Service life prediction method and device of MOSFET device, terminal equipment and storage medium

Abstract

The invention discloses a service life prediction method and device of an MOSFET device, terminal equipment and a storage medium, and can input time sequence data of threshold voltage and on-state resistance of the MOSFET device in a time period into a service life prediction model, thereby capturing a long-term dependency relationship in a time sequence through the service life prediction model. The method comprises the following steps: analyzing the resistance and voltage of each time point in a time sequence to extract the time sequence characteristics of the threshold voltage and the on-state resistance which are subjected to non-linear change along with time, distributing different weights to the characteristics, and finally, combining each time sequence characteristic and the corresponding weight to obtain the time sequence characteristics of the threshold voltage and the on-state resistance. The voltage trend change and the resistance trend change of the device in the degradation process are accurately captured, so that the residual life of the device is obtained. According to the method, the nonlinear change trend of the threshold voltage and the on-state resistance in the device along with time can be effectively considered, the residual life of the device can be accurately predicted, and the accuracy of life prediction is improved.

Description

Technical Field

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

[0002] With the development of intelligent power system, smart meter is an important part of modern power monitoring and management. Its core power electronic device, metal oxide semiconductor field effect transistor (MOSFET), plays a vital role in power conversion, control and data processing. However, in long-term continuous operation, MOSFET devices in smart meters will be exposed to various environmental stresses and load fluctuations, which will lead to their gradual degradation. By predicting the life of MOSFET devices, potential problems can be discovered in time, preventive maintenance can be carried out, and the stable operation of the meter can be guaranteed.

[0003] Traditional MOSFET life prediction methods calculate and predict life based on empirical formulas or statistical models, and often focus on some macroscopic, easily measurable parameters for statistics, such as working hours, workload, etc., rather than specific performance parameters inside the device. Affected by multiple factors such as temperature and load fluctuations, performance parameters such as threshold voltage and on-state resistance inside MOSFET devices will change nonlinearly over time, and traditional prediction methods cannot effectively cope with such nonlinear changes, and cannot accurately capture the actual degradation of MOSFET devices, resulting in inaccurate device life prediction results. Summary of the invention

[0004] The embodiments of the present invention provide a method, apparatus, terminal device and storage medium for predicting the life of a MOSFET device. The long-term dependency in a time series is captured by a life prediction model, so that the timing characteristics of the nonlinear change of the threshold voltage and the on-state resistance over time can be extracted. This can effectively solve the problem in the prior art that the nonlinear change trend of the performance parameters such as the threshold voltage and the on-state resistance inside the MOSFET device that change over time cannot be effectively dealt with, and the actual degradation of the MOSFET device cannot be accurately captured.

[0005] An embodiment of the present invention provides a method for predicting the life of a MOSFET device, comprising:

[0006] Acquire operation data corresponding to the MOSFET device within a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage;

[0007] Inputting the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, generate the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics;

[0008] Wherein, the generation of the life prediction model includes:

[0009] Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

[0010] Preferably, the life prediction model comprises: a hidden layer;

[0011] The method of capturing the long-term dependencies in the time series based on the operation data and generating the time series features corresponding to each moment includes:

[0012] The hidden layer is used to capture the long-term dependency between the current moment and the time series of the running data based on the gating mechanism at each moment, and generate the time series features corresponding to the current moment based on the long-term dependency corresponding to the current moment.

[0013] Preferably, the hidden layer comprises: a forget gate, an input gate and an output gate;

[0014] The gating mechanism is used to capture the long-term dependency between the current moment and the time series of the running data, and to generate the time series features corresponding to the current moment based on the long-term dependency corresponding to the current moment, including:

[0015] The forget gate is used to generate a first forgetting degree corresponding to the on-state resistance at a previous moment at a current moment and a second forgetting degree corresponding to the threshold voltage at a previous moment at a current moment based on the time series data in the operation data;

[0016] The input gate is used to generate a first update degree of the on-state resistance at the current moment for the timing characteristics at the previous moment based on the time series data in the operation data, and to generate a second update degree of the threshold voltage at the previous moment for the threshold voltage at the current moment;

[0017] The output gate is used to generate a first contribution degree of the on-resistance at the current moment to the current output based on the time series data in the operation data, and to generate a second contribution degree of the threshold voltage at the current moment to the current output;

[0018] The output gate is further used to generate a resistance timing characteristic at the current moment according to the on-state resistance, the first forgetting degree, the first updating degree and the first contribution degree at the current moment, and to generate a voltage timing characteristic at the current moment according to the threshold voltage, the second forgetting degree, the second updating degree and the second contribution degree at the current moment;

[0019] Among them, the resistance timing feature is the resistance change state between the current moment and the previous moment; the voltage timing feature is the voltage change state between the current moment and the previous moment; the resistance change state includes: resistance increase or resistance decrease; the voltage change state includes: voltage increase or voltage decrease.

[0020] Preferably, the lifespan prediction model comprises: an attention layer;

[0021] The method of assigning different weights to the time series features at each moment includes:

[0022] The attention layer is used to use the time series feature corresponding to the last moment in the time series data as a reference time series feature;

[0023] Calculate the similarity between the time series features at each moment and the reference time series features to obtain the similarity of each time series feature; wherein the similarity represents the contribution of the time series feature to the prediction of the remaining life;

[0024] Each similarity is normalized into a probability distribution through an activation function, and then the weight corresponding to the time series feature of each moment is generated based on the probability value of the probability distribution; wherein the weight represents the degree of attention of the model to the moment.

[0025] Preferably, the life prediction model comprises: a fully connected layer and an output layer;

[0026] The generating of the remaining life corresponding to the MOSFET device according to each timing feature and the corresponding weight includes:

[0027] The fully connected layer is used to perform weighted summation according to each time series feature and the weight corresponding to each time series feature, so as to generate a sequence feature for representing the global understanding of the model on the time series data;

[0028] Splicing the sequence feature with the time series feature corresponding to the last moment in the time series data to generate a spliced ​​feature vector; wherein the spliced ​​feature vector includes: global features and local features;

[0029] The output layer is used to generate the remaining life corresponding to the MOSFET device according to the spliced ​​feature vector.

[0030] Preferably, during each iterative training, an operation data sample is input into the life prediction model, so that the life prediction model generates a remaining life prediction result corresponding to the operation data sample according to the time series feature corresponding to each moment in the operation data sample and the weight of each time series feature;

[0031] The remaining life prediction results are compared with the actual remaining life results, and the network parameters of the life prediction model are adjusted according to the comparison results.

[0032] Preferably, it also includes:

[0033] The parameters of the life prediction model are updated according to the deviation value of the following loss function:

[0034]

[0035] Among them, MSE is the deviation value of the loss function, y i is the actual remaining life of a running data sample, is the predicted remaining life of an operating data sample, and n is the number of operating data samples.

[0036] Based on the above method embodiments, the present invention provides corresponding device embodiments.

[0037] An embodiment of the present invention provides a life prediction device for a MOSFET device, comprising: an operation data acquisition module and a remaining life prediction module;

[0038] The operation data acquisition module is used to acquire the operation data corresponding to the MOSFET device within a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage;

[0039] The remaining life prediction module is used to input the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, and then generate the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics;

[0040] Wherein, the generation of the life prediction model includes:

[0041] Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

[0042] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0043] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the life prediction method of a MOSFET device described in the above-mentioned embodiment of the invention is implemented.

[0044] Based on the above method embodiments, the present invention provides corresponding storage medium item embodiments.

[0045] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a MOSFET device life prediction method described in the above-mentioned invention embodiment.

[0046] The following beneficial effects are achieved by implementing the present invention:

[0047] The embodiment of the present invention provides a life prediction method, device, terminal equipment and storage medium for a MOSFET device. The present invention can input the time series data of the threshold voltage and on-state resistance of the MOSFET device within a time period into the life prediction model, so as to capture the long-term dependence in the time series through the life prediction model, that is, by analyzing the resistance and voltage at each time point in the time series, the time series characteristics of the nonlinear change of the threshold voltage and the on-state resistance over time can be extracted, and different weights are assigned to these characteristics, so that the model can pay more attention to those time points or characteristics that are more critical to the life prediction, and finally, by combining the time series characteristics at each moment and their corresponding weights, the voltage trend change and resistance trend change of the device in the degradation process are accurately captured, so as to obtain the remaining life of the MOSFET device. Compared with the prior art, the present invention can effectively consider the nonlinear change trend of the threshold voltage and on-state resistance inside the MOSFET device over time, and accurately predict the remaining life of the device, so that the actual degradation of the device can be more accurately captured, and the accuracy of the life prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of a method for predicting the life of a MOSFET device provided by an embodiment of the present invention.

[0049] Figure 2 It is a schematic structural diagram of a life prediction device for a MOSFET device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] like Figure 1 FIG. 1 is a flow chart of a method for predicting the life of a MOSFET device provided by an embodiment of the present invention. The method for predicting the life of a MOSFET device comprises:

[0052] Step S1: Acquire operation data corresponding to the MOSFET device in a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage;

[0053] Step S2: inputting the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, and then generating the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics;

[0054] Wherein, the generation of the life prediction model includes:

[0055] Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

[0056] The present invention can utilize the characteristics of the life prediction model in processing time series data, and combine it with the attention mechanism to deeply mine and analyze the operating data of MOSFET devices, so as to accurately predict their remaining life.

[0057] For step S1, the present invention can collect the operation data of the MOSFET device within a period of time from the smart meter, which mainly includes the time series data of the on-resistance and the time series data of the threshold voltage. Schematically, the above time series data reflect the dynamic changes of the device during operation.

[0058] In a preferred embodiment, the present invention can perform denoising on the original time series data of on-resistance and time series data of threshold voltage, such as using wavelet transform denoising or median filtering to reduce the influence of environmental noise on the data, thereby obtaining pre-processed time series data of on-resistance and time series data of threshold voltage;

[0059] The preprocessed data is normalized, for example, the min-max normalization method is used to scale each time series data to the range of 0 to 1 to ensure the comparability of data of different dimensions.

[0060] Therefore, by collecting time series data containing key parameters such as on-resistance and threshold voltage, the state changes of MOSFET devices in actual operation can be more comprehensively reflected, which can provide accurate feature inputs for the model and improve the accuracy of lifetime prediction.

[0061] For step S2, in a preferred embodiment, the life prediction model of the present invention is an LSTM network model, and the training process of the life prediction model includes:

[0062] The first step is to collect samples of the training set, and then perform power cycle tests on the device. During the test, the gate voltage is a 15V square wave signal with a frequency of 1kHz and a duty cycle of 40%. At the same time, the drain-source voltage is biased to 4V DC, and a 0.2Ω resistor load is connected to the output of the device. During the experiment, the aging process of the device is monitored by measuring the housing temperature and using it as a control variable for the thermal cycle. During the aging process, parameters such as the leakage current and drain-source voltage of the MOSFET device are recorded in real time. Based on these data, the threshold voltage and on-state resistance R are calculated. DS(on) , taking these two parameters as the main indicators of device degradation. For ease of analysis, the data was resampled and filtered to ensure consistent sampling frequency and remove noise. The final data set includes the aging data of the on-resistance time series data and the threshold voltage time series data, which can show the ΔR of each device respectively. DS(on) The degradation trajectory and threshold voltage degradation trajectory can be used to study the degradation behavior of MOSFET and develop and verify the lifetime prediction model.

[0063] Next, according to the specific needs of MOSFET device life prediction, a multi-layer LSTM network structure is built. The LSTM network structure includes an input layer, several recursive hidden layers, a fully connected layer, and an output layer. Schematically, the number of LSTM hidden layers and the number of neurons are determined by hyperparameter search to optimize the model's ability to extract time series features. In order to capture long-term dependencies, forward connections and attention mechanisms are introduced in each hidden layer to enhance the network's ability to capture complex time series patterns. Each recursive hidden layer in LSTM handles long-term dependencies in time series data by introducing forget gates, input gates, and output gates.

[0064] Among them, the specific update formulas corresponding to the forget gate, input gate and output gate are:

[0065] Forget gate: used to control the memory cell C of the previous moment t-1 The degree of forgetting at the current moment is expressed as:

[0066] f t =σ(W f ·[h t-1 , x t ]+b f )

[0067] Input gate: used to control the current input x t and the previous moment output h t-1 The contribution to the update of the current memory cell is:

[0068] i t =σ(W f ·[h t-1 , x t ]+b f )

[0069]

[0070]

[0071] Output gate: used to control the current memory cell C t For the current output h t The influence of is:

[0072] o t =σ(W o ·[h t-1 , x t ]+b o )

[0073] h t =o t *tanh(C t )

[0074] Among them, W f , W i , W C , W O are the weight matrices of the forget gate, input gate, candidate memory cell, and output gate respectively; b f , b i , b C , b o is the corresponding bias term; σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function; h t is the hidden state output at the current moment.

[0075] Furthermore, the LSTM model first processes the input time series data samples step by step through its hidden layer, and uses the hidden state of the previous moment and the current input state for calculation to capture the long-term dependency characteristics in the time series. During the training process, the gradient of the loss function relative to the model parameters can also be calculated through the back-propagation algorithm, and then the weights and biases of the model can be updated through the gradient descent method.

[0076] In order to further enhance the model's learning ability on time series, the time backpropagation algorithm can also be used to extend the gradient calculation to multiple time steps, thereby adjusting the model parameters more accurately.

[0077] Finally, during each iterative training, an operation data sample may be input into the life prediction model, so that the life prediction model generates a remaining life prediction result corresponding to the operation data sample according to the time series feature corresponding to each moment in the operation data sample and the weight of each time series feature;

[0078] The remaining life prediction results are compared with the actual remaining life results, and the network parameters of the life prediction model are adjusted according to the comparison results.

[0079] In a preferred embodiment, the embodiment of the present invention can also use the MSE mean square error loss function in the training process, by calculating the square of the prediction error and taking the average value, the formula is:

[0080]

[0081] Among them, MSE is the deviation value of the loss function, y i is the actual remaining life of a running data sample, is the predicted remaining life of a running data sample, and n is the number of running data samples. By minimizing the loss function, the model parameters can be gradually adjusted to improve the accuracy of the prediction. After multiple iterations of training, the model parameters tend to be stable, and finally an LSTM model that can effectively predict the remaining life of MOSFET devices is formed, that is, the life prediction model.

[0082] In a preferred embodiment, the present invention can also use accelerated aging data to verify the predictive performance of the model, specifically:

[0083] Compare the prediction results of the model with the actual data recorded in the aging experiment one by one. By comparing the error data obtained, calculate the prediction accuracy of the model at different time points and different working conditions. By analyzing the distribution of errors, you can check the time periods and working conditions where the prediction deviation is large, and identify areas where the model may be deficient or inaccurate during the prediction process. Based on these analysis results, adjust the model's hyperparameters, such as learning rate, number of LSTM layers, and number of neurons, or introduce regularization measures to reduce overfitting. Re-enter the accelerated aging data set to verify whether the adjusted model has improved in these key areas. Repeat multiple comparisons and verifications until the MSE value and error distribution of the model reach the expected performance standard and the optimal life prediction model is obtained.

[0084] Furthermore, the trained life prediction model can be deployed in the smart meter to obtain the operating data of the MOSFET device in real time; the real-time operating data can be input into the life prediction model to perform life prediction and generate a remaining life estimate.

[0085] Illustratively, when the life prediction model captures the long-term dependency in the time series based on the operating data, specifically:

[0086] The life prediction model includes: a hidden layer; the hidden layer is used to capture the long-term dependency between the current moment and the time series of the operating data based on the gating mechanism at each moment, and generate the time series features corresponding to the current moment based on the long-term dependency corresponding to the current moment.

[0087] Specifically, the gating mechanism of the hidden layer may be:

[0088] The forget gate is used to generate a first forgetting degree corresponding to the on-state resistance at a previous moment at a current moment and a second forgetting degree corresponding to the threshold voltage at a previous moment at a current moment based on the time series data in the operation data;

[0089] The input gate is used to generate a first update degree of the on-state resistance at the current moment for the timing characteristics at the previous moment based on the time series data in the operation data, and to generate a second update degree of the threshold voltage at the previous moment for the threshold voltage at the current moment;

[0090] The output gate is used to generate a first contribution degree of the on-resistance at the current moment to the current output based on the time series data in the operation data, and to generate a second contribution degree of the threshold voltage at the current moment to the current output;

[0091] The output gate is further used to generate a resistance timing characteristic at the current moment according to the on-state resistance, the first forgetting degree, the first updating degree and the first contribution degree at the current moment, and to generate a voltage timing characteristic at the current moment according to the threshold voltage, the second forgetting degree, the second updating degree and the second contribution degree at the current moment;

[0092] Among them, the resistance timing feature is the resistance change state between the current moment and the previous moment; the voltage timing feature is the voltage change state between the current moment and the previous moment; the resistance change state includes: resistance increase or resistance decrease; the voltage change state includes: voltage increase or voltage decrease.

[0093] It can be understood that through the action of the gating mechanism (forget gate, input gate, output gate), the model can capture the dynamic changes of the on-state resistance and threshold voltage of the MOSFET device during operation, and these dynamic changes reflect the degradation state and performance fluctuations of the device, that is, the aging degree and performance degradation of the device. The resistance timing characteristics at each moment and the voltage timing characteristics at each moment are the key information for predicting its remaining life.

[0094] Specifically, the embodiment of the present invention can generate resistance time series features and voltage time series features based on the resistance and voltage changes between the current moment and the previous moment, and the above resistance features and voltage features not only contain the information of the current moment, but also integrate the context information of the previous time step, thereby achieving a comprehensive understanding of the time series data. Moreover, through the gating mechanism of the hidden layer, the model can capture the long-term dependencies in the time series, so that the model can learn the mutual influence between the devices at different time steps or different moments, so as to more accurately predict their remaining life.

[0095] Specifically, the forget gate is used to control the forgetting degree of the memory cells (including information such as on-resistance and threshold voltage) at the previous moment at the current moment. By calculating the first forgetting degree and the second forgetting degree, the forget gate can determine which information should be retained and which information should be forgotten, thereby optimizing the model's learning process of time series features.

[0096] The input gate is used to control the update contribution of the current input (i.e. the timing characteristics of the current moment) and the output of the previous moment to the current memory cell (on-state resistance and threshold voltage). By calculating the first update degree and the second update degree, the input gate can determine how the new information affects the state of the current memory cell, thereby ensuring that the model can learn the key changes in the time series.

[0097] The output gate is used to control the contribution of the current memory cell (including the on-state resistance and threshold voltage after forgetting and updating) to the current output. By calculating the first contribution and the second contribution, the output gate can determine which information should be used to generate the final prediction result.

[0098] Therefore, the forget gate of the embodiment of the present invention controls the degree of forgetting of information so that the model ignores unimportant historical information during the prediction process, thereby more accurately capturing the changing trend of resistance and voltage. The input gate helps the model update the state of the memory cell by introducing new information to reflect the latest resistance and voltage data. The output gate generates the final resistance timing characteristics and voltage timing characteristics by integrating the information of the current memory cells, and these characteristics are directly used to predict the life of the MOSFET device. The present invention can accurately capture the changing state of resistance and voltage (including resistance increase / decrease, voltage increase / decrease) through the forget gate, input gate and output gate, so that the remaining life of the MOSFET device can be more accurately predicted, thereby also improving the reliability of the smart meter and the stability of the power system.

[0099] Furthermore, the life prediction model assigns different weights to the time series features at each moment, specifically:

[0100] The lifespan prediction model comprises: an attention layer;

[0101] The attention layer is used to use the time series feature corresponding to the last moment in the time series data as a reference time series feature;

[0102] Calculate the similarity between the time series features at each moment and the reference time series features to obtain the similarity of each time series feature; wherein the similarity represents the contribution of the time series feature to the prediction of the remaining life;

[0103] Each similarity is normalized into a probability distribution through an activation function, and then the weight corresponding to the time series feature of each moment is generated based on the probability value of the probability distribution; wherein the weight represents the degree of attention of the model to the moment.

[0104] It can be understood that the weight generation and allocation process of the present invention first takes the time series feature corresponding to the last moment in the time series data as the reference time series feature, and then calculates the similarity between the time series features at each moment and the reference time series feature to obtain the similarity of each time series feature, and this similarity represents the contribution of the time series feature to the remaining life prediction. Then, each similarity is normalized into a probability distribution through an activation function (such as a Softmax function), and these probability values ​​are the weights corresponding to the time series features at each moment, and the size of the weight represents the degree of attention the model pays to the moment.

[0105] The embodiment of the present invention uses the attention mechanism to assign different weights to the time series features at each moment, so that the model can pay more attention to the key moments that contribute more to the remaining life prediction, thereby improving the accuracy of the prediction. When faced with complex and changeable working conditions and noise data, the attention mechanism can help the model screen out important time series features, reduce the interference of noise data, and enhance the robustness of the model.

[0106] In a preferred embodiment, the present invention can introduce an attention mechanism, so that the state vector corresponding to each hidden layer (i.e., the timing characteristics at each moment) is similarly calculated with a reference vector to determine the importance of the time step in the entire sequence. The higher the similarity score, the greater the contribution of the time step to the life prediction of the MOSFET device. After calculating the similarity scores of all time steps, these scores are normalized into probability distributions through the Softmax function. These probability values ​​are the attention weights, which indicate the degree of attention the model pays to each time step.

[0107] After normalization by the Softmax function, the attention weight a at all times t It forms a probability distribution, namely:

[0108]

[0109] These attention weights reflect the degree of attention the model pays to each time step. The larger the weight, the more attention the model pays to the information of that time step.

[0110] In a preferred embodiment, the present invention can calculate and normalize the attention weights of all moments, apply these weights to the LSTM hidden state vector of each time step, and then perform weighted summation according to each time series feature and the weight corresponding to each time series feature to generate a context vector, and this context vector integrates the important information of all moments in the time series, such as the information of the key moments that are given higher weights by the attention mechanism. Schematically, the generated context vector represents the model's global understanding of the entire sequence, and is input into the subsequent layers of the model as a sequence feature representation that characterizes the model's global understanding of the time series data.

[0111] Specifically, the life prediction model further includes: a fully connected layer and an output layer; and the remaining life corresponding to the MOSFET device is generated according to each timing feature and the corresponding weight. The specific process is as follows:

[0112] The fully connected layer is used to perform weighted summation according to each time series feature and the weight corresponding to each time series feature, so as to generate a sequence feature for representing the global understanding of the model on the time series data;

[0113] Splicing the sequence feature with the time series feature corresponding to the last moment in the time series data to generate a spliced ​​feature vector; wherein the spliced ​​feature vector includes: global features and local features;

[0114] The output layer is used to generate the remaining life corresponding to the MOSFET device according to the spliced ​​feature vector.

[0115] It can be understood that by combining the generated context vector (i.e., the sequence features mentioned above) with the last hidden state of the LSTM layer, the combined vector, i.e., the concatenated feature vector, not only contains the model's global information about the entire time series, but also incorporates the key moment features highlighted by the attention mechanism, thereby ensuring that the model can utilize both global features and key local information when predicting the life of MOSFET devices, significantly improving the accuracy and reliability of the prediction.

[0116] After the combined feature vector (i.e., the concatenated feature vector) is input into the output layer, the concatenated feature vector is converted into a predicted value of the output space based on the mapping of the output layer, thereby generating an accurate prediction result for the remaining life of the MOSFET device.

[0117] In summary, the present invention introduces an attention mechanism to dynamically adjust the model's attention to inputs at different times, thereby more accurately capturing the degradation characteristics of MOSFET devices during operation. At the same time, through the processing of the fully connected layer and the output layer, an accurate prediction of the remaining life of the MOSFET device is achieved. This not only improves the accuracy of the prediction, but also provides real-time response for fault prevention of smart meters and stability of the power system.

[0118] The present invention fully mines the working data of MOSFET devices in smart meters under various stress environments, and combines the high-level abstract representation of device degradation characteristics with a deep learning model, such as generating the resistance timing characteristics and voltage timing characteristics corresponding to each moment, thereby improving the accuracy of life analysis of MOSFET devices and obtaining accurate life results.

[0119] like Figure 2 As shown, based on the above-mentioned various embodiments of the life prediction method of MOSFET devices, the present invention provides a corresponding device embodiment;

[0120] An embodiment of the present invention provides a life prediction device for a MOSFET device, comprising: an operation data acquisition module and a remaining life prediction module;

[0121] The operation data acquisition module is used to acquire the operation data corresponding to the MOSFET device within a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage;

[0122] The remaining life prediction module is used to input the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, and then generate the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics;

[0123] Wherein, the generation of the life prediction model includes:

[0124] Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

[0125] It should be noted that the device embodiments described above are merely schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without paying creative labor.

[0126] Those skilled in the art can clearly understand that, for the sake of convenience and simplicity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0127] Based on the above-mentioned embodiments of the life prediction method of various MOSFET devices, the present invention provides corresponding embodiments of terminal equipment items.

[0128] An embodiment of the present invention provides a terminal device, including 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 method for predicting the life of a MOSFET device as described in any method embodiment of the present invention is implemented.

[0129] The terminal device may be a computing terminal device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0130] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0131] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0132] Based on the above-mentioned embodiments of the life prediction method of various MOSFET devices, the present invention provides corresponding storage medium item embodiments.

[0133] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a life prediction method for a MOSFET device described in any method embodiment of the present invention.

[0134] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, the steps of each method embodiment described above can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0135] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting the life of a MOSFET device, characterized in that: include: Acquire operation data corresponding to the MOSFET device within a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage; Inputting the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, generate the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics; Wherein, the generation of the life prediction model includes: Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

2. A method for predicting the life of a MOSFET device as claimed in claim 1, characterized in that: The life prediction model comprises: a hidden layer; The method of capturing the long-term dependencies in the time series based on the operation data and generating the time series features corresponding to each moment includes: The hidden layer is used to capture the long-term dependency between the current moment and the time series of the running data based on the gating mechanism at each moment, and generate the time series features corresponding to the current moment based on the long-term dependency corresponding to the current moment.

3. A method for predicting the life of a MOSFET device as claimed in claim 2, characterized in that: The hidden layer includes: a forget gate, an input gate and an output gate; The gating mechanism is used to capture the long-term dependency between the current moment and the time series of the running data, and to generate the time series features corresponding to the current moment based on the long-term dependency corresponding to the current moment, including: The forget gate is used to generate a first forgetting degree corresponding to the on-state resistance at a previous moment at a current moment and a second forgetting degree corresponding to the threshold voltage at a previous moment at a current moment based on the time series data in the operation data; The input gate is used to generate a first update degree of the on-state resistance at the current moment for the timing characteristics at the previous moment based on the time series data in the operation data, and to generate a second update degree of the threshold voltage at the previous moment for the threshold voltage at the current moment; The output gate is used to generate a first contribution degree of the on-resistance at the current moment to the current output based on the time series data in the operation data, and to generate a second contribution degree of the threshold voltage at the current moment to the current output; The output gate is further used to generate a resistance timing characteristic at the current moment according to the on-state resistance, the first forgetting degree, the first updating degree and the first contribution degree at the current moment, and to generate a voltage timing characteristic at the current moment according to the threshold voltage, the second forgetting degree, the second updating degree and the second contribution degree at the current moment; Among them, the resistance timing feature is the resistance change state between the current moment and the previous moment; the voltage timing feature is the voltage change state between the current moment and the previous moment; the resistance change state includes: resistance increase or resistance decrease; the voltage change state includes: voltage increase or voltage decrease.

4. A method for predicting the life of a MOSFET device as claimed in claim 3, characterized in that: The lifespan prediction model comprises: an attention layer; The method of assigning different weights to the time series features at each moment includes: The attention layer is used to use the time series feature corresponding to the last moment in the time series data as a reference time series feature; Calculate the similarity between the time series features at each moment and the reference time series features to obtain the similarity of each time series feature; wherein the similarity represents the contribution of the time series feature to the prediction of the remaining life; Each similarity is normalized into a probability distribution through an activation function, and then the weight corresponding to the time series feature of each moment is generated based on the probability value of the probability distribution; wherein the weight represents the degree of attention of the model to the moment.

5. A method for predicting the life of a MOSFET device as claimed in claim 4, characterized in that: The life prediction model includes: a fully connected layer and an output layer; The generating of the remaining life corresponding to the MOSFET device according to each timing feature and the corresponding weight includes: The fully connected layer is used to perform weighted summation according to each time series feature and the weight corresponding to each time series feature, so as to generate a sequence feature for representing the global understanding of the model on the time series data; Splicing the sequence feature with the time series feature corresponding to the last moment in the time series data to generate a spliced ​​feature vector; wherein the spliced ​​feature vector includes: global features and local features; The output layer is used to generate the remaining life corresponding to the MOSFET device according to the spliced ​​feature vector.

6. A method for predicting the life of a MOSFET device as claimed in claim 1, characterized in that: In each iterative training, an operation data sample is input into the life prediction model, so that the life prediction model generates a remaining life prediction result corresponding to the operation data sample according to the time series feature corresponding to each moment in the operation data sample and the weight of each time series feature; The remaining life prediction results are compared with the actual remaining life results, and the network parameters of the life prediction model are adjusted according to the comparison results.

7. A method for predicting the life of a MOSFET device as claimed in claim 1, characterized in that: Also includes: The parameters of the life prediction model are updated according to the deviation value of the following loss function: Among them, MSE is the deviation value of the loss function, y i is the actual remaining life of a running data sample, is the predicted remaining life of an operating data sample, and n is the number of operating data samples.

8. A life prediction device for a MOSFET device, characterized in that: include: Run the data acquisition module and the remaining life prediction module; The operation data acquisition module is used to acquire the operation data corresponding to the MOSFET device within a time period; wherein the operation data includes: time series data of on-resistance and time series data of threshold voltage; The remaining life prediction module is used to input the operation data into the life prediction model, so that the life prediction model can capture the long-term dependency in the time series according to the operation data, and generate the timing characteristics corresponding to each moment, after assigning different weights to the timing characteristics of each moment, and then generate the remaining life corresponding to the MOSFET device according to each timing characteristic and the corresponding weight; wherein the timing characteristics include resistance timing characteristics and voltage timing characteristics; Wherein, the generation of the life prediction model includes: Taking each operation data sample of the MOSFET device and the actual remaining life corresponding to each operation data sample as input, and taking the predicted remaining life of each operation data sample as output, the life prediction model to be trained is iteratively trained until the model converges, thereby generating a trained life prediction model; wherein the operation data samples include time series data samples of on-resistance and time series data samples of threshold voltage.

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 method for predicting the life of a MOSFET device 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 a life prediction method for a MOSFET device according to any one of claims 1 to 7.

Images