A neural network model-based real-time system wake-up latency compensation method, system, electronic device, and computer-readable storage medium

Abstract

The application provides a real-time system wake-up delay compensation method and system based on a neural network model, an electronic device and a computer readable storage medium. The method comprises the following steps: obtaining a real delay wake-up time value and system wake-up related data when a real-time system is woken up from a low-power mode, and obtaining a training set and a test set after preprocessing; optimizing a neural network model by using the training set and the test set, and obtaining an optimized neural network model; collecting system wake-up related data when a to-be-tested real-time system is woken up from a low-power mode, preprocessing the system wake-up related data, inputting the preprocessed system wake-up related data into the optimized neural network model to obtain a delay wake-up time prediction value of the to-be-tested real-time system, and adjusting a sleep time of the to-be-tested real-time system according to the delay wake-up time prediction value, so as to compensate for the delay and ensure the wake-up real-time performance. The application uses a neural network to predict the low-power system wake-up delay, and improves the real-time performance through a compensation mechanism.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of system control, and particularly relates to a real-time system wake-up time delay compensation method and system based on a neural network model, an electronic device, and a computer readable storage medium. BACKGROUND

[0002] Tickless is a low-power mechanism in a real-time operating system. Tick usually refers to a periodic interrupt generated by a system timer, that is, a "tick" signal, and Tickless is understood as running with reduced or no such tick clock beats. In this mode, the system can adjust the generation frequency of the timer interrupt according to the actual needs of the current task, and reduce the interrupt as much as possible when there is no active task, so that the system enters a low-power state. At the same time, the operating system calculates the time of the next task or event that needs to be processed, and sets a corresponding timer. The processor enters a low-power sleep mode and is awakened only when the timer expires.

[0003] However, the real-time system often has a time delay when it is awakened from the low-power mode, which affects the real-time performance of the system. SUMMARY

[0004] To solve the above technical problems, the application provides a real-time system wake-up time delay compensation method and system based on a neural network model, an electronic device, and a computer readable storage medium.

[0005] The first aspect of the application discloses a real-time system wake-up time delay compensation method based on a neural network model, which comprises the following steps:

[0006] S1, obtaining a delay wake-up time real value when a real-time system is awakened from a low-power mode and system wake-up related data, and pre-processing the obtained data to obtain a training set and a test set; the system wake-up related data includes: sleep time, hardware information weighted value, system load value, task information weighted value received by the system before wake-up, and task information weighted value received by the system after wake-up;

[0007] S2, inputting the training set into a constructed neural network model for training, updating the weights and biases of the neural network model through a back propagation algorithm, and obtaining a trained neural network model;

[0008] S3, inputting the test set into the trained neural network model to obtain a delay wake-up time prediction value, calculating the error between the delay wake-up time prediction value and the corresponding delay wake-up time real value in the test set, and optimizing the structure and hyperparameters of the neural network model according to the error to obtain an optimized neural network model;

[0009] S4, the system wake-up related data of the to-be-tested real-time system when being woken up from the low-power mode is collected, preprocessed, and input into the optimized neural network model to obtain a delay wake-up time prediction value of the to-be-tested real-time system, and the sleep time of the to-be-tested real-time system is adjusted according to the delay wake-up time prediction value of the to-be-tested real-time system, so as to compensate for the delay and ensure the wake-up real-time performance.

[0010] Optionally, in step S1, the task information includes task priority, task stack size, task execution time and task resource requirement; and the hardware information includes hardware main frequency and peripheral performance.

[0011] Optionally, in step S1, the acquired data is preprocessed, specifically including:

[0012] The acquired data is cleaned to remove abnormal values and noise data.

[0013] The cleaned data is normalized, and the normalized data is divided into a training set and a test set.

[0014] Optionally, the normalization manner is Min-Max scaling or Z-Score standardization.

[0015] Optionally, in step S2, the neural network model adopts one or a combination of multiple MLP neural networks, RNN neural networks, LSTM neural networks and GRU neural networks.

[0016] Optionally, in step S3, the mean square error or the mean absolute error is used when calculating the error, and the optimization of the structure and the hyperparameter of the neural network model includes adjusting the learning rate, increasing the training rounds or expanding the size of the hidden layer.

[0017] The second aspect of the present application discloses a real-time system wake-up delay compensation system based on a neural network model, which comprises:

[0018] The first processing module is configured to acquire a delay wake-up time true value and system wake-up related data when a real-time system is woken up from a low-power mode, and preprocess the acquired data to obtain a training set and a test set; the system wake-up related data includes sleep time, hardware information weighted value, system load value, task information weighted value received by the system before wake-up and task information weighted value received by the system after wake-up.

[0019] The second processing module is configured to input the training set into the constructed neural network model for training, update the weight and bias of the neural network model through a back propagation algorithm, and obtain a trained neural network model.

[0020] The third processing module is configured to input the test set into the trained neural network model to obtain a delay wake-up time prediction value, calculate an error between the delay wake-up time prediction value and a corresponding delay wake-up time true value in the test set, and optimize the structure and hyperparameters of the neural network model according to the error to obtain an optimized neural network model.

[0021] The fourth processing module is configured to collect system wake-up related data of the to-be-tested real-time system when the to-be-tested real-time system is woken up from the low-power consumption mode, pre-process the system wake-up related data, input the pre-processed system wake-up related data into the optimized neural network model to obtain a delay wake-up time prediction value of the to-be-tested real-time system, and adjust a sleep time of the to-be-tested real-time system according to the delay wake-up time prediction value of the to-be-tested real-time system to compensate for the delay and ensure the real-time wake-up.

[0022] The third aspect of the present application discloses an electronic device. The electronic device comprises a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the real-time system wake-up delay compensation method based on a neural network model are implemented.

[0023] The fourth aspect of the present application discloses a computer readable storage medium. The computer readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the real-time system wake-up delay compensation method based on a neural network model are implemented.

[0024] In summary, the scheme provided by the present application has the following technical effects: the present application uses a neural network to predict the delay of a low-power consumption system wake-up, and improves the real-time performance through a compensation mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0026] Figure 1 The flow chart of the real-time system wake-up delay compensation method based on a neural network model of the embodiment of the present application;

[0027] Figure 2 The data collection schematic diagram of the embodiment of the present application;

[0028] Figure 3 The neural network model structure schematic diagram of the embodiment of the present application;

[0029] Figure 4A neural network training flowchart for an embodiment of the present application;

[0030] Figure 5 A structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of the present application.

[0032] When a real-time system wakes up from a low-power mode under a tickless mechanism, time delay often occurs, which affects the real-time performance of the system. The main reasons include hardware-related delay and software-related delay. The hardware-related reasons mainly include three reasons: clock recovery time, peripheral device initialization, and corresponding time of power supply. The clock recovery time is that when the system wakes up from the low-power mode, the clock source needs to be restarted. For example, the main clock (such as HSE) of an STM32 microcontroller may be turned off in the stop mode, and the clock needs to be restarted when waking up, which introduces a certain delay. The basic delay is the start-up time of the HSI oscillator, and if the voltage regulator works in the low-power mode, the time of the voltage regulator switching from the low-power mode to the normal mode also needs to be added. The peripheral device initialization is that the peripheral devices (such as GPIO, ADC, USART, etc.) may be turned off in the low-power mode, and these devices need to be reinitialized when waking up, which increases the wake-up delay. The last point is that the power supply may reduce the frequency in the low-power mode, and needs to be restored to the normal working state when waking up, which takes a certain time. The software-related reason is mainly the stop of the system tick interrupt, that is, in some low-power modes, the system tick interrupt may be stopped, and the system tick interrupt needs to be restarted after waking up, which introduces additional delay.

[0033] Therefore, the present application aims to design a real-time system wake-up delay compensation method based on a neural network model to compensate for the time delay caused by unknown reasons when the real-time system wakes up from the low-power mode, so that the real-time system can also guarantee good real-time performance while retaining the low-power capability.

[0034] The first aspect of the present application discloses a real-time system wake-up delay compensation method based on a neural network model, the method comprising:

[0035] S1, obtaining a real value of a delay wake-up time when a real-time system is woken up from a low-power mode and system wake-up related data, and pre-processing the obtained data to obtain a training set and a test set; the system wake-up related data includes: a sleep time, a hardware information weighted value, a system load value, a task information weighted value received by the system before wake-up, and a task information weighted value received by the system after wake-up;

[0036] Optionally, in step S1, the task information includes: a task priority, a task stack size, a task execution time, and a task resource requirement; and the hardware information includes a hardware main frequency and peripheral performance.

[0037] In this step, the dynamic characteristics of the task and the hardware (such as priority and main frequency) are combined as model inputs, which improves the prediction accuracy.

[0038] Optionally, in step S1, the obtained data is pre-processed, specifically including:

[0039] The obtained data is cleaned to remove abnormal values and noise data;

[0040] The cleaned data is normalized, and the normalized data is divided into a training set and a test set.

[0041] Optionally, the normalization method is Min-Max scaling or Z-Score standardization (standard deviation standardization).

[0042] S2, inputting the training set into the constructed neural network model for training, updating the weights and biases of the neural network model through a back propagation algorithm, and obtaining a trained neural network model;

[0043] Optionally, in step S2, the neural network model adopts one or a combination of MLP neural network, RNN neural network, LSTM neural network, and GRU neural network.

[0044] When the neural network model adopts the MLP neural network, it is applicable to the collected pre-wake-up tasks and post-wake-up tasks, and there is no time sequence correlation between the pre-wake-up tasks and the post-wake-up tasks, that is, the implementation of the post-wake-up tasks generally does not depend on the data or completion of the pre-wake-up tasks. For application scenarios with more front and back associations, a time sequence neural network model (RNN neural network, LSTM neural network, and GRU neural network) can be replaced to improve the prediction accuracy.

[0045] S3, inputting the test set into the trained neural network model to obtain a delay wake-up time prediction value, calculating an error between the delay wake-up time prediction value and a corresponding delay wake-up time real value in the test set, and optimizing the structure and hyperparameters of the neural network model according to the error to obtain an optimized neural network model;

[0046] Optionally, in step S3, the mean square error or the mean absolute error is used when calculating the error, and the optimization of the structure and hyperparameters of the neural network model includes adjusting the learning rate, increasing the number of training rounds, or expanding the size of the hidden layer.

[0047] In this step, the neural network model uses offline training when training and optimizing, and the model is not trained synchronously during the real-time operation of the system, which further causes additional energy consumption of the system during operation.

[0048] S4, the system wake-up related data of the to-be-tested real-time system when it is woken up from the low-power consumption mode is collected, preprocessed, and input into the optimized neural network model to obtain the delay wake-up time prediction value of the to-be-tested real-time system, and the sleep time of the to-be-tested real-time system is adjusted according to the delay wake-up time prediction value of the to-be-tested real-time system to compensate for the delay and ensure the wake-up real-time performance.

[0049] In this step, the trained model is deployed to the real-time system, and the system sleep time is dynamically adjusted according to the prediction value to offset the delay and ensure the wake-up real-time performance.

[0050] As an implementable manner, the implementation process of the technical scheme of the present application is as shown in Figure 1 , which includes:

[0051] Step 1: Data collection and processing

[0052] Collect data: Collect a certain number of delay wake-up times (i.e. the time when the system wakes up late, such as the system should normally wake up at 5 seconds, but actually wakes up at 5.1 seconds, so the delay wake-up time is 0.1 seconds) and system wake-up related data (system wake-up related data includes sleep time, hardware information weighting value, system load value, task information weighting value received by the system before wake-up, and task information weighting value received by the system after wake-up) when the system wakes up from the low-power consumption mode. Task information generally includes task priority, task stack size, task execution time, and task resource demand. Hardware information generally includes hardware frequency and peripheral performance. System wake-up related data can be directly obtained or calculated by system logs, sensor data, etc.

[0053] Data cleaning: clean the collected data to remove abnormal values and noise data to ensure the quality and reliability of the data.

[0054] Data normalization: normalize the cleaned data to scale the data to a fixed range (such as between 0 and 1) to facilitate model training.

[0055] Divide the training set and the test set: divide the normalized data into a training set and a test set, usually according to a certain proportion (such as 80% of the training set and 20% of the test set).

[0056] The invention collects relevant data as shown in Figure 2 The hardware information data is collected and a weighted value is calculated during the system startup phase. The task information is collected and a weighted value is calculated when the task is about to enter sleep before wake-up. The task information and the delay wake-up time are collected after the task is started after wake-up. The sleep time T calculated by a specific function in the system tickless is collected during the system sleep phase, and the system load information is collected. The above five wake-up related data are collected as a set, and a total of N (N >= 200) sets of data are collected for constructing the input data set of the model. Correspondingly, the system also needs to collect the N sets of delay wake-up time data for constructing the output data set of the model.

[0057] Step 2: Model construction and training

[0058] Model construction preparation: use a deep learning framework (such as TensorFlow or PyTorch) to build a neural network model. The input layer of the model receives the processed wake-up related data, the hidden layer can use one layer, and the output layer outputs the predicted delay wake-up time value.

[0059] Set hyperparameters: set the hyperparameters of the model, such as the number of neurons in the hidden layer, the learning rate, the batch size, the number of training rounds, etc. These hyperparameters can be adjusted according to experiments and experience to optimize the performance of the model.

[0060] Train the model: input the training set data into the constructed neural network model for training, and update the weights and biases of the model through the backpropagation algorithm, so that the model can learn the delay features in the data.

[0061] As shown in Figure 3 , the neural network model of the embodiment of the invention is divided into three layers, the input layer has 5 neurons, representing that the input variable dimension of the model is 5, each dimension respectively represents the weighted value of the pre-wake-up task information, the weighted value of the post-wake-up task information, the sleep time T, the weighted value of the hardware information and the system load. The number of neurons in the hidden layer uses the conventional hidden layer neuron setting method, that is, it is directly set to twice the number of neurons in the input layer, which is set to 10. The output layer has only one neuron, representing the delay wake-up time.

[0062] As shown in Figure 4 , during the training process of the model, the loss function and the gradient of the weight value are calculated according to the predicted output value of the model. Then, through the reverse gradient calculation formula, the multiple weight values in the neural network model are updated and adjusted, so that the model can better fit the relationship between the input and the output.

[0063] Step 3: Model evaluation and optimization

[0064] Evaluate model performance: Evaluate the trained model using the test set data, calculate the error between the predicted delay wake-up time and the true value of the delay wake-up time in the test set, such as mean square error (MSE), mean absolute error (MAE), etc. The error should be as small as possible to prove the high prediction accuracy of the model.

[0065] Analyze the prediction results: Observe the prediction results of the model, analyze the differences and trends between the predicted delay wake-up time and the true value. If there is a large delay or error in the prediction value of the model, further adjust the structure and hyperparameters of the model, or try other optimization methods.

[0066] Optimize the model: According to the evaluation results and analysis, optimize the model. For example, increase the number of neurons in the hidden layer, adjust the learning rate, increase the number of training rounds, etc. to improve the prediction accuracy of the model and reduce the delay.

[0067] In particular, the main work of optimizing the model is to set the hyperparameters in this step. The performance of the model after the first training is difficult to meet the requirements, and this step of optimizing the model needs to adjust the model structure and other methods to make the model better understand the relationship between the input data and the output data.

[0068] Step 4: Prediction and application

[0069] Real-time prediction: Deploy the optimized neural network model to the real-time system, collect system wake-up related data in real time, perform preprocessing and feature extraction, and input it into the model for prediction to obtain the predicted delay wake-up time caused by the wake-up.

[0070] Real-time compensation: The system compensates the predicted value of the model to the timer used for wake-up, so that the system reduces the sleep time to ensure the real-time accuracy of the system wake-up.

[0071] The second aspect of the present application discloses a real-time system wake-up delay compensation system based on a neural network model, which comprises:

[0072] The first processing module is configured to obtain the true value of the delay wake-up time when the real-time system is woken up from the low-power mode and the system wake-up related data, and preprocess the obtained data to obtain the training set and the test set. The system wake-up related data includes: the weighted value of the task information before and after wake-up, the sleep time, the weighted value of the hardware information, and the system load value.

[0073] The second processing module is configured to input the training set into the constructed neural network model for training, update the weights and biases of the neural network model through a back propagation algorithm, and obtain a trained neural network model.

[0074] The third processing module is configured to input the test set into the trained neural network model, obtain a delay wake-up time prediction value, calculate an error between the delay wake-up time prediction value and a corresponding delay wake-up time true value in the test set, and optimize the structure and hyperparameters of the neural network model according to the error, to obtain an optimized neural network model.

[0075] The fourth processing module is configured to collect system wake-up related data of a to-be-tested real-time system when the to-be-tested real-time system is woken up from a low-power consumption mode, pre-process the system wake-up related data, input the pre-processed system wake-up related data into the optimized neural network model to obtain a delay wake-up time prediction value of the to-be-tested real-time system, and adjust a sleep time of the to-be-tested real-time system according to the delay wake-up time prediction value of the to-be-tested real-time system, to compensate for a delay and ensure real-time wake-up.

[0076] The third aspect of the present application discloses an electronic device. The electronic device comprises a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the real-time system wake-up delay compensation method based on a neural network model are implemented.

[0077] Figure 5 As shown in a structural diagram of an electronic device according to an embodiment of the present application, Figure 5 The electronic device comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the electronic device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, near field communication (NFC) or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0078] Those skilled in the art can understand, Figure 5The structures shown in the figures are only structural diagrams of parts related to the technical solutions of the present disclosure, and do not constitute a limitation on the electronic device to which the solutions of the present application are applied. The specific electronic device can include more or fewer components than those shown in the figures, or combine certain components, or have a different arrangement of components.

[0079] The fourth aspect of the present application discloses a computer readable storage medium. The computer readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the real-time system wake-up latency compensation method based on a neural network model are implemented.

[0080] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time system wake-up delay compensation method based on a neural network model, characterized in that, The method includes: S1: Obtain the actual value of the delayed wake-up time and system wake-up-related data when the real-time system is woken up from low-power mode, and preprocess the obtained data to obtain the training set and test set; the system wake-up-related data includes: sleep time, hardware information weighting value, system load value, weighting value of task information received by the system before wake-up and weighting value of task information received by the system after wake-up; in step S1, the task information includes: task priority, task stack size, task execution time and task resource requirements; the hardware information includes hardware clock frequency and peripheral performance; S2, input the training set into the constructed neural network model for training, and update the weights and biases of the neural network model through the backpropagation algorithm to obtain the trained neural network model; S3. Input the test set into the trained neural network model to obtain the predicted value of the delayed wake-up time. Calculate the error between the predicted value of the delayed wake-up time and the corresponding true value of the delayed wake-up time in the test set. Optimize the structure and hyperparameters of the neural network model based on this error to obtain the optimized neural network model. S4. Collect system wake-up related data when the real-time system under test is woken up from low-power mode, and after preprocessing, input it into the optimized neural network model to obtain the predicted value of the delayed wake-up time of the real-time system under test. Adjust the sleep time of the real-time system under test according to the predicted value of the delayed wake-up time of the real-time system under test to compensate for the delay and ensure the real-time wake-up.

2. The method according to claim 1, characterized in that, In step S1, the acquired data is preprocessed, specifically including: The acquired data is cleaned to remove outliers and noisy data; The cleaned data is normalized, and the normalized data is then divided into training and testing sets.

3. The method according to claim 2, characterized in that, The normalization process is either Min-Max scaling or Z-Score standardization.

4. The method according to claim 1, characterized in that, In step S2, the neural network model uses either an MLP neural network or an RNN neural network.

5. The method according to claim 1, characterized in that, In step S3, the mean squared error or mean absolute error is used to calculate the error. The optimization of the structure and hyperparameters of the neural network model includes adjusting the learning rate, increasing the number of training rounds, or expanding the size of the hidden layer.

6. A real-time system wake-up delay compensation system based on a neural network model, characterized in that, The system includes: The first processing module is configured to acquire the actual value of the delayed wake-up time and system wake-up-related data when the real-time system is woken up from low-power mode, and preprocess the acquired data to obtain training and testing sets. The system wake-up-related data includes: sleep time, hardware information weighting value, system load value, weighting value of task information received by the system before wake-up, and weighting value of task information received by the system after wake-up. The task information includes: task priority, task stack size, task execution time, and task resource requirements. The hardware information includes hardware clock speed and peripheral performance. The second processing module is configured to input the training set into the constructed neural network model for training, and update the weights and biases of the neural network model through the backpropagation algorithm to obtain the trained neural network model. The third processing module is configured to input the test set into the trained neural network model, obtain the predicted value of the delayed wake-up time, calculate the error between the predicted value of the delayed wake-up time and the corresponding true value of the delayed wake-up time in the test set, and optimize the structure and hyperparameters of the neural network model based on the error to obtain the optimized neural network model. The fourth processing module is configured to collect system wake-up related data when the real-time system under test is woken up from low-power mode, preprocess the data, input it into the optimized neural network model to obtain the predicted value of the delayed wake-up time of the real-time system under test, and adjust the sleep time of the real-time system under test according to the predicted value of the delayed wake-up time of the real-time system under test to compensate for the delay and ensure the real-time wake-up.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the steps of the real-time system wake-up delay compensation method based on a neural network model according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the real-time system wake-up delay compensation method based on a neural network model according to any one of claims 1 to 5.

Images