Method and device for predicting service life of MOSFET device of intelligent electric meter, terminal equipment and storage medium

Abstract

The invention discloses a service life prediction method and device of an intelligent ammeter MOSFET device, terminal equipment and a storage medium. The method comprises the steps of firstly obtaining accelerated aging data of a to-be-tested MOSFET device in a preset time period; inputting the accelerated aging data into a preset life prediction model, performing convolution and pooling on the accelerated aging data, and extracting local features; according to the local features, capturing long-term dependency relationships among parameters in the accelerated aging data at different moments in forward propagation and backward propagation processes, and generating corresponding context features; and the context features are mapped after being spliced, and the residual life of the MOSFET device to be tested is obtained. According to the invention, the demand of high-precision prediction of the residual life of the MOSFET of the intelligent electric meter can be realized by capturing the dynamic change characteristics of the related parameters of the device in the long-time high-frequency switching process.

Description

Technical Field

[0001] The present invention relates to the technical field of life prediction, and particularly to a method, device, terminal device and storage medium for predicting the life of an MOSFET device in an intelligent meter. Background Art

[0002] In a power system, intelligent meters are usually used for power monitoring and management. With the long-term operation of the meters, the metal oxide semiconductor field effect transistors (MOSFETs) in the meters will be affected by various environmental stresses and load fluctuations, resulting in the gradual degradation of their performance, and then affecting the reliability of the intelligent meters and the stable operation of the power system.

[0003] Traditional life prediction methods usually rely on empirical models or statistical formulas to predict the life by assuming a linear relationship between the life and influencing factors. However, in actual situations, the life is affected by complex non-linear interactions of multiple factors. For example, during the long-term high-frequency switching process of the MOSFET, key parameters such as its voltage and on-resistance will change, and these changes are potential omens of device degradation. Therefore, traditional life prediction methods have great limitations in capturing the characteristics of these dynamic changes and are difficult to meet the requirements of intelligent meters for high-precision life prediction. Summary of the Invention

[0004] The present invention provides a method, device, terminal device and storage medium for predicting the life of an MOSFET device in an intelligent meter, which can capture the dynamic change characteristics of relevant parameters during the long-term high-frequency switching process of the MOSFET and meet the requirement of accurately predicting the remaining life of the MOSFET in the intelligent meter.

[0005] An embodiment of the present invention provides a method for predicting the life of an MOSFET device in an intelligent meter, including:

[0006] Obtaining the accelerated aging data of the MOSFET device to be tested in the intelligent meter within a preset time period; wherein, the accelerated aging data includes: drain current, drain-source voltage, gate voltage, case temperature, flange temperature, on-resistance and sampling frequency at each moment within the preset time period;

[0007] Inputting the accelerated aging data into a preset life prediction model, so that the preset life prediction model performs convolution and pooling on the accelerated aging data to extract local features; according to the local features, capture the long-term dependence relationship between the parameters in the accelerated aging data at different moments during the forward propagation and backward propagation processes respectively, and generate corresponding context features based on the long-term dependence relationship; after splicing the context features, map the spliced context features to obtain the remaining life of the MOSFET device to be tested.

[0008] Further, before inputting the above acceleration aging data into the preset life prediction model, it also includes:

[0009] According to a preset sliding window, smooth the above acceleration aging data according to the moving average method, and normalize the smoothed acceleration aging data.

[0010] Further, the training of the above preset life prediction model includes:

[0011] Obtain a number of acceleration aging sample data with real labels; wherein, the above real labels are used to represent the real remaining life of the MOSFET devices corresponding to the above acceleration aging sample data;

[0012] Divide a number of acceleration aging sample data into pre-training samples and formal training samples;

[0013] Pre-train the life prediction model according to the above pre-training samples, determine the hyperparameters of the preset life prediction model, and obtain the pre-trained life prediction model;

[0014] Train the pre-trained life prediction model according to the above formal training samples, determine the weights and bias parameters of the preset life prediction model, and obtain the trained preset life prediction model.

[0015] Further, the above pre-training the life prediction model according to the above pre-training samples, determining the hyperparameters of the preset life prediction model, and obtaining the pre-trained life prediction model includes:

[0016] Generate a number of particles according to a preset hyperparameter threshold range; wherein, initially, each particle corresponds to an initial particle position and an initial particle velocity; the particle position is used to represent the hyperparameter value in the life prediction model;

[0017] Repeat the model pre-training operation until the current iterative pre-training times are not less than the preset iterative times, and generate the pre-trained life prediction model according to the hyperparameter value corresponding to the current global optimal particle position;

[0018] Wherein, the above model pre-training operation includes:

[0019] Obtain the current particle positions and current particle velocities of all particles, and generate the current fitness of each particle according to the current particle positions of all particles and the above pre-training samples; wherein, the initial particle position is the above initial particle position, and the initial particle velocity is the above initial particle velocity;

[0020] Update the current individual optimal particle position, the current individual optimal fitness, the current global optimal particle position, and the current global optimal particle fitness according to the current fitness of all particles;

[0021] Determine whether the current iteration number is less than the preset iteration number. If so, for each particle, calculate the particle velocity at the next moment according to the current particle position, the current particle velocity, the current latest individual optimal particle position, and the current latest global optimal particle position; Calculate the particle position at the next moment according to the particle velocity at the next moment and the current particle position.

[0022] Furthermore, generate the current fitness of each particle according to the current particle positions of all particles and the above pre-training samples, including:

[0023] Generate several current life prediction models according to the current particle positions of all particles;

[0024] Input the above pre-training samples into all current life prediction models respectively. For each current life prediction model, predict the current first predicted life;

[0025] Calculate the current fitness of each particle according to the current first predicted life and the corresponding true label.

[0026] Furthermore, the above update of the current individual optimal particle position, the current individual optimal fitness, the current global optimal particle position, and the current global optimal particle fitness according to the current fitness of all particles includes:

[0027] For each particle, compare the current fitness with the individual optimal fitness of the current individual optimal particle position; Among them, each particle corresponds to a current individual optimal particle position and a current individual optimal fitness;

[0028] If the current fitness is less than the current individual optimal fitness, use the current particle position as the updated individual optimal particle position and the current fitness as the updated individual optimal fitness; Otherwise, do not update the current individual optimal particle position and the current individual optimal fitness;

[0029] Extract the current minimum fitness from the current fitnesses of all particles, and compare the current minimum fitness with the global optimal fitness of the current global optimal particle position;

[0030] When the current minimum fitness is less than the current global optimal fitness, the particle position corresponding to the current minimum fitness is used as the updated global optimal particle position, and the current minimum fitness is used as the updated global optimal fitness; otherwise, the current global optimal particle position and the current global optimal fitness are not updated.

[0031] Further, training the pre-trained life prediction model according to the above formal training samples to determine the weights and bias parameters of the preset life prediction model, and obtaining the trained preset life prediction model, including:

[0032] Input the formal training samples into the pre-trained life prediction model for iterative training until the loss function converges, and generate the above preset life prediction model;

[0033] Among them, in each iterative training, according to the current pre-training samples, the current second predicted life is predicted; according to the current second predicted life and the corresponding true label, the current loss function is calculated, and it is judged whether the current loss function converges; if the current loss function converges, the current life prediction model is used as the above preset life prediction model; otherwise, the weights and bias parameters of the current life prediction model are adjusted.

[0034] Based on the above method item embodiments, the present invention correspondingly provides device item embodiments;

[0035] The present invention provides a life prediction device for an intelligent meter MOSFET device, including:

[0036] An accelerated aging data acquisition module and a remaining life prediction module;

[0037] The above accelerated aging data acquisition module is used to acquire the accelerated aging data of the MOSFET device to be tested of the intelligent meter within a preset time period; wherein, the above accelerated aging data includes: drain current, drain-source voltage, gate voltage, case temperature, flange temperature, on-resistance, and sampling frequency at each moment within the above preset time period;

[0038] The above remaining life prediction module is used to input the above accelerated aging data into a preset life prediction model, so that the preset life prediction model performs convolution and pooling on the above accelerated aging data to extract local features; according to the above local features, respectively capture the long-term dependence relationships between the parameters in the above accelerated aging data at different moments during the forward propagation and backward propagation processes, and generate corresponding context features based on the above long-term dependence relationships; after splicing the above context features, map the spliced context features to obtain the remaining life of the MOSFET device to be tested.

[0039] Based on the above method item embodiments, the present invention correspondingly provides a terminal device item embodiment;

[0040] The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the above memory and configured to be executed by the above processor. When the above processor executes the above computer program, it implements a method for predicting the life of an intelligent meter MOSFET device according to any one of the embodiments of the present invention.

[0041] Based on the above method item embodiments, the present invention correspondingly provides a storage medium item embodiment;

[0042] The present invention provides a storage medium, including a processor, a memory, and a computer program stored in the above memory and configured to be executed by the above processor. When the above processor executes the above computer program, it implements a method for predicting the life of an intelligent meter MOSFET device according to any one of the embodiments of the present invention.

[0043] The embodiments of the present invention have the following beneficial effects:

[0044] The present invention provides a method, device, terminal device, and storage medium for predicting the life of an intelligent meter MOSFET device. The above method includes: first, obtaining the accelerated aging data of the MOSFET device to be tested in the intelligent meter within a preset time period; wherein, the above accelerated aging data includes: drain current, drain-source voltage, gate voltage, case temperature, flange temperature, on-resistance, and sampling frequency at each moment within the above preset time period; then inputting the above accelerated aging data into a preset life prediction model, so that the preset life prediction model performs convolution and pooling on the above accelerated aging data to extract local features; according to the above local features, respectively capture the long-term dependence relationships between the parameters in the above accelerated aging data at different moments during the forward propagation and backward propagation processes, and generate corresponding context features based on the above long-term dependence relationships; after splicing the above context features, map the spliced context features to obtain the remaining life of the MOSFET device to be tested. Therefore, the present invention first extracts the local features of the accelerated aging data through a preset life prediction model, and then captures the long-term dependence relationships in both the forward and backward directions of the time series of the accelerated aging data during the two-way propagation process to obtain the relevant features of the mutual influence of the parameters in the accelerated aging data in the time series. Finally, the captured context features are mapped through a fully connected layer to obtain the final life prediction result, meeting the requirements for high-precision life prediction of MOSFET devices. Description of the Drawings

[0045] Figure 1It is a schematic flow chart of a method for predicting the life of a MOSFET device in an intelligent electric meter provided by an embodiment of the present invention.

[0046] Figure 2 It is a schematic diagram of the principle of a preset life prediction model provided by an embodiment of the present invention.

[0047] Figure 3 It is a schematic structural diagram of a device for predicting the life of a MOSFET device in an intelligent electric meter provided by an embodiment of the present invention. Detailed implementation manners

[0048] Next, the technical solutions in the present invention will be clearly and completely described in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] As Figure 1 shown, a method for predicting the life of a MOSFET device in an intelligent electric meter provided by an embodiment of the present invention includes:

[0050] Step S101: Obtain the accelerated aging data of the MOSFET device to be tested in the intelligent electric meter within a preset time period; wherein, the above-mentioned accelerated aging data includes: drain current, drain-source voltage, gate voltage, case temperature, flange temperature, on-resistance, and sampling frequency at each moment within the above-mentioned preset time period;

[0051] Specifically, the above-mentioned accelerated aging data is the data obtained after performing a thermal cycle overstress test on the MOSFET device. This test aims to simulate the influence brought by the mismatch of the thermal expansion coefficients between different materials by generating thermo-mechanical stress inside the component. The specific experimental method includes performing a power cycle test on the device without using an external radiator.

[0052] Specifically, when performing a thermal cycle overstress test on the MOSFET device, the gate voltage is a square wave signal of 15V, the frequency is 1kHz, and the duty cycle is 40%. At the same time, the drain-source voltage is biased with a direct current of 4V, and a 0.2Ω resistive load is connected to the output end of the device. By measuring the case temperature of the MOSFET device and using it as the control variable of the thermal cycle, the aging process of the MOSFET device is monitored. During the experiment, the drain current, drain-source voltage, gate voltage, case temperature, flange temperature, and sampling frequency of the MOSFET device at each moment are recorded in real time. At the same time, based on the above data, the on-resistance R DS(on) is calculated as the main indicator of device degradation.

[0053] Specifically, the above sampling frequency is divided into the sampling frequency in the steady state and the sampling frequency in the transient state. The sampling frequency in the transient state is relatively high, about 400 kHz, which can capture the transient characteristics when the MOSFET is switched.

[0054] Step S102: Input the above acceleration aging data into a preset life prediction model, so that the preset life prediction model performs convolution and pooling on the acceleration aging data to extract local features; according to the local features, capture the long-term dependence relationships among the parameters in the acceleration aging data at different times during the forward propagation and backward propagation processes respectively, and generate corresponding context features based on the long-term dependence relationships; splice the context features and then map the spliced context features to obtain the remaining life of the MOSFET device to be tested.

[0055] Specifically, the above preset life prediction model is mainly obtained by cascading a CNN and a deep multi-layer BiLSTM. When inputting data into the preset life prediction model, first, the convolution kernel in the CNN convolution layer slides on the input data to perform local weighted summation operations to capture short-term dependencies and local patterns in the sequence and generate local features. Subsequently, the BiLSTM is used to learn from both the past and future directions of the data through bidirectional propagation, so as to more comprehensively capture the long-term dependence relationships in the time series and generate context features.

[0056] Specifically, in the BiLSTM, there are LSTM layers in two directions. Each LSTM layer controls the flow of information through a gating mechanism (input gate, forget gate, and output gate) to avoid the problem of gradient disappearance. Among them, the forget gate is used to control the forgetting degree of the memory cell C in the acceleration aging data at the previous moment, and its update formula is: t-1 At the current moment, its update formula is:

[0057] f t =σ(W f ·[h t-1 ,x t +b f )

[0058] In the formula, f t represents the forgetting degree at the current moment, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, h t-1 represents the hidden state output of the output gate at the previous moment, x t represents the acceleration aging data at the current moment, and b f represents the bias term of the forget gate.

[0059] The input gate is used to represent the acceleration aging data x at the current moment t, and the hidden state output h of the output gate at the previous moment t-1 The update contribution to the memory cell C at the current moment t is given by the update formula:

[0060] i t = σ(W f · [h t-1 , x t + b f )

[0061]

[0062] In the formula, i t represents the input gate, which is used to control the degree to which data at the current moment enters the memory cell. represents the candidate memory value at the current moment, tanh represents the hyperbolic tangent activation function, W c represents the weight matrix of the candidate memory cell, and b c represents the bias term of the candidate memory cell.

[0063] The output gate, which is used to control the memory cell C in the data of accelerated aging at the current moment t , and its influence on the hidden state output h of the output gate at the current moment t is given by the update formula:

[0064] o t = σ(W o · [h t-1 , x t + b o )

[0065] h t = o t * tanh(C t )

[0066] In the formula, o t represents the output gate, which is used to control the output of the hidden state h t at the current moment, W o represents the weight matrix of the output gate, and b o represents the bias term of the output gate.

[0067] Specifically, the BiLSTM processes sequence data through two LSTM layers in two directions, from front to back and from back to front respectively, and finally concatenates the context features in the two directions. The specific update formula is:

[0068]

[0069] In the formula, represents the context feature obtained from front to back, Represents the context features obtained from back to front, and [;] represents the concatenation operation. Represents the accelerated aging data processed by the forward-backward LSTM layer at the current moment. Represents the hidden state output of the output gate of the forward-backward LSTM layer at the previous moment. Represents the accelerated aging data processed by the backward-forward LSTM layer at the current moment. Represents the hidden state output of the output gate of the backward-forward LSTM layer at the previous moment.

[0070] Schematically, the schematic diagram of the preset life prediction model is as Figure 2 shown. Figure 2 The "initialized data set" in represents the accelerated aging data after smoothing and normalization.

[0071] Preferably, since the change of local features in the aging process of MOSFET devices is of great significance in life prediction, capturing local features through CNN can improve the prediction accuracy of the preset life prediction model for the remaining life. At the same time, the BiLSTM in the preset life prediction model can not only utilize the forward historical information, but also integrate the reverse information flow, which makes BiLSTM perform well in the life prediction task and is especially suitable for the life prediction of MOSFET devices in smart meters.

[0072] In a preferred embodiment, before inputting the above-mentioned accelerated aging data into the preset life prediction model, it further includes:

[0073] According to the preset sliding window, smooth the above-mentioned accelerated aging data according to the moving average method, and normalize the smoothed accelerated aging data.

[0074] Specifically, the accelerated aging data is smoothed by the following formula:

[0075]

[0076] In the formula, x s represents the s-th accelerated aging data after smoothing, s represents the length of the preset sliding window, x i represents the i-th accelerated aging data before smoothing, and n represents the number of accelerated aging data.

[0077] Specifically, after smoothing, the maximum value and the minimum value are extracted from the smoothed accelerated aging data, and the data is normalized by the maximum value and the minimum value. After normalization, the aging data can be mapped to the value range [0,1] to ensure the comparability between data with different dimensions. The smoothed accelerated aging data is normalized by the following formula:

[0078]

[0079] where x t represents the t-th accelerated aging data after normalization, min(x s ) represents the minimum value, and max(x s ) represents the maximum value.

[0080] In this preferred embodiment, after sequentially performing smoothing processing and normalization on the original accelerated aging data, the input data for the preset life prediction model is obtained.

[0081] In another preferred embodiment, the training of the above preset life prediction model includes:

[0082] Obtaining a number of accelerated aging sample data with true labels; wherein, the above true labels are used to represent the true remaining life of the MOSFET devices corresponding to the above accelerated aging sample data;

[0083] Dividing a number of accelerated aging sample data into pre-training samples and formal training samples;

[0084] Specifically, the MOSFET aging data set provided by the NASA Prognostics Center of Excellence can be used as the above-mentioned number of accelerated aging sample data with true labels. There can be aging data of multiple different MOSFET devices under thermal overstress in the accelerated aging sample data, and multiple MOSFET devices can respectively show different degradation trajectories of on-resistance. The number of MOSFET devices selected in the present invention is 6.

[0085] Pre-training the life prediction model according to the above pre-training samples, determining the hyperparameters of the preset life prediction model, and obtaining the pre-trained life prediction model;

[0086] Specifically, the above hyperparameters include: the number of neurons in the BiLSTM hidden layer, the number of neurons in the fully connected hidden layer, and the learning rate.

[0087] Specifically, using the particle swarm optimization algorithm, regarding the number of neurons in the BiLSTM hidden layer, the number of neurons in the fully connected hidden layer, and the learning rate as the positions of particles, through continuous updating of the particle positions during the iteration process, the optimal combination of these hyperparameters is searched, thereby improving the prediction performance of the model.

[0088] Specifically, the particle swarm optimization algorithm finds the optimal solution by simulating the behavior of bird flocks or fish schools. Each particle represents a potential solution, and the position and velocity of the particle are the core variables of the algorithm. Initially, the positions and velocities of the particles are randomly generated. The individual optimal position and the global optimal position are, initially, the initial positions of the particles. For each particle, the hyperparameter configuration corresponding to its position is applied to the life prediction model. If the fitness value of the current particle is better than the fitness value of its individual optimal position, then the individual optimal position is updated. If the fitness value of the current particle is better than the fitness value of the global optimal position, then the global optimal position is updated. The fitness value reflects the quality of the hyperparameter combination corresponding to the particle. By comparing the fitness value of the current particle with the fitness values of its individual optimal and global optimal positions, it is determined whether to update these optimal positions.

[0089] According to the above formal training samples, the pre-trained life prediction model is trained to determine the weights and bias parameters of the preset life prediction model, and the trained preset life prediction model is obtained.

[0090] Specifically, through the backpropagation algorithm, the pre-trained life prediction model is iteratively trained to optimize the weights and bias parameters of the preset life prediction model, and the trained preset life prediction model is obtained.

[0091] In another preferred embodiment, the above-mentioned pre-training of the life prediction model according to the above pre-training samples to determine the hyperparameters of the preset life prediction model and obtain the pre-trained life prediction model includes:

[0092] Generate a number of particles according to the preset hyperparameter threshold range; wherein, initially, each particle corresponds to an initial particle position and an initial particle velocity; the particle position is used to represent the hyperparameter value in the life prediction model;

[0093] Specifically, in the present invention, the threshold range of the number of BiLSTM hidden layer neurons in the hyperparameters is set to [10, 100], the threshold range of the number of fully connected hidden layer neurons is set to [10, 100], and the threshold range of the learning rate is set to [0.0001, 0.001].

[0094] Specifically, after setting the hyperparameter threshold range, randomly generate the positions and velocities of N particles. The particle positions correspond to hyperparameter value combinations, and initialize the individual optimal position and the global optimal position of each particle.

[0095] Repeat the model pre-training operation until the current number of iterative pre-training times is not less than the preset number of iterations, and generate the pre-trained life prediction model according to the hyperparameter values corresponding to the current global optimal particle position;

[0096] Among them, the above model pre-training operation includes:

[0097] Obtain the current particle positions and current particle velocities of all particles, and generate the current fitness of each particle according to the current particle positions of all particles and the above pre-trained samples; wherein, the initial particle position is the above initial particle position, and the initial particle velocity is the above initial particle velocity;

[0098] Update the current personal best particle position, current personal best fitness, current global best particle position, and current global best particle fitness according to the current fitness of all particles;

[0099] Determine whether the current iteration number is less than the preset iteration number. If so, for each particle, calculate the particle velocity at the next moment according to the current particle position, current particle velocity, current latest personal best particle position, and current latest global best particle position; Calculate the particle position at the next moment according to the particle velocity at the next moment and the current particle position.

[0100] Specifically, if the updated particle position exceeds the hyperparameter threshold range, boundary processing is required, and the exceeded part is truncated to the boundary of the above hyperparameter threshold range.

[0101] Specifically, calculate the particle velocity of each particle at the next moment through the following formula:

[0102]

[0103] In the formula, represents the particle velocity of the i-th particle in the d-th dimension at the next moment, w represents the inertia weight, represents the particle velocity of the i-th particle in the d-th dimension at the current moment, c 1 and c 2 both represent learning factors, r 1 and r 2 represent random numbers in the interval [0, 1], pBest i,d represents the current latest personal best position of the i-th particle in the d-th dimension, represents the particle position of the i-th particle in the d-th dimension at the current moment, gBest d represents the current latest global best position.

[0104] Specifically, calculate the particle position of each particle at the next moment through the following formula:

[0105]

[0106] In the formula, represents the particle position of the i-th particle in the d-th dimension at the next moment.

[0107] Preferably, through the above two update formulas for particle position and particle velocity, the particles gradually approach the individual optimal position and the global optimal position, thereby finding the optimal solution in the search space. Boundary processing ensures that the particle positions are always within the defined range. Finally, the optimal hyperparameter combination is determined, and the final global optimal position is the optimal hyperparameter combination, including the learning rate, the number of neurons in the BiLSTM hidden layer, and the number of neurons in the fully connected hidden layer.

[0108] In this preferred embodiment, the life prediction model is pre-trained by the particle swarm optimization algorithm to determine the optimal hyperparameters of the preset life prediction model.

[0109] In another preferred embodiment, according to the current particle positions of all particles and the above pre-training samples, the current fitness of each particle is generated, including:

[0110] According to the current particle positions of all particles, a number of current life prediction models are generated;

[0111] The above pre-training samples are respectively input into all current life prediction models, and for each current life prediction model, the current first predicted life is obtained;

[0112] According to the current first predicted life and the corresponding true label, the current fitness of each particle is calculated.

[0113] Specifically, during the entire pre-training process, the MSE (mean squared error) loss function value is used as the fitness value. MSE is obtained by calculating the square of the prediction error and taking the average to get the fitness. The fitness is calculated by the following formula:

[0114]

[0115] In the formula, MSE represents the fitness, y i represents the true label corresponding to the i-th pre-training sample, represents the first predicted life corresponding to the i-th pre-training sample, n 1 represents the number of pre-training samples.

[0116] In this preferred embodiment, the MSE (mean squared error) loss function value is calculated through the first predicted life and the corresponding true label to obtain the fitness value.

[0117] In another preferred embodiment, the above updates the current individual optimal particle position, the current individual optimal fitness, the current global optimal particle position, and the current global optimal particle fitness according to the current fitness of all particles, including:

[0118] For each particle, compare the current fitness with the individual best fitness of the current individual best particle position; wherein, each particle corresponds to a current individual best particle position and a current individual best fitness.

[0119] In the case where the current fitness is less than the current individual best fitness, take the current particle position as the updated individual best particle position and take the current fitness as the updated individual best fitness; otherwise, do not update the current individual best particle position and the current individual best fitness.

[0120] Extract the current minimum fitness from the current fitnesses of all particles, and compare the current minimum fitness with the global best fitness of the current global best particle position.

[0121] In the case where the current minimum fitness is less than the current global best fitness, take the particle position corresponding to the current minimum fitness as the updated global best particle position and take the current minimum fitness as the updated global best fitness; otherwise, do not update the current global best particle position and the current global best fitness.

[0122] Illustratively, assume that currently there are particles A, B, and C, where the current fitness of particle A is 0.6 and the current individual best fitness is 0.4; the current fitness of particle B is 0.5 and the current individual best fitness is 0.5; the current fitness of particle C is 0.3 and the current individual best fitness is 0.7; the current global best fitness is 0.35. Then for particle A, its individual best fitness remains unchanged at 0.4; for particle B, its individual best fitness remains unchanged at 0.5; for particle C, its individual best fitness is updated to 0.3. For the current global best fitness, it is updated to the current fitness of particle C, i.e., 0.3.

[0123] In this preferred embodiment, update the current individual best particle position, the current individual best fitness, the current global best particle position, and the current global best particle fitness based on the comparison results of the current fitnesses of all particles, the current individual best particle positions, the current individual best fitnesses, the current global best particle positions, and the current global best particle fitnesses.

[0124] In another preferred embodiment, training the pre-trained life prediction model according to the above formal training samples to determine the weights and bias parameters of the preset life prediction model, and obtaining the trained preset life prediction model includes:

[0125] Input the formal training samples into the pre-trained life prediction model for iterative training until the loss function converges, and generate the above-mentioned preset life prediction model;

[0126] Among them, in each iterative training, according to the current pre-training samples, predict the current second predicted life; according to the current second predicted life and the corresponding true label, calculate the current loss function, and determine whether the current loss function converges; if the current loss function converges, use the current life prediction model as the above-mentioned preset life prediction model; otherwise, adjust the weights and bias parameters of the current life prediction model.

[0127] Specifically, use the MSE (Mean Squared Error) loss function as the above-mentioned loss function, and in each formal training, determine whether the model is trained completed by judging whether it converges.

[0128] Preferably, when each second predicted life is obtained, it can be compared with the corresponding true label to evaluate the prediction accuracy of the model at different time points and different working conditions. Analyze the distribution of errors, and focus on the time periods and working conditions with relatively large prediction deviations to identify possible deficiencies or inaccurate areas in the prediction process of the model.

[0129] Preferably, deploy the trained preset life prediction model in the smart meter. By implementing the obtained accelerated aging data, the remaining life of the MOSFET device can be continuously monitored, providing reliable data support for the preventive maintenance and life management of the power system. At the same time, through continuous monitoring and real-time life prediction, the performance of the preset life prediction model in the real working environment can be verified, and the accuracy and stability of the model in long-term use can be detected. According to the detection results, timely feedback adjustment and optimization of the preset life prediction model are carried out to ensure that the model can adapt to environmental changes and new data, so as to continuously provide high-quality life prediction services and ultimately achieve precise control of the life of the MOSFET device.

[0130] In this preferred embodiment, the pre-trained life prediction model is trained with formal training samples to further determine the weights and bias parameters of the preset life prediction model, and obtain the trained preset life prediction model.

[0131] Based on the above method item embodiments, the present invention correspondingly provides device item embodiments.

[0132] As Figure 3 shown, an embodiment of the present invention provides a life prediction device for a MOSFET device in a smart meter, including:

[0133] An accelerated aging data acquisition module and a remaining life prediction module;

[0134] The above-mentioned accelerated aging data acquisition module is used to acquire the accelerated aging data of the MOSFET device to be tested in the intelligent meter within a preset time period; wherein, the above-mentioned accelerated aging data includes: drain current, drain-source voltage, gate voltage, case temperature, flange temperature, on-resistance, and sampling frequency at each moment within the above-mentioned preset time period;

[0135] The above-mentioned remaining life prediction module is used to input the above-mentioned accelerated aging data into a preset life prediction model, so that the preset life prediction model performs convolution and pooling on the above-mentioned accelerated aging data to extract local features; according to the above-mentioned local features, respectively capture the long-term dependence relationships between the parameters in the above-mentioned accelerated aging data at different moments during the forward propagation and backward propagation processes, and generate corresponding context features based on the above-mentioned long-term dependence relationships; after splicing the above-mentioned context features, map the spliced context features to obtain the remaining life of the MOSFET device to be tested.

[0136] It should be noted that the device embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationships between the modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement without creative efforts. The above schematic diagram is only an example of a life prediction device for an intelligent meter MOSFET device, and does not constitute a limitation on a life prediction device for an intelligent meter MOSFET device, and may include more or fewer components than shown in the figure, or combine some components, or different components.

[0137] Based on the above method item embodiments, the present invention correspondingly provides terminal device item embodiments.

[0138] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the above memory and configured to be executed by the above processor. When the above processor executes the above computer program, it implements the life prediction method for an intelligent meter MOSFET device in any one of the above embodiments of the present invention.

[0139] Exemplarily, in this embodiment, the above computer program can be divided into one or more modules. The above one or more modules are stored in the above memory and executed by the above processor to implement the present invention. The above one or more module elements can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the above computer program in the above device;

[0140] The above terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The above device may include, but is not limited to, a processor and a memory;

[0141] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The above processor is the control center of the above device, and connects various parts of the entire device through various interfaces and lines;

[0142] The above memory can be used to store the above computer program and / or module. The above processor realizes various functions of the above device by running or executing the computer program and / or module stored in the above memory, and by calling the data stored in the memory. The above memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.; in addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0143] Based on the above method item embodiment, the present invention correspondingly provides a storage medium item embodiment.

[0144] Another embodiment of the present invention provides a storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the storage medium is located to execute the method for predicting the life of an intelligent meter MOSFET device according to any one of the embodiments of the present invention.

[0145] In this embodiment, the storage medium is a computer-readable storage medium. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0146] Compared with the prior art, by implementing the above various embodiments of the present invention, it is possible to capture the dynamic change characteristics of relevant parameters during the long-term high-frequency switching process of the MOSFET, and meet the requirement of accurately predicting the remaining life of the intelligent meter MOSFET.

[0147] The above is the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A life prediction method for a smart meter MOSFET device, characterized in that: include: Acquire accelerated aging data of the MOSFET device to be tested of the smart meter within a preset period of time; wherein the accelerated aging data includes: drain current, drain-source voltage, gate voltage, shell temperature, flange temperature, on-resistance and sampling frequency at each moment within the preset period of time; The accelerated aging data is input into a preset life prediction model so that the preset life prediction model performs convolution and pooling on the accelerated aging data to extract local features; according to the local features, the long-term dependencies between the parameters in the accelerated aging data at different times are captured during forward propagation and backward propagation, and corresponding context features are generated based on the long-term dependencies; after splicing the context features, the spliced ​​context features are mapped to obtain the remaining life of the MOSFET device to be tested.

2. The life prediction method of a smart meter MOSFET device according to claim 1, characterized in that: Before inputting the accelerated aging data into a preset life prediction model, the method further includes: According to a preset sliding window, the accelerated aging data is smoothed according to a sliding average method, and the smoothed accelerated aging data is normalized.

3. The life prediction method of a smart meter MOSFET device according to claim 2, characterized in that: The training of the preset life prediction model includes: Acquire a number of accelerated aging sample data with real labels; wherein the real labels are used to indicate the real remaining life of the MOSFET device corresponding to the accelerated aging sample data; Divide a number of accelerated aging sample data into pre-training samples and formal training samples; Pre-training the life prediction model according to the pre-training samples, determining the hyperparameters of the preset life prediction model, and obtaining the pre-trained life prediction model; The pre-trained life prediction model is trained according to the formal training samples, the weight and bias parameters of the preset life prediction model are determined, and the trained preset life prediction model is obtained.

4. The life prediction method of a smart meter MOSFET device according to claim 3, characterized in that: The pre-training of the life prediction model according to the pre-training samples, determining the hyper parameters of the preset life prediction model, and obtaining the pre-trained life prediction model includes: Generate a number of particles according to a preset hyperparameter threshold range; wherein, initially, each particle corresponds to an initial particle position and an initial particle velocity; the particle position is used to represent the hyperparameter value in the life prediction model; Repeat the model pre-training operation until the current number of iterative pre-training is not less than the preset number of iterations, and generate a pre-trained life prediction model according to the hyperparameter values ​​corresponding to the current global optimal particle position; The model pre-training operation includes: Obtaining the current particle positions and current particle velocities of all particles, and generating the current fitness of each particle according to the current particle positions of all particles and the pre-training samples; wherein the initial particle position is the initial particle position, and the initial particle velocity is the initial particle velocity; According to the current fitness of all particles, the current individual optimal particle position, the current individual optimal fitness, the current global optimal particle position and the current global optimal particle fitness are updated; Determine whether the current number of iterations is less than the preset number of iterations. If so, for each particle, calculate the particle velocity at the next moment based on the current particle position, the current particle velocity, the current latest individual optimal particle position, and the current latest global optimal particle position; calculate the particle position at the next moment based on the particle velocity at the next moment and the current particle position.

5. The life prediction method of a smart meter MOSFET device according to claim 4, characterized in that: According to the current particle positions of all particles and the pre-training samples, the current fitness of each particle is generated, including: According to the current particle positions of all particles, several current life prediction models are generated; Inputting the pre-trained samples into all current life prediction models respectively, and predicting a current first predicted life for each current life prediction model; According to the current first predicted lifetime and the corresponding true label, the current fitness of each particle is calculated.

6. The life prediction method of a smart meter MOSFET device according to claim 5, characterized in that: The method of updating the current individual optimal particle position, the current individual optimal fitness, the current global optimal particle position and the current global optimal particle fitness according to the current fitness of all particles includes: For each particle, the current fitness is compared with the individual optimal fitness of the current individual optimal particle position; wherein each particle corresponds to a current individual optimal particle position and a current individual optimal fitness; When the current fitness is less than the current individual optimal fitness, the current particle position is used as the updated individual optimal particle position, and the current fitness is used as the updated individual optimal fitness; otherwise, the current individual optimal particle position and the current individual optimal fitness are not updated; Extract the current minimum fitness from the current fitness of all particles, and compare the current minimum fitness with the global optimal fitness of the current global optimal particle position; When the current minimum fitness is less than the current global optimal fitness, the particle position corresponding to the current minimum fitness is used as the updated global optimal particle position, and the current minimum fitness is used as the updated global optimal fitness; otherwise, the current global optimal particle position and the current global optimal fitness are not updated.

7. The life prediction method of a smart meter MOSFET device according to claim 6, characterized in that: The pre-trained life prediction model is trained according to the formal training sample to determine the weight and bias parameters of the preset life prediction model to obtain the trained preset life prediction model, including: Inputting the formal training samples into the pre-trained life prediction model for iterative training until the loss function converges, thereby generating the preset life prediction model; Among them, in each iterative training, the current second predicted life span is predicted based on the current pre-training sample; the current loss function is calculated based on the current second predicted life span and the corresponding true label, and it is determined whether the current loss function converges; if the current loss function converges, the current life prediction model is used as the preset life prediction model; otherwise, the weight and bias parameters of the current life prediction model are adjusted.

8. A life prediction device for a smart meter MOSFET device, characterized in that: include: Accelerated aging data acquisition module and remaining life prediction module; The accelerated aging data acquisition module is used to acquire the accelerated aging data of the MOSFET device to be tested of the smart meter within a preset period of time; wherein the accelerated aging data includes: the drain current, drain-source voltage, gate voltage, shell temperature, flange temperature, on-resistance and sampling frequency at each moment within the preset period of time; The remaining life prediction module is used to input the accelerated aging data into a preset life prediction model so that the preset life prediction model performs convolution and pooling on the accelerated aging data to extract local features; according to the local features, in the process of forward propagation and backward propagation, the long-term dependencies between the parameters in the accelerated aging data at different times are captured, and corresponding context features are generated based on the long-term dependencies; after splicing the context features, the spliced ​​context features are mapped to obtain the remaining life of the MOSFET device to be tested.

9. A terminal 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, a life prediction method for a MOSFET device of a smart meter as claimed in any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the life prediction method for a MOSFET device of a smart meter as claimed in any one of claims 1 to 7.

Images