Method, device and equipment for evaluating running state of smoothing reactor and medium

Abstract

The invention discloses a smoothing reactor operation state evaluation method, device and equipment and a medium, and the method comprises the steps: obtaining time sequence data measured by a mutual inductor of a target smoothing reactor, and extracting a time sequence electrical characteristic quantity according to the time sequence data; acquiring sensor data measured by a sensor of the target smoothing reactor, and generating a time sequence non-electrical quantity image according to the sensor data; vectorizing the time sequence electrical characteristic quantity image and the time sequence non-electrical quantity image to obtain an electrical quantity vector and a non-electrical quantity vector, and training a reactor evaluation model according to the electrical quantity vector and the non-electrical quantity vector; and evaluating and analyzing the running state of the target smoothing reactor according to the trained reactor evaluation model. According to the method, time delay and dynamic recursion connection are introduced by introducing the liquid neural network, so that time sequence characteristics are better captured, the sensitivity to a complex mode is kept, and the reliability of a smoothing reactor operation state evaluation result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment evaluation, and in particular to a method, device, equipment and medium for evaluating the operating state of a smoothing reactor. Background Art

[0002] Dry-type air-core smoothing reactor is one of the key devices in DC converter station. Its functions include preventing the steep voltage wave generated by the DC line side or DC switch from entering the high-voltage valve hall, thereby preventing the converter valve from being damaged by overvoltage. In addition, it also has the function of smoothing the ripple in the DC current, and when the load is light, it can prevent current interruption, thereby effectively reducing the commutation failure rate.

[0003] At present, common smoothing reactor fault monitoring methods mainly include manual regular inspection and diagnostic protection based on a single threshold. However, manual inspection has operational limitations and relatively low efficiency; while the diagnostic protection method based on a single threshold has problems such as high false alarm rate and missed alarm rate.

[0004] In addition to manual inspections, the operating status of smoothing reactors can be evaluated using neural networks. However, traditional neural networks find it difficult to capture dynamic changes in data in time series tasks. When processing complex input data, they may overfit and find it difficult to effectively learn the underlying structure in the data. State perception of smoothing reactors requires a large amount of multi-category data with strong temporal continuity. Therefore, traditional neural network evaluation methods have the problems of low efficiency and high false alarm and missed alarm rates.

[0005] It can be seen that how to accurately evaluate the state of the smoothing reactor has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0006] The present invention provides a method, device, equipment and medium for evaluating the operating state of a smoothing reactor, so as to achieve accurate evaluation of the operating state of the smoothing reactor.

[0007] In order to solve the above technical problems, an embodiment of the present invention provides a method for evaluating the operating status of a smoothing reactor, comprising:

[0008] The time series data of the mutual inductor measurement of the target smoothing reactor is obtained, and the time series electrical characteristic quantity is extracted according to the time series data.

[0009] Sensor data measured by a sensor of a target smoothing reactor is acquired, and a time series non-electrical quantity image is generated according to the sensor data.

[0010] The time series electrical characteristic quantity and the time series non-electrical quantity image are respectively vectorized to obtain an electrical quantity vector and a non-electrical quantity vector, and the reactor evaluation model is trained according to the electrical quantity vector and the non-electrical quantity vector. The training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the reactor evaluation model according to the loss function.

[0011] The operating state of the target smoothing reactor is evaluated and analyzed according to the trained reactor evaluation model.

[0012] Furthermore, the step of inputting the non-electrical quantity vector into an adaptive convolutional neural network comprises:

[0013] If there are bad pixels in the non-electrical quantity vector, the non-electrical quantity vector is input into the graph convolutional neural network for propagation; if there are no bad pixels in the non-electrical quantity vector, the non-electrical quantity vector is input into the convolutional neural network for propagation.

[0014] Furthermore, the pooling layer adopts an average pooling operation to downsample the output of the convolutional layer. The operation process of the pooling layer is:

[0015]

[0016] Among them, K is the number of elements in the window, x i+m,j+n is the element of the input feature map in the pooling window, is the non-electrical feature of the output after pooling, m is the offset of the pooling window along the height direction, and n is the offset of the pooling window along the width direction.

[0017] Furthermore, the step of inputting the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network comprises:

[0018] The pooled non-electrical quantity vector and the electrical quantity vector are vector-added, and the addition result is input into the liquid neural network. The input data of the liquid neural network is:

[0019]

[0020] Among them, Xe(t) is the electrical quantity vector value at time t, is the pooled non-electrical quantity vector value at time t.

[0021] Furthermore, the loss function is:

[0022]

[0023] Among them, C is the number of categories, yi is the true label, is the predicted label, λ is the regularization strength, and wj is the network weight.

[0024] The optimizing training of the reactor evaluation model according to the loss function comprises:

[0025] The loss function is calculated during the training process, and the parameters of the reactor evaluation model are iteratively updated according to the loss function calculation result and the chain rule.

[0026] Furthermore, the liquid neural network realizes recursive connection between neurons through a weight matrix.

[0027] Further, the evaluating and analyzing the operating state of the target smoothing reactor according to the trained reactor evaluation model includes:

[0028] The state characteristic quantity of the target smoothing reactor is detected according to the trained reactor evaluation model, and the operating state of the target smoothing reactor is evaluated and analyzed according to the state characteristic quantity, and the state characteristic quantity is:

[0029] X(t)=X(Xe1(t),Xe2(t)…Xem(t),Xne1(t),Xne2(t)…Xnen(t))

[0030] Among them, X(t) is the characteristic quantity of the smoothing reactor at time t, Xe i (t) is the i-th electrical characteristic quantity of the smoothing reactor at time t, Xne i (t) is the i-th non-electrical characteristic quantity of the smoothing reactor at time t.

[0031] Another embodiment of the present invention provides a smoothing reactor operating state assessment device, comprising:

[0032] The time series data acquisition module is used to acquire the time series data of the mutual inductor measurement of the target smoothing reactor, and extract the time series electrical characteristic quantity according to the time series data.

[0033] The sensor data acquisition module is used to acquire sensor data measured by a sensor of a target smoothing reactor, and to generate a time series non-electrical quantity image according to the sensor data.

[0034] The model training module is used to vectorize the time-series electrical characteristic quantity and the time-series non-electrical quantity image respectively to obtain an electrical quantity vector and a non-electrical quantity vector, and train the reactor evaluation model according to the electrical quantity vector and the non-electrical quantity vector. The training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the reactor evaluation model according to the loss function.

[0035] A state evaluation module is used to evaluate and analyze the operating state of the target smoothing reactor according to the trained reactor evaluation model.

[0036] Yet another embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the above-mentioned method for evaluating the operating state of a smoothing reactor when executing the computer program.

[0037] Yet another embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein when a device where the computer-readable storage medium is located executes the computer program, the above-mentioned method for evaluating the operating status of a smoothing reactor is implemented.

[0038] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0039] (1) By introducing time delay and dynamic recursive connection into the liquid neural network, the network can store and reflect past input information, so as to better capture the timing characteristics, and enable the network to automatically adjust its internal state when the input data changes, maintaining sensitivity to complex patterns, thereby improving the reliability of the smoothing reactor operating status assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of the steps of a method for evaluating the operating status of a smoothing reactor provided by an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a temperature measurement system provided by an embodiment of the present invention;

[0042] Figure 3 An adaptive convolutional neural network architecture diagram provided for an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of back propagation of a liquid neural network provided by an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of recursive connection of liquid neurons provided by an embodiment of the present invention;

[0045] Figure 6 A structural block diagram of a smoothing reactor operating state evaluation device provided by an embodiment of the present invention;

[0046] Figure 7 A structural diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0049] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0050] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0051] An embodiment of the present invention provides a method for evaluating the operating status of a smoothing reactor. For details, see Figure 1 , Figure 1 The flowchart of the method for evaluating the operating state of a smoothing reactor in one embodiment of the present invention includes steps S11 to S14:

[0052] Step S11, acquiring time series data of a transformer measurement of a target smoothing reactor, and extracting a time series electrical characteristic quantity according to the time series data.

[0053] According to the voltage of the line where the target smoothing reactor equipment is located, the current transformer or the measuring device of the protection equipment is used to collect the operation recording data of the target smoothing reactor in real time.

[0054] In order to solve the problems that traditional denoising algorithms are difficult to eliminate noise within the effective frequency band, the denoising effect is related to the sampling frequency, and the adaptability is poor, this embodiment uses a data denoising algorithm based on empirical wavelet transform to clean and denoise the original running recording data.

[0055] Specifically, the empirical wavelet function is defined as:

[0056]

[0057] Among them, ω i is the middle frequency of the ith frequency band of the empirical wavelet transform, γ is the transition parameter, which controls the smoothness of the frequency band boundary to avoid sharp changes in the frequency transition area, and β is the transition function, which is used to smooth the transition area between frequency bands to avoid signal distortion.

[0058] After cleaning and denoising the operation recording data obtained from the time series measurement results, the operation recording data is preprocessed to obtain time series electrical characteristic quantities such as amplitude, phase angle, harmonic component, and active power as input data of the neural network.

[0059] Step S12, acquiring sensor data measured by a sensor of the target smoothing reactor, and generating a time series non-electrical quantity image according to the sensor data.

[0060] The input data of the adaptive convolutional neural network is the temperature, ultrasound, ultra-high frequency and other image data measured in the non-electrical feature quantity measurement system. Therefore, by acquiring the data measured by the temperature sensor and the like, a time series non-electrical quantity image is generated. Specifically, in this embodiment, Figure 2 The temperature measurement system in is used as a temperature sensor to obtain temperature data.

[0061] Step S13, respectively vectorize the time series electrical characteristic quantity and the time series non-electrical quantity image to obtain an electrical quantity vector and a non-electrical quantity vector, and train the inductor evaluation model according to the electrical quantity vector and the non-electrical quantity vector. The training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the inductor evaluation model according to the loss function.

[0062] An adaptive convolutional neural network is built, the parameters of the adaptive convolutional neural network are initialized, the time series non-electrical quantity image is processed, the non-electrical characteristic quantity of the smoothing reactor during operation, that is, the non-electrical quantity vector, is extracted, and the non-electrical quantity vector is input into the adaptive convolutional neural network.

[0063] like Figure 3 As shown in FIG. 1 , the adaptive convolutional neural network of this embodiment includes 1 input layer, 3 convolutional layers, 1 pooling layer and 2 fully connected layers. The input data of the adaptive convolutional neural network is the non-electrical quantity vector obtained by the temperature, ultrasound, ultra-high frequency and other image data measured in the non-electrical feature quantity measurement system. Taking the temperature quantity as an example, Figure 2 As shown, the infrared radiation emitted by the target object is collected by an optical collection system such as an optical imaging lens, and is reflected on the infrared detector through spectral filtering, spatial filtering and other processing. The infrared radiation information is converted into an electrical signal using a photosensitive element, and an infrared thermal image is displayed on the display screen. The signal is then amplified and converted into image data, which is then vectorized and transmitted to the adaptive convolutional neural network.

[0064] When the non-electrical quantity vector is input into the adaptive convolutional neural network, if the image data corresponding to the non-electrical quantity vector has a certain number of bad pixels, the graph convolutional neural network is used for propagation. The input data of the graph convolutional neural network is a graph structure, including nodes and edges. The connection relationship between nodes in the graph is defined by edges. The structure of the graph data can be irregular, and nodes can be connected to different numbers of other nodes. The output of the graph convolutional neural network is:

[0065] H l+1 =σ(AH l W l )

[0066] Among them, H l is the feature matrix of the l-th layer node, A is the normalized adjacency matrix, which represents the connection relationship between nodes and can be expressed as Indicates that D is the degree matrix, and its diagonal elements are the number of edges connected to each node. l is the weight matrix of the lth layer, σ represents the activation function, usually the ReLU function, which introduces nonlinear features.

[0067] If there are no bad pixels in the image data corresponding to the non-electrical quantity vector, the data is input into the convolutional neural network, and the number of convolution kernels used is n, and different non-electrical feature quantities are extracted respectively. The size of each convolution kernel is A×A, and the step length of sliding on the input data is b, where A and b are hyperparameters that determine the size of the receptive field during convolution and the size of the output feature map. The output of the convolution layer is a feature map, the elements of which are:

[0068]

[0069] Where X represents the input data, W represents the weight matrix of the convolution kernel, b is the bias term, Z is the feature map output by the convolution layer, m is the offset of the convolution kernel window in the height direction, n is the offset of the convolution kernel window in the width direction, i is the index position of the output feature map in the height direction, and j is the index position of the output feature map in the width direction.

[0070] The pooling layer of the adaptive convolutional neural network uses an average pooling operation to downsample the feature map output by the convolutional layer, thereby discarding redundant features and retaining important features. Global maximum pooling and global average pooling operations are used to retain the global features of the input data and increase the generalization ability of the neural network. Global maximum pooling and global average pooling can be expressed as:

[0071]

[0072] Among them, K is the number of elements in the window, x i+m,j+n is the element of the input feature map in the pooling window, is the non-electrical feature of the output after pooling, m is the offset of the pooling window along the height direction, and n is the offset of the pooling window along the width direction.

[0073] The feature maps after pooling share one perceptron. In this embodiment, the perceptron uses two fully connected layers. The feature dimension is first reduced through one fully connected layer, and then the dimension is restored through another fully connected layer. The two output results of the perceptron are spliced ​​to obtain the output result of the non-electrical feature quantity of the smoothing reactor.

[0074] A liquid neural network is constructed, and the pooled non-electrical quantity vector and electrical quantity vector obtained after passing through the pooling layer and the fully connected layer are input into the liquid neural network.

[0075] The generalization of traditional neural network models is facing certain challenges and requires a large number of samples for training. Liquid neural networks can handle complex and dynamically changing input data, especially in time series prediction and control tasks, and can continue to learn in a real-time environment without retraining the entire network. In power systems, the equipment failure rate is low and the number of failure samples is small. The training effect of traditional neural networks may not be good, and liquid neural networks are more suitable for this application scenario.

[0076] The liquid neural network of this embodiment includes 1 input layer, 3 hidden layers and 1 fully connected layer. The input of the liquid neural network is:

[0077]

[0078] Among them, X e (t) is the electrical quantity vector value at time t, is the pooled non-electrical quantity vector value at time t.

[0079] The input data of the liquid neural network is from the input layer to the hidden layer, between different hidden layers, and from the hidden layer to the fully connected layer. The specific formula is:

[0080]

[0081] Among them, W l is the weight matrix from the input layer to the hidden layer, is the input signal of layer l, b l is the bias term for forward propagation, is the output of the forward propagation of each layer of the liquid neural network, σ l is the activation function. In the present invention, the activation function from the input layer to the hidden layer is ReLU, the activation function between the hidden layers is Sigmod, and the activation function from the hidden layer to the output layer is Softmax.

[0082] The training process also includes: Figure 4 As shown, Figure 4 A schematic diagram of back propagation of a liquid neural network provided in an embodiment of the present invention is shown, back propagation is performed and a cross entropy loss function is calculated, and the parameters of the neural network are updated according to the loss function calculation result and the chain rule.

[0083] In view of the gradient vanishing phenomenon when the traditional least squares loss function is used to deal with classification problems and the overfitting problem of the cross entropy loss function, the loss function of the liquid neural network is selected as the L2 regularized cross entropy loss:

[0084]

[0085] Where C is the number of categories, y i is the true label, is the predicted label, λ is the regularization strength, and w j is the network weight.

[0086] The excellent generalization and processing ability of liquid neural networks for time series data lies in the fact that their structure and connection methods are not fixed, such as Figure 5 As shown in the figure, there are recursive connections between different neurons in each layer. The recursive connections between different neurons in the liquid neural network are realized through the weight matrix. Each neuron not only receives external input, but also receives feedback from other neurons. These feedbacks are adjusted through the weight matrix to form a dynamic neuron state update. In the input layer, the recursive connection net of neurons recurr for:

[0087]

[0088] in, is the weight matrix of the recursive connection of the input layer neurons, Y 1 is the state of the input layer neurons at the current moment, is the bias term of the recursive connection of the input layer neurons, net recurr is the output of the recursive connection of the input layer neurons, and σ is the activation function, which is generally Softmax.

[0089] Output and recursive connections in hidden layers net recurr for:

[0090]

[0091] in, is the weight matrix of the recursive connection of neurons in the lth hidden layer, Y is the state of neurons in the lth hidden layer at the current moment, is the bias term for the recursive connection of the lth hidden layer neurons, net recurr is the output of the recursive connection of the lth hidden layer neurons.

[0092] The state of a liquid neural network neuron is determined by its input, previous state, and recursive connection at each time step. From time t to time t+Δt, the update rule of the neuron state is:

[0093]

[0094] Where y is the state of the neuron, τ is the time constant associated with the neuron, and net out With net recurr are the neuron output and the activation value of the recurrent connection, E l and is the connection strength matrix between forward propagation and neuron recursion, and k is the number of neurons.

[0095] After one training, the neuron state will be updated as:

[0096]

[0097] Where k represents the learning rate, which is a hyperparameter set before training. Represents the update speed, y and y * Represents the neuron state before and after the update.

[0098] Repeat the above training process several times, iterate the training set data multiple times without initializing the parameters of the neural network, stop the iteration after the preset calculation accuracy is met, and dynamically adjust the activation state of the neurons, and use the adaptive adjustment of the current network structure and parameters of the time series historical data to complete the training of the reactor evaluation model. Specifically, the training set data of the Sum group is divided into multiple batches, each batch contains 32 sampling data, so as to improve efficiency and accelerate convergence. For each batch of data, the non-electrical feature quantity is refined by the adaptive convolutional neural network and input into the liquid neural network, the activation state of the neurons is calculated, and the predicted output of the network is obtained by passing through the hidden layer and the output layer in turn. According to the predicted output and the true label, the loss function and its gradient are calculated, and the error of each layer of neurons is calculated by using the chain rule, and then it is reversely transferred from the output layer to the input layer to update the weight and bias. Repeat the above process Sum / 32 times until the predetermined calculation accuracy is met, and the parameters after the last iteration of the neural network are saved as the final parameters for the evaluation of the operating status of the smoothing reactor.

[0099] Step S14: evaluating and analyzing the operating state of the target smoothing reactor according to the trained reactor evaluation model.

[0100] The recorded data and images measured during the operation of the smoothing reactor are used as a test set data for a smoothing reactor operating state evaluation method based on liquid neural network. The operating state of the smoothing reactor is used as the detection and prediction result of the algorithm, specifically including normal operating state, turn-to-turn short-circuit operating state, lightning impulse operating state and abnormal operating state caused by commutation failure of the converter valve.

[0101] Specifically, this embodiment uses the state characteristic quantity obtained by the reactor evaluation model to evaluate and analyze the smoothing reactor, and the state characteristic quantity is:

[0102] X(t)=X(Xe1(t),Xe2(t)…Xem(t),Xne1(t),Xne2(t)…Xnen(t))

[0103] Among them, X(t) is the characteristic quantity of the smoothing reactor at time t, Xe i (t) is the i-th electrical characteristic quantity of the smoothing reactor at time t, Xne i (t) is the i-th non-electrical characteristic quantity of the smoothing reactor at time t.

[0104] The liquid neural network described in the present invention includes two core features: dynamic architecture and continuous learning and adaptability. The dynamic architecture makes the neurons of the liquid neural network more expressive than traditional neural networks, thereby improving the interpretability of the model; and continuous learning and adaptability enable the liquid neural network to continue to adapt to changing data after training, which is closer to the working mechanism of the biological brain. In addition, the liquid neural network also shows stronger robustness to noise and interference in the input signal. Therefore, the liquid neural network does not require a large number of pre-training samples. Through recursive connections between neurons, real-time learning can be performed in the received input data, thereby realizing rapid detection of the operating status of the smoothing reactor, significantly reducing the complexity of the algorithm and improving the confidence of the algorithm.

[0105] The embodiment of the present invention further provides a smoothing reactor operating state evaluation device, which is used to execute the smoothing reactor operating state evaluation method as described above. Figure 6 This is a structural block diagram of a smoothing reactor operating state evaluation device according to an embodiment of the present invention, wherein the device comprises:

[0106] The time series data acquisition module 21 is used to acquire the time series data of the mutual inductor measurement of the target smoothing reactor, and extract the time series electrical characteristic quantity according to the time series data.

[0107] The sensor data acquisition module 22 is used to acquire sensor data measured by a sensor of a target smoothing reactor, and to generate a time series non-electrical quantity image according to the sensor data.

[0108] The model training module 23 is used to vectorize the time-series electrical characteristic quantity and the time-series non-electrical quantity image respectively to obtain an electrical quantity vector and a non-electrical quantity vector, and train the inductor evaluation model according to the electrical quantity vector and the non-electrical quantity vector. The training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the inductor evaluation model according to the loss function.

[0109] The state evaluation module 24 is used to evaluate and analyze the operating state of the target smoothing reactor according to the trained reactor evaluation model.

[0110] The technical features and technical effects of the device proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and are not described in detail here. Each module in the above device can be implemented in whole or in part by software, hardware, and a combination thereof. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0111] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program; wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the smoothing reactor operating status evaluation method as described above.

[0112] The embodiment of the present invention further provides a computer device, Figure 7 A structural block diagram of a preferred embodiment of a computer device provided by the present invention, the computer device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the above-mentioned method for evaluating the operating status of a smoothing reactor when executing the computer program.

[0113] Preferably, the computer program can be divided into one or more modules / units (such as computer program 1, computer program 2, ...), and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0114] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor. The processor is the control center of the computer device, and various parts of the computer device are connected using various interfaces and lines.

[0115] The memory mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, and a flash card (Flash Card), etc., or the memory can also be other volatile solid-state storage devices.

[0116] It should be noted that the above-mentioned computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that Figure 7 The structural block diagram is merely an example of a computer device and does not constitute a limitation of the computer device, and may include more or less components than shown in the figure, or combine certain components, or different components.

[0117] In summary, the smoothing reactor operation status assessment method, device, equipment and medium provided by the embodiments of the present invention have the following beneficial effects compared with the prior art:

[0118] (1) By introducing time delay and dynamic recursive connection into the liquid neural network, the network can store and reflect past input information, so as to better capture the timing characteristics, and enable the network to automatically adjust its internal state when the input data changes, maintaining sensitivity to complex patterns, thereby improving the reliability of the smoothing reactor operating status assessment results.

[0119] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for evaluating the operating status of a smoothing reactor, characterized in that: include: Acquire time series data of a transformer measurement of a target smoothing reactor, and extract a time series electrical characteristic quantity according to the time series data; Acquiring sensor data measured by a sensor of a target smoothing reactor, and generating a time-series non-electrical quantity image according to the sensor data; The time series electrical characteristic quantity and the time series non-electrical quantity image are respectively vectorized to obtain an electrical quantity vector and a non-electrical quantity vector, and a reactor evaluation model is trained according to the electrical quantity vector and the non-electrical quantity vector. The training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the reactor evaluation model according to a loss function. The operating state of the target smoothing reactor is evaluated and analyzed according to the trained reactor evaluation model.

2. A method for evaluating the operating status of a smoothing reactor according to claim 1, characterized in that: The step of inputting the non-electrical quantity vector into an adaptive convolutional neural network comprises: If there are bad pixels in the non-electrical quantity vector, the non-electrical quantity vector is input into the graph convolutional neural network for propagation; if there are no bad pixels in the non-electrical quantity vector, the non-electrical quantity vector is input into the convolutional neural network for propagation.

3. A method for evaluating the operating status of a smoothing reactor according to claim 1, characterized in that: The pooling layer uses an average pooling operation to downsample the output of the convolutional layer. The operation process of the pooling layer is: Among them, K is the number of elements in the window, x i+m,j+n is the element of the input feature map in the pooling window, is the non-electrical feature of the output after pooling, m is the offset of the pooling window along the height direction, and n is the offset of the pooling window along the width direction.

4. A method for evaluating the operating status of a smoothing reactor according to claim 1, characterized in that: The step of inputting the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network comprises: The pooled non-electrical quantity vector and the electrical quantity vector are vector-added, and the result of the addition is input into the liquid neural network. The input data of the liquid neural network is: Among them, X e (t) is the electrical quantity vector value at time t, is the pooled non-electrical quantity vector value at time t.

5. A method for evaluating the operating status of a smoothing reactor according to claim 1, characterized in that: The loss function is: Where C is the number of categories, y i is the true label, is the predicted label, λ is the regularization strength, and w j is the network weight; The optimizing training of the reactor evaluation model according to the loss function comprises: The loss function is calculated during the training process, and the parameters of the reactor evaluation model are iteratively updated according to the loss function calculation result and the chain rule.

6. A method for evaluating the operating status of a smoothing reactor as claimed in claim 5, characterized in that: The liquid neural network realizes recursive connection between neurons through a weight matrix.

7. A method for evaluating the operating status of a smoothing reactor according to claim 1, characterized in that: The step of evaluating and analyzing the operating state of the target smoothing reactor according to the trained reactor evaluation model includes: The state characteristic quantity of the target smoothing reactor is detected according to the trained reactor evaluation model, and the operating state of the target smoothing reactor is evaluated and analyzed according to the state characteristic quantity, and the state characteristic quantity is: X(t)=X(X e1 (t),X e2 (t)…X em (t),X ne1 (t),X ne2 (t)…X nen (t)) Among them, X(t) is the characteristic quantity of the smoothing reactor at time t, X ei (t) is the i-th electrical characteristic quantity of the smoothing reactor at time t, X nei (t) is the i-th non-electrical characteristic quantity of the smoothing reactor at time t.

8. A smoothing reactor operating status evaluation device, characterized in that: include: A time series data acquisition module, used to acquire time series data of the mutual inductor measurement of the target smoothing reactor, and extract time series electrical characteristic quantities according to the time series data; A sensor data acquisition module, used to acquire sensor data measured by a sensor of a target smoothing reactor, and generate a time series non-electrical quantity image according to the sensor data; A model training module, used for respectively vectorizing the time series electrical characteristic quantity and the time series non-electrical quantity image to obtain an electrical quantity vector and a non-electrical quantity vector, and training the reactor evaluation model according to the electrical quantity vector and the non-electrical quantity vector, wherein the training process is designed to input the non-electrical quantity vector into an adaptive convolutional neural network, propagate along the input layer, the convolution layer and the pooling layer in sequence to obtain a pooled non-electrical quantity vector, input the pooled non-electrical quantity vector and the electrical quantity vector into a liquid neural network, and optimize the reactor evaluation model according to a loss function; A state evaluation module is used to evaluate and analyze the operating state of the target smoothing reactor according to the trained reactor evaluation model.

9. A computer device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for evaluating the operating state of a smoothing reactor according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the method for evaluating the operating status of a smoothing reactor according to any one of claims 1 to 7 is implemented.

Images