherwise, load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an abnormality, and output a prediction result.

[0014] Optionally, in an implementation of the first aspect of the present invention, the step S2: preprocessing the data and performing time series data processing includes:

[0015] S2.1: Data collection and integration: clarify the data source, store data from different sources in a unified manner, ensure consistent format, handle missing values, duplicate values, and outliers, and ensure data integrity;

[0016] S2.2: Data format standardization: unify the formats of different data sources, convert data into types suitable for analysis, and encode categorical data;

[0017] S2.3: Data alignment and matching: aligning timestamps of different time series data, aligning coordinate systems of geospatial data, and matching entities in different data sources;

[0018] S2.4: Feature engineering: extract useful features from raw data, select the features that are most helpful for model prediction, reduce redundancy, and standardize or normalize numerical features;

[0019] S2.5: Data quality assessment, data consistency check, ensure data consistency across different data sources, completeness check: check data completeness, handle missing values, verify data accuracy, and correct erroneous data;

[0020] S2.6: Sort different time series data by timestamp.

[0021] Optionally, in an implementation of the first aspect of the present invention, S3: using timestamps as indexes and employing data fusion technology to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment, includes:

[0022] Use convolutional neural network (CNN) to extract image data features, and recurrent neural network (RNN) to extract time series data features;

[0023] Use feature splicing to fuse data from different sources;

[0024] The fused data is used to train a deep learning model to generate series data of comprehensive health indicators of wind farm equipment.

[0025] Optionally, in an implementation of the first aspect of the present invention, S4: intercepting sequence data within a fixed time period p, performing dimensionality reduction on the sequence data using PCA, extracting main components, reducing feature dimensions, and dividing the data into a training set, a validation set, and a test set according to a preset ratio, includes:

[0026] Determine the starting and ending points of time period p;

[0027] Using a sliding window method to intercept the sequence data between the starting point and the end point;

[0028] The intercepted time series data is converted into a matrix form suitable for PCA, with each row as a sample and each column as a feature;

[0029] Perform Z-score standardization on matrix data;

[0030] The PCA algorithm was applied to reduce the dimensionality of the truncated sequence data, and the ElbowPlot method was used to select the optimal number of the first k principal components.

[0031] Optionally, in an implementation of the first aspect of the present invention, the architecture of the dynamic LSTM sequence model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer;

[0032] Wherein, the input layer receives and preprocesses input time series data;

[0033] The LSTM layer extracts features of sequence data and generates a hidden state sequence;

[0034] The attention mechanism layer assigns weights to the hidden states of each time step, generates weighted representations, and enhances the model's attention to important time steps;

[0035] The fully connected layer selects an activation function according to the task type and maps the context vector to the target space;

[0036] The output layer generates the final prediction result.

[0037] Optionally, in an implementation of the first aspect of the present invention, S6: evaluating the LSTM model using a mean square error evaluation index, and when a threshold index is not met, calculating a sliding step size q based on a deviation index, includes:

[0038] The LSTM model is evaluated by the mean square error evaluation index, where the error evaluation formula is M SE :

[0039]

[0040] Among them, y i represents the true value, Represents the predicted value, n represents the number of samples, ε represents the error coefficient, and when y i =0,ε=0.001, when y i ≠0, ε=0;

[0041] The sliding step length q is calculated based on the deviation index. The formula is:

[0042]

[0043] Where δ represents the threshold, [] represents the rounding operation, and η is the adjustment coefficient.

[0044] Optionally, in an implementation of the first aspect of the present invention, the loss function is:

[0045]

[0046] Where N is the number of samples, y i,c is the one-hot encoding of the true label, is the predicted probability that the i-th sample belongs to category C, where C is the number of categories. If the sample belongs to category C, then y i,c =1, otherwise 0, α t,i is the attention weight of the i-th input at time step t, and λ is the regularization coefficient.

[0047] In a second aspect, an embodiment of the present application provides a wind farm equipment health assessment system based on multi-source data fusion and adaptive dynamic modeling, which is applied to the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling as described in the first aspect. The system includes:

[0048] Data acquisition module: collects time-stamped operation and monitoring data from various devices, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones;

[0049] Data preprocessing module: preprocess the data and perform time series data processing;

[0050] Data fusion module: Using timestamps as indexes, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment;

[0051] Data processing module: intercepts sequence data within a fixed time period p, uses PCA to reduce the dimension of the sequence data, extracts the main components, reduces the feature dimensions, and divides the data into training set, validation set and test set according to the preset ratio;

[0052] Model training module: The extracted main components are constructed into a dynamic LSTM sequence model according to the time step and trained; the dynamic LSTM sequence model is validated on the time series data of the validation set to evaluate the model performance and adjust the hyperparameters; and the generalization ability of the model is finally evaluated on the test set;

[0053] Model evaluation module: Evaluate the dynamic LSTM sequence model using the mean square error evaluation index. When the threshold index is not met, calculate the sliding step size q based on the deviation index, set p = p + q, and return to S4;

[0054] Anomaly detection module: Otherwise, the dynamic LSTM sequence model is loaded, anomaly detection is performed on the time series data to be detected through the dynamic LSTM sequence model, whether an anomaly exists is analyzed, and a prediction result is output.

[0055] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to implement the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling as described in the first aspect when executing the instructions.

[0059] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a program, wherein the program instructs a device to execute the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling as described in the first aspect.

[0060] The technical solution provided by the present invention provides a wind farm equipment health assessment method and system based on multi-source data fusion and adaptive dynamic modeling. The method collects multi-source data, pre-processes the data, and fuses the data from different sources using timestamps as indexes using data fusion technology. Sequence data within a fixed time period p is intercepted and dimensionality reduced using PCA to extract the main components. A dynamic LSTM sequence model is trained. When a threshold indicator is not met, a sliding step size q is calculated based on the deviation index, p is set to p = p + q, and the training step is returned to the training step. The prediction result is output. An adaptive dynamic LSTM model is constructed using multi-source data and a sliding time window with variable width to analyze time series data related to anomalies, improving the model's performance and prediction accuracy.

[0061] Beneficial effects:

[0062] (1) Considering that photovoltaic stations involve multiple devices, the data generated by different devices are processed by data fusion, which improves the monitoring capability of the comprehensive operating status of the entire photovoltaic station.

[0063] (2) An adaptive dynamic LSTM model is constructed with a sliding time window of variable width to analyze the time series data related to anomalies, thereby improving the performance of the model and the accuracy of prediction.

[0064] (3) An attention mechanism is added to assign weights to different parts of the input sequence, capture important moments in the time series, speed up convergence, and improve the accuracy of the model.

[0065] (4) Based on the auto-encoding algorithm, the operating characteristic parameters of the whole machine are extracted, and combined with the dynamic threshold algorithm, real-time identification of fan faults is achieved, supporting data-driven equipment health assessment and degradation trend prediction.

[0066] (5) Integrate grid load data, environmental parameters (such as atmospheric boundary layer model) and operation and maintenance logs to establish a dynamic correlation model between wind farm efficiency and environmental adaptability to support power generation optimization and ecological compatibility decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A schematic diagram of a module for a wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling provided in one embodiment of the present application.

[0068] Figure 2 A schematic diagram of a wind farm equipment health assessment system module based on multi-source data fusion and adaptive dynamic modeling provided in one embodiment of the present application.

[0069] Figure 3 A schematic diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0071] It should be noted that, in the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art in the art to which this application relates. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0072] It should be noted that, in the embodiments of the present application, words such as "first" and "second" are only used for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order. Features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way.

[0073] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0074] Example 1

[0075] The present application provides a wind farm equipment health assessment method and system based on multi-source data fusion and adaptive dynamic modeling, including: S1: collecting operation and monitoring data with timestamps from different devices, the devices including photovoltaic modules, electrical equipment, detection equipment, meteorological sensors, video surveillance equipment and drones; S2: pre-processing the data and performing time series data processing; S3: using timestamps as indexes, using data fusion technology to fuse data from different sources to generate wind farm equipment comprehensive health index sequence data; S4: intercepting sequence data within a fixed time period p, using PCA to reduce the dimension of the sequence data, extracting the main components, reducing the feature dimensions, and dividing it into training data according to a preset ratio. set, validation set and test set; S5: construct the extracted main components into a dynamic LSTM sequence model according to the time step and train it; verify the time series data of the validation set through the dynamic LSTM sequence model, evaluate the model performance, and adjust the hyperparameters; finally evaluate the generalization ability of the model on the test set; S6: evaluate the dynamic LSTM sequence model through the mean square error evaluation index. When the threshold index is not met, calculate the sliding step q according to the deviation index, set p=p+q, and return to S4; S7: otherwise, load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an anomaly, and output the prediction result.

[0076] An adaptive dynamic LSTM model is constructed through multi-source data and a sliding time window with variable width. Combined with a dynamic threshold algorithm, it enables real-time identification of wind turbine faults, supports data-driven equipment health assessment and degradation trend prediction, and incorporates an attention mechanism to capture important moments in the time series, accelerate convergence, improve model accuracy, and support power generation optimization and ecological compatibility decision-making.

[0077] Figure 1 A schematic flow chart of a wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling provided in one embodiment of the present application.

[0078] like Figure 1 As shown in FIG, a wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling includes:

[0079] S1: Collects time-stamped operation and monitoring data from various devices, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones.

[0080] It is understood that in this embodiment, data is obtained from different devices, which may include photovoltaic modules, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones.

[0081] Multi-source heterogeneous data integration: Integrate multimodal data sources such as SCADA real-time operation data, meteorological sensors (temperature / pressure / wind speed), video surveillance (AI / AR / pan-tilt), and equipment status detection systems to build a unified spatiotemporal database.

[0082] S2: Preprocess the data and perform time series data processing.

[0083] It is understood that, in this embodiment, the step S2: pre-processing the data and performing time series data processing includes the following steps:

[0084] S2.1: Data collection and integration: clarify the data source, store data from different sources in a unified manner, ensure consistent format, handle missing values, duplicate values, and outliers, and ensure data integrity;

[0085] S2.2: Data format standardization: unify the formats of different data sources, convert data into types suitable for analysis, and encode categorical data;

[0086] S2.3: Data alignment and matching: aligning timestamps of different time series data, aligning coordinate systems of geospatial data, and matching entities in different data sources;

[0087] S2.4: Feature engineering: extract useful features from raw data, select the features that are most helpful for model prediction, reduce redundancy, and standardize or normalize numerical features;

[0088] S2.5: Data quality assessment, data consistency check, ensure data consistency across different data sources, completeness check: check data completeness, handle missing values, verify data accuracy, and correct erroneous data;

[0089] S2.6: Sort different time series data by timestamp.

[0090] Specifically, data is preprocessed, which mainly includes missing value processing, outlier detection and processing, data standardization and normalization, and time series data processing.

[0091] Missing value handling: In a PV plant, different devices may generate different data types. Therefore, care must be taken to handle missing data due to device failure or communication issues. Missing data must be identified and handled. Missing values ​​are then filled using the more conventional interpolation method.

[0092] Outlier detection and processing: Outliers can be caused by equipment failure, sensor errors, or other abnormal conditions. Methods for handling outliers include deleting them, replacing them with appropriate values, or using interpolation methods to fill in the gaps. This is also a relatively common approach.

[0093] Time series data processing: This involves processing time series data, ensuring the correctness of timestamps, and performing time alignment. Since this analysis primarily involves determining whether there are anomalies at the PV station, the timestamp is a crucial parameter.

[0094] Data standardization and normalization: Data standardization and normalization are key steps in data preprocessing. The main purpose is to make data with different features have the same scale so as to better adapt to model training and improve algorithm performance.

[0095] The goal of data normalization is to scale the data to a specific range, usually [0,1]. The method used here is the L1 norm, also known as (Lassoregularization). The L1 norm refers to the sum of the absolute values ​​of each element in the vector, also known as the Manhattan norm, because it measures the Manhattan distance between two points of the vector elements. For an n-dimensional vector x = [x1, x2, ... x n ],

[0096] S3: Using timestamp as index, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment.

[0097] It is understandable that, in this embodiment, the step S3: using timestamps as indexes and employing data fusion technology to fuse data from different sources to generate a series of wind farm equipment comprehensive health index data, includes:

[0098] Use convolutional neural network (CNN) to extract image data features, and recurrent neural network (RNN) to extract time series data features;

[0099] Use feature splicing to fuse data from different sources;

[0100] The fused data is used to train a deep learning model to generate series data of comprehensive health indicators of wind farm equipment.

[0101] Specifically, data fusion techniques (such as Kalman filtering and deep learning fusion models) are used to fuse data from different sources to generate a comprehensive health indicator. Kalman filtering is suitable for linear systems, while deep learning fusion models can handle nonlinear relationships.

[0102] S4: intercept the sequence data within a fixed time period p, use PCA to reduce the dimension of the sequence data, extract the main components, reduce the feature dimensions, and divide it into a training set, a validation set, and a test set according to a preset ratio.

[0103] Specifically, principal component analysis (PCA) is a commonly used dimensionality reduction technique that can extract key features from high-dimensional data and reduce redundant information. LSTM (Long Short-Term Memory) is a deep learning model suitable for time series modeling that can capture long-term dependencies in data. Combining PCA with LSTM can reduce model complexity while improving the accuracy and efficiency of time series forecasting. This method has broad application prospects in finance, meteorology, healthcare, and other fields. PCA is used to reduce the dimensionality of standardized data, extracting the principal components and reducing feature dimensionality. The reduced data is then structured into sequences suitable for LSTM input, based on time steps.

[0104] It is understood that, in this embodiment, the S4: intercepting sequence data within a fixed time period p, using PCA to reduce the dimension of the sequence data, extracting the main components, reducing the feature dimensions, and dividing the sequence data into a training set, a validation set, and a test set according to a preset ratio, includes:

[0105] Determine the starting and ending points of time period p;

[0106] Using a sliding window method to intercept the sequence data between the starting point and the end point;

[0107] The intercepted time series data is converted into a matrix form suitable for PCA, with each row as a sample and each column as a feature;

[0108] Perform Z-score standardization on matrix data;

[0109] The PCA algorithm was applied to reduce the dimensionality of the truncated sequence data, and the ElbowPlot method was used to select the optimal number of the first k principal components.

[0110] S5: Construct the extracted main components into a dynamic LSTM sequence model according to the time step and train it; verify the dynamic LSTM sequence model on the time series data of the validation set, evaluate the model performance, and adjust the hyperparameters; finally evaluate the generalization ability of the model on the test set.

[0111] It can be understood that, in this embodiment, the architecture of the dynamic LSTM sequence model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer;

[0112] Wherein, the input layer receives and preprocesses input time series data;

[0113] The LSTM layer extracts features of sequence data and generates a hidden state sequence;

[0114] The attention mechanism layer assigns weights to the hidden states of each time step, generates weighted representations, and enhances the model's attention to important time steps;

[0115] The fully connected layer selects an activation function according to the task type and maps the context vector to the target space;

[0116] The output layer generates the final prediction result.

[0117] Specifically, a multi-layer LSTM network is constructed, with features extracted using PCA as input and predictions as output. Dynamic adjustment mechanisms, such as adaptive learning rate and dropout, are introduced to improve model generalization. Loss functions (such as mean squared error and cross entropy) are selected based on task requirements.

[0118] Specifically, the architecture of the dynamic LSTM sequence model consists of an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input data shape is (batch_size, time_steps, feature_dim), where: batch_size: batch size; time_steps: time steps; feature_dim: feature dimension at each time step.

[0119] LSTM layer: A core component used to capture long-term dependencies in time series. Parameter settings: units: The number of hidden layer neurons, which determines model complexity. return_sequences: Whether to return the output of each time step (True for multi-step predictions, False for single-step predictions). dropout: Prevents overfitting by randomly dropping some neurons.

[0120] Attention Mechanism Layer: The attention mechanism layer is a multi-head attention mechanism, an extension of the self-attention mechanism. It captures information from different subspaces of the input data by computing multiple attention heads in parallel. The query, key, and value are projected into different subspaces using multiple linear transformations. Attention is computed separately in each subspace. The outputs of multiple attention heads are concatenated and linearly transformed to obtain the final output.

[0121] Fully connected layer: maps the LSTM output to the target dimension, usually used for classification or regression tasks. Output layer: selects the activation function based on the task (such as softmax for classification and linear for regression).

[0122] Preprocess time series data related to anomalies, such as light intensity, temperature, and current, which may show significant changes when an anomaly occurs. Label the anomaly data, either as a binary label (normal / abnormal) or a multi-class label (different types of anomalies).

[0123] from tensorflow.keras.models import Sequential

[0124] from tensorflow.keras.layers import LSTM,Dense

[0125] model=Sequential()

[0126] model.add(LSTM(units=50,activation='relu',input_shape=(n_steps,n_features)))

[0127] model.add(Dense(units=1))#output layer

[0128] model.compile(optimizer='adam',loss='mse')#Mean square error (MSE) in regression task

[0129] Construct the training set time series data into a format suitable for LSTM input, typically a three-dimensional array, including the number of samples, time steps, and features. Use the training set to train the LSTM model and monitor the model's performance on the validation set.

[0130] The evaluation of LSTM here uses three methods: confusion matrix, accuracy and recall.

[0131] A confusion matrix is ​​a two-dimensional table that shows the performance of a classification model on different classes. For a binary classification problem, the confusion matrix contains four entries: True Positives (TP): positive examples correctly classified as positive. True Negatives (TN): negative examples correctly classified as negative. False Positives (FP): negative examples incorrectly classified as positive. False Negatives (FN): positive examples incorrectly classified as negative.

[0132] Predicted as positive Predicted as negative class Actually positive TP FN Actually negative FP TN

[0133] Recall is an indicator used to measure the ability of a model to capture positive examples. For a binary classification problem, the recall rate is:

[0134]

[0135] When detecting anomalies in photovoltaic stations, a high recall rate means that the model can better capture actual abnormal events and reduce the missed alarm rate.

[0136] Accuracy is an indicator for evaluating the performance of a classification model. It refers to the proportion of samples that are correctly classified by the model. For a binary classification problem, the accuracy calculation formula is:

[0137]

[0138] In the anomaly detection problem, accuracy is not the only evaluation metric. In extremely imbalanced datasets (where normal samples far outnumber abnormal samples), the model may tend to predict all samples as normal, resulting in high accuracy but ineffective detection of anomalies. In this case, other metrics such as recall are more practical.

[0139] Model Adjustment: LSTM models can capture regularities in normal data by learning time series patterns, allowing them to identify anomalies that deviate from these patterns during testing. During model training, the model needs to be continuously adjusted and optimized based on specific scenarios to improve the accuracy and robustness of anomaly detection.

[0140] S6: Evaluate the dynamic LSTM sequence model using the mean square error evaluation index. When the threshold index is not met, calculate the sliding step size q based on the deviation index, set p = p + q, and return to S4.

[0141] It is understood that, in this embodiment, the S6: evaluating the LSTM model using the mean square error evaluation index, and when the threshold index is not met, calculating the sliding step size q according to the deviation index, includes:

[0142] The LSTM model is evaluated by the mean square error evaluation index, where the error evaluation formula is M SE :

[0143]

[0144] Among them, y i represents the true value, Represents the predicted value, n represents the number of samples, ε represents the error coefficient, and when y i =0,ε=0.001, when y i ≠0, ε=0;

[0145] The sliding step length q is calculated based on the deviation index. The formula is:

[0146]

[0147] Where δ represents the threshold, [] represents the rounding operation, and η is the adjustment coefficient.

[0148] Specifically, the loss function is:

[0149]

[0150] Where N is the number of samples, y i,c is the one-hot encoding of the true label, is the predicted probability that the i-th sample belongs to category C, where C is the number of categories. If the sample belongs to category C, then y i,c =1, otherwise 0, α t,i is the attention weight of the i-th input at time step t, and λ is the regularization coefficient.

[0151] It can be understood that in this embodiment, the data for constructing the model is continuously updated through a rolling time window with variable width, old data is removed, and new data is continuously added, thereby avoiding the problem of the model being incompatible with the statistical laws of the data.

[0152] S7: Otherwise, load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an abnormality, and output a prediction result.

[0153] An autoencoding algorithm extracts characteristic operational parameters for the entire turbine, combined with a dynamic threshold algorithm to identify wind turbine faults in real time, supporting data-driven equipment health assessment and degradation trend prediction. Cross-system collaborative analysis: Integrating grid load data, environmental parameters (such as atmospheric boundary layer models), and operation and maintenance logs, a dynamic correlation model is established between wind farm efficiency and environmental adaptability, supporting power generation optimization and ecological compatibility decision-making.

[0154] Specifically, real-time data collection is required to ensure that real-time data is obtained from various devices (PV panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, drones, etc.). This data may include information related to the operating status of the PV station, such as light intensity, temperature, and current.

[0155] Perform preprocessing on real-time data similar to that used for training data, including missing value handling, outlier detection and handling, data standardization and normalization, and time series data processing. Ensure that the real-time data is in the same format as the data used for model training.

[0156] Use the trained LSTM model to make predictions on real-time data. Ensure the input data is in the correct time series format and use the probability values ​​or category labels output by the model to determine whether anomalies exist. If the model is designed for multi-category anomaly detection, it can also identify different types of anomalies.

[0157] Anomaly determination and processing: Based on the probability value or category label output by the model, a threshold is set. Samples exceeding this threshold are considered anomalies. Depending on actual needs, different processing methods can be adopted, such as recording anomaly information, issuing an alert, and performing automatic repairs.

[0158] Multi-category anomaly handling: If the model is designed for multi-category anomaly detection, different types of anomalies can be identified based on the category labels output by the model. Different handling strategies can be adopted for different types of anomalies.

[0159] Regularly evaluate the performance of the model and monitor its performance in real-world applications. If the model's performance degrades or drifts, consider retraining the model or adjusting model parameters.

[0160] The model output is compared with the set anomaly threshold. If the threshold is exceeded, an anomaly alarm is triggered, and the anomaly information is transmitted to the inspection-related smart devices. Alarms can also be generated through various means, including but not limited to: sending emails or text messages to notify relevant personnel; triggering automated systems to implement specific remediation or isolation measures; and displaying warning messages on the monitoring interface for immediate operator attention.

[0161] Considering that photovoltaic stations involve multiple devices, the data generated by different devices are processed through data fusion, which improves the monitoring capability of the comprehensive operating status of the entire photovoltaic station.

[0162] Data standardization and normalization methods are used to ensure that data generated by different devices are compared on the same scale. This helps eliminate dimensional differences between data from different devices and improves model stability and convergence speed.

[0163] By adopting deep learning models such as LSTM, we can better capture complex patterns in time series data, thereby improving the accuracy of anomaly detection. Model evaluation metrics such as confusion matrix, precision, and recall help evaluate model performance.

[0164] The establishment of an abnormality alarm system enables timely notification of relevant personnel when abnormalities are discovered, allowing necessary measures to be taken. This enhances the real-time response capability to abnormal situations in photovoltaic plants.

[0165] Smart devices are used to capture data from various devices in photovoltaic plants, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones. The collected data is preprocessed, including missing value handling, outlier detection and handling, data standardization and normalization, and time series data processing. Missing values ​​are filled using interpolation methods, outliers can be deleted or replaced, and time series data is time-stamped and aligned.

[0166] Use the Long Short-Term Memory (LSTM) model to analyze time series data related to anomalies. The LSTM model architecture includes an input layer, an LSTM layer, an attention layer, and an output layer. It is used to capture patterns in time series data, especially the significant changes observed when anomalies occur.

[0167] Use the trained model to predict real-time data and determine whether there are any anomalies in the PV plant. If an anomaly is detected, the anomaly alarm system is triggered, which responds by sending notifications or executing automated measures.

[0168] Example 2

[0169] like Figure 2 The figure shows a schematic diagram of a wind farm equipment health assessment system module based on multi-source data fusion and adaptive dynamic modeling provided by the present application. The present application provides a wind farm equipment health assessment system based on multi-source data fusion and adaptive dynamic modeling, which is applied to the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling as described in Example 1, and includes: a data acquisition module 11, a data preprocessing module 12, a data fusion module 13, a data processing module 14, a model training module 15, a model evaluation module 16, and an anomaly detection module 17.

[0170] Specifically, in this embodiment, the data acquisition module 11 is used to collect operation and monitoring data with timestamps from different devices, including photovoltaic modules, electrical equipment, detection equipment, meteorological sensors, video surveillance equipment and drones.

[0171] Specifically, in this embodiment, the data preprocessing module 12 is used to preprocess the data and perform time series data processing.

[0172] Specifically, in this embodiment, the data fusion module 13 is used to fuse data from different sources using a timestamp as an index and adopt a data fusion technology to generate a series of wind farm equipment comprehensive health index data.

[0173] Specifically, in this embodiment, the data processing module 14 is used to intercept the sequence data within a fixed time period p, use PCA to reduce the dimension of the sequence data, extract the main components, reduce the feature dimensions, and divide it into a training set, a validation set, and a test set according to a preset ratio.

[0174] Specifically, in this embodiment, the model training module 15 is used to construct the extracted main components into a dynamic LSTM sequence model according to the time step and train it; verify the time series data of the validation set through the dynamic LSTM sequence model, evaluate the model performance, and adjust the hyperparameters; and finally evaluate the generalization ability of the model on the test set.

[0175] Specifically, in this embodiment, the model evaluation module 16 is used to evaluate the dynamic LSTM sequence model through the mean square error evaluation index. When the threshold index is not met, the sliding step size q is calculated according to the deviation index, p is set to p+q, and the process returns to S4.

[0176] Specifically, in this embodiment, the anomaly detection module 17 is used to load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an abnormal situation, and output a prediction result.

[0177] Figure 3 This is an electronic device provided by an embodiment of the present application. Figure 3 As shown, the electronic device includes at least the following parts: a processor 101 and a memory 100 , a communication interface 103 , and a bus 102 .

[0178] In the embodiment of the present application, the memory 100 is used to store instructions executable by the processor 101. The processor 101 is configured to execute the instructions to implement the following Figure 2 The figure shows a wind farm equipment health assessment system module based on multi-source data fusion and adaptive dynamic modeling.

[0179] In an embodiment of the present application, a computer-readable storage medium includes instructions, and the instructions instruct a device to execute the method of the first aspect. For example, the instructions instruct the device to execute Figure 1 The process steps shown in the figure are a wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling.

[0180] The program running in the electronic device involved in one embodiment of the present application can be a program that controls a central processing unit (CPU) and the like to realize the functions of the above-mentioned embodiment involved in one embodiment of the present invention (a program that enables a computer to function). Then, the information processed by these devices is temporarily stored in a random access memory (RAM) during its processing, and then stored in various ROMs such as read-only memory (Flash ROM) and hard disk drive (HDD), and is read, modified, and written by the CPU as needed.

[0181] It should be noted that a portion of the electronic device of the above embodiment may also be implemented by a computer. In this case, a program for implementing the control function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer and executed.

[0182] It should be noted that the "computer" mentioned here refers to a computer built into an electronic device, employing hardware including an operating system (OS) and peripheral devices. Furthermore, "computer-readable recording medium" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computers.

[0183] Furthermore, "computer-readable recording media" may include: media that dynamically store programs for a short period of time, such as communication lines when transmitting programs via networks such as the Internet or communication lines such as telephone lines; and media that store programs for a fixed period of time, such as volatile memory within computers acting as servers or clients in this context. Furthermore, the aforementioned program may be a program for implementing a portion of the aforementioned functions, or a program that can achieve the aforementioned functions by combining with a program already stored in a computer.

[0184] Furthermore, the electronic device in the above-described embodiments can also be implemented as a collection (device group) consisting of multiple devices. Each device constituting the device group may have a portion or all of the functions or functional blocks of the electronic device in the above-described embodiments. A device group only needs to have all the functions or functional blocks of the electronic device.

[0185] Those skilled in the art should recognize that the above embodiments are merely intended to illustrate the present application and are not intended to limit the present application. As long as they are within the spirit of the present application, appropriate changes and modifications to the above embodiments are within the scope of protection claimed in the present application.

Claims

1. A wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling, characterized by: The method comprises: S1: Collects time-stamped operation and monitoring data from various devices, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones; S2: Preprocess the data and perform time series data processing; S3: Using timestamps as indexes, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment; S4: intercepting sequence data within a fixed time period p, using PCA to reduce the dimension of the sequence data, extracting the main components, reducing the feature dimensions, and dividing the sequence data into a training set, a validation set, and a test set according to a preset ratio; S5: Construct the extracted main components into a dynamic LSTM sequence model according to the time step and train it; use the dynamic LSTM sequence model to verify the time series data of the validation set, evaluate the model performance, and adjust the hyperparameters; finally evaluate the generalization ability of the model on the test set; S6: Evaluate the dynamic LSTM sequence model using the mean square error evaluation index. If the threshold index is not met, calculate the sliding step size q based on the deviation index, set p=p+q, and return to S4. S7: Otherwise, load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an anomaly, and output a prediction result; S6: Evaluate the dynamic LSTM sequence model using a mean square error evaluation index. When the threshold index is not met, calculate the sliding step size q based on the deviation index, including: The LSTM model is evaluated by the mean square error evaluation index, where the error evaluation formula is: : ， in, represents the true value, represents the predicted value, represents the number of samples, represents the error coefficient, and when , ,when , ; The sliding step length q is calculated based on the deviation index. The formula is: ； in, Indicates the threshold, [] indicates the rounding operation, is the adjustment coefficient, is the deviation index.

2. A wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to claim 1, characterized in that: S2: pre-processing the data and performing time series data processing, the steps include: S2.1: Data collection and integration: clarify the data source, store data from different sources in a unified manner, ensure consistent format, handle missing values, duplicate values, and outliers, and ensure data integrity; S2.2: Data format standardization: unify the formats of different data sources, convert data into types suitable for analysis, and encode categorical data; S2.3: Data alignment and matching: aligning timestamps of different time series data, aligning coordinate systems of geospatial data, and matching entities in different data sources; S2.4: Feature engineering: extract useful features from raw data, select the features that are most helpful for model prediction, reduce redundancy, and standardize or normalize numerical features; S2.5: Data quality assessment, data consistency check, ensure data consistency across different data sources, completeness check: check data completeness, handle missing values, verify data accuracy, and correct erroneous data; S2.6: Sort different time series data by timestamp.

3. A wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to claim 1, characterized in that: S3: Using timestamps as indexes, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment, including: Use convolutional neural network (CNN) to extract image data features, and recurrent neural network (RNN) to extract time series data features; Use feature splicing to fuse data from different sources; The fused data is used to train a deep learning model to generate series data of comprehensive health indicators of wind farm equipment.

4. The wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to claim 1 is characterized in that: S4: intercepting sequence data within a fixed time period p, using PCA to reduce the dimension of the sequence data, extracting the main components, reducing the feature dimensions, and dividing the sequence data into a training set, a validation set, and a test set according to a preset ratio, including: Determine the starting and ending points of time period p; Using a sliding window method to intercept the sequence data between the starting point and the end point; The intercepted time series data is converted into a matrix form suitable for PCA, with each row as a sample and each column as a feature; Perform Z-score standardization on matrix data; The PCA algorithm was applied to reduce the dimensionality of the truncated sequence data, and the ElbowPlot method was used to select the optimal number of the first k principal components.

5. A wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to claim 4, characterized in that: The architecture of the dynamic LSTM sequence model includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer; Wherein, the input layer receives and preprocesses input time series data; The LSTM layer extracts features of sequence data and generates a hidden state sequence; The attention mechanism layer assigns weights to the hidden states of each time step, generates weighted representations, and enhances the model's attention to important time steps; The fully connected layer selects an activation function according to the task type and maps the context vector to the target space; The output layer generates the final prediction result.

6. A wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to claim 5, characterized in that: The loss function of the dynamic LSTM sequence model is: ， Where N is the number of samples, is the one-hot encoding of the true label, is the predicted probability that the i-th sample belongs to category C, where C is the number of categories. If the sample belongs to category C, then =1, otherwise 0, is the attention weight of the i-th input at time step t, and λ is the regularization coefficient.

7. A wind farm equipment health assessment system based on multi-source data fusion and adaptive dynamic modeling, applied to the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to any one of claims 1 to 6, characterized in that: The system comprises: Data acquisition module: collects time-stamped operation and monitoring data from various devices, including photovoltaic panels, electrical equipment, testing equipment, meteorological sensors, video surveillance equipment, and drones; Data preprocessing module: preprocess the data and perform time series data processing; Data fusion module: Using timestamps as indexes, data fusion technology is used to fuse data from different sources to generate series data of comprehensive health indicators of wind farm equipment; Data processing module: intercepts sequence data within a fixed time period p, uses PCA to reduce the dimension of the sequence data, extracts the main components, reduces the feature dimensions, and divides the data into training set, validation set and test set according to the preset ratio; Model training module: The extracted main components are constructed into a dynamic LSTM sequence model according to the time step and trained; the dynamic LSTM sequence model is validated on the time series data of the validation set to evaluate the model performance and adjust the hyperparameters; the generalization ability of the model is finally evaluated on the test set; Model evaluation module: evaluates the dynamic LSTM sequence model using the mean square error evaluation index. When the threshold index is not met, the sliding step size q is calculated based on the deviation index, p=p+q is set, and the module returns to S4. Anomaly detection module: otherwise, load the dynamic LSTM sequence model, perform anomaly detection on the time series data to be detected through the dynamic LSTM sequence model, analyze whether there is an anomaly, and output the prediction result; S6: Evaluate the dynamic LSTM sequence model using a mean square error evaluation index. When the threshold index is not met, calculate the sliding step size q based on the deviation index, including: The LSTM model is evaluated by the mean square error evaluation index, where the error evaluation formula is: : ， in, represents the true value, represents the predicted value, represents the number of samples, represents the error coefficient, and when , ,when , ; The sliding step length q is calculated based on the deviation index. The formula is: ； in, Indicates the threshold, [] indicates the rounding operation, is the adjustment coefficient, is the deviation index.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to implement the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling as described in any one of claims 1 to 6 when executing the instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, and the program instructs a device to execute the wind farm equipment health assessment method based on multi-source data fusion and adaptive dynamic modeling according to any one of claims 1 to 6.

Images