Method for training and measuring AI model in real time by using parameter measurement

Abstract

The invention discloses a method for training and measuring an AI model in real time by using parameter measurement, which is combined with a traditional measurement method to automatically generate a sample as a model training sample. An online learning method is used for perfecting an existing artificial intelligence model, and the artificial intelligence model is used for measuring a signal which cannot be accurately measured by a traditional measurement method and has a small signal-to-noise ratio, namely, a signal with overlarge noise, so that self-perfection and growth of a measurement system are realized, and new signals are continuously recognized, recognized and measured in the using process; and the measurement precision is continuously improved. According to the method, the characteristic that the approximate trend of the signal is determined under the condition of overlarge noise is utilized, the sample required by AI model training is automatically generated by using the accurate measurement result of the measured signal with the large signal-to-noise ratio in traditional measurement, the problems of automatic acquisition and automatic marking of the training sample in AI model training for parameter measurement are solved, and the accuracy of the AI model training is improved. And automatic training is realized, and the AI model is automatically updated, so that the measurement system has a self-improvement capability and is more intelligent.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and signal analysis technology, and more specifically, relates to a method for real-time training and measuring an AI model using parameter measurement. Background Art

[0002] Signal parameter measurement is essentially the process of extracting information from a signal based on the data characteristics of the signal itself. Traditional signal parameter measurement uses methods such as directly obtaining signal characteristics and extracting signal characteristics after statistics to measure signal parameters. During the statistical process, if the relative noise contained in the signal is too large, the statistical results obtained by the traditional histogram-based signal parameter measurement method will be inaccurate, resulting in the inability to measure statistically relevant measurement parameters.

[0003] When the noise is too large, the general trend of the signal is certain. At this time, AI (artificial intelligence) models can be used to identify the signal and extract the statistical features of the signal for measurement. However, there are also problems when using trained AI models to measure signals. Trained AI models have fixed measurement accuracy and do not have growth properties. If you need to increase the accuracy of model measurement, you need to manually create a model training set and manually label the training set, then train the model, and finally update the model. The process is cumbersome and requires a lot of manpower.

[0004] In combination with the above, signal measurement outside the traditional signal parameter measurement range is a problem in current digital signal processing. Automatically improving the measurement accuracy of artificial intelligence measurement systems has also become a problem in artificial intelligence applications and digital signal measurement systems. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for real-time training of AI models and measurements using parameter measurements, so as to achieve self-improvement and growth of the measurement system, continuously recognize, identify and measure new signals during use, and continuously improve measurement accuracy.

[0006] To achieve the above-mentioned object of the invention, the present invention uses a method for real-time training and measuring of an AI model using parameter measurement, which is characterized by comprising the following steps:

[0007] (1) For the measured signal measured for the nth time, use Data n Indicates that the peak-to-peak value of the signal is directly obtained by inputting it into the measurement system using the traditional measurement method, and using Peak to Peak n express;

[0008] (2) If the peak-to-peak value of the signal is PeaktoPeak nIf the value is greater than the measurement threshold of the parameter measurement, then step (3) is executed; otherwise, step (8) is executed, wherein the measurement threshold is represented by Threshold, and the measurement threshold Threshold is determined by the signal-to-noise ratio of the measured signal. The lower the signal-to-noise ratio of the measured signal, the larger the measurement threshold Threshold is.

[0009] (3) Data of the measured signal n Perform traditional measurements to obtain measurement results, using MeasureResult n Indicates that the output measurement result MeasureResult n Execute step (4) at the same time;

[0010] (4) Save the measurement results MeasureResult n , measured signal Data n As AI model training sample n , and the measurement result MeasureResult n As a sample n Labels;

[0011] (5) Sample the labeled training samples n Input into the online learning module to adjust the current AI model Current Conduct online learning and training, and the training result is a new AI model Model New ;

[0012] (6) Use the test set to test the new AI model New Conduct testing;

[0013] (7) Determine the trained new AI model New Is the measurement accuracy better than the current AI model? Current , if the new AI model Model New Better than the current AI model Current , then the new AI model Model New As the current AI model Current Otherwise, keep the current AI model Current , n=n+1, return to step (1);

[0014] (8) Read the current AI model Current ;

[0015] (9) Use the current AI model Current Data of the measured signal nPerform measurement and obtain measurement results MeasureResult nwithAI And output, n=n+1, and return to step (1).

[0016] The object of the invention of the present invention is achieved in this way:

[0017] The present invention uses parameter measurement to train AI models in real time and combines the measurement method with traditional measurement methods to automatically generate samples as model training samples, and uses online learning methods to improve the existing artificial intelligence model and use the artificial intelligence model to measure signals with low signal-to-noise ratio, i.e., excessive noise, which cannot be accurately measured by traditional measurement methods, thereby achieving self-improvement and growth of the measurement system, continuously recognizing, identifying, and measuring new signals during use, and continuously improving measurement accuracy.

[0018] The present invention has the following beneficial effects:

[0019] (1) The present invention combines traditional measurement methods and AI model measurement methods, and uses the measurement results of traditional measurement methods to automatically label signal samples, generate training samples for the AI ​​model in real time, and train the AI ​​model. Compared with the single signal feature extraction in traditional measurements, the trained AI model can extract multi-dimensional features related to signal period, top value, bottom value, etc., thereby integrating multiple features to achieve more accurate measurement of the signal. In particular, the present invention can accurately identify multiple features of the signal when the measured signal contains large noise, solving the problem that the traditional measurement method cannot accurately measure the signal using a single statistical feature when the noise is relatively large, and broadens the measurement range of the measurement system;

[0020] (2) The present invention uses a method for real-time training of the AI ​​model to enable the AI ​​model to continuously improve the recognition accuracy and measurement accuracy of the measurement system during use, making the measurement system more accurate and easier to use.

[0021] (3) The present invention utilizes the characteristic that the general trend of the signal is determined when the noise is too large, and uses the traditional measurement of the peak-to-peak value of the signal with a large signal-to-noise ratio. n The measured signal Data is greater than the measurement threshold of the parameter measurement n Accurate measurement results MeasureResult n It automatically generates samples required for AI model training, solves the problem of automatic acquisition and automatic labeling of training samples in AI model training for parameter measurement, and realizes automatic training and automatic updating of AI models, making the measurement system capable of self-improvement and more intelligent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1It is the histogram effect diagram of traditional measurement when the signal is greater than the measurement threshold;

[0023] Figure 2 It is the histogram effect diagram of traditional measurement when the signal is less than the measurement threshold;

[0024] Figure 3 It is a flowchart of a specific implementation method of the present invention for real-time training and measuring of AI models using parameter measurement. DETAILED DESCRIPTION

[0025] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0026] Figure 1 , 2 The histograms of the traditional measurement are respectively greater than and less than the measurement threshold. The bold lines represent the noise of the signal. It is assumed that the two curves have the same noise. Figure 1 It can be seen that the statistical histogram of the waveform has obvious peak values ​​in the upper half and the lower half, that is, the top value and the bottom value. Through this histogram statistical feature, the signal characteristics such as the period of the signal can be indirectly calculated to obtain the measurement values ​​of various measurement items. Figure 2 It can be seen that the statistical histogram of the waveform can no longer clearly distinguish the peaks of the upper and lower parts. At this time, the maximum and minimum values ​​can only be used to replace the top and bottom values ​​of the signal, which will bring great errors. However, if a trained AI model is used to measure this signal, the measurement values ​​such as the period, top value, and bottom value can be directly obtained through the model.

[0027] However, there are also problems when using trained AI models to measure signals. Trained AI models have fixed measurement accuracy and do not have growth properties. If you need to increase the accuracy of model measurements, you need to manually create a model training set and manually label the training set, then train the model, and finally update the model. The process is cumbersome and requires a lot of manpower.

[0028] The present invention utilizes the characteristic that the general trend of the signal is determined when the noise is too large, and uses the traditional measurement of the peak-to-peak value of the signal with a large signal-to-noise ratio. n The measured signal Data is greater than the measurement threshold of the parameter measurement n Accurate measurement results MeasureResult nIt automatically generates samples required for AI model training, solves the problem of automatic acquisition and automatic labeling of training samples in AI model training for measuring signal parameters with a small signal-to-noise ratio, and realizes automatic training and automatic updating of AI models, making the measurement system capable of self-improvement and more intelligent.

[0029] Figure 3 It is a flowchart of a specific implementation method of the present invention for real-time training and measuring of AI models using parameter measurement.

[0030] In this embodiment, if Figure 3 As shown, the method of using parameter measurement to train an AI model in real time and measure the parameters in the present invention includes the following steps:

[0031] Step S1: Use the traditional measurement method to obtain the peak-to-peak value of the measured signal PeaktoPeak n

[0032] For the nth measured signal, use Data n Indicates that the peak-to-peak value of the signal is directly obtained by inputting it into the measurement system using the traditional measurement method, and using Peak to Peak n express.

[0033] The measured signal is input into the parameter measurement system implemented by the present invention, and the simplest peak-to-peak value measurement is first performed. This step can be performed by taking the measured signal Data n The maximum and minimum values ​​of the peak-to-peak value are obtained by taking the difference between the two. n .

[0034] Step S2: Determine the peak-to-peak value PeaktoPeak n Is it greater than the parameter measurement threshold?

[0035] The present invention uses the peak-to-peak value of the measured signal Data n Divide the signal to determine the Data n Should it be processed by the traditional measurement method or the AI ​​measurement method? Both methods can give a set of measurement results. That is, if the peak-to-peak value of the signal is PeaktoPeak nIf the value is greater than the measurement threshold of the parameter measurement, then step S3 is executed; otherwise, step S8 is executed, wherein the measurement threshold is represented by Threshold, and the measurement threshold Threshold is determined by the signal-to-noise ratio of the measured signal. The lower the signal-to-noise ratio of the measured signal, the larger the measurement threshold Threshold. The lower the signal-to-noise ratio of the device or measurement system, the smaller the proportion of noise in the signal, the higher the signal-to-noise ratio of the signal, the better the signal quality, and the easier it is to measure. The smaller the measurement threshold Threshold of the device or measurement system; the higher the signal-to-noise ratio of the device or measurement system, the larger the proportion of noise in the signal, the lower the signal-to-noise ratio of the signal, the worse the signal quality, and the more difficult it is to measure. The larger the measurement threshold Threshold of the device or measurement system. The measured signal with a high signal-to-noise ratio is measured by traditional measurement, and the measured signal with a low signal-to-noise ratio is measured by traditional measurement.

[0036] Step S3: Data of the measured signal n Perform traditional measurements to obtain measurement results, using MeasureResult n Indicates that the output measurement result MeasureResult n At the same time, step S4 is executed to update the current AI model Model Current .

[0037] The present invention utilizes the characteristic that the general trend of the signal is determined when the noise is too large, and uses the traditional measurement of the peak-to-peak value of the signal with a large signal-to-noise ratio. n The measured signal Data is greater than the measurement threshold of the parameter measurement n Accurate measurement results MeasureResult n It automatically generates samples required for AI model training, solves the problem of automatic acquisition and automatic labeling of training samples in AI model training for measuring signal parameters with a small signal-to-noise ratio, and realizes automatic training and automatic updating of AI models, making the measurement system capable of self-improvement and more intelligent.

[0038] Step S4: Save the measurement result MeasureResult n And as a sample n Tags

[0039] Save the measurement results MeasureResult n , measured signal Data n As AI model training sample n , and the measurement result MeasureResult n As a sample nAmong them, the measurement result of traditional measurement MeasureResult n It contains many test items, including maximum value, minimum value, peak-to-peak value, average value, frequency, period, cycle maximum value, cycle minimum value, cycle peak-to-peak value, rise time, fall time, top value, bottom value, amplitude, effective value, standard deviation, positive overshoot, positive preshoot, negative overshoot, negative preshoot, positive pulse width, negative pulse width, area, median, number of data points, data rate, number of cycles, high time, low time, setup time, hold time, time outside the level, number of peak points, total number of samples, total number of waveforms and other measurement parameters. Each measurement parameter can be used as a mark item to mark the corresponding sample.

[0040] Step S5: Train the current AI model Current For the new AI model Model NeW

[0041] The labeled training sample Sample n Input into the online learning module to adjust the current AI model Current Conduct online learning and training, and the training result is a new AI model Model New .

[0042] Step S6: Use the test set to test the new AI model New Conduct a test.

[0043] Step S7: Determine the trained new AI model New Is the measurement accuracy better than the current AI model? Current , if the new AI model Model New Better than the current AI model Current , then the new AI model Model New As the current AI model Current Otherwise, keep the current AI model Current , return to step S1.

[0044] After multiple iterations, the current AI model is retained. Current The accuracy of identifying the measured signal will become higher and higher, and the measurement accuracy will become higher and higher.

[0045] Step S8: Read the current AI model Current .

[0046] Step S9: Use the current AI model Current Data of the measured signal n Perform measurement and obtain measurement results MeasureResultn withAI And output, return to step S1. Measurement result MeasureResult n withAI Marked as "This measurement result is given by the AI ​​model" to assist users in judging the credibility of the measurement value.

[0047] In the present invention, the measured signal may be measured by a traditional measurement method or an AI model measurement method. Regardless of which method is used, the measurement system will eventually give a set of measurement results. If the result is given by the AI ​​model, the measurement result of the traditional measurement may be empty, because the AI ​​model can broaden the scope of measurement, but cannot completely cover all measured signals.

[0048] In this embodiment, in order to improve the operating efficiency of the system, the training process from step S4 to step S7 is optimized to obtain multiple samples Sample n , for example, the jth sample Sample j To the kth sample Sample k The set SET = {Sample j , Sample j+1 , ..., Sample k-1 , Sample k}) and use the set SET to set the current AI model Model Current Train and update the current AI model Current The frequency of training and updating depends on the specific application scenario. Current The online learning method used can also be replaced by other machine learning or deep learning methods to implement the application of artificial intelligence models in different ways. Although the above describes the illustrative specific implementations of the present invention to facilitate the understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific implementations. For those of ordinary skill in the art, as long as the various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A method for real-time training and measuring of an AI model using parameter measurement, characterized in that: The following steps are involved: (1) For the measured signal measured for the nth time, use Data n Indicates that the peak-to-peak value of the signal is directly obtained by inputting it into the measurement system using the traditional measurement method, and using Peak to Peak n express; (2) If the peak-to-peak value of the signal is PeaktoPeak n If the value is greater than the measurement threshold of the parameter measurement, then step (3) is executed; otherwise, step (8) is executed, wherein the measurement threshold is represented by Threshold, and the measurement threshold Threshold is determined by the signal-to-noise ratio of the measured signal. The lower the signal-to-noise ratio of the measured signal, the larger the measurement threshold Threshold is. (3) Data of the measured signal n Perform traditional measurements to obtain measurement results, using MeasureResult n Indicates that the output measurement result MeasureResult n Execute step (4) at the same time; (4) Save the measurement results MeasureResult n , measured signal Data n As AI model training sample n , and the measurement result MeasureResult n As a sample n Labels; (5) The labeled training samples n Input into the online learning module to adjust the current AI model Current Conduct online learning and training, and the training result is a new AI model Model New ; (6) Use the test set to test the new AI model New Conduct testing; (7) Determine the trained new AI model New Is the measurement accuracy better than the current AI model? Current , if the new AI model Model New Better than the current AI model Current , then the new AI model Model New As the current AI model current Otherwise, keep the current AI model Current , return to step (1); (8) Read the current AI model Current ; (9) Use the current AI model Current Data of the measured signal n Perform measurement and obtain measurement results MeasureResult n withAI And output, return to step (1).

2. The method for real-time training of AI models and measuring using parameter measurement according to claim 1, characterized in that: The training process from step (4) to step (7) is optimized to obtain multiple samples Sample n The set SET after that is used for the current AI model Model Current Train and update.

Images