A method and system for interpolating and enhancing the sampling rate of a source measurement unit based on time series prediction

Abstract

The present invention relates to a method and system for interpolating and boosting the sampling rate of a source measurement unit based on time series prediction, including: data acquisition, including: collecting current and voltage time series as a data set; dividing the data set into a training set and a test set; data preprocessing, including: filling in missing values and normalizing; model training, including: training a time series prediction model; time series prediction, including: inputting real-time actual measurement data in the form of slices into the trained time series prediction model, so as to output the predicted values corresponding to the moments of the slices; after obtaining all the predicted values, arranging the actual measurement data and the predicted data in time sequence to obtain the final predicted data. The method of combining time series prediction and interpolation in the present invention has the advantages of higher interpolation accuracy and a wider range of applicable scenarios. The trained time series prediction model of the present invention will be stored in the SOC chip of the SMU in the form of an IP core for convenient subsequent invocation.

Description

Technical Field

[0001] The present invention relates to a method and system for interpolating and enhancing the sampling rate of a source measurement unit based on time series prediction, and belongs to the technical field of enhancing the sampling rate of a source measurement unit. Background Art

[0002] A source measurement unit (SMU) is an instrument widely used in the field of electronic measurement for measuring current, voltage, etc. In practical applications, the sampling rate is crucial for the performance and application scenarios of the SMU. However, due to the cost of transmitting and storing measurement data, as well as hardware resource limitations, there are usually limitations on the sampling rate. This limitation may lead to the loss of capturing details in some rapidly changing signal scenarios, thus affecting the accurate understanding of the dynamic behavior of the system.

[0003] Traditional methods for increasing the sampling rate include: upgrading the SMU hardware: This may include increasing the speed of the analog-to-digital converter (ADC), increasing the storage capacity, and optimizing the data transmission channel. However, hardware upgrades are usually accompanied by an increase in cost and technical difficulty. Multi-channel acquisition: By simultaneously using multiple channels for data acquisition, the overall sampling rate can be increased. This method is suitable for certain applications, but requires the SMU to support parallel acquisition of multiple relatively independent signals. Data compression: Using data compression algorithms can reduce the need for data transmission and storage, thus allowing a higher sampling rate. However, compression is usually accompanied by some information loss, so a trade-off is required.

[0004] With the progress of machine learning and data science, time series prediction techniques have been widely applied. These techniques can predict future data points by analyzing past data patterns. In the aspect of time series data, such as data where current and voltage change over time, time series prediction has demonstrated excellent performance in various fields. Interpolation is a technique for calculating unknown points through estimation between known data points. In the field of digital signal processing, interpolation is often used to generate continuous curves between discrete data points. In this field, the application of interpolation techniques can fill the gaps between actual sampling values with predicted values during the process of increasing the sampling rate.

[0005] In summary, the following main technical problems and secondary technical problems exist in the prior art;

[0006] Main Technical Problems:

[0007] 1. Sampling rate limitation problem: The sampling rate limitation problem is a major challenge in current SMU systems, especially in application scenarios that require accurate measurement of high-frequency and rapidly changing signals.

[0008] 2. Data loss problem: In the current system with limited sampling rate, there is a problem of data loss, which leads to the loss of some information on the dynamic changes of the signal, affecting the capture and analysis of signal details and reducing the measurement accuracy.

[0009] 3. Real-time and sensitivity problems: The current system is restricted by the sampling rate, resulting in insufficient real-time and sensitivity. Increasing the sampling rate will enable the system to respond more quickly to the immediate changes in the signal.

[0010] Minor technical problems:

[0011] 1. Hardware cost and resource problems: Increasing the sampling rate may involve hardware upgrades, which brings additional costs and secondary technical problems of increased system resources.

[0012] 2. Data processing requirement problems: Increasing the sampling rate brings the need for processing and storing larger amounts of data, which are secondary technical problems that need to be solved. Summary of the invention

[0013] In view of the deficiencies of the prior art, the present invention provides a method and system for interpolating and enhancing the sampling rate of a source measurement unit based on time series prediction;

[0014] Traditional methods for enhancing the sampling rate of SMU are often restricted by hardware resources and real-time performance. However, the method of combining time series prediction with interpolation can, to a certain extent, make up for these deficiencies, increase the sampling rate, and thus capture and analyze high-dynamic-range electronic signals more accurately. The purpose of this invention is to address the challenges in the above background technology. By integrating time series prediction and interpolation techniques, it provides a method and system for increasing the sampling rate in SMU measurements. This will bring higher sensitivity and accuracy to the field of electronic measurement, especially when dealing with high-frequency and rapidly changing signal time differences.

[0015] Term explanations:

[0016] 1. Power amplifier chip: Abbreviated as power amplifier, mainly used to increase the amplitude of an electrical signal, thereby increasing the power of the signal. Such chips are usually used to amplify audio, radio frequency (wireless communication), microwave and other signals.

[0017] 2. Switch chip: A switch chip, usually used in applications such as power management and switch control in digital circuits. They can switch between input and output and create or break electrical connections in the circuit.

[0018] 3. LNA chip: Low-noise amplifier chip. The LNA chip is used to amplify weak signals and minimize the noise in the signal during the amplification process. This kind of chip is often used at the front end of receiving signals, such as in communication receivers.

[0019] 4. D / A conversion chip: A digital-to-analog conversion chip that converts digital signals into corresponding analog signals and is often used in fields such as audio processing and graphic display.

[0020] 5. A / D conversion chip: An analog-to-digital conversion chip that performs the opposite operation, converting analog signals into digital form and is widely used in aspects such as sensor data acquisition and audio digitization.

[0021] 6. LED chip: LED chips are used to control and drive light-emitting diodes (LEDs). They contain current regulation and protection circuits to ensure that the LEDs operate in a stable and safe manner.

[0022] 7. IGBT chip: An IGBT chip is a power semiconductor device that combines the characteristics of a field-effect transistor (FET) and a bipolar junction transistor (BJT). It is widely used in high-power, high-voltage switching power supplies and inverters.

[0023] 8. DC / DC chip: DC / DC chips are used in direct current (DC) power systems to provide the required power voltage by converting voltage levels. They are commonly used in electronic devices, power adapters, etc.

[0024] 9. MOSFET chip: A metal-oxide-semiconductor field-effect transistor chip, which is a common field-effect transistor used in electronic switches, amplifiers, and other applications, providing characteristics such as high-speed switching and low power consumption and is widely used in the fields of integrated circuits and power electronics.

[0025] The technical solution of the present invention is as follows:

[0026] A method for interpolating and boosting the sampling rate of a source measurement unit based on time series prediction, including:

[0027] Data acquisition, including: acquiring current and voltage time series as a data set; dividing the data set into a training set and a test set;

[0028] Data preprocessing, including: filling in missing values and normalizing;

[0029] Model training, including: training a time series prediction model;

[0030] Time series prediction, including: inputting real-time true measurement data in the form of slices into the trained time series prediction model to output the predicted values corresponding to the moments of the slices; after obtaining all the predicted values, arranging the real measurement data and the predicted data in time sequence to obtain the final predicted data.

[0031] Preferably according to the present invention, in data preprocessing, interval sampling is performed on the original data to simulate the sampling data obtained by the measuring device; after slicing the sampled data, it is randomly arranged and shuffled.

[0032] The remaining sampling points of the original data are used to verify the prediction effect during the process of training the time series prediction model, and then the time series prediction model is optimized and iterated.

[0033] In the test stage after the time series prediction model is trained, the continuous sampling points in the sampled data are input into the time series prediction model to obtain predicted values; the predicted values are inserted into the sampling data at intervals.

[0034] Preferably according to the present invention, collecting current and voltage time series includes:

[0035] Building a data acquisition experimental platform using a PXIe-4139 source measurement unit with a maximum sampling rate of 1.8 MS / s.

[0036] For power amplifier chips, Switch chips, LNA chips, D / A conversion chips, DC / DC chips, A / D conversion chips, conducting open / short circuit tests, leakage current tests, static current tests, dynamic current tests, conversion level tests, and output drive current tests.

[0037] For LED chips, IGBT chips, DC / DC chips, MOSFET chips, conducting open / short circuit tests, leakage current tests, input characteristic tests, and output characteristic tests.

[0038] After slicing and selecting the time periods with obvious changes in the measurement results and splicing them, test curve sample libraries are formed according to different loads and different test parameter items. The curve sample libraries include measurement curves obtained under different loads and different measurement parameters. This test curve sample library is used as the dataset for subsequent training of the time series prediction model.

[0039] Preferably according to the present invention, training a time series prediction model includes:

[0040] Using the model.fit() method, inputting the training set and corresponding labels to train the time series prediction model; the training set includes the historically collected current and voltage time series; the parameters of the time series prediction model are trained through the gradient descent optimization algorithm to minimize the prediction error; the Adam optimizer and binary cross-entropy loss function are used during the training process; the training and optimization process of the time series prediction model is iterated repeatedly, adjusting the weights and biases to minimize the loss function, so that the time series prediction model can better fit the training data.

[0041] Preferably according to the present invention, the time series prediction model includes an input layer, two GRU layers, a Dropout layer, and a fully connected layer.

[0042] The GRU layer is the Gated Recurrent Unit (GRU).

[0043] The mathematical representation of the Gated Recurrent Unit is as follows:

[0044] Z t = σ(W z x t + U z h t-1 + b z ) (1)

[0045] r t = σ(W r x t + U r h t-1 + b r ) (2)

[0046]

[0047]

[0048] In equations (1) to (4), x t is the input at the current time step, h t-1 is the hidden state at the previous time step, Z t is the output of the update gate, r t is the output of the reset gate, is the candidate value of the new hidden state, W z is the weight matrix of the update gate, U z is the bias parameter, b z , b r , b h are the bias terms, and ⊙ represents element-wise multiplication.

[0049] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the method for interpolating and enhancing the sampling rate of the source measurement unit based on time series prediction.

[0050] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the method for interpolating and enhancing the sampling rate of the source measurement unit based on time series prediction.

[0051] A system for interpolating and enhancing the sampling rate of the source measurement unit based on time series prediction includes:

[0052] The data acquisition module is configured to: collect current and voltage time series as a data set; divide the data set into a training set and a test set;

[0053] The data preprocessing module is configured to: supplement missing values and perform normalization processing;

[0054] The model training module is configured to: train a time series prediction model;

[0055] The time series prediction module is configured to: input real-time true measurement data in the form of slices into the trained time series prediction model, so as to output the predicted values corresponding to the moments of the slices; after obtaining all the predicted values, arrange the real measurement data and the predicted data in time series to obtain the final predicted data.

[0056] The beneficial effects of the present invention are as follows:

[0057] 1. Nonlinear relationship modeling: The GRU deep learning model can better capture the nonlinear relationships in the time series. Compared with traditional linear interpolation methods, the predicted values obtained through the deep learning model can better reflect the complex patterns in the data.

[0058] 2. Long-term dependence: The deep learning model can capture long-term dependence, that is, the model can learn the long-distance dependence relationships in the time series. This enables the high-sampling-rate sequence obtained through interpolation to better retain the dynamic characteristics of the original time series.

[0059] 3. Adaptability: The deep learning model can adapt to the complex patterns and changes in the data during the training process, thus better performing interpolation. Traditional interpolation methods may require manually selecting different interpolation strategies according to the different characteristics of the data, while the deep learning model can automatically learn adaptability.

[0060] 4. High-precision prediction: The deep learning model can usually provide higher prediction accuracy in time series prediction. Therefore, the high-sampling-rate sequence obtained through interpolation is also more likely to maintain a high accuracy.

[0061] 5. Self-adaptability: The deep learning model has stronger adaptability to different data sets. Therefore, when performing interpolation on different time series data, it is more likely to obtain good results.

[0062] 6. Noise reduction effect: The high-sampling-rate sequence predicted through the deep learning model may have a certain noise reduction effect because the model can filter out some noise and outliers during the learning process. Description of the Drawings

[0063] Figure 1 is a schematic diagram of the architecture of the gated recurrent unit;

[0064] Figure 2Schematic diagram of the architecture of the time series prediction model;

[0065] Figure 3 Schematic diagram of the implementation process of obtaining the final prediction data using the time series prediction model. Detailed implementation manners

[0066] The present invention will be further defined below in conjunction with the accompanying drawings of the specification and embodiments, but not limited thereto.

[0067] Embodiment 1

[0068] A method for interpolating and enhancing the sampling rate of a source measurement unit based on time series prediction, comprising:

[0069] Data acquisition, including: acquiring current and voltage time series as a data set; dividing the data set into a training set and a test set;

[0070] The data acquired by the above data acquisition experimental platform has 1.8M sampling points within 1 second. 20% of the data is extracted as the test set according to different loads and different test parameter items in the above test curve sample library. 900K sampling points are obtained by interval sampling to simulate the data acquired by an SMU with a sampling rate of 900K, and the remaining 900K sampling points are used to evaluate the training effect of the model in the subsequent test set. The remaining 80% of the data set is used as the training set.

[0071] Data preprocessing, including: filling in missing values and normalizing;

[0072] After filling in missing values and normalizing the training set and the test set, the above sampling data is divided into samples by combining a sliding window and slicing. Among them, the width and sliding step of the sliding window are set as parameters, which is convenient for finding the window width and sliding step with the best prediction effect for the voltage and current time series when adjusting the parameters in the later stage. The obtained training samples are randomly arranged and shuffled to increase the entropy of each batch, so that the model can learn more voltage and current time-dependent information.

[0073] Model training, including: training a time series prediction model;

[0074] Time series prediction, as Figure 3 shown, including: inputting real-time real measurement data in the form of slices into the trained time series prediction model, so as to output the predicted value corresponding to the time of the slice; after obtaining all the predicted values, arranging the real measurement data and the predicted data in time sequence, thus obtaining the final predicted data with the sampling rate doubled.

[0075] Assume that there are n data points in the real measurement data. A new data point is predicted by a prediction algorithm based on the 0th to kth data points of the real data, and then another new data point is obtained based on the 1st to k+1th data points, and so on. A total of n-k predicted data points can be obtained by predicting with n real data points. The obtained set of predicted data points is inserted into the real data according to the prediction order. When the sampling rate is high enough, the real data plus the predicted data can be approximately regarded as 2n data points, thus doubling the sampling rate.

[0076] Based on the real sampling data, the present invention predicts the data at the next moment that cannot be sampled by the current sampling rate of the device. After obtaining the predicted value, it is interpolated into the real sampling data, thereby indirectly increasing the sampling rate of the measuring device. Compared with the commonly used linear interpolation, Lagrange interpolation, Newton interpolation, and Hermite interpolation, the method of combining time series prediction and interpolation in the present invention has the advantages of higher interpolation accuracy and a wider range of applicable scenarios. The trained time series prediction model of this algorithm will be stored in the SOC chip of the SMU (Source Measurement Unit) in the form of an IP core for convenient subsequent calling.

[0077] The most crucial step of the present invention is to use the predicted value output by the time series prediction model as the data source for interpolation, and then double the sampling rate of the measuring device through interpolation. The time series prediction model is trained with historical measurement data to enable it to learn the variation rules of measurement data in various measurement scenarios. The interpolated data obtained through this time series prediction model can better capture the non-linear relationship in the time series and the relationship between data points that are far apart compared to traditional interpolation methods. It has strong adaptability to different data distributions and dynamic changes. In contrast, traditional interpolation methods may require manual selection of appropriate interpolation strategies, which may not be flexible enough when facing different data. It can filter out the noise in the data to a certain extent, which helps to improve the quality of the interpolated data.

[0078] Embodiment 2

[0079] A method for interpolating to improve the sampling rate of the source measurement unit based on time series prediction according to Embodiment 1, which is characterized in that:

[0080] In data preprocessing, the original data with a sampling rate of 1.8 MS / s is sampled at intervals to simulate the sampling data obtained by a measuring device with a sampling rate of 900 KS / s; after slicing the sampled data, it is randomly arranged and shuffled;

[0081] To increase the entropy of each batch and enable the model to learn more voltage-current time-dependent information. The remaining sampling points of the original data are used to verify the prediction effect during the training of the time series prediction model, and then the time series prediction model is optimized and iterated;

[0082] In the test phase after the completion of the training of the time series prediction model, consecutive sampling points in the sampled data are input into the time series prediction model to obtain predicted values; these predicted values are inserted into the sampled data at intervals. Thus, the effect of alternating real measurement data and predicted data is achieved, and further the effect of doubling the sampling rate is achieved.

[0083] Collect current and voltage time series, including:

[0084] Build a data acquisition experimental platform using a PXIe-4139 source measurement unit with a maximum sampling rate of 1.8 MS / s; the data acquisition experiment consists of a source measurement unit hardware board and data acquisition software. Insert the PXIe-4139 into the PXIe-1062Q chassis slot, install the driver software for driving, and open the test panel through the NIMAX software to control the board; the NIMAX software provides three modes: chart, waveform, and scan, which are applied to different scenarios for measurement. For example, the transient response process of the load can be obtained through single-step acquisition in waveform mode, and the V-I characteristics of the load can be obtained through scan mode. In addition to using the NIMAX software for data acquisition, the PXIe-4139 also supports development based on LabVIEW. Call the NIDCpower module in LabVIEW to control the board and achieve program control of the board through graphical programming.

[0085] For power amplifier chips, Switch chips, LNA chips, D / A conversion chips, DC / DC chips, A / D conversion chips, conduct open / short circuit tests, leakage current tests, static current tests, dynamic current tests, conversion level tests, and output drive current tests;

[0086] For LED chips, IGBT chips, DC / DC chips, MOSFET chips, conduct open / short circuit tests, leakage current tests, input characteristic tests, and output characteristic tests;

[0087] Select time periods with obvious changes in the measurement results of the slices (the change in the voltage / current curve is visible to the naked eye, and the change amplitude exceeds 5% of the current amplitude), and then splice them to enhance the time series characteristics of the voltage and current of the measured load, facilitating the subsequent GRU model to learn the time-dependent relationship between voltage and current. And build a test curve sample library according to different loads and different test parameter items. The test curve sample library includes measurement curves obtained for different loads under different measurement parameters. Taking the leakage current test as an example, the magnitude of the leakage current is obtained by measuring the current through the output voltage. This test will record the voltage and the transient response of the current during the power-on process of the load, and the curve change is very obvious. The voltage / current curves of different loads form the sample library under the leakage current test. Similarly, through the open / short circuit test, the voltage change curves under different load output current modes can be obtained, which are also part of the sample library. This test curve sample library is used as the dataset for the subsequent training of the time series prediction model.

[0088] Training a time series prediction model, including:

[0089] Use the model.fit() method to train the time series prediction model by passing in the training set and the corresponding labels; the training set includes the historical current and voltage time series; the parameters of the time series prediction model are trained using the gradient descent optimization algorithm to minimize the prediction error; the Adam optimizer and binary cross-entropy loss function are used during the training process; the training and optimization process of the time series prediction model is iterated repeatedly, adjusting the weights and biases to minimize the loss function, making the time series prediction model better fit the training data. After the time series prediction model is trained, it is evaluated using test data and used to predict future data.

[0090] The time series prediction model includes an input layer, two GRU layers, a Dropout layer, and a fully connected layer;

[0091] A function to build a GRU (Gated Recurrent Unit) model using the Keras library. This function takes a list containing information about the number of units as input and then returns a GRU model with the corresponding configuration. The main components of this function include two GRU layers; a Dropout layer is used to prevent the model from overfitting to the training data by randomly setting the outputs of some neurons to zero during the training process; a fully connected layer and the Sigmoid activation function are used. The specific model architecture diagram is as Figure 2 shown.

[0092] The GRU layer is the Gated Recurrent Unit (GRU);

[0093] The Gated Recurrent Unit (GRU) is a variant of the Recurrent Neural Network (RNN) used to process sequential data. Compared with traditional RNNs, GRUs have better performance in learning long-term dependencies and are relatively easy to prevent the vanishing gradient problem during training. The GRU layer introduces a reset gate that determines whether the model should ignore the previous hidden state. It allows the model to discard outdated information to better adapt to the current input. The output of the reset gate ranges from 0 to 1. The update gate of the GRU controls how much past memory information should be passed to the current state. Like the reset gate, the output of the update gate ranges from 0 to 1. The more open the update gate is, the more past information is retained. Based on the outputs of the reset gate and the update gate, a new hidden state is calculated. The new hidden state is a combination of the current input and the previous hidden state. The output of the GRU layer can be the hidden state itself or a transformation of the hidden state. This depends on the specific task and model design. The GRU layer reduces the number of parameters compared to the standard RNN and has better performance in learning long-term dependencies. The schematic diagram of the GRU layer is as shown in Figure 1 shown below:

[0094] The mathematical representation of the Gated Recurrent Unit is as follows:

[0095] Z t = σ(W z x t + U z h t-1 + b z )(1)

[0096] r t = σ(W r x t + U r h t-1 + b r )(2)

[0097]

[0098]

[0099] In equations (1) to (4), x t is the input at the current time step, h t-1 is the hidden state at the previous time step, Z t is the output of the update gate, r t is the output of the reset gate, is the candidate value for the new hidden state, W z is the weight matrix of the update gate, U z is the bias parameter, b z , b r , bh is the bias term, and ⊙ represents element-wise multiplication.

[0100] In the experimental environment, by using this invention, the sampling rate of the data sampled by a measurement device with an original sampling rate of 900 KS / s was successfully increased to 1.8 MS / s, and the measurement accuracy for high-frequency signals was significantly improved. As shown in Table 1:

[0101] Table 1

[0102]

[0103] T represents the true data measured by the measurement device, and P represents the predicted value obtained by the algorithm of this invention. Seven true data at sampling times 12, 14, 16, 18, 20, 22, and 24 were used as a slice to input into a trained GRU model to obtain the predicted value P for the 25th sampling time. Similarly, predicted values at sampling times 27, 29, 31, 33... were obtained. The predicted values were interpolated into the corresponding sampling times to form an effect where true data and predicted data appear alternately. At this time, the sampling rate of the final data was also increased from 900K of the original sampled data to 1.8M.

[0104] Example 3

[0105] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the method for interpolating and improving the sampling rate of the source measurement unit based on time series prediction described in Example 1 or 2.

[0106] Example 4

[0107] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the method for interpolating and improving the sampling rate of the source measurement unit based on time series prediction described in Example 1 or 2.

[0108] Example 5

[0109] A system for interpolating and improving the sampling rate of the source measurement unit based on time series prediction includes:

[0110] A data acquisition module configured to: collect current and voltage time series as a data set; divide the data set into a training set and a test set;

[0111] A data preprocessing module configured to: supplement missing values and perform normalization processing;

[0112] A model training module configured to: train a time series prediction model;

[0113] The time series prediction module is configured to: input the real-time actual measurement data in the form of slices into the trained time series prediction model, so as to output the predicted values corresponding to the moments of the slices; after obtaining all the predicted values, arrange the actual measurement data and the predicted data in time sequence to obtain the final predicted data.

Claims

1. A method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction, characterized in that: include: Data collection, including: collecting current and voltage time series as a data set; dividing the data set into a training set and a test set; Data preprocessing, including: filling missing values ​​and normalization; Model training, including: training time series prediction models; Time series prediction, including: inputting real-time measurement data into the trained time series prediction model in the form of slices, thereby outputting the predicted value at the moment corresponding to the slice; after obtaining all the predicted values, arranging the real measurement data and the predicted data in time sequence to obtain the final predicted data; The time series prediction model includes an input layer, two GRU layers, a Dropout layer, and a fully connected layer; The GRU layer is a gated recurrent unit; The mathematical representation of a gated recurrent unit is shown below: Z t =σ(W z x t +U z h t-1 +b z ) (1) r t =σ(W r x t +U r h t-1 +b r ) (2) In formula (1) to formula (4), x t is the input of the current time step, h t-1 is the hidden state of the previous time step, Z t is the output of the update gate, r t is the output of the reset gate, is the candidate value of the new hidden state, W z is the weight matrix of the update gate, U z is the bias parameter, b z , b r , b h is the bias term, and ⊙ represents element-by-element multiplication.

2. The method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction according to claim 1, characterized in that: In data preprocessing, the original data is sampled at intervals to simulate the sampling data obtained by the measuring equipment; the sampled data is sliced ​​and then randomly arranged to disrupt the order; The remaining sampling points of the original data are used to verify the prediction effect in the process of training the time series prediction model, and then optimize and iterate the time series prediction model; In the testing phase after the time series prediction model training is completed, the continuous sampling points in the sampled data are input into the time series prediction model to obtain the predicted value; the predicted value interval is inserted into the sampled data.

3. The method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction according to claim 1, characterized in that: Acquire current and voltage time series, including: Use source measurement units to build a data acquisition experimental platform; For power amplifier chips, switch chips, LNA chips, D / A conversion chips, DC / DC chips, A / D conversion chips, open and short circuit tests, leakage current tests, static current tests, dynamic current tests, conversion level tests, and output drive current tests; For LED chips, IGBT chips, DC / DC chips, MOSFET chips, open and short circuit tests, leakage current tests, input characteristic tests, and output characteristic tests; The slices are selected for the time periods where the measurement results change significantly and then stitched together, and a test curve sample library is established according to different load and test parameter items. The curve sample library includes measurement curves obtained under different loads and different measurement parameters. The test curve sample library is used as a data set for subsequent time series prediction model training.

4. The method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction according to claim 1, characterized in that: Train time series forecasting models, including: Use the model.fit() method to pass in the training set and corresponding labels to train the time series prediction model; the training set includes the historically collected current and voltage time series; the parameters of the time series prediction model are trained using the gradient descent optimization algorithm to minimize the prediction error; the Adam optimizer and binary cross entropy loss function are used during the training process; the training and optimization process of the time series prediction model is iterated repeatedly, and the weights and biases are adjusted to minimize the loss function so that the time series prediction model can better fit the training data.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction according to any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for improving the sampling rate of a source measurement unit by interpolation based on time series prediction according to any one of claims 1 to 4 are implemented.

7. A system for improving the sampling rate of a source measurement unit by interpolation based on time series prediction, characterized in that: include: The data acquisition module is configured to: acquire current and voltage time series as a data set; Divide the dataset into training and testing sets; The data preprocessing module is configured to: fill in missing values ​​and normalize; The model training module is configured to: train a time series prediction model; The time series prediction module is configured to: input the real-time measurement data into the trained time series prediction model in the form of slices, thereby outputting the predicted value at the time corresponding to the slice; After obtaining all the predicted values, the actual measured data and the predicted data are arranged in time series to obtain the final predicted data; The time series prediction model includes an input layer, two GRU layers, a Dropout layer, and a fully connected layer; The GRU layer is a gated recurrent unit; The mathematical representation of a gated recurrent unit is shown below: Z t =σ(W z x t +U z h t-1 +b z ) (1) r t =σ(W r x t +U r h t-1 +b r ) (2) In formula (1) to formula (4), x t is the input of the current time step, h t-1 is the hidden state of the previous time step, Z t is the output of the update gate, r t is the output of the reset gate, is the candidate value of the new hidden state, W z is the weight matrix of the update gate, U z is the bias parameter, b z , b r , b h is the bias term, and ⊙ represents element-by-element multiplication.

Images