New energy boost transformer abnormity early warning system and method

Abstract

The invention discloses a new energy boost transformer abnormity early warning system and method, and the system comprises a data collection module which employs a sensor technology, a data collection card carries out the collection of the data of a new energy boost transformer, and carries out the early warning of the new energy boost transformer. The sensors comprise a temperature sensor, a voltage sensor, a current sensor, a vibration sensor and an oil level sensor. Feature vectors are selected by adopting a wavelet transform algorithm, a boost variation anomaly positioning model is established by adopting an improved artificial neural network to determine an anomaly position, parameters of an anomaly detection model are determined by adopting a genetic algorithm, and an anomaly type is judged. The MFE-TS model is established to optimize the objective function so as to improve the accuracy of new energy boost variation abnormity identification, the man-machine interaction module is adopted to display the abnormal position and the abnormal type on the user interface, and the new energy boost variation abnormity early warning capability is improved.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and more specifically to a new energy boost transformer abnormality warning system and method. Background Art

[0002] A new energy step-up transformer refers to a transformer used to increase the voltage in a new energy power generation system. New energy power generation systems usually include wind power generation, solar power generation, etc. The voltage of the electricity generated by these power generation methods is relatively low and cannot be directly connected to the power grid or meet the needs of long-distance transmission. Therefore, a step-up transformer is needed to increase the voltage to a level suitable for grid connection or transmission. The working principle of the step-up transformer is to use electromagnetic induction to increase the voltage by changing the turns ratio of the primary coil and the secondary coil. In a new energy power generation system, a step-up transformer is usually installed between the generator and the power grid to ensure that the electricity can be transmitted efficiently and safely. Due to the intermittent and volatile nature of new energy power generation, the step-up transformer also needs to have a certain degree of adaptability and stability to cope with changes in power output. The new energy step-up transformer is an indispensable key equipment in a new energy power generation system. It ensures that the electricity can be effectively transmitted and utilized by increasing the voltage, thereby promoting the widespread application of new energy.

[0003] In the prior art, during the abnormality detection of the new energy boost transformer, due to the complexity and variability of the power system operating environment, the early warning system will frequently give false alarms, resulting in a high false alarm rate of the early warning system, and at the same time will lead to a decrease in the trust in the early warning system. On the other hand, the lack of abnormal positioning function means that the abnormal location of the power generation boost transformer cannot be timely determined, resulting in low efficiency of the boost transformer abnormality warning. The lack of human-computer interaction function to display the abnormal results and abnormal location of the boost transformer cannot accurately and clearly display the abnormal warning results of the boost transformer. Therefore, we propose a new energy boost transformer abnormality warning system and method to solve the above problems. Summary of the invention

[0004] In view of the deficiencies of the above-mentioned technologies, the present invention discloses a new energy boost transformer abnormality warning system and method, which selects feature vectors by adopting a wavelet transform algorithm, establishes a boost transformer abnormality positioning model by adopting an improved artificial neural network to determine the abnormal position, determines the parameters of the abnormality detection model by adopting a genetic algorithm, judges the abnormality type, establishes an MFE-TS model to optimize the objective function to improve the accuracy of new energy boost transformer abnormality identification, and adopts a human-computer interaction module to enable the abnormal position and abnormality type to be displayed on the user interface, so that the user can observe the abnormal situation more intuitively.

[0005] A new energy boost transformer abnormality warning system comprises the following steps: It is characterized in that: the boost transformer abnormality warning system comprises: The data acquisition module uses sensor technology and a data acquisition card to collect data from the new energy boost transformer. The sensors include a temperature sensor, a voltage sensor, a current sensor, a vibration sensor, and an oil level sensor. The data acquisition card converts the analog signal output by the sensor into a digital signal to ensure that the hardware device is compatible with the computer system. The data preprocessing module uses data cleaning tools to remove outliers, fill missing values, and correct deviations from the collected new energy boost transformer parameters; Alarm module, using sound and light alarm to warn of abnormalities in new energy boost transformers; An anomaly detection module, the anomaly detection module includes a feature extraction unit, a feature analysis modeling unit and an anomaly calculation unit, the output end of the feature extraction unit is connected to the input end of the feature analysis modeling unit, and the output end of the feature analysis modeling unit is connected to the input end of the anomaly calculation unit; An abnormality locating module, using an abnormality locator to analyze the abnormal data of the new energy boost transformer to determine the abnormal location, the abnormality locator includes an abnormal state identification unit and an abnormal location determination unit, the output end of the abnormal state identification unit is connected to the input end of the abnormal location determination unit; A human-computer interaction module, using a graphical user interface to display the abnormality judgment result and abnormal position of the new energy boost transformer, the graphical user interface includes a real-time update unit, an abnormality judgment result display unit and an abnormal position identification unit, and the output end of the real-time update unit is respectively connected to the input end of the abnormality judgment result display unit and the input end of the abnormal position identification unit; The output end of the data acquisition module is connected to the input end of the data preprocessing module, the output end of the data preprocessing module is connected to the input end of the abnormality calculation unit, the output end of the abnormality calculation unit is connected to the input end of the abnormality positioning module and the input end of the alarm module, and the output end of the abnormality calculation unit and the output end of the abnormality positioning module are connected to the input end of the human-computer interaction module.

[0006] Furthermore, the feature extraction unit includes feature selection, feature learning and feature fusion. The feature selection uses the LASSO feature selection algorithm to select the most relevant features from all features to reduce the feature dimension. The feature learning uses an autoencoder to automatically learn the feature representation of data to improve the level of abstract features. The feature fusion uses feature splicing technology to fuse different types of features to form a comprehensive feature to improve the accuracy and robustness of anomaly detection. Furthermore, the working method of the feature analysis modeling unit is: Step 1: Use wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector of the boost transformer operation state. The expression of the abnormal feature vector is: In formula (1), W represents the abnormal feature vector, Expressed as the winding vibration frequency value of the step-up transformer, It is the subscript of the vibration frequency value of the step-up transformer winding. It is expressed as the average value of the vibration frequency of the step-up transformer winding, and n is expressed as the number of vibration frequencies of the step-up transformer winding. is represented as an abnormal feature value, Expressed as the step-up transformer winding vibration frequency capability, It is expressed as the efficiency of step-up to step-down; Step 2: Based on the determined abnormal feature vector and the improved artificial neural network, a boost transformer abnormal location model is established. The improved artificial neural network designs the network structure of the abnormal detection model through the neuron activation function, and then trains the abnormal detection model through the back propagation algorithm. The back propagation algorithm is trained according to the abnormal feature vector to obtain the abnormal detection model; Step 3: Finally, a genetic algorithm is used to determine the anomaly detection model parameters. The anomaly detection model parameters include the scaling factor, bias and optimal fitness value of the anomaly detection model. The genetic algorithm determines the scaling factor and bias of the anomaly detection model by population initialization, performs initialization processing on the anomaly detection model, and then obtains the optimal fitness value of the anomaly detection model through the fitness function. The expression of the optimal fitness value of the anomaly detection model is: In formula (2), Denotes the optimal fitness value of the anomaly detection model, Expressed as the recall rate of abnormal evaluation, is the target variable value represented by the fitness function, It is expressed as the harmonic mean of the fault detection model. Furthermore, the abnormality calculation unit integrates the operating equipment of the boost transformer by establishing the abnormality detection model of IndRNN-VAE, and the abnormality detection model judges the state of the boost transformer equipment as normal and abnormal through a binary classification method. The abnormality detection model then extracts the abnormal data of the boost transformer through an artificial ant colony search algorithm, and the artificial ant colony search algorithm represents the abnormal state matrix of the boost transformer equipment through a pheromone matrix, and then updates the pheromone matrix through an objective function, and finally obtains the abnormal data of the boost transformer through an iterative search.

[0007] Further, the abnormality identification unit uses a Gaussian mixture model algorithm to group the boost transformer abnormal data, the Gaussian mixture model algorithm models the boost transformer abnormal data distribution into a Gaussian distribution combination, and the Gaussian mixture model algorithm evaluates the fitting effect of the Gaussian distribution combination through the Bayesian information criterion to obtain a cluster group. The abnormal state identification unit also uses a support vector machine classification model to determine the abnormality type represented by the cluster group. The support vector machine classification model determines the abnormality type corresponding to the cluster group by analyzing the characteristics and attributes of the cluster group. The abnormality identification unit also uses the establishment of an MFE-TS model to optimize the objective function, thereby improving the accuracy of abnormality identification. The established MFE-TS model optimization objective function is: In formula (3), it is represented by the original parameter X of the boost transformer at the time point The observed value of are the weight parameters of the two reconstruction losses, L represents the capacity of the parameters, It is expressed as the number of parameters, thereby optimizing the abnormality identification and improving the accuracy of abnormality identification. Furthermore, the abnormal position determination unit adopts a random forest algorithm model to realize the determination of the abnormal position. The random forest algorithm model combines the key feature filtering mechanism with the decision tree, records the value range of the key feature in the tree node, uses the value range of the key feature to filter the abnormal points, and uses the random forest algorithm model to record the value range of the dual features in each node of the decision tree, which is used for filtering abnormal samples in the classification stage to determine the abnormal type and abnormal position, and obtain samples of abnormal combination. By adopting the establishment of a confidence function model, the confidence function model obtains the confidence. When the confidence is greater than or equal to the associated threshold, the abnormal position is determined. The confidence function calculation formula is: In formula (4), Expressed as confidence, Expressed as the probability of anomaly occurrence, The sample size expressed as the abnormal location and abnormal type, It is expressed as the frequency of occurrence of the abnormal type. The interval length is expressed as the confidence level. Furthermore, the alarm module uses a set threshold to determine whether the parameter exceeds the normal range. The alarm module receives information from the abnormal calculation unit, notifies relevant personnel through the change of light color, and uses color nodes and network topology diagrams to display the abnormal location.

[0008] Further, the method is applied to a new energy boost transformer abnormality warning system as described in any one of claims 1 to 7, and comprises: Step 1: Collect equipment parameters of the boost transformer through temperature sensors, voltage sensors, current sensors, vibration sensors and oil level sensors; Step 2: preprocessing the collected equipment parameters of the boost transformer, wherein the data preprocessing includes removing abnormal values, filling missing values ​​and correcting deviations; Step 3: The feature extraction unit uses the LASSO feature selection algorithm to select the most relevant features from all the features. The feature analysis modeling unit uses the wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector. The improved artificial neural network and genetic algorithm are used to establish an abnormality detection model and determine the parameters of the abnormality detection model. The abnormality calculation unit uses the IndRNN-VAE abnormality detection model to integrate the operating equipment of the boost transformer and then judge the abnormality of the boost transformer. Step 4: The abnormality identification unit groups the abnormal data of the boost transformer by using the Gaussian mixture model algorithm, and then evaluates the Gaussian distribution model into a Gaussian distribution combination through the Bayesian information criterion, thereby identifying the abnormal type, and using the support phase connection classification model to determine the abnormal location; Step 5. The sound and light alarm warns of abnormalities in the new energy boost transformer. The human-computer interaction module displays the abnormality judgment results through a graphical user interface. At the same time, the parameters of the boost transformer are updated through a real-time update unit, and the abnormal location is displayed according to the color node and network topology diagram.

[0009] Positive beneficial effects The wavelet transform algorithm is used to select feature vectors, the improved artificial neural network is used to establish the boost transformer abnormal location model to determine the abnormal position, the genetic algorithm is used to determine the parameters of the abnormal detection model, and the abnormal type is judged. The MFE-TS model is established to optimize the objective function to improve the accuracy of new energy boost transformer abnormality identification, and the human-computer interaction module is used to display the abnormal position and abnormal type on the user interface, so that users can observe the abnormal situation more intuitively. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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 prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 It is a schematic diagram of the overall architecture of a new energy boost transformer abnormality warning system of the present invention; Figure 2 This is a structural diagram of a feature extraction unit in a new energy boost transformer abnormality warning system of the present invention; Figure 3 A working method of a characteristic modeling and analysis unit in a new energy boost transformer abnormality warning system of the present invention; Figure 4 It is a diagram of the architecture of a graphical user interface in a new energy boost transformer abnormal warning system of the present invention; Figure 5 The present invention is a schematic flow chart of a new energy boost transformer abnormality warning method. DETAILED DESCRIPTION

[0011] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0012] like Figure 1-5 As shown, a new energy boost transformer abnormality warning system and method, the boost transformer abnormality warning system includes: The data acquisition module uses sensor technology and a data acquisition card to collect data from the new energy boost transformer. The sensors include a temperature sensor, a voltage sensor, a current sensor, a vibration sensor, and an oil level sensor. The data acquisition card converts the analog signal output by the sensor into a digital signal to ensure that the hardware device is compatible with the computer system. The data preprocessing module uses data cleaning tools to remove outliers, fill missing values, and correct deviations from the collected new energy boost transformer parameters; Alarm module, using sound and light alarm to warn of abnormalities in new energy boost transformers; An anomaly detection module, the anomaly detection module includes a feature extraction unit, a feature analysis modeling unit and an anomaly calculation unit, the output end of the feature extraction unit is connected to the input end of the feature analysis modeling unit, and the output end of the feature analysis modeling unit is connected to the input end of the anomaly calculation unit; An abnormality locating module, using an abnormality locator to analyze the abnormal data of the new energy boost transformer to determine the abnormal location, the abnormality locator includes an abnormal state identification unit and an abnormal location determination unit, the output end of the abnormal state identification unit is connected to the input end of the abnormal location determination unit; A human-computer interaction module, using a graphical user interface to display the abnormality judgment result and abnormal position of the new energy boost transformer, the graphical user interface includes a real-time update unit, an abnormality judgment result display unit and an abnormal position identification unit, and the output end of the real-time update unit is respectively connected to the input end of the abnormality judgment result display unit and the input end of the abnormal position identification unit; The output end of the data acquisition module is connected to the input end of the data preprocessing module, the output end of the data preprocessing module is connected to the input end of the abnormality calculation unit, the output end of the abnormality calculation unit is connected to the input end of the abnormality positioning module and the input end of the alarm module, and the output end of the abnormality calculation unit and the output end of the abnormality positioning module are connected to the input end of the human-computer interaction module. In the above specific embodiments, the temperature sensor, voltage sensor, current sensor, vibration sensor and oil level sensor collect various parameters of the new energy boost transformer in real time. The data collection also includes data collection software. The data collection software connects and communicates with the interface of each sensor to collect sensor data. When collecting data, a data collection protocol should be established. The data acquisition card is used to collect analog signals from sensors, instruments or other data sources and convert them into digital signals. The data acquisition card has analog signal input and multiple analog input channels. It can collect multiple analog signals at the same time and collect a large range of parameters of the boost transformer. The data acquisition card is internally provided with a filter and an isolator for preprocessing the analog signal. The method of preprocessing the collected data using the data cleaning tool is as follows: 1. Processing missing values, including filling missing values ​​by interpolation, thereby retaining data integrity and improving data quality; 2. Processing abnormal values, classifying and processing abnormal values, and then detecting abnormalities of the boost transformer; 3. Processing data deviation values, understanding the source of the deviation, analyzing the cause of the deviation, and then correcting the deviation. According to the characteristics of the source of the deviation, a correction formula or a correction system is used to correct the data.

[0013] The sound and light alarm includes a sound alarm and a light alarm. The sound and light alarm includes multiple speakers for emitting sound signals. The internal circuit of the sound and light alarm generates audio signals, which are preset alarm sounds. The audio signals can also be amplified by an amplifier circuit to drive the speakers to emit sounds loud enough to be heard in a larger range. The light source in the sound and light alarm is an LED lamp, a xenon lamp or other types of bulbs. The generation of light signals is controlled by circuits, which control the flashing frequency and brightness of the light source to attract attention. The sound and light alarm is connected to each sensor to detect specific conditions: the sound and light alarm usually contains one or more electromagnetic oscillators, which excite the oscillator plate in the oscillator through the change of current to make it vibrate and produce sound. When current passes through the oscillator, the magnetic field causes the oscillator plate to vibrate, thereby generating a higher frequency sound signal. As a further embodiment of the present invention, the feature extraction unit includes feature selection, feature learning and feature fusion. The feature selection adopts the LASSO feature selection algorithm to select the most relevant features from all features to reduce the feature dimension. The feature learning adopts an automatic encoder to automatically learn the feature representation of data to improve the level of abstract features. The feature fusion adopts feature splicing technology to fuse different types of features to form a comprehensive feature to improve the accuracy and robustness of anomaly detection.

[0014] It should be added that the feature extraction unit works in collaboration with edge computing and terminal devices to share the pressure of data processing, realize data preprocessing and feature extraction, and then aggregate the processed data to the central cloud for more in-depth analysis and processing. At the same time, the processing process also includes feature monitoring. The feature extraction unit can also provide feature monitoring and alarm functions to promptly discover and resolve feature anomalies and problems, reduce fault detection and repair, and improve the efficiency of feature extraction.

[0015] As a further embodiment of the present invention, the working method of the feature analysis modeling unit is: Step 1: Use wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector of the boost transformer operation state. The expression of the abnormal feature vector is: In formula (1), W represents the abnormal feature vector, Expressed as the winding vibration frequency value of the step-up transformer, It is the subscript of the vibration frequency value of the step-up transformer winding. It is expressed as the average value of the vibration frequency of the step-up transformer winding, and n is expressed as the number of vibration frequencies of the step-up transformer winding. is represented as an abnormal feature value, Expressed as the step-up transformer winding vibration frequency capability, It is expressed as the efficiency of the step-up transformer; in specific applications, by dividing by the standard deviation, the dimension differences between different vibration frequency values ​​are eliminated, and the standardized values ​​are used to identify those frequency values ​​that deviate significantly from the average value, which are signs of abnormal operating conditions. By multiplying by the efficiency (\eta), the effect of the transformer efficiency on the vibration frequency is taken into account, which helps to more accurately evaluate the operating status of the transformer.

[0016] The wavelet transform algorithm is used to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector of the boost transformer operation state. The operation data of the boost transformer is collected during the calculation, including but not limited to the winding vibration frequency, temperature, dissolved gas analysis in oil and other parameters. According to the operation characteristics and data characteristics of the boost transformer, a suitable wavelet basis is selected for transformation. Common wavelet bases include Haar, Daubechies, Symlet, Coiflet, etc. The collected data is subjected to multi-scale wavelet decomposition to decompose the signal into approximate coefficients (low-frequency components) and detail coefficients (high-frequency components). The energy of the decomposed approximate coefficients and detail coefficients is calculated respectively. Then, a feature vector is constructed, and the energy value of each layer of decomposition is used as the feature vector. Then, the average value and standard deviation of each energy in the feature vector are calculated based on the calculated average energy and standard deviation. Each energy value in the feature vector is standardized by calculating the standardized energy. Then, the abnormal threshold is determined, and the abnormal threshold is determined based on experience or statistical methods, and the identified abnormal energy value is constructed into an abnormal feature vector. The abnormal feature vector is used for subsequent analysis, such as clustering, classification or anomaly detection, to identify the abnormal operation state of the boost transformer. Step 2: Based on the determined abnormal feature vector and the improved artificial neural network, a boost transformer abnormal location model is established. The improved artificial neural network designs the network structure of the abnormal detection model through the neuron activation function, and then trains the abnormal detection model through the back propagation algorithm. The back propagation algorithm is trained according to the abnormal feature vector to obtain the abnormal detection model; According to the determined abnormal feature vector, the improved artificial neural network (ANN) is combined to establish the boost transformer abnormal location model. The abnormal feature vector is determined during the application process. The abnormal feature vector is extracted from the operation data of the boost transformer through signal processing technologies such as wavelet transform. These feature vectors contain numerical information reflecting the abnormal state of the equipment. The network structure is determined by improving the artificial neural network design. The structure of the artificial neural network is designed, including input layer, hidden layer and output layer. The input layer receives the abnormal feature vector, the hidden layer extracts and transforms the feature, and the output layer provides the abnormal location result. Select a suitable neuron activation function to enhance the nonlinear processing ability of the network. Common activation functions include Sigmoid, ReLU, Tanh, etc. For the anomaly detection model, an activation function that can capture abnormal patterns is required. Design the network structure according to the dimension of the abnormal feature vector and the requirements of abnormal location. For example, if the abnormal feature vector has n dimensions, then the input layer has n neurons. In the training process, the network weights are initialized, the network weights and biases are randomly initialized, or a pre-training method is used. Use the backpropagation algorithm to train the network. The backpropagation algorithm is a gradient descent algorithm that adjusts the network weights to minimize the error between the output layer and the true label. In each iteration, the gradient from the output layer to the hidden layer and the gradient from the hidden layer to the input layer are calculated. The weights and biases of the network are updated based on the calculated gradients to reduce the prediction error. Over multiple iterations, the network gradually learns how to identify abnormal patterns from abnormal feature vectors. This process involves providing a training dataset that contains normal and abnormal samples. The network is trained using normal samples to enable it to recognize normal patterns. The network is further trained using abnormal samples to enable it to distinguish between normal and abnormal patterns. The network performance is verified to ensure its ability to recognize abnormal patterns. After the training is completed, the performance of the model is evaluated using methods such as cross-validation, leave-one-out, or splitting the dataset. The trained model is applied to new data to locate the abnormal state of the boost transformer. If the model identifies an anomaly, the abnormal feature vector is further analyzed to determine the specific location and cause of the anomaly.

[0017] Step 3: Finally, a genetic algorithm is used to determine the anomaly detection model parameters. The anomaly detection model parameters include the scaling factor, bias and optimal fitness value of the anomaly detection model. The genetic algorithm determines the scaling factor and bias of the anomaly detection model by population initialization, performs initialization processing on the anomaly detection model, and then obtains the optimal fitness value of the anomaly detection model through the fitness function. The expression of the optimal fitness value of the anomaly detection model is: In formula (2), Denotes the optimal fitness value of the anomaly detection model, Expressed as the recall rate of abnormal evaluation, is the target variable value represented by the fitness function, It is expressed as the harmonic mean of the fault detection model. Using a genetic algorithm (GA) to determine the parameters of anomaly detection models is an optimization method that is particularly suitable for problems that are difficult to optimize directly. In anomaly detection models, parameters include scaling factors, biases, etc. The following is the calculation process of the best fitness value of anomaly detection models: First, the population is initialized and the scaling factor and bias are represented: First, it is necessary to determine how to represent the scaling factor and bias. These parameters are represented as real numbers or vectors. Then the population is generated: Using the population initialization step of the genetic algorithm, a set of parameter values ​​is randomly generated as the initial population. The size of this population depends on the complexity of the problem and the computing resources. When calculating the fitness function, the fitness function is the core of the genetic algorithm, which evaluates the pros and cons of each individual (i.e., parameter set). When selecting the fitness function calculation, select the appropriate fitness function based on the performance requirements of the model and the characteristics of the data. Then evaluate the individual: For each individual (parameter set) in the population, use the anomaly detection model to make predictions and calculate its fitness value. The fitness value of the individual is calculated based on the above-mentioned indicators such as accuracy, precision, recall rate or F1 score. Select the individual with the highest fitness value in the population as the current best fitness value.

[0018] The specific implementation process of the genetic algorithm includes: Selection: Based on the fitness value, individuals are selected for reproduction.

[0019] Crossover: Randomly select two individuals for crossover operation to produce new offspring.

[0020] Mutation: Randomly mutate individuals to increase the diversity of the population.

[0021] Iteration: The selection, crossover, mutation, and evaluation steps are repeated until a termination condition is reached (such as reaching the maximum number of iterations or the fitness value no longer improves significantly). In this way, the genetic algorithm helps find the best scaling factor and bias to optimize the performance of the anomaly detection model.

[0022] In the above specific embodiment, a genetic algorithm is used to determine the anomaly detection model parameters, and at the same time, the accuracy of the best fitness value of the anomaly detection model needs to be verified.

[0023] Genetic Algorithm (GA) uses a variety of hardware structures to improve computational efficiency and optimize algorithm performance when determining the parameters of anomaly detection models. For example, a multi-core processor is used to use a multi-core central processing unit (CPU) to parallelly process multiple operations in the genetic algorithm, such as selection, crossover, mutation, and fitness evaluation. Field programmable gate arrays (FPGAs) are designed to be customized for the specific needs of genetic algorithms to achieve hardware acceleration of specific operations. Application-specific integrated circuits (ASICs) are designed to perform key operations in genetic algorithms and provide higher performance than general-purpose processors. Cluster and distributed computing are used, such as cluster computing, which connects multiple computers to form a computing cluster and runs multiple genetic algorithm instances in parallel to accelerate the parameter optimization process. Through distributed computing, computer resources in multiple geographical locations connected by the Internet are used to achieve distributed operation of genetic algorithms. Through cloud computing resources, the resources of cloud computing service providers are used to dynamically adjust computing power as needed to provide a powerful computing environment for genetic algorithms. Adopting field programmable system-on-chip (SoPC): SoPC combines the flexibility of FPGAs and the performance of ASICs to design specialized hardware to execute specific parts of genetic algorithms.

[0024] In other specific embodiments, a multi-core CPU cluster can also be used: for the parallelization of large-scale genetic algorithms, each CPU core independently processes an individual in the population. Apply cloud computing platforms, etc., such as running genetic algorithms on cloud platforms, and using the elasticity of virtual machines to process large-scale data sets and complex optimization problems. During the experiment, the above-mentioned hardware structure is used to input anomaly detection data information and parameters to improve computing power and application capabilities. This study uses K-means clustering algorithm, isolated deep forest algorithm and half-space deep algorithm to compare with genetic algorithm, and then can verify the accuracy of the best fitness value of the anomaly detection model. The accuracy of the best fitness value of the anomaly detection model of the four algorithms is: As can be seen from Table 1, the use of genetic algorithm to determine the parameters of the anomaly detection model makes the accuracy of the best fitness value of the anomaly detection model the highest, while improving the accuracy of selecting the best fitness value of the anomaly detection model, and then the use of genetic algorithm can better determine the parameters of the anomaly detection model.

[0025] As a further embodiment of the present invention, the abnormality calculation unit integrates the operating equipment of the boost transformer by establishing an abnormality detection model of IndRNN-VAE. The abnormality detection model judges the state of the boost transformer equipment as normal and abnormal through a binary classification method. The abnormality detection model then extracts the abnormal data of the boost transformer through an artificial ant colony search algorithm. The artificial ant colony search algorithm represents the abnormal state matrix of the boost transformer equipment through a pheromone matrix, and then updates the pheromone matrix through an objective function. Finally, the abnormal data of the boost transformer is obtained through an iterative search.

[0026] The abnormal calculation unit uses an abnormal detection model that combines IndRNN-VAE (Inductive Recurrent Neural Network Variational Autoencoder) and artificial ant colony search algorithm to integrate the operating equipment of the boost transformer. The IndRNN-VAE abnormal detection model is based on the architecture of recurrent neural network (RNN) and uses variational autoencoder (Variational Autoencoder) during the working process. The IndRNN-VAE model is trained using data from a normal operating boost transformer. The model learns the normal distribution of the data and is able to generate new data that is similar to the distribution of the real data. When the model receives new operating data, it encodes this data into the latent space. The model determines whether the data point is abnormal based on the data distribution in the latent space. Abnormal data is usually far away from the normal data distribution in the latent space. The artificial ant colony search algorithm uses a pheromone matrix to represent the abnormal state matrix of the boost transformer device. Each element in the pheromone matrix represents the transition from device state A to device state B. The objective function is used to evaluate the update strategy of the pheromone matrix, which is usually based on the target of the abnormal data. The objective function includes minimizing the difference between abnormal data and normal data, or maximizing the importance of abnormal data. In each iteration, the pheromone matrix is ​​updated according to the current data point and the evaluation result of the objective function. The update rule is based on the successful path of the ant during the search process. The concentration of pheromone on the path will increase, while the concentration of pheromone on the failed path will decrease. Through iterative search, the algorithm gradually optimizes the pheromone matrix to find the best path corresponding to the abnormal data. The search process converges to one or more optimal paths, which represent the abnormal data set of the boost transformer equipment. The operating data of the boost transformer is input into the IndRNN-VAE model. The IndRNN-VAE model analyzes the data and identifies abnormal data points in the latent space. Initialize a pheromone matrix representing the abnormal state matrix. The artificial ant colony search algorithm searches for abnormal data by iteratively updating the pheromone matrix. The algorithm outputs the abnormal data of the boost transformer, which can help further diagnose and repair the equipment. The abnormal calculation unit combines the powerful anomaly detection capability of IndRNN-VAE and the effective search strategy of the artificial ant colony search algorithm to accurately identify and locate the abnormal data of the boost transformer equipment.

[0027] The anomaly detection model of IndRNN-VAE is an application layer protocol, which is mainly used for data transmission on the Internet. If the anomaly detection model is to be integrated with the boost transformer, it needs to be implemented through the network protocol of the anomaly detection model of IndRNN-VAE. The specific principles are as follows: 1. Run the Web server on the boost transformer and store the anomaly detection model on the Web server; 2. Write the IndRNN-VAE request interface in the anomaly detection model to receive the request information from the boost transformer and return the detection result; 3. On the boost transformer, call the IndRNN-VAE client, send a request to the IndRNN-VAE interface of the anomaly detection model, and transmit the relevant data to the anomaly detection model; 4. After the anomaly detection model receives the data transmitted by the boost transformer through the IndRNN-VAE request interface, it uses the binary classification method to judge the status of the boost transformer as normal and abnormal, and returns the result to the boost transformer through the IndRNN-VAE protocol; 5. After receiving the result returned by the anomaly detection model, the boost transformer performs corresponding processing and reflects the detection result in the actual application.

[0028] As a further embodiment of the present invention, the abnormality identification unit uses a Gaussian mixture model algorithm to group the boost transformer abnormal data, the Gaussian mixture model algorithm models the boost transformer abnormal data distribution into a Gaussian distribution combination, and the Gaussian mixture model algorithm evaluates the fitting effect of the Gaussian distribution combination through the Bayesian information criterion to obtain a cluster group. The abnormal state identification unit also uses a support vector machine classification model to determine the abnormality type represented by the cluster group. The support vector machine classification model determines the abnormality type corresponding to the cluster group by analyzing the characteristics and attributes of the cluster group. The abnormality identification unit also uses the establishment of an MFE-TS model to optimize the objective function, thereby improving the accuracy of abnormality identification. The established MFE-TS model optimization objective function is: In formula (3), it is represented by the original parameter X of the boost transformer at the time point The observed value of are the weight parameters of the two reconstruction losses, L represents the capacity of the parameters, It is expressed as the number of parameters, and then the anomaly recognition is optimized to improve the accuracy of anomaly recognition. The anomaly recognition unit adopts the MFE-TS (minimizing feature entropy-time series) model to optimize the objective function to improve the accuracy of anomaly recognition. The MFE-TS model is an anomaly detection method based on time series data, which optimizes the performance of anomaly detection by minimizing feature entropy. Feature entropy is an indicator of data uncertainty and is usually used in information theory. First, data preprocessing is performed to preprocess the operating data of the boost transformer, including data cleaning, normalization or standardization to ensure the quality and consistency of the data. Features are extracted from the preprocessed time series data, which may be statistical features (such as mean, variance, maximum value, minimum value, etc.) or model-based features (such as the latent space features output by the IndRNN-VAE model). Then, feature entropy calculation is performed, and for each feature, its corresponding feature entropy is calculated. Then, the objective function is constructed, which is the key to evaluating and optimizing the anomaly detection model. In the MFE-TS model, the objective function is usually to minimize the sum of feature entropies. Then, the MFE-TS model is trained using time series data, and the model learns how to identify anomalies based on feature entropies. During the training process, the model adjusts parameters to minimize the objective function. When performing data anomaly detection, the trained model is used to process new time series data. The model calculates the feature entropy of each data point and determines whether the data point is abnormal based on the value of the feature entropy. During the detection process, the model may further adjust and optimize the objective function based on new data points. Online learning or incremental learning methods are involved, so that the model can adapt to data changes. The model outputs the results of anomaly detection and marks abnormal data points or events. During the experiment, initialization calculations are first performed to initialize model parameters and objective functions. Feature calculations are then performed to calculate the features of each time series data point. Feature entropy calculations are then performed to calculate the entropy value of each feature. The objective function is evaluated on the parameter data information to evaluate the objective function value under the current model parameters. The model parameters are adjusted according to the evaluation results of the objective function. The calculations are repeated until the objective function value converges or the predetermined number of iterations is reached. In this way, the MFE-TS model optimizes the objective function of anomaly detection by minimizing feature entropy, thereby improving the accuracy of anomaly identification.

[0029] By adopting the established MFE-TS model, anomaly identification can be optimized and the accuracy of anomaly identification can be improved. During the experiment, servers with multiple high-performance CPU cores, such as Dell PowerEdge or HP ProLiant servers, were used, equipped with high-performance GPUs such as NVIDIA Tesla V100 or Quadro RTX 8000. At least 4TB of high-speed SSD and multiple HDDs were used for data storage. Such a hardware environment can provide the necessary computing resources for the training and anomaly detection of the MFE-TS model, ensuring that the model can run efficiently, thereby improving the accuracy of anomaly identification. By performing computationally intensive tasks such as data preprocessing, feature calculation, model training, and objective function optimization. At the same time, the KNN model and the random outlier selection model were compared with the MFE-TS model to verify the accuracy of anomaly identification. The verification results are: It can be seen from Table 2 that in the process of abnormal identification of boost transformer, the establishment of MFE-TS model can improve the accuracy of identification, and the error value is the smallest. The error value of KNN model is greater than the error value of random outlier selection model, and the error value of random outlier selection model is greater than the error value of MFE-TS model. Therefore, the selection of MFE-TS model can reduce the error of abnormal identification and improve the accuracy of abnormal identification.

[0030] As a further embodiment of the present invention, the abnormal position determination unit adopts a random forest algorithm model to realize the determination of the abnormal position. The random forest algorithm model combines the key feature filtering mechanism with the decision tree, records the value range of the key feature in the tree node, uses the value range of the key feature to filter the abnormal points, and adopts the random forest algorithm model to record the value range of the dual features in each node of the decision tree, which is used for filtering abnormal samples in the classification stage to determine the abnormal type and abnormal position, and obtain samples of abnormal combination. By adopting the establishment of a confidence function model, the confidence function model obtains the confidence, and the confidence is greater than or equal to the associated threshold to determine the abnormal position, and the confidence function calculation formula is: In formula (4), Expressed as confidence, Expressed as the probability of anomaly occurrence, The sample size expressed as the abnormal location and abnormal type, It is expressed as the frequency of occurrence of the abnormal type. Expressed as the length of the confidence interval. In the above specific embodiment, the establishment of the random forest algorithm model requires the construction of a decision tree, and the steps are: 1. Randomly extract samples from the original data set, usually using self-service sampling; 2. Select the best features for purification, and record the value range of the dual features in each node of the decision tree, which is used to filter abnormal samples in the classification stage to determine the abnormal type and abnormal position. After obtaining the sample of the abnormal combination, the confidence of the abnormal combination is calculated. When the confidence is high, it is considered that the credibility of the abnormal combination is high, and then the abnormal position and abnormal type are determined. The results of calculating the confidence value in different ways are: As can be seen from Table 3, four experimental groups were selected and two different methods were used to calculate the confidence. The absolute error value of the random deep forest model algorithm in the four groups of experiments was the smallest. Therefore, the random deep forest model algorithm was selected with higher accuracy through the confidence function.

[0031] As a further embodiment of the present invention, the alarm module uses a set threshold to determine whether the parameter exceeds the normal range. The alarm module receives information from the abnormal calculation unit, notifies relevant personnel through changes in light color, and uses color nodes and network topology diagrams to display the abnormal location.

[0032] It should be added that the alarm module is real-time monitoring and anomaly detection. Real-time monitoring means that in the Internet of Things system, through communication between devices, the status and data changes of the device are obtained in real time, and real-time analysis is performed, so that the step-up transformer equipment can be monitored in real time. During the real-time monitoring process, when the system detects that the status or data changes of the device exceed the expected range or specified standards, an alarm notification is automatically issued. At the same time, a threshold is set in the alarm module, and the threshold is used to determine whether the alarm is triggered. When the operating parameters of the step-up transformer equipment exceed the threshold, an alarm will be triggered.

[0033] As a further embodiment of the present invention, the method comprises: Step 1: Collect equipment parameters of the boost transformer through temperature sensors, voltage sensors, current sensors, vibration sensors and oil level sensors; Step 2: preprocessing the collected equipment parameters of the boost transformer, wherein the data preprocessing includes removing abnormal values, filling missing values ​​and correcting deviations; Step 3: The feature extraction unit uses the LASSO feature selection algorithm to select the most relevant features from all the features. The feature analysis modeling unit uses the wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector. The improved artificial neural network and genetic algorithm are used to establish an abnormality detection model and determine the parameters of the abnormality detection model. The abnormality calculation unit uses the IndRNN-VAE abnormality detection model to integrate the operating equipment of the boost transformer and then judge the abnormality of the boost transformer. Step 4: The abnormality identification unit groups the abnormal data of the boost transformer by using the Gaussian mixture model algorithm, and then evaluates the Gaussian distribution model into a Gaussian distribution combination through the Bayesian information criterion, thereby identifying the abnormal type, and using the support phase connection classification model to determine the abnormal location; Step five, the sound and light alarm warns the abnormality of the new energy boost transformer, and the human-computer interaction module displays the abnormal judgment result by using a graphical user interface, and updates the parameters of the boost transformer through a real-time update unit, and displays the abnormal position according to the color node and the network topology diagram. Although the above describes the specific embodiments of the present invention, it should be understood by those skilled in the art that these specific embodiments are only examples, and those skilled in the art may make various omissions, substitutions and changes to the details of the above methods and systems without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same way to achieve substantially the same results falls within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. A new energy boost transformer abnormality warning system, characterized by: The boost transformer abnormality warning system comprises: The data acquisition module uses sensor technology and a data acquisition card to collect data from the new energy boost transformer. The sensors include a temperature sensor, a voltage sensor, a current sensor, a vibration sensor, and an oil level sensor. The data acquisition card converts the analog signal output by the sensor into a digital signal to ensure that the hardware device is compatible with the computer system. The data preprocessing module uses data cleaning tools to remove outliers, fill missing values, and correct deviations from the collected new energy boost transformer parameters; Alarm module, using sound and light alarm to warn of abnormalities in new energy boost transformers; An anomaly detection module, the anomaly detection module includes a feature extraction unit, a feature analysis modeling unit and an anomaly calculation unit, the output end of the feature extraction unit is connected to the input end of the feature analysis modeling unit, and the output end of the feature analysis modeling unit is connected to the input end of the anomaly calculation unit; An abnormality locating module, using an abnormality locator to analyze the abnormal data of the new energy boost transformer to determine the abnormal location, the abnormality locator includes an abnormal state identification unit and an abnormal location determination unit, the output end of the abnormal state identification unit is connected to the input end of the abnormal location determination unit; A human-computer interaction module, using a graphical user interface to display the abnormality judgment result and abnormal position of the new energy boost transformer, the graphical user interface includes a real-time update unit, an abnormality judgment result display unit and an abnormal position identification unit, and the output end of the real-time update unit is respectively connected to the input end of the abnormality judgment result display unit and the input end of the abnormal position identification unit; The output end of the data acquisition module is connected to the input end of the data preprocessing module, the output end of the data preprocessing module is connected to the input end of the abnormality calculation unit, the output end of the abnormality calculation unit is connected to the input end of the abnormality positioning module and the input end of the alarm module, and the output end of the abnormality calculation unit and the output end of the abnormality positioning module are connected to the input end of the human-computer interaction module.

2. According to claim 1, a new energy boost transformer abnormal warning system is characterized by: The feature extraction unit includes feature selection, feature learning and feature fusion. The feature selection adopts LASSO feature selection algorithm to select the most relevant features from all features to reduce feature dimensions. The feature learning adopts autoencoder to automatically learn the feature representation of data to improve the level of abstract features. The feature fusion adopts feature splicing technology to fuse different types of features to form comprehensive features to improve the accuracy and robustness of anomaly detection.

3. According to claim 1, a new energy boost transformer abnormality warning system is characterized by: The working method of the feature analysis modeling unit is: Step 1: Use wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector of the boost transformer operation state. The expression of the abnormal feature vector is: In formula (1), W represents the abnormal feature vector, Expressed as the winding vibration frequency value of the step-up transformer, It is the subscript of the vibration frequency value of the step-up transformer winding. It is expressed as the average value of the vibration frequency of the step-up transformer winding, and n is expressed as the number of vibration frequencies of the step-up transformer winding. is represented as an abnormal feature value, Expressed as the step-up transformer winding vibration frequency capability, It is expressed as the efficiency of step-up to step-down; Step 2: Based on the determined abnormal feature vector and the improved artificial neural network, a boost transformer abnormal location model is established. The improved artificial neural network designs the network structure of the abnormal detection model through the neuron activation function, and then trains the abnormal detection model through the back propagation algorithm. The back propagation algorithm is trained according to the abnormal feature vector to obtain the abnormal detection model; Step 3: Finally, a genetic algorithm is used to determine the anomaly detection model parameters. The anomaly detection model parameters include the scaling factor, bias and optimal fitness value of the anomaly detection model. The genetic algorithm determines the scaling factor and bias of the anomaly detection model by population initialization, performs initialization processing on the anomaly detection model, and then obtains the optimal fitness value of the anomaly detection model through the fitness function. The expression of the optimal fitness value of the anomaly detection model is: In formula (2), Denotes the optimal fitness value of the anomaly detection model, Expressed as the recall rate of abnormal evaluation, is the target variable value represented by the fitness function, Expressed as the harmonic mean of the fault detection model.

4. The new energy boost transformer abnormality warning system and method according to claim 1 is characterized by: The abnormality calculation unit integrates the operating equipment of the boost transformer by establishing an abnormality detection model of IndRNN-VAE. The abnormality detection model judges the state of the boost transformer equipment as normal and abnormal through a binary classification method. The abnormality detection model then extracts the abnormal data of the boost transformer through an artificial ant colony search algorithm. The artificial ant colony search algorithm represents the abnormal state matrix of the boost transformer equipment through a pheromone matrix, and then updates the pheromone matrix through an objective function. Finally, the abnormal data of the boost transformer is obtained through an iterative search.

5. The new energy boost transformer abnormality warning system according to claim 1 is characterized by: The abnormality identification unit uses a Gaussian mixture model algorithm to group the boost transformer abnormal data. The Gaussian mixture model algorithm models the boost transformer abnormal data distribution into a Gaussian distribution combination. The Gaussian mixture model algorithm evaluates the fitting effect of the Gaussian distribution combination through the Bayesian information criterion to obtain a cluster group. The abnormal state identification unit also uses a support vector machine classification model to determine the abnormality type represented by the cluster group. The support vector machine classification model determines the abnormality type corresponding to the cluster group by analyzing the characteristics and attributes of the cluster group. The abnormality identification unit also uses the establishment of an MFE-TS model to optimize the objective function, thereby improving the accuracy of abnormality identification. The established MFE-TS model optimization objective function is: In formula (3), it is represented by the original parameter X of the boost transformer at the time point The observed value of are the weight parameters of the two reconstruction losses, L represents the capacity of the parameters, It is expressed as the number of parameters, thereby optimizing anomaly recognition and improving the accuracy of anomaly recognition.

6. The new energy boost transformer abnormality warning system according to claim 1 is characterized by: The abnormal position determination unit uses a random forest algorithm model to determine the abnormal position. The random forest algorithm model combines the key feature filtering mechanism with the decision tree, records the value range of the key feature in the tree node, uses the value range of the key feature to filter the abnormal points, and uses the random forest algorithm model to record the value range of the dual features in each node of the decision tree, which is used to filter the abnormal samples in the classification stage to determine the abnormal type and abnormal position, and obtain the sample of the abnormal combination. By establishing a confidence function model, the confidence function model obtains the confidence. When the confidence is greater than or equal to the associated threshold, the abnormal position is determined. The confidence function calculation formula is: In formula (4), Expressed as confidence, Expressed as the probability of anomaly occurrence, The sample size expressed as the abnormal location and abnormal type, It is expressed as the frequency of occurrence of the abnormal type. The length of the interval expressed as a confidence level.

7. The new energy boost transformer abnormality warning system according to claim 1 is characterized by: The alarm module uses a set threshold to determine whether the parameter exceeds the normal range. The alarm module receives information from the abnormal calculation unit, notifies relevant personnel through changes in light color, and uses color nodes and network topology diagrams to display the abnormal location.

8. The new energy boost transformer abnormality early warning method according to claim 1 is characterized by: The method applied to a new energy boost transformer abnormality warning system as described in any one of claims 1 to 7 comprises: Step 1: Collect equipment parameters of the boost transformer through temperature sensors, voltage sensors, current sensors, vibration sensors and oil level sensors; Step 2: preprocessing the collected equipment parameters of the boost transformer, wherein the data preprocessing includes removing abnormal values, filling missing values ​​and correcting deviations; Step 3: The feature extraction unit uses the LASSO feature selection algorithm to select the most relevant feature from all the features. The feature analysis modeling unit uses the wavelet transform algorithm to analyze the parameters of the boost transformer operation to obtain the abnormal feature vector. The improved artificial neural network and genetic algorithm are used to establish an abnormality detection model and determine the parameters of the abnormality detection model. The abnormality calculation unit uses the IndRNN-VAE abnormality detection model to integrate the operating equipment of the boost transformer and then judge the abnormality of the boost transformer. Step 4: The abnormality identification unit groups the abnormal data of the boost transformer by using the Gaussian mixture model algorithm, and then evaluates the Gaussian distribution model into a Gaussian distribution combination through the Bayesian information criterion, thereby identifying the abnormal type, and using the support phase connection classification model to determine the abnormal location; Step 5. The sound and light alarm warns of abnormalities in the new energy boost transformer. The human-computer interaction module displays the abnormality judgment results through a graphical user interface. At the same time, the parameters of the boost transformer are updated through a real-time update unit, and the abnormal location is displayed according to the color node and network topology diagram.

Images