Electrical equipment life prediction system and method based on deep learning

Abstract

The invention discloses an electrical equipment life prediction system and method based on deep learning, and belongs to the field of electrical equipment life prediction, and the method comprises the steps: constructing a data set, and carrying out the preprocessing; performing convolution operation on the time sequence data in the data set, extracting local features to obtain a convolutional neural network, capturing a long and short term dependency relationship of the time sequence data in the data set by using LSTM to obtain a recurrent neural network, and fusing output features of the convolutional neural network and the recurrent neural network to obtain comprehensive feature representation; constructing a deep learning model based on the fusion of a convolutional neural network and a recurrent neural network, adding an attention mechanism, adopting a mean square error as a loss function, optimizing model parameters, carrying out model training by utilizing a training data set, evaluating model performance by adopting a verification set, and adjusting hyper-parameters; and inputting real-time data into the trained model, predicting the residual life of the electrical equipment, and designing an online learning mechanism. According to the invention, the accuracy and real-time performance of prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of electrical equipment life prediction, and more specifically, to a system and method for electrical equipment life prediction based on deep learning. Background Art

[0002] As a key component of the energy supply sector, the power industry is a top priority in my country's future energy strategy. The construction of a new power system based on green and low-carbon transformation is a necessary support for achieving my country's carbon peak and carbon neutrality goals. Renewable energy generation technologies such as wind power generation, photovoltaic power generation, and hydropower generation have made great contributions to carbon emission reduction, and renewable energy generation technologies are also moving towards the goal of precision and technology. Some power generation sites have special site conditions. For example, offshore wind power generation has the particularity of offshore conditions, such as salt spray corrosion, humidity, distance from land, and difficulty in equipment maintenance. Due to geological reasons, some hydropower stations will flood the soil during operation due to periodic water storage and release, causing the biomass in the soil to rot, and corrosive gases such as sulfur, nitrogen, and ammonia will gush out, resulting in frequent electrical equipment failures. These factors have put higher requirements on the protection of electrical equipment, so it is necessary to monitor the health status and predict the life of the running electrical equipment in a timely manner.

[0003] The inventor of the present application found that the life prediction of electrical equipment in the prior art is limited to considering the influencing factors under the general environment, and cannot take into account the influence of marine environmental conditions and some special corrosive environments. In addition, due to the harsh operating environment, data collection during the operation of electrical equipment is difficult, and the obtained data has a large error with the actual data, so the number of effective samples is very small. At present, deep learning models usually require a large amount of high-quality training data. However, the acquisition of electrical equipment life prediction data may be limited, especially the historical data with fault labels is less. In the process of collecting data, it will be affected by factors such as sensor failure and environmental interference, resulting in noise and outliers in the data. Healthy data in life prediction may account for the majority, while fault data is less, resulting in data imbalance, which will affect the training effect of the model. Deep learning models usually have complex network structures and require a lot of computing resources and time for training and reasoning. During the model training process, the selection and tuning of hyperparameters require a lot of experiments and experience, and the optimization process is complex and time-consuming. Deep learning models may perform well on training data, but perform poorly on unseen test data and are difficult to generalize. The model has poor robustness to abnormal data and emergencies, resulting in unstable prediction results. Electrical equipment life prediction requires real-time processing and prediction, but the complex calculations of deep learning models may cause delays and make it difficult to meet real-time requirements. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an electrical equipment life prediction system and method based on deep learning.

[0005] The object of the present invention is achieved through the following solutions:

[0006] The embodiment of the present invention provides a method for predicting the life of electrical equipment based on deep learning, including: collecting operating data and environmental data of electrical equipment through sensors, constructing a data set, processing missing values ​​and abnormal values ​​of the data set, and normalizing the data set; performing convolution operation on time series data in the data set, extracting local features to obtain a convolutional neural network, using LSTM to capture the long-term and short-term dependencies of the time series data in the data set to obtain a recurrent neural network, fusing the output features of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature representation; constructing a deep learning model based on the fusion of the convolutional neural network and the recurrent neural network, adding an attention mechanism, using mean square error as a loss function, optimizing model parameters, using a training data set to train the model, using a validation set to evaluate model performance, and adjusting hyperparameters; inputting real-time data into the trained model, predicting the remaining life of the electrical equipment, designing an online learning mechanism, and regularly updating the model parameters according to new data to ensure the real-time and accuracy of the model.

[0007] Furthermore, the operating data includes current, voltage, temperature and humidity; and the environmental data includes weather data and load data.

[0008] Furthermore, the processing of missing values ​​and outliers in the data set and normalizing the data set includes:

[0009] For time series data, linear interpolation is used to fill missing values; for numerical data, the mean or median is used to fill missing values; for categorical data, the mode is used to fill missing values; and the data is scaled to the range of [0,1] through minimum-maximum normalization and standardization.

[0010] Furthermore, the convolution operation is performed on the time series data in the data set to extract local features to obtain a convolutional neural network, including: setting multiple convolution kernels to extract local features in the time series, reducing the dimension of the convolution layer output through maximum pooling or average pooling operations, and retaining important features; flattening the multi-dimensional features output by the pooling layer into a one-dimensional vector for input into the fully connected layer, and outputting a value if the equipment life at a certain moment in the future is predicted.

[0011] Furthermore, the use of LSTM to capture the long-term and short-term dependencies of time series data in the data set to obtain a recurrent neural network includes: capturing the long-term and short-term dependencies of time series data through a long short-term memory network layer, setting the number of neurons in the output layer according to the prediction target; if the equipment life at a certain moment in the future is predicted, a value is output.

[0012] Furthermore, the output features of the convolutional neural network and the recurrent neural network are fused to obtain a comprehensive feature representation, including: splicing the flattened layer outputs of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature vector containing the features of both, defining the fused network structure, the fully connected layer contains a number of neurons, further processing and combining the fused feature vectors, and the output layer sets the number of neurons in the output layer according to the prediction target; if the equipment life is predicted at a certain moment in the future, a value is output.

[0013] Furthermore, the construction is based on a deep learning model that combines convolutional neural networks and recurrent neural networks, adding an attention mechanism, including:

[0014] Through the attention mechanism, the attention weight of each time step is calculated to enhance the data features of important time steps. The expression is as follows:

[0015]

[0016] Among them, e t represents the attention score, We represents the attention weight matrix, and be represents the bias term;

[0017] And normalize the attention weights:

[0018]

[0019] Among them, α t represents the normalized attention weight;

[0020] Then calculate the weighted sum of the output features of the recurrent neural network as the output of the attention mechanism, concatenate the flattened layer outputs of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature vector containing the features of both, and define the fully connected layer of the fused network structure:

[0021]

[0022] in, represents the output of the fully connected layer, W fc represents the weight matrix of the fully connected layer, b fc represents the bias term;

[0023] Output layer:

[0024]

[0025] Among them, Wout represents the weight matrix of the output layer, and bout represents the bias term of the output layer.

[0026] Furthermore, the method adopts mean square error as loss function, optimizes model parameters, uses training data set to train the model, uses validation set to evaluate model performance, and adjusts hyperparameters, including: calculating mean square error between predicted value and true value as loss function, setting number of training rounds and number of training samples in each round, iteratively training the model on the training set, and in each epoch, the optimizer adjusts model parameters according to back propagation algorithm to minimize training loss; at the end of each round of training, records training loss and validation loss to ensure that the model is not overfitted or underfitted during training; and visualizes the loss value of each round of training and validation through graphs to observe the training status of the model.

[0027] Furthermore, the method of inputting real-time data into a trained model, predicting the remaining life of electrical equipment, designing an online learning mechanism, and regularly updating model parameters according to new data to ensure the real-time and accuracy of the model includes: organizing the preprocessed data according to the model input format, and if the model requires fixed-size time window data, dividing the real-time data according to the window size; inputting the preprocessed real-time data into a trained model, calling the prediction function, and the model outputting the predicted value of the remaining life of the electrical equipment; regularly collecting new data and using the data to retrain the entire model; directly fine-tuning the existing model when new data arrives; acquiring data from sensors or monitoring devices in real time, caching new data every hour to form a batch data set; and triggering the model update mechanism after the data cache reaches a set size or time.

[0028] The beneficial effects of the present invention include:

[0029] The present invention can effectively process complex nonlinear relationships and multi-dimensional data and improve the accuracy of life prediction by utilizing a deep learning model. The local features of the data are extracted by a convolutional neural network (CNN) and the long short-term memory network (LSTM) is used to capture the long-term and short-term dependencies of the time series data, making full use of the information in the data. By making predictions through real-time input data, the operating status and remaining life of the equipment can be understood in a timely manner, and scientific maintenance and overhaul plans can be made to avoid sudden equipment failures. Based on the prediction results, the use and maintenance strategies of the equipment can be optimized, management efficiency can be improved, and unnecessary maintenance and downtime can be reduced. The failure time of the equipment can be predicted in advance, and preventive maintenance can be performed to avoid serious damage caused by equipment failure and extend the service life of the equipment. By introducing an online learning mechanism, the model can be continuously updated according to new data, adapt to changes in the environment and equipment status, and maintain the accuracy and real-time nature of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 A flowchart of a method for predicting the life of electrical equipment based on deep learning is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0033] like Figure 1 As shown in FIG. 1 , an embodiment of the present invention is provided, which provides a method for predicting the life of electrical equipment based on deep learning, including:

[0034] S1: Collect the operating data and environmental data of electrical equipment through sensors, build a data set, process missing values ​​and outliers in the data set, and normalize the data set.

[0035] Preferably, the operating data includes current, voltage, temperature and humidity; and the environmental data includes weather data and load data.

[0036] Preferably, for time series data, linear interpolation is used to fill missing values; for numerical data, the mean or median is used for filling; for categorical data, the mode is used for filling; and the data is scaled to the range of [0,1] through minimum-maximum normalization and standardization.

[0037] S2: Perform convolution operation on the time series data in the dataset, extract local features to obtain a convolutional neural network, use LSTM to capture the long-term and short-term dependencies of the time series data in the dataset to obtain a recurrent neural network, and fuse the output features of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature representation.

[0038] Preferably, multiple convolution kernels are set to extract local features in the time series, and the dimension of the convolution layer output is reduced through maximum pooling or average pooling operations, while retaining important features; the multi-dimensional features output by the pooling layer are flattened into a one-dimensional vector for input into the fully connected layer, and a value is output if the equipment life at a certain moment in the future is predicted.

[0039] Furthermore, the input layer accepts data of shape (10,4), each window contains 10 time points, and each time point has 4 features (current, voltage, temperature, vibration).

[0040] Convolution layer: number of convolution kernels: 32;

[0041] Convolution kernel size: 3;

[0042] Activation function: ReLU

[0043] Convolution operation:

[0044] Among them, Y is the convolution output, X is the input matrix, K is the convolution kernel matrix, and H and W are the height and width of the convolution kernel respectively.

[0045] Preferably, the long-term and short-term dependencies of the time series data are captured through the long short-term memory network layer, and the number of neurons in the output layer is set according to the prediction target; if the equipment life at a certain moment in the future is predicted, a value is output.

[0046] Furthermore, the input layer accepts data of shape (10,4), each window contains 10 time points, and each time point has 4 features (current, voltage, temperature, vibration). The recurrent layer uses a long short-term memory network (LSTM) layer to capture the long-term and short-term dependencies of time series data. Assuming that the number of hidden units in the LSTM layer is 50, the formula of the LSTM unit is as follows:

[0047] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0048] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0049]

[0050] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0051] h t =o t *tanh(C t )

[0052] Among them, ft is the forget gate, it is the input gate, C~t is the candidate memory, Ct is the cell state, ot is the output gate, ht is the hidden state, σ is the sigmoid activation function, and tanh is the tanh activation function.

[0053] Preferably, the flattened layer outputs of the convolutional neural network and the recurrent neural network are concatenated to obtain a comprehensive feature vector containing the features of both, and the fused network structure is defined. The fully connected layer contains a number of neurons, and the fused feature vectors are further processed and combined. The output layer sets the number of neurons in the output layer according to the prediction target; if the equipment life is predicted at a certain moment in the future, a value is output.

[0054] Furthermore, the output features of the convolutional neural network (CNN) and the recurrent neural network (RNN) are fused. Fully connected layer: The fused features are input into the fully connected layer for processing. The number of nodes in the fully connected layer is 128, and the ReLU activation function is used:

[0055] Y=W·X+b

[0056] Among them, W is the output layer weight matrix, b is the output layer bias term, Xo\mathbf{X}_oXo is the input vector, and Y^ is the predicted value.

[0057] The mean square error (MSE) is used as the loss function:

[0058]

[0059] Optimize model parameters using the gradient descent optimization algorithm:

[0060]

[0061] The data set is divided into a training set and a validation set, the model is trained on the training set, and the model parameters are updated through the back-propagation algorithm.

[0062] S3: Build a deep learning model based on the fusion of convolutional neural network and recurrent neural network, add attention mechanism, use mean square error as loss function, optimize model parameters, use training data set for model training, use validation set to evaluate model performance, and adjust hyperparameters.

[0063] Preferably, the attention weight of each time step is calculated through the attention mechanism to enhance the data features of important time steps. The expression is as follows:

[0064]

[0065] Among them, e t represents the attention score, We represents the attention weight matrix, and be represents the bias term;

[0066] And normalize the attention weights:

[0067]

[0068] Among them, α t represents the normalized attention weight;

[0069] Then calculate the weighted sum of the output features of the recurrent neural network as the output of the attention mechanism, concatenate the flattened layer outputs of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature vector containing the features of both, and define the fully connected layer of the fused network structure:

[0070]

[0071] in, represents the output of the fully connected layer, W fc represents the weight matrix of the fully connected layer, b fc represents the bias term;

[0072] Output layer:

[0073]

[0074] Among them, Wout represents the weight matrix of the output layer, and bout represents the bias term of the output layer.

[0075] Furthermore, the mean square error between the predicted value and the true value is calculated as the loss function, the number of training rounds and the number of training samples in each round are set, and the model is iteratively trained on the training set. In each epoch, the optimizer adjusts the model parameters according to the back-propagation algorithm to minimize the training loss. At the end of each round of training, the training loss and validation loss are recorded to ensure that the model is not overfitted or underfitted during the training process. The loss values ​​of each round of training and validation are visualized through charts to observe the training status of the model.

[0076] S4, inputs real-time data into the trained model, predicts the remaining life of electrical equipment, designs an online learning mechanism, and regularly updates the model parameters based on new data to ensure the real-time and accuracy of the model.

[0077] Preferably, the preprocessed data is organized according to the model input format. If the model requires fixed-size time window data, the real-time data is divided according to the window size; the preprocessed real-time data is input into the trained model, the prediction function is called, and the model outputs the remaining life prediction value of the electrical equipment; new data is collected regularly and used to retrain the entire model; when new data arrives, the existing model is directly fine-tuned; data is acquired from sensors or monitoring devices in real time, and new data is cached every hour to form a batch data set; after the data cache reaches a set size or time, the model update mechanism is triggered.

[0078] Furthermore, given the following time series data, the time series length is 10 (each window contains 10 time points):

[0079] Table 1

[0080]

[0081] Input layer: shape (10,4), convolution layer: 32 3x3 convolution kernels, ReLU activation function, pooling layer: maximum pooling, pooling window size is 2, LSTM layer: 50 hidden units, weighted summation, get important feature representation; hidden layer: 128 nodes, ReLU activation function, output layer: 1 node, linear activation function, training set and validation set division, training set: 80% data, validation set: 20% data; The data set is divided into training set and validation set in a ratio of 80:20, in the training The model is trained on the training set, and the model parameters are updated through the back-propagation algorithm. The learning rate is set to 0.001. The model performance is evaluated on the validation set, and the mean square error (MSE) is calculated. The hyperparameters such as learning rate, number of convolution kernels, and number of LSTM hidden units are adjusted according to the validation set results. Multiple iterations of training and evaluation are performed until the model performance reaches a satisfactory level. It is assumed that after the trained model is input into the data window, the predicted remaining life of the equipment is 0.78 (normalized value), which is the actual value of 78 hours after denormalization.

[0082] The present invention can timely discover potential quality problems and defects by accurately classifying the reliability of wafer chips, thereby improving product design and production processes in a targeted manner and improving product quality and stability. By reducing production losses and defective product rates caused by wafer chip failures, production costs can be reduced, and production efficiency and profit margins can be improved. By analyzing and processing wafer test data, production processes and process control can be optimized, the stability and reliability of production lines can be improved, and waste and changes in production can be reduced.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0084] According to one aspect of an embodiment of the present invention, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in the above various optional implementations.

[0085] As another aspect, an embodiment of the present invention further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment.

[0086] In addition to the above examples, those skilled in the art may obtain other embodiments based on the above disclosure or by using the knowledge or technology in the related field to make changes. The features of each embodiment may be interchangeable or replaced. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention and should be within the scope of protection of the claims attached to the present invention.

Claims

1. A method for predicting the life of electrical equipment based on deep learning, characterized in that: The following steps are involved: Collect the operating data and environmental data of electrical equipment through sensors, build data sets, process missing values ​​and outliers in the data sets, and normalize the data sets; Perform convolution operation on the time series data in the data set, extract local features to obtain a convolutional neural network, use LSTM to capture the long-term and short-term dependencies of the time series data in the data set to obtain a recurrent neural network, and fuse the output features of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature representation; Build a deep learning model based on the fusion of convolutional neural network and recurrent neural network, add attention mechanism, use mean square error as loss function, optimize model parameters, use training data set to train the model, use validation set to evaluate model performance, and adjust hyperparameters; Input real-time data into the trained model to predict the remaining life of electrical equipment, design an online learning mechanism, and regularly update the model parameters based on new data to ensure the real-time and accuracy of the model.

2. The electrical equipment life prediction system and method based on deep learning according to claim 1 is characterized in that: The operating data includes current, voltage, temperature and humidity; the environmental data includes weather data and load data.

3. The electrical equipment life prediction system and method based on deep learning according to claim 1 is characterized in that: The processing of missing values ​​and outliers of the data set and normalization of the data set includes: For time series data, linear interpolation is used to fill missing values; for numerical data, the mean or median is used to fill missing values; for categorical data, the mode is used to fill missing values; and the data is scaled to the range of [0,1] through minimum-maximum normalization and standardization.

4. The electrical equipment life prediction system and method based on deep learning according to claim 1, characterized in that: The convolution operation is performed on the time series data in the data set to extract local features to obtain a convolutional neural network, including: Set multiple convolution kernels to extract local features in the time series. Reduce the dimension of the convolution layer output through maximum pooling or average pooling operations and retain important features. Flatten the multi-dimensional features output by the pooling layer into a one-dimensional vector for input into the fully connected layer. If the equipment life is predicted at a certain moment in the future, output a value.

5. The electrical equipment life prediction system and method based on deep learning according to claim 3 is characterized in that: The LSTM is used to capture the long-term and short-term dependencies of the time series data in the data set to obtain a recurrent neural network, including: The long-term and short-term dependencies of time series data are captured through the long short-term memory network layer. The number of neurons in the output layer is set according to the prediction target. If the equipment life at a certain moment in the future is predicted, a value is output.

6. The electrical equipment life prediction system and method based on deep learning according to claim 1, characterized in that: The output features of the convolutional neural network and the recurrent neural network are fused to obtain a comprehensive feature representation, including: The flattened layer outputs of the convolutional neural network and the recurrent neural network are concatenated to obtain a comprehensive feature vector containing the features of both. The fused network structure is defined. The fully connected layer contains several neurons to further process and combine the fused feature vectors. The output layer sets the number of neurons in the output layer according to the prediction target. If the equipment life is predicted at a certain moment in the future, a value is output.

7. The electrical equipment life prediction system and method based on deep learning according to claim 1, characterized in that: The construction is based on a deep learning model that integrates convolutional neural networks and recurrent neural networks, adding an attention mechanism, including: Through the attention mechanism, the attention weight of each time step is calculated to enhance the data features of important time steps. The expression is as follows: Among them, et represents the attention score, We represents the attention weight matrix, and be represents the bias term; And normalize the attention weights: Among them, αt represents the normalized attention weight; Then calculate the weighted sum of the output features of the recurrent neural network as the output of the attention mechanism, concatenate the flattened layer outputs of the convolutional neural network and the recurrent neural network to obtain a comprehensive feature vector containing the features of both, and define the fully connected layer of the fused network structure: in, represents the output of the fully connected layer, Wfc represents the weight matrix of the fully connected layer, and bfc represents the bias term; Output layer: Among them, Wout represents the weight matrix of the output layer, and bout represents the bias term of the output layer.

8. The electrical equipment life prediction system and method based on deep learning according to claim 1, characterized in that: The method uses mean square error as the loss function, optimizes model parameters, uses the training data set to train the model, uses the validation set to evaluate model performance, and adjusts hyperparameters, including: Calculate the mean square error between the predicted value and the true value as the loss function, set the number of training rounds and the number of training samples in each round, iteratively train the model on the training set, and in each epoch, the optimizer adjusts the model parameters according to the back-propagation algorithm to minimize the training loss; at the end of each round of training, record the training loss and validation loss to ensure that the model is not overfitting or underfitting during the training process; visualize the loss values ​​of each round of training and validation through charts to observe the training status of the model.

9. The electrical equipment life prediction system and method based on deep learning according to claim 1, characterized in that: The real-time data is input into the trained model to predict the remaining life of the electrical equipment, design an online learning mechanism, and regularly update the model parameters according to the new data to ensure the real-time and accuracy of the model, including: The preprocessed data is organized according to the model input format. If the model requires fixed-size time window data, the real-time data is split according to the window size. The preprocessed real-time data is input into the trained model, the prediction function is called, and the model outputs the remaining life prediction value of the electrical equipment. New data is collected regularly and used to retrain the entire model. When new data arrives, the existing model is directly fine-tuned. Data is acquired from sensors or monitoring devices in real time, and new data is cached every hour to form a batch data set. When the data cache reaches the set size or time, the model update mechanism is triggered.

10. A system for predicting the life of electrical equipment based on deep learning, comprising a computer device, characterized in that: A computer program is stored in the memory of the computer device, and when the computer program is loaded by a processor in the computer device, the steps of the method according to any one of claims 1 to 9 are executed.

Images