Method and device for verifying the validity of monitoring logic of valve base electronic equipment

Abstract

The present invention discloses a method and device for verifying the validity of the monitoring logic of a valve-based electronic device, wherein the method comprises: after a VBE sends an optical trigger signal to a test chassis, receiving an optical signal fed back by the test chassis; the test chassis simulates the function of a thyristor of a converter valve; the optical signal is sequentially input into a trained feature extraction model, a feature denoising model, and a classifier model, and an optical signal recognition result is output; the optical signal recognition result includes an optical signal category, and the optical signal category reflects the state of the thyristor; the classifier model is pre-trained based on an extreme learning machine, and an adaptive Hadamard transform is performed on the denoised data during the training process; receiving optical signal description information sent by the test chassis; comparing the optical signal recognition result and the optical signal description information, and determining the monitoring logic verification result of the VBE according to the comparison result. The present invention can realize the monitoring logic validity verification of the VBE valve-based electronic device without being connected to a real working converter valve.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage direct current power transmission, and in particular to a method and a device for verifying the validity of monitoring logic of a valve-based electronic equipment. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] High-voltage direct current transmission is the main transmission method for most of the ultra-high voltage transmission projects currently under construction and in operation due to its long transmission distance, large capacity and low loss. As the most core component in the high-voltage direct current transmission project, the converter valve is a power device based on thyristor series technology, which completes the conversion of AC into DC (rectification) or AC into DC (inversion) under the command of VBE (Valve Base Electronics). As a monitoring and control device for the converter valve, VBE can monitor and feedback the operating status of the valve in real time. The internal software of VBE is designed with monitoring logic for thyristors, which is used to monitor the triggering logic and function of the thyristors on the convection valve. In order to ensure the normal operation of the thyristor monitoring logic of VBE, VBE needs to be fully tested.

[0004] At present, the commonly used VBE testing method is still mainly based on manual operation, which includes appearance and hardware inspection of cabinets, chassis and boards, continuity test of various optical fibers on VBE and communication links between chassis, function test of VBE body and various signals, and completion of VBE thyristor monitoring logic test. However, this method relies on manual experience and judgment, which is not only inefficient, but also easily affected by subjective factors, resulting in limited accuracy and consistency of the results.

[0005] The existing technology also has automated testing for VBE, but it is necessary to connect VBE with a real working converter valve to realize automated testing of VBE. At the same time, developers or testers are also required to build and write test-related work content, which is highly dependent on experienced testers. Summary of the invention

[0006] The embodiment of the present invention provides a method for verifying the validity of the monitoring logic of a valve-based electronic device, which is used to verify the validity of the monitoring logic of a VBE thyristor when the device is disconnected from the actual working converter valve, thereby reducing the dependence on test personnel while ensuring the accuracy of the test. The method includes:

[0007] After the valve base electronic device VBE to be tested sends an optical trigger signal to the test chassis, an optical signal fed back by the test chassis is received; the test chassis simulates the function of the thyristor of the converter valve;

[0008] The optical signal is input into a trained feature extraction model, and the feature-extracted data is output; the feature extraction model pre-uses the historical optical signal fed back to the VBE by the thyristor of the converter valve to train the multi-layer fully connected neural network, and the exploratory gradient descent is used to update the neural network parameters during the training process;

[0009] Input the feature-extracted data into a trained feature denoising model, and output the denoised data; the feature denoising model is pre-trained based on an autoencoder for inverse reconstruction, the autoencoder includes an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data;

[0010] The denoised data is input into a trained classifier model, and an optical signal recognition result is output; the optical signal recognition result includes an optical signal category, and the optical signal category reflects the state of the thyristor; the classifier model is pre-trained based on an extreme learning machine, and an adaptive Hadamard transform is performed on the denoised data during the training process. The processed data is passed to an output layer, a loss function is calculated at the output layer, and the training extreme learning machine model parameters are updated through a back propagation algorithm according to the loss function;

[0011] receiving optical signal description information sent by the test chassis; the optical signal description information includes an optical signal category;

[0012] The optical signal identification result and the optical signal description information are compared, and the monitoring logic verification result of the VBE to be tested is determined according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal.

[0013] The embodiment of the present invention further provides a valve base electronic device monitoring logic validity verification device, which is used to verify the validity of the VBE thyristor monitoring logic when it is disconnected from the actual working converter valve, while reducing the dependence on the tester and ensuring the accuracy of the test. The device includes:

[0014] A data acquisition module, used to receive an optical signal fed back by a test chassis after the VBE to be tested sends an optical trigger signal to the test chassis; the test chassis simulates the function of a thyristor of a converter valve;

[0015] A feature extraction module is used to input the optical signal into a trained feature extraction model and output feature-extracted data; the feature extraction model pre-uses the historical optical signal fed back to the VBE by the thyristor of the converter valve to train a multi-layer fully connected neural network, and uses exploratory gradient descent to update the neural network parameters during the training process;

[0016] A feature denoising module is used to input the feature-extracted data into a trained feature denoising model and output the denoised data; the feature denoising model is pre-trained based on an autoencoder of inverse reconstruction, the autoencoder includes an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data;

[0017] An optical signal category recognition module is used to input the denoised data into a trained classifier model and output an optical signal recognition result; the optical signal recognition result includes an optical signal category, and the optical signal category reflects the state of the thyristor; the classifier model is pre-trained based on an extreme learning machine, and the denoised data is subjected to an adaptive Hadamard transform during the training process. The processed data is transmitted to an output layer, a loss function is calculated at the output layer, and the training extreme learning machine model parameters are updated through a back propagation algorithm according to the loss function;

[0018] The result output module is used to receive the optical signal description information sent by the test chassis; the optical signal description information includes the optical signal category; compare the optical signal identification result and the optical signal description information, and determine the monitoring logic verification result of the VBE to be tested according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal.

[0019] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned valve base electronic device monitoring logic validity verification method when executing the computer program.

[0020] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for verifying the validity of the valve base electronic device monitoring logic is implemented.

[0021] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned valve base electronic device monitoring logic validity verification method is implemented.

[0022] In the embodiment of the present invention, the test chassis is used to simulate the function of the thyristor of the converter valve, the VBE to be tested sends an optical trigger signal to the test chassis, and the test chassis simulates the feedback optical signal of the thyristor of the converter valve. Without being connected to the real working converter valve, the effectiveness verification of the monitoring logic of the valve-based electronic equipment can be realized. During the verification, the trained feature extraction model is used to extract features, the trained feature denoising model is used to denoise features, and the trained classifier model is used to classify and identify features to determine the optical signal recognition result. By comparing the optical signal category in the optical signal recognition result with the optical signal description information sent by the test chassis, it is determined whether the monitoring logic of the VBE to be tested is normal or abnormal. Compared with the technical solutions in the prior art, the embodiment of the present invention uses artificial intelligence technology to perform test verification, which can reduce the dependence on testers. At the same time, the method uses an exploratory gradient descent neural network for feature extraction, which improves the stability and model performance of the feature extraction model. The autoencoder algorithm based on inverse reconstruction is used to minimize the dependency between features and the influence of noise, improve the encoding quality and decoding accuracy, and adopts adaptive Hadamard transform processing to enhance the classification accuracy and adaptability of the model to complex data, which can ensure the accuracy of the test. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0024] Figure 1 It is a flow chart of a method for verifying the validity of valve base electronic equipment monitoring logic in an embodiment of the present invention;

[0025] Figure 2 Schematic diagram of the feature extraction model training process in an embodiment of the present invention;

[0026] Figure 3 Schematic diagram of the feature denoising model training process in an embodiment of the present invention;

[0027] Figure 4 It is a diagram showing an example of a method for verifying the validity of the monitoring logic of a valve base electronic device in an embodiment of the present invention;

[0028] Figure 5 Schematic diagram of a valve base electronic monitoring logic validity verification device in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0030] As a monitoring and control device for the converter valve, the effectiveness of the thyristor monitoring logic of the VBE is directly related to the stability and reliability of the entire converter valve operation. However, existing verification methods mostly rely on manual experience and judgment, which is not only inefficient, but also easily affected by subjective factors, resulting in limited accuracy and consistency of the results. In addition, due to the large number of thyristor stages on the converter valve and the complexity of the actual operation, it is difficult for the manual verification method to perform a large number of tests in a short period of time and at the same time meet the requirements of efficient and accurate testing. The automated testing method in the prior art relies on experienced testers to edit the test script. To this end, a method for verifying the effectiveness of the monitoring logic of a valve-based electronic device is proposed in an embodiment of the present invention.

[0031] Figure 1 FIG. 1 is a flow chart of a method for verifying the validity of valve base electronic equipment monitoring logic according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0032] Step 101, after the VBE to be tested sends an optical trigger signal to the test chassis, receiving an optical signal fed back by the test chassis; the test chassis simulates the function of the thyristor of the converter valve;

[0033] Step 102: Input the optical signal into a trained feature extraction model and output feature-extracted data; the feature extraction model pre-uses historical optical signals fed back to the VBE by the converter valve thyristor to train a multi-layer fully connected neural network, and uses exploratory gradient descent to update the neural network parameters during the training process;

[0034] Step 103: input the feature-extracted data into a trained feature denoising model, and output the denoised data; the feature denoising model is pre-trained based on an autoencoder of inverse reconstruction, the autoencoder includes an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data;

[0035] Step 104, input the denoised data into the trained classifier model, and output the optical signal recognition result; the optical signal recognition result includes the optical signal category, and the optical signal category reflects the thyristor state; the classifier model is pre-trained based on the extreme learning machine, and the denoised data is subjected to adaptive Hadamard transform processing during the training process. The processed data is transmitted to the output layer, and the loss function is calculated at the output layer. The training extreme learning machine model parameters are updated through the back propagation algorithm according to the loss function;

[0036] Step 105, receiving the optical signal description information sent by the test chassis; the optical signal description information includes the optical signal category;

[0037] Step 106 , compare the optical signal identification result and the optical signal description information, and determine the monitoring logic verification result of the VBE to be tested according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal.

[0038] from Figure 1 It can be seen from the flow chart that in the embodiment of the present invention, the test chassis is used to simulate the function of the thyristor of the converter valve, the VBE to be tested sends an optical trigger signal to the test chassis, and the test chassis simulates the feedback optical signal of the thyristor of the converter valve. Without being connected to the real working converter valve, the effectiveness verification of the monitoring logic of the valve-based electronic equipment can be realized. During the verification, the trained feature extraction model is used to extract features, the trained feature denoising model is used to denoise features, and the trained classifier model is used to classify and identify features to determine the optical signal recognition result. By comparing the optical signal category in the optical signal recognition result with the optical signal description information sent by the test chassis, it is determined whether the monitoring logic of the VBE to be tested is normal or abnormal. Compared with the technical solutions in the prior art, the embodiment of the present invention uses artificial intelligence technology to perform test verification, which can reduce the dependence on testers. At the same time, the method uses an exploratory gradient descent neural network for feature extraction, which improves the stability and model performance of the feature extraction model. The autoencoder algorithm based on inverse reconstruction is used to minimize the dependency between features and the influence of noise, improve the encoding quality and decoding accuracy, and adopts adaptive Hadamard transform processing to enhance the classification accuracy and adaptability of the model to complex data, which can ensure the accuracy of the test.

[0039] The following is a detailed explanation of the method for verifying the validity of the valve base electronic equipment monitoring logic in an embodiment of the present invention.

[0040] In the embodiment of the present invention, it is necessary to first obtain sample data, train the model, and then put the model into use.

[0041] 1. Data Collection

[0042] In the embodiment of the present invention, the test chassis simulates the function of the thyristor of the converter valve. After the test chassis simulates the unlocking operation of the converter valve, the VBE turns on the monitoring logic and determines the state of the thyristor by receiving the optical signal fed back by the test chassis.

[0043] In one embodiment, the feedback signal and the protective trigger signal returned by the converter valve to the VBE are collected in a continuous time, and both signals are optical signals and are transmitted through optical fibers.

[0044] In one embodiment, before inputting the optical signal into the trained feature extraction model, the following steps may also be included:

[0045] Converting optical signals into vector data;

[0046] For example, a light signal is represented as: ,in, represents the optical signal vector output from VBE, r ai is the i-th component of the optical signal vector, n g is the dimension of the light signal.

[0047] Vector data is stored in a structured format, for example, in JSON format.

[0048] Furthermore, inputting the optical signal into the trained feature extraction model may include:

[0049] Input the vector data into the trained feature extraction model.

[0050] In the training phase, after the data is acquired, it can be labeled by a program or manually, and classified according to the characteristics of the optical signal and its function in the monitoring logic. The labeled categories are divided into "valid" or "invalid", or labeled as "normal state", "fault state", and the "fault state" can be specifically "short circuit fault", "overcurrent fault", "overvoltage fault", "reverse breakdown fault", etc.

[0051] 2. Data Expansion

[0052] In the training phase, since the experimental data obtained is limited, in one embodiment, before the VBE to be tested sends an optical trigger signal to the test chassis, the method may further include:

[0053] Obtaining historical optical signals fed back by the converter valve thyristor to the VBE, labeling, preprocessing, and feature extraction of the historical optical signals fed back by the converter valve thyristor to the VBE to obtain training samples;

[0054] The training samples are expanded by using a generative adversarial network to obtain expanded training samples;

[0055] Based on the expanded training samples, feature extraction model, feature denoising model and classifier model are trained.

[0056] Since the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The embodiment of the present invention uses a generative adversarial network based on quantum coding to generate samples, thereby achieving data expansion. The generative adversarial network based on quantum information coding consists of two parts: a generator and a discriminator. The generator is responsible for creating realistic virtual data, and the task of the discriminator is to distinguish the generated data from the real data. After receiving the random noise signal, the generator optimizes the representation of information through quantum information compression technology to generate new data samples. The discriminator evaluates the authenticity of the sample and is trained through the quantum error correction loss function to enhance its discrimination ability.

[0057] The specific training process of the generative adversarial network based on quantum information coding is as follows:

[0058] 2.1. Initialize the network parameters of the generator and the discriminator, the parameters including the initial configuration of the quantum state and the classical network weights. In one embodiment, the initialization parameters are implemented in a Gaussian distribution and can be expressed as:

[0059]

[0060] In the formula, and Represent the initial parameters of the generator and discriminator respectively; N (0,I) represents a Gaussian distribution with a mean of 0 and a covariance matrix that is the identity matrix.

[0061] 2.2 Generator Receive random noise z, use quantum information compression method to process the noise, and generate quantum state ψ, which can be expressed as:

[0062] ψ = Q(z;θ q )

[0063] Where z is the random noise of the generator input; θ q Represents the parameters of the quantum state generation function; Q() is the quantum state generation function.

[0064] Furthermore, let μ be the mean of the noise, σ 2 is the variance of the noise, and the sampling of random noise z follows the following distribution:

[0065] z~ N (μ',σ' 2 )

[0066] In the formula, μ' and σ' 2 are the adjusted mean and variance respectively, and the adjustment method can be expressed as:

[0067] μ'=μ+βcg × S(f;σ s )

[0068] σ' 2= σ 2 × exp(S(f;σ s )

[0069] In the formula, β cg is the adjustment factor, S( ) is the adaptive noise adjustment function, σ s is the parameter of the adaptive noise adjustment function. Preferably, β cg Set to 0.3, f is the input of the noise adjustment function, specifically a random noise vector.

[0070] Furthermore, the calculation method of the adaptive noise adjustment function S( ) can be expressed as:

[0071] S(f;σ s )=tanh(W s ×f+b s )

[0072] Where W s and b s are the weight and bias parameters of the adaptive noise adjustment function, and tanh( ) is the hyperbolic tangent function.

[0073] In one embodiment, the calculation method of the quantum state generation function Q() can be expressed as:

[0074] Q(z;θ q )=U(θ' q )×H(z)

[0075] In the formula, H(z) represents the random noise Through a series of Hadamard gates, it is converted into a superposition state of the quantum bit; U(θ' q ) represents the parameter θ q Applied quantum logic gates; θ' q are the adjusted quantum gate parameters.

[0076] In one embodiment, an adjustment module based on an attention mechanism is used to dynamically adjust θ according to the characteristics of the input data. q Specifically, consider the input random noise z and random noise vector f, θ q The adjustment method using the attention-based adjustment module can be expressed as:

[0077] θ' q =θ q +α cf ×A(f;θ a )

[0078] In the formula, θ' q is the adjusted quantum gate parameter, α cf is the learning rate, A is the attention mechanism function of the attention-based adjustment module, θ a is the parameter of the attention module. Preferably, α cf Set to 0.01.

[0079] Furthermore, the calculation method of the attention mechanism function can be expressed as:

[0080] A(f;θ a )=Softmax(W a ×f+b a )

[0081] Where W a and b a are the weight and bias parameters of the attention module; Softmax() is the Softmax function used to normalize the weight distribution.

[0082] 2.3. The quantum state ψ undergoes a quantum classical conversion function C, which converts quantum information into classical data that can be used for discrimination, which can be expressed as:

[0083] x'=C(ψ;θ c )

[0084] Where x' is the classical data, specifically the vector data in the same format as the original collected data; θ c represents the parameters of the quantum-classical conversion function; C( ) is the quantum-to-classical conversion operation.

[0085] In one embodiment, the specific implementation of the conversion function C( ) from the quantum state ψ to the classical data x' is:

[0086] C(ψ;θ c )=M(ψ) ©θ c

[0087] Where M(ψ) represents a series of quantum measurement operations on the quantum state ψ to obtain a classical bit string; © represents the classical bit string obtained by the parameter Perform a specific arithmetic operation (such as dot product or weighted sum), θ c Used to adjust the weight of the measurement results.

[0088] 2.4. The discriminator evaluates the generated classical data and the input real data, and uses the quantum error correction loss function L QEC The constraints for the training process can be expressed as:

[0089] L QEC(x,x';θ d ) = -log(D(x;θ d ))-log(1-D(x';θ d ))

[0090] In the formula, θ d represents the parameters of the discriminator; D is the discriminator function, the probability that the output sample is a true sample; x is the real data sample.

[0091] Furthermore, the discriminator function D(x;θ d ) can be calculated as:

[0092] D(x;θ d )=Sig(W×x+b)

[0093] Where Sig( ) is the Sigmoid activation function, W and b are the weight matrix and bias term of the discriminator, respectively, which are used to convert the input real data x into a probability value between 0 and 1.

[0094] 2.5. Adjust the parameters θ of the generator G according to the output of the discriminator and the quantum error correction loss function g , the way to update the parameters can be expressed as:

[0095]

[0096] In the formula, is the generator parameter of the current iteration; η vb is the learning rate; is the gradient of the loss function with respect to the generator parameters. Preferably, η vb Set to 0.005.

[0097] Furthermore, the gradient The calculation method can be expressed as:

[0098]

[0099] In the formula, Denotes the loss function L QEC Partial derivatives for classical data x'; Represents the classic data x' for the generator parameter θ g The partial derivative of .

[0100] 2.6. Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0101] After data expansion, the expanded sample data can be used to form a training set, a test set, and a validation set, and the training set, test set, and validation set can be used to train, test, and validate the model, respectively.

[0102] The embodiment of the present invention includes a feature extraction model, a feature denoising model, and a classifier model. The training process of each model is introduced below.

[0103] 3. Feature Extraction Model Training

[0104] The feature extraction model pre-utilizes the historical optical signal fed back to the VBE by the thyristor of the converter valve to train the multi-layer fully connected neural network. During the training process, the exploratory gradient descent is used to update the neural network parameters.

[0105] In one embodiment, the feature extraction model includes an input layer, multiple hidden layers and an output layer, each hidden layer has a different number of neurons, and each layer uses a preset activation function to process input data; the exploratory gradient descent is used to update the neural network parameters during the training process, which may include: before each iteration, before the neural network parameters are updated, the historical gradient linear regression is used to predict the gradient changes of this iteration; the neural network parameters are updated using the gradient changes of this iteration; wherein, a first maximum number of iterations is preset.

[0106] For example, a 6-layer fully connected neural network is used for feature extraction. The structure of the 6-layer fully connected neural network is as follows:

[0107] a-Input layer: receives and transmits sample data after sample expansion.

[0108] b- Hidden layers: The network contains four hidden layers, each consisting of a certain number of neurons.

[0109] b1-first hidden layer: has 128 neurons, used for preliminary feature extraction;

[0110] b2-second hidden layer: has 256 neurons to further abstract features;

[0111] b3-the third hidden layer: has 512 neurons, further deepening feature extraction;

[0112] b4-the fourth hidden layer: has 1024 neurons, further deepening feature extraction.

[0113] b5 - output layer: has 512 neurons.

[0114] In the prior art, when some solutions use neural networks for feature extraction, in certain neural network structures, they may encounter problems such as gradient vanishing, gradient exploding, or falling into local optimal solutions, which affects the stability of training and the performance of the model. The embodiment of the present invention uses a neural network based on exploratory gradient descent as a feature extraction model. In the traditional neural network training process, the exploratory gradient descent method is used to not only consider the current gradient information, but also adopt a forward exploration mechanism to predict the possible change trend of the gradient before the gradient is updated, and adjust the learning step size and direction to achieve faster convergence speed and better local minimum, thereby enhancing the generalization ability of the feature.

[0115] Figure 2 This is a schematic diagram of the feature extraction model training process in an embodiment of the present invention, referring to Figure 2 Introduce the training process of the neural network algorithm based on exploratory gradient descent.

[0116] Step 1: Initialize the parameters of the neural network, including the weight and bias parameters of the neural network. In one embodiment, the initialization method is expressed as:

[0117] W p,i =σ p × N (0,1)+ε p

[0118] b p,i =β p × N (0,1)

[0119] Where W p,i is the weight matrix of the i-th layer of the neural network; b p,i is the bias vector of the i-th layer of the neural network; σ p is the scaling factor of the neural network weights, β p is the scaling factor of the bias of the neural network; N (0,1) represents a normal distribution with a mean of 0 and a standard deviation of 1; ε p is a tiny noise. Preferably, ε p is a noise vector whose elements are all 0.001.

[0120] Step 2: The input data passes through each layer of the neural network.

[0121] Step 3: Each layer is processed by linear transformation and nonlinear activation function.

[0122] Step 4: Calculate the output layer feature representation. The output calculation method of each layer is:

[0123] a p,i =Rez(W p,i ×ap,i-1 +b p,i )

[0124] In the formula, a p,i represents the activation output of the i-th layer of the neural network; a p,i-1 is the output of the i-1th layer of the neural network; Rez() is the adaptive ReLU (Rectified Linear Unit) activation function.

[0125] In one embodiment, let the input of the adaptive ReLU activation function be x zsy , the calculation method of the adaptive ReLU activation function is expressed as:

[0126] Rez(x zsy ) = γ p ×max(0,x zsy )+ξ p ×min(0,x zsy )

[0127] In the formula, γ p and p are the adaptive coefficients of the positive and negative parts respectively, max( ) is the maximum value function, and min( ) is the minimum value function. Preferably, γ p Set to 0.3, ξ p Set to 0.7.

[0128] Step 5: Use an exploratory gradient descent update strategy to predict possible gradient changes before each update of the neural network parameters. Adjust the step size and direction of the gradient update according to the prediction results to optimize the parameter adjustment during training and avoid falling into the local optimum too early. The calculation method is expressed as:

[0129]

[0130] In the formula, ΔW p,i and Δb p,i are the update amounts of the i-th layer weight and i-th layer bias of the neural network respectively; η p is the learning rate of the neural network; is the loss function with respect to parameter W p,i The gradient of is the loss function with respect to parameter b p,i The gradient of is the loss function of the neural network; α p is the exploration intensity coefficient; is the prediction function of the gradient. Preferably, η p Set to 0.03, Cross entropy loss is used.

[0131] In one embodiment, the exploratory gradient prediction function uses historical gradient data to perform linear regression prediction, and the calculation method is expressed as:

[0132]

[0133] In the formula, Represents the loss function at the t-1th iteration with respect to the parameter W p,i The gradient of t is the attenuation factor; α p is the exploration intensity coefficient; Tgz is the total number of iterations. Preferably, β t Set to 0.98, alpha p Set to 2.

[0134] Furthermore, the weight and bias parameters of the neural network are updated by using the error back propagation method.

[0135] Step 6: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0136] 4. Feature denoising model training

[0137] The feature denoising model is pre-trained based on the inversely reconstructed autoencoder, which includes an encoder and a decoder. The encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data. The feature denoising model is used to perform noise reduction on the feature-extracted data.

[0138] In one embodiment, the feature denoising model can be pre-trained based on an inversely reconstructed autoencoder in the following manner:

[0139] Each iteration is performed as follows until the second maximum number of iterations is reached:

[0140] Initializing parameters of the autoencoder, wherein the parameters of the autoencoder include weights and bias parameters of the autoencoder;

[0141] Input the sample data into the encoder and decoder in sequence, and obtain the output data reconstructed by the decoder;

[0142] According to the sample data and the output data reconstructed by the decoder, the inverse reconstruction loss of the autoencoder is calculated using the following formula:

[0143]

[0144] Where, L ris the inverse reconstruction loss; Ntr is the number of samples input in the current batch; || || represents the L2 norm; λ r is the regularization parameter; D KL (P r ||Q r ) is the KL divergence, x r,i is the input data of the encoder; is the output data reconstructed by the decoder;

[0145] Based on the gradient descent method, the parameters of the autoencoder are updated according to the inverse reconstruction loss.

[0146] During the specific implementation, the data after feature extraction is input into the feature denoising model to train the feature denoising model. The embodiment of the present invention adopts an autoencoder based on inverse reconstruction as the feature denoising model, and the autoencoder includes two parts: an encoder and a decoder. The encoder is responsible for compressing the input data into a low-dimensional representation, and the decoder reconstructs the data from this low-dimensional representation. In order to enhance the adaptability of the feature dimensionality reduction of the model, the embodiment of the present invention adopts an inverse reconstruction strategy to process the high-dimensional data after input feature extraction, so as to minimize the dependency between different features, thereby improving the quality of encoding and the accuracy of decoding.

[0147] Figure 3 This is a schematic diagram of the feature denoising model training process in an embodiment of the present invention, referring to Figure 3 , the training process of the autoencoder algorithm based on inverse reconstruction is introduced.

[0148] Step 1: Initialize the parameters of the autoencoder, including the weight and bias parameters of the autoencoder. In one embodiment, the initialization method is expressed as:

[0149]

[0150] Where W r,init is the initial weight matrix of the autoencoder; b r,init is the initial bias of the autoencoder; n r,in and n r,iout Represent the number of neurons in the input and output layers respectively; U(-1,1) represents a random number uniformly distributed in the interval.

[0151] Step 2: Input the feature-extracted data into the autoencoder, where encoding: converts it into a low-dimensional space representation, and the low-dimensional representation is then decoded: reconstructed into an output with the same dimension as the original data. In the forward propagation process, the encoder output calculation method is expressed as:

[0152] z r =Sig(W r,enc × r +b r,enc );

[0153] And, the reconstructed output of the decoder is:

[0154]

[0155] In the formula, z r is the low-dimensional feature representation of the code; Sig( ) is the Sigmoid activation function; W r,enc and W r,dec are the weights of the encoder and decoder respectively; b r,enc and b r,dec are the biases of the encoder and decoder respectively; x r is the input data; is the output of the reconstruction.

[0156] Step 3: Calculate the reverse reconstruction loss. The calculation method of the reverse reconstruction loss function is defined as:

[0157]

[0158] Where, L r is the inverse reconstruction loss function; Ntr is the number of samples input in the current batch; || || represents the L2 norm; λ r is the regularization parameter; D KL (P r ||Q r ) is the KL (Kullback-Leibler Divergence, abbreviated as KL, KL divergence is sometimes also called relative entropy) divergence, x r,i and They represent the input data and reconstructed output of the i-th iteration respectively.

[0159] In one embodiment, KL divergence measures the difference between the original data distribution and the reconstructed data distribution, and is used to ensure that the key features of the data are not lost during the dimensionality reduction process. The calculation method is expressed as:

[0160]

[0161] In the formula, p r,i and q r,i are the probability distributions of the original data and the reconstructed data in the i-th dimension, respectively, and Mtr is the feature dimension before feature dimensionality reduction.

[0162] Step 4: Use the gradient descent method to update the weights and biases. Specifically, according to the gradient calculated by the loss function, the weights and biases of the encoder and decoder are updated using the gradient descent algorithm. The update method is expressed as:

[0163]

[0164] Where η r is the learning rate; is the updated encoder weight; is the updated encoder bias; is the updated decoder weight; is the updated decoder bias.

[0165] Furthermore, the gradient For example, the calculation method is expressed as:

[0166]

[0167] In the formula, represents the partial derivative of the loss function with respect to the reconstructed output, is the partial derivative during decoding, is the partial derivative during encoding.

[0168] Step 5: Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1200 times.

[0169] 5. Classifier model training

[0170] The classifier model is pre-trained based on the extreme learning machine. During the training process, the denoised data is processed by adaptive Hadamard transform. The processed data is passed to the output layer, where the loss function is calculated. The training extreme learning machine model parameters are updated through the back propagation algorithm based on the loss function.

[0171] In one embodiment, the classifier model is trained as follows:

[0172] Each iteration is performed as follows until the third maximum number of iterations is reached:

[0173] Initializing parameters of an extreme learning machine, wherein the parameters of the extreme learning machine include weights and bias parameters of the extreme learning machine;

[0174] According to the following formula, the input sample data is processed by adaptive Hadamard transform to obtain the processed data:

[0175]

[0176]

[0177] In the formula, is the processed data output by the linear transformation of the lth layer, is the processed data output after activation of the lth layer, is the Hadamard matrix, is the Sigmoid activation function, represents the Hadamard product;

[0178] Use the cross entropy loss function to calculate the loss using the processed data;

[0179] Based on the calculated loss, the back-propagation algorithm is used to calculate the gradient and update the parameters of the extreme learning machine.

[0180] During implementation, the reduced-dimensional data is input into a classifier to train the classifier model. The embodiment of the present invention adopts an extreme learning machine algorithm based on an adaptive Hadamard transform as a classification algorithm. In order to improve the classification accuracy and adaptability of the extreme learning machine, the embodiment of the present invention adopts an adaptive Hadamard transform strategy based on the traditional extreme learning machine. In the hidden layer, the Hadamard transform is used instead of the traditional linear weighted sum to increase the model's ability to process nonlinear features.

[0181] Specifically, the training process of the extreme learning machine algorithm based on adaptive Hadamard transform is as follows:

[0182] 5.1. Initialize the weight and bias parameters of the extreme learning machine. In one embodiment, the weight and bias of the extreme learning machine are initialized using a strategy based on an adaptive Hadamard transform, which is expressed as:

[0183]

[0184] In the formula, represents the weight of the lth layer of the extreme learning machine, represents the bias of the lth layer of the extreme learning machine, and denote the number of nodes in the lth and lth layers respectively, Generate a A matrix whose elements are uniformly distributed in [0, 1].

[0185] 5.2. The input data after dimension reduction is processed by adaptive Hadamard transform to generate the activation output of the hidden layer. Specifically, let the input data be X u , which is expressed by Hadamard transform:

[0186]

[0187] In the formula, is the linear transformation output of the lth layer, is the output after activation of the lth layer, H u is the Hadamard matrix, Sig() is the Sigmoid activation function, and ⊙ represents the Hadamard product.

[0188] In one embodiment, the Hadamard matrix H u The generation method is expressed as:

[0189]

[0190] In the formula, bit(i&j) counts the number of "1" in the binary representation of the i-th sample and the j-th sample, d u is the dimension of the Hadamard matrix. Preferably, d u Choose to be a power of 2 close to the input dimension.

[0191] 5.3. The output of the hidden layer is passed to the output layer, and the loss value of the output layer is calculated. The embodiment of the present invention uses a cross entropy loss function to calculate the loss, which is expressed as:

[0192]

[0193] In the formula, represents the loss function of the extreme learning machine, N is the number of samples, K is the number of categories, and y ik is the true label of the i-th iteration, is the predicted probability of the i-th iteration.

[0194] Furthermore, the back propagation algorithm is used to calculate the gradient and update the weight. The calculation method of the weight update amount is expressed as:

[0195]

[0196] In the formula, is the update amount of the extreme learning machine's weight, and η is the learning rate of the extreme learning machine. Preferably, η is set to 0.001.

[0197] 5.4. The error gradient is calculated through the back propagation algorithm according to the loss function, and the weight of the model is adjusted using the dynamic update mechanism to enhance the stability and speed of learning. The update method is expressed as:

[0198]

[0199] In the formula, is the momentum term, and γ is the momentum coefficient. Preferably, γ is set to a value close to 1, such as 0.9.

[0200] 5.5. Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0201] In summary, the model training is completed.

[0202] When the above three models are formally used, the optical signal fed back by the VBE to be tested receiving the test chassis is input into the feature extraction model for feature extraction, the feature extracted data is input into the trained feature denoising model, the denoised data is output, the denoised data is input into the trained classifier model, and the output optical signal recognition result includes the optical signal category, which reflects the thyristor status. The thyristor status includes but is not limited to the normal state and the following fault states: short circuit, overcurrent, overvoltage, and reverse breakdown fault.

[0203] Afterwards, the optical signal description information received from the test chassis is compared with the predicted optical signal identification result, and the monitoring logic verification result of the VBE to be tested is determined according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal.

[0204] Figure 4 FIG. 1 is an example of a method for verifying the validity of the valve base electronic equipment monitoring logic in an embodiment of the present invention, referring to Figure 4 , the method includes a model training process and a model use process. Both the model training process and the model use process are described from the perspectives of input, steps, and output. The model training process includes collecting and labeling the original training data, using the original training data set to perform a data expansion model, obtaining a trained data expansion model and an expanded training data set, and then using the expanded training data set to train a feature extraction model, obtaining a trained feature extraction model and a training data set after feature extraction, and so on, finally obtaining a trained data expansion model, feature extraction model, feature denoising model, and classifier model. In the process of using the model, the aforementioned trained data expansion model, feature extraction model, feature denoising model, and classifier model are used to process the collected new samples in turn to evaluate the effectiveness of the monitoring logic.

[0205] In summary, the embodiments of the present invention:

[0206] 1. To address the problems of poor model generalization ability and limited accuracy caused by insufficient number of training samples, the embodiments of the present invention use a generative adversarial network based on quantum coding to generate samples, optimize sample quality through quantum information compression technology and quantum error correction loss function, and enhance the generalization ability of the model.

[0207] 2. In order to solve the problems of gradient disappearance, gradient explosion or falling into local optimum that may occur in the traditional neural network during the feature extraction process, the embodiment of the present invention adopts an exploratory gradient descent neural network to adjust the learning step size and direction by predicting the gradient change trend, thereby improving training stability and model performance.

[0208] 3. To address the problems of strong feature dependency and low encoding quality of high-dimensional data, the embodiment of the present invention uses an autoencoder algorithm based on inverse reconstruction to improve encoding quality and decoding accuracy by minimizing the dependency between features.

[0209] 4. In order to improve the ability of the extreme learning machine to process nonlinear features, the embodiment of the present invention adopts an adaptive Hadamard transform strategy to enhance the classification accuracy and adaptability of the model to complex data.

[0210] Therefore, the embodiments of the present invention have the following beneficial effects:

[0211] 1. Through an AI-based approach, the validity of the thyristor monitoring logic of all thyristor stages of the converter valve can be verified from the VBE side in a short time, and the verification results are both accurate and efficient.

[0212] 2. Through quantum coding technology, the embodiment of the present invention effectively solves the problem of insufficient sample quantity. The generated samples have high fidelity, which significantly improves the quality and efficiency of model training.

[0213] 3. The exploratory gradient descent method makes the feature extraction process more stable, avoids common problems in traditional neural network training, and improves the generalization ability of features.

[0214] 4. The reverse reconstruction strategy is adopted to optimize the feature dimensionality reduction process, reduce the mutual interference between features, and improve the model's ability to identify and utilize key features.

[0215] 5. The use of adaptive Hadamard transform enhances the extreme learning machine's ability to process nonlinear features, improves the classification accuracy and the adaptability of the model.

[0216] The embodiment of the present invention also provides a valve base electronic device monitoring logic validity verification device, as described in the following embodiment. Since the principle of solving the problem by the device is similar to the valve base electronic device monitoring logic validity verification method, the implementation of the device can refer to the implementation of the valve base electronic device monitoring logic validity verification method, and the repeated parts will not be repeated.

[0217] Figure 5 FIG. 1 is a schematic diagram of a valve base electronic equipment monitoring logic validity verification device according to an embodiment of the present invention. Figure 5 As shown, the device comprises:

[0218] The data acquisition module 501 is used to receive the optical signal fed back by the test chassis after the VBE to be tested sends an optical trigger signal to the test chassis; the test chassis simulates the function of the thyristor of the converter valve;

[0219] The feature extraction module 502 is used to input the optical signal into the trained feature extraction model and output the feature-extracted data; the feature extraction model pre-uses the historical optical signal fed back to the VBE by the thyristor of the converter valve to train the multi-layer fully connected neural network, and the exploratory gradient descent is used to update the neural network parameters during the training process;

[0220] A feature denoising module 503 is used to input the feature-extracted data into a trained feature denoising model and output the denoised data; the feature denoising model is pre-trained based on an autoencoder of inverse reconstruction, the autoencoder includes an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data;

[0221] The optical signal category recognition module 504 is used to input the denoised data into the trained classifier model and output the optical signal recognition result; the optical signal recognition result includes the optical signal category, and the optical signal category reflects the state of the thyristor; the classifier model is pre-trained based on the extreme learning machine, and the denoised data is subjected to adaptive Hadamard transform processing during the training process. The processed data is transmitted to the output layer, and the loss function is calculated at the output layer. The training extreme learning machine model parameters are updated through the back propagation algorithm according to the loss function;

[0222] The result output module 505 is used to receive the optical signal description information sent by the test chassis; the optical signal description information includes the optical signal category; compare the optical signal identification result and the optical signal description information, and determine the monitoring logic verification result of the VBE to be tested according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal.

[0223] In one embodiment, the apparatus further comprises:

[0224] A data format conversion module, used to convert the optical signal into vector data before the feature extraction module 502 inputs the optical signal into the trained feature extraction model; the vector data is stored in JSON format;

[0225] The feature extraction module 502 is specifically used to input vector data into a trained feature extraction model.

[0226] In one embodiment, the feature extraction model includes an input layer, a plurality of hidden layers and an output layer, each hidden layer has a different number of neurons, and each layer processes input data using a preset activation function;

[0227] During the training process, exploratory gradient descent is used to update the neural network parameters, including:

[0228] Before updating the neural network parameters in each iteration, the historical gradient linear regression is used to predict the gradient change of this iteration;

[0229] The neural network parameters are updated using the gradient change of this iteration; wherein a first maximum number of iterations is preset.

[0230] In one embodiment, the feature denoising model is pre-trained based on an inversely reconstructed autoencoder in the following manner:

[0231] Each iteration is performed as follows until the second maximum number of iterations is reached:

[0232] Initializing parameters of the autoencoder, wherein the parameters of the autoencoder include weights and bias parameters of the autoencoder;

[0233] Input the sample data into the encoder and decoder in sequence, and obtain the output data reconstructed by the decoder;

[0234] According to the sample data and the output data reconstructed by the decoder, the inverse reconstruction loss of the autoencoder is calculated using the following formula:

[0235]

[0236] Where, L r is the inverse reconstruction loss; Ntr is the number of samples input in the current batch; || || represents the L2 norm; λ r is the regularization parameter; D KL (P r ||Q r ) is the KL divergence, x r,i is the input data of the encoder; is the output data reconstructed by the decoder;

[0237] Based on the gradient descent method, the parameters of the autoencoder are updated according to the inverse reconstruction loss.

[0238] In one embodiment, the classifier model is trained as follows:

[0239] Each iteration is performed as follows until the third maximum number of iterations is reached:

[0240] Initializing parameters of an extreme learning machine, wherein the parameters of the extreme learning machine include weights and bias parameters of the extreme learning machine;

[0241] According to the following formula, the input sample data is processed by adaptive Hadamard transform to obtain the processed data:

[0242]

[0243] In the formula, is the processed data output by the linear transformation of the lth layer, is the processed data output after activation of the lth layer, is the weight of the lth layer of the extreme learning machine, H u is the Hadamard matrix, X u For input data, is the bias of the lth layer of the extreme learning machine, Sig() is the Sigmoid activation function, and ⊙ represents the Hadamard product;

[0244] Use the cross entropy loss function to calculate the loss using the processed data;

[0245] Based on the calculated loss, the back-propagation algorithm is used to calculate the gradient and update the parameters of the extreme learning machine.

[0246] In one embodiment, the apparatus further comprises:

[0247] The data expansion module is used to obtain the historical optical signal fed back to the VBE by the thyristor of the converter valve before the VBE to be tested sends the optical trigger signal to the test chassis, and to label, preprocess and extract features of the historical optical signal fed back to the VBE by the thyristor of the converter valve to obtain training samples; to expand the training samples using a generative adversarial network to obtain expanded training samples; and to obtain a feature extraction model, a feature denoising model and a classifier model based on the expanded training samples.

[0248] In one embodiment, the thyristor state includes a normal state and the following fault states: short circuit, overcurrent, overvoltage, and reverse breakdown fault.

[0249] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned valve base electronic device monitoring logic validity verification method when executing the computer program.

[0250] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for verifying the validity of the valve base electronic device monitoring logic is implemented.

[0251] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned valve base electronic device monitoring logic validity verification method is implemented.

[0252] In the embodiment of the present invention, the test chassis is used to simulate the function of the thyristor of the converter valve, the VBE to be tested sends an optical trigger signal to the test chassis, and the test chassis simulates the feedback optical signal of the thyristor of the converter valve. Without being connected to the real working converter valve, the effectiveness verification of the monitoring logic of the valve-based electronic equipment can be realized. During the verification, the trained feature extraction model is used to extract features, the trained feature denoising model is used to denoise features, and the trained classifier model is used to classify and identify features to determine the optical signal recognition result. By comparing the optical signal category in the optical signal recognition result with the optical signal description information sent by the test chassis, it is determined whether the monitoring logic of the VBE to be tested is normal or abnormal. Compared with the technical solutions in the prior art, the embodiment of the present invention uses artificial intelligence technology to perform test verification, which can reduce the dependence on testers. At the same time, the method uses an exploratory gradient descent neural network for feature extraction, which improves the stability and model performance of the feature extraction model. The autoencoder algorithm based on inverse reconstruction is used to minimize the dependency between features and the influence of noise, improve the encoding quality and decoding accuracy, and adopts adaptive Hadamard transform processing to enhance the classification accuracy and adaptability of the model to complex data, which can ensure the accuracy of the test.

[0253] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0254] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0255] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0256] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0257] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for verifying the validity of valve base electronic equipment monitoring logic, characterized in that: include: After the valve base electronic device VBE to be tested sends an optical trigger signal to the test chassis, an optical signal fed back by the test chassis is received; the test chassis simulates the function of the thyristor of the converter valve; Input the optical signal into the trained feature extraction model and output the feature-extracted data; The feature extraction model pre-uses the historical: optical signal fed back to the VBE by the thyristor of the converter valve to train the multi-layer fully connected neural network, and uses exploratory gradient descent to update the neural network parameters during the training process; Input the feature-extracted data into the trained feature denoising model and output the denoised data; The feature denoising model is pre-trained based on an autoencoder for inverse reconstruction, the autoencoder comprising an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data; The denoised data is input into a trained classifier model, and an optical signal recognition result is output; the optical signal recognition result includes an optical signal category, and the optical signal category reflects the state of the thyristor; the classifier model is pre-trained based on an extreme learning machine, and an adaptive Hadamard transform is performed on the denoised data during the training process. The processed data is passed to the output layer, and a loss function is calculated at the output layer. The parameters of the trained extreme learning machine model are updated through a back propagation algorithm according to the loss function; the thyristor state includes a normal state and the following fault states: short circuit, overcurrent, overvoltage, and reverse breakdown fault; receiving optical signal description information sent by the test chassis; The optical signal description information includes the optical signal category; Comparing the optical signal identification result with the optical signal description information, and determining the monitoring logic verification result of the VBE to be tested according to the comparison result; The monitoring logic verification result includes whether the monitoring logic is normal or abnormal; Before the VBE to be tested sends an optical trigger signal to the test chassis, the method further includes: obtaining the optical signal fed back to the VBE by the historical converter valve thyristor, labeling, preprocessing, and feature extraction of the optical signal fed back to the VBE by the historical converter valve thyristor, and obtaining a training sample; using a generative adversarial network to expand the training sample to obtain an expanded training sample; training based on the expanded training sample to obtain a feature extraction model, a feature denoising model, and a classifier model; the training process of the generative adversarial network is as follows: First, initialize the network parameters of the generator and discriminator; the initialization parameters are implemented using Gaussian distribution; Then, the generator receives random noise and uses quantum information compression method to process the noise and generate quantum state ψ: ψ= Q(z;θ q ); Where z is the random noise of the generator input; θ q represents the parameters of the quantum state generation function; Q() is the quantum state generation function; the calculation method of the quantum state generation function Q() is expressed as: Q(z;θ q )=U(θ’ q )×H(z); In the formula, H(z) represents the random noise Through a series of Hadamard gates, it is converted into a superposition state of the quantum bit; U(θ' q ) represents the parameter θ q Applied quantum logic gates; θ' q is the adjusted quantum gate parameter; where θ q Use the attention-based adjustment module for adjustment; The sampling of random noise z follows the following distribution: z~ N (m',s' 2 ); Among them, μ is the mean of the noise, σ 2 is the variance of the noise, μ' and σ' 2 are the mean and variance of the adjusted noise respectively, and the adjustment method is expressed as: μ'=μ+β cg × S(f;σ s ); in 2= s 2 × exp(S(f;σ s )); In the formula, β cg is the adjustment factor, S( ) is the adaptive noise adjustment function, f is the random noise vector, σ s is the parameter of the adaptive noise adjustment function; the calculation method of the adaptive noise adjustment function S( ) is expressed as: S(f;σ s )=tanh(W s ×f+b s ); Where W s and b s are the weight and bias parameters of the adaptive noise adjustment function, tanh() is the hyperbolic tangent function; Afterwards, the quantum state ψ passes through a quantum classical conversion function C, which converts the quantum information into classical data that can be used for judgment, expressed as: x'=C(ψ;θ c ); Where x' is the classical data, specifically the vector data in the same format as the original collected data; θ c represents the parameters of the quantum-classical conversion function; C( ) is the quantum-to-classical conversion operation; Afterwards, the discriminator evaluates the generated classical data and the input real data, and uses the quantum error correction loss function L QEC The constraints for the training process are expressed as: L QEC (x,x';θ d )= -log(D(x;θ d ))-log(1-D(x';θ d )); In the formula, θ d represents the parameters of the discriminator; D is the discriminator function, the probability that the output sample is a true sample; x is a true data sample; Afterwards, according to the output of the discriminator and the quantum error correction loss function, the parameters θ of the generator G are adjusted. g , update parameters, expressed as: ； In the formula, is the generator parameter of the current iteration; η vb is the learning rate; is the gradient of the loss function with respect to the generator parameters; Repeat the above steps until the preset stop iteration condition is met; The classifier model is trained as follows: Each iteration is performed as follows until the third maximum number of iterations is reached: Initializing parameters of an extreme learning machine, wherein the parameters of the extreme learning machine include weights and bias parameters of the extreme learning machine; According to the following formula, the input sample data is processed by adaptive Hadamard transform to obtain the processed data: ； In the formula, is the processed data output by the linear transformation of the lth layer, is the processed data output after activation of the lth layer, is the weight of the lth layer of the extreme learning machine, H u is the Hadamard matrix, X u For input data, is the bias of the lth layer of the extreme learning machine, Sig() is the Sigmoid activation function, and ⊙ represents the Hadamard product; Use the cross entropy loss function to calculate the loss using the processed data; Based on the calculated loss, the back-propagation algorithm is used to calculate the gradient and update the parameters of the extreme learning machine.

2. The method for verifying the validity of valve base electronic equipment monitoring logic according to claim 1, characterized in that: Before the optical signal is input into the trained feature extraction model, it also includes: Convert the optical signal into vector data; the vector data is stored in JSON format; The optical signal is fed into a trained feature extraction model, including: Input the vector data into the trained feature extraction model.

3. The method for verifying the validity of valve base electronic equipment monitoring logic according to claim 1, characterized in that: The feature extraction model includes an input layer, multiple hidden layers and an output layer, each hidden layer has a different number of neurons, and each layer processes input data using a preset activation function; During the training process, exploratory gradient descent is used to update the neural network parameters, including: Before updating the neural network parameters in each iteration, the historical gradient linear regression is used to predict the gradient change of this iteration; The neural network parameters are updated using the gradient change of this iteration; wherein a first maximum number of iterations is preset.

4. The method for verifying the validity of valve base electronic equipment monitoring logic according to claim 1, characterized in that: The feature denoising model is pre-trained based on the inverse reconstruction autoencoder in the following manner: Each iteration is performed as follows until the second maximum number of iterations is reached: Initializing parameters of the autoencoder, wherein the parameters of the autoencoder include weights and bias parameters of the autoencoder; Input the sample data into the encoder and decoder in sequence, and obtain the output data reconstructed by the decoder; According to the sample data and the output data reconstructed by the decoder, the inverse reconstruction loss of the autoencoder is calculated using the following formula: ； Where, L r is the inverse reconstruction loss; Ntr is the number of samples input in the current batch; || || represents the L2 norm; λ r is the regularization parameter; D KL (P r ||Q r ) is the KL divergence, x r,i is the input data of the encoder; is the output data reconstructed by the decoder; Based on the gradient descent method, the parameters of the autoencoder are updated according to the inverse reconstruction loss.

5. A valve base electronic equipment monitoring logic validity verification device, characterized in that: include: A data acquisition module, used for receiving an optical signal fed back by the test chassis after the VBE to be tested sends an optical trigger signal to the test chassis; The test chassis simulates the function of the thyristor of the converter valve; A feature extraction module is used to input the optical signal into a trained feature extraction model and output feature-extracted data; The feature extraction model pre-uses the historical: optical signal fed back to the VBE by the thyristor of the converter valve to train the multi-layer fully connected neural network, and uses exploratory gradient descent to update the neural network parameters during the training process; The feature denoising module is used to input the feature-extracted data into the trained feature denoising model and output the denoised data; The feature denoising model is pre-trained based on an autoencoder for inverse reconstruction, the autoencoder comprising an encoder and a decoder, the encoder is used to compress the feature-extracted data into low-dimensional representation data, and the decoder is used to reconstruct the low-dimensional representation data into data with the same dimension as the feature-extracted data; An optical signal category recognition module is used to input the denoised data into a trained classifier model and output an optical signal recognition result; the optical signal recognition result includes the optical signal category, and the optical signal category reflects the thyristor state; the classifier model is pre-trained based on an extreme learning machine, and the denoised data is subjected to adaptive Hadamard transform processing during the training process. The processed data is transmitted to the output layer, and the loss function is calculated at the output layer. The parameters of the trained extreme learning machine model are updated through a back propagation algorithm according to the loss function; the thyristor state includes a normal state and the following fault states: short circuit, overcurrent, overvoltage, and reverse breakdown fault; A result output module, used for receiving the optical signal description information sent by the test chassis; The optical signal description information includes the optical signal category; Comparing the optical signal identification result and the optical signal description information, and determining the monitoring logic verification result of the VBE to be tested according to the comparison result; the monitoring logic verification result includes whether the monitoring logic is normal or abnormal; The data expansion module is used to obtain the optical signal fed back by the thyristor of the converter valve to the VBE before the VBE to be tested sends the optical trigger signal to the test chassis, and to label, preprocess, and extract features of the optical signal fed back by the thyristor of the converter valve to the VBE to obtain training samples; to expand the training samples using a generative adversarial network to obtain expanded training samples; to obtain a feature extraction model, a feature denoising model, and a classifier model based on the expanded training samples; the training process of the generative adversarial network is as follows: First, initialize the network parameters of the generator and discriminator; the initialization parameters are implemented using Gaussian distribution; Then, the generator receives random noise and uses quantum information compression method to process the noise and generate quantum state ψ: ψ= Q(z;θ q ); Where z is the random noise of the generator input; θ q represents the parameters of the quantum state generation function; Q() is the quantum state generation function; the calculation method of the quantum state generation function Q() is expressed as: Q(z;θ q )=U(θ’ q )×H(z); In the formula, H(z) represents the random noise Through a series of Hadamard gates, it is converted into a superposition state of the quantum bit; U(θ' q ) represents the parameter θ q Applied quantum logic gates; θ' q is the adjusted quantum gate parameter; where θ q Use the attention-based adjustment module for adjustment; The sampling of random noise z follows the following distribution: z~ N (m',s' 2 ); Among them, μ is the mean of the noise, σ 2 is the variance of the noise, μ' and σ' 2 are the mean and variance of the adjusted noise respectively, and the adjustment method is expressed as: μ'=μ+β cg × S(f;σ s ); in 2= s 2 × exp(S(f;σ s )); In the formula, β cg is the adjustment factor, S( ) is the adaptive noise adjustment function, f is the random noise vector, σ s is the parameter of the adaptive noise adjustment function; the calculation method of the adaptive noise adjustment function S( ) is expressed as: S(f;σ s )=tanh(W s ×f+b s ); Where W s and b s are the weight and bias parameters of the adaptive noise adjustment function, tanh() is the hyperbolic tangent function; Afterwards, the quantum state ψ passes through a quantum classical conversion function C, which converts the quantum information into classical data that can be used for judgment, expressed as: x'=C(ψ;θ c ); Where x' is the classical data, specifically the vector data in the same format as the original collected data; θ c represents the parameters of the quantum-classical conversion function; C( ) is the quantum-to-classical conversion operation; Afterwards, the discriminator evaluates the generated classical data and the input real data, and uses the quantum error correction loss function L QEC The constraints for the training process are expressed as: L QEC (x,x';θ d )= -log(D(x;θ d ))-log(1-D(x';θ d )); In the formula, θ d represents the parameters of the discriminator; D is the discriminator function, the probability that the output sample is a true sample; x is a true data sample; Afterwards, according to the output of the discriminator and the quantum error correction loss function, the parameters θ of the generator G are adjusted. g , update parameters, expressed as: ； In the formula, is the generator parameter of the current iteration; η vb is the learning rate; is the gradient of the loss function with respect to the generator parameters; Repeat the above steps until the preset stop iteration condition is met; The classifier model is trained as follows: Each iteration is performed as follows until the third maximum number of iterations is reached: Initializing parameters of an extreme learning machine, wherein the parameters of the extreme learning machine include weights and bias parameters of the extreme learning machine; According to the following formula, the input sample data is processed by adaptive Hadamard transform to obtain the processed data: ； In the formula, is the processed data output by the linear transformation of the lth layer, is the processed data output after activation of the lth layer, is the weight of the lth layer of the extreme learning machine, H u is the Hadamard matrix, X u For input data, is the bias of the lth layer of the extreme learning machine, Sig() is the Sigmoid activation function, and ⊙ represents the Hadamard product; Use the cross entropy loss function to calculate the loss using the processed data; Based on the calculated loss, the back-propagation algorithm is used to calculate the gradient and update the parameters of the extreme learning machine.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for verifying the validity of the valve base electronic device monitoring logic according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for verifying the validity of the valve base electronic device monitoring logic according to any one of claims 1 to 4 is implemented.

8. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method for verifying the validity of the valve base electronic device monitoring logic according to any one of claims 1 to 4 is implemented.

Images