Neural network model training method for predicting occurrence of lag

Abstract

The invention relates to a training method of a neural network model for predicting occurrence of lag. The method comprises the following steps: acquiring system state information of equipment; based on the system state information, obtaining a lagging event information sample or a smooth event information sample; the lagging event information sample or the smooth event information sample is input into a system optimization neural network model, a prediction result is obtained, the prediction result is a probability value, and the probability value reflects the possibility of lagging of the system in a period of time in the future; and training the system optimization neural network model based on the prediction result back propagation algorithm. According to the technology, the response speed of the equipment can be obviously improved, the user experience is obviously enhanced, and the continuous and comprehensive improvement of the equipment performance is realized. On the whole, the technology brings smoother, more efficient and adaptive operation experience to the intelligent equipment.

Description

Technical Field

[0001] The present invention relates to the field of system optimization, and specifically provides a training method for a neural network model for predicting the occurrence of lag. Background Art

[0002] With the rise of smart devices and the bloated software, system lag has become an important factor affecting the user experience of smart devices.

[0003] In the existing technology, most methods rely on traditional manual adjustments or rule-based algorithms to avoid or reduce system freezes. However, this method often cannot fully cover all situations that may cause freezes, and lacks a dynamic, adaptive adjustment mechanism. As a result, the performance response of the device may not be satisfactory, and it is impossible to truly achieve a continuous and comprehensive optimization effect.

[0004] Accordingly, the art needs a new system optimization solution to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects, the present invention is proposed to provide a solution or at least partially solve the problem that the system optimization in the technical problem is too one-sided and lacks dynamic and adaptive capabilities.

[0006] In a first aspect, the present invention provides a training method for a neural network model for predicting the occurrence of lag, the method comprising: obtaining system status information of a device; obtaining a lag event information sample or a smooth event information sample based on the system status information; inputting the lag event information sample or the smooth event information sample into a system optimization neural network model to obtain a prediction result, wherein the prediction result is a probability value reflecting the possibility of the system lag occurring in a future period of time; and training the system optimization neural network model based on a back propagation algorithm of the prediction result.

[0007] In one of the above technical solutions, "obtaining a jam event information sample or a smooth event information sample based on the system status information" includes: when a jam is detected, obtaining the system status information of the previous time window before the jam as a jam event information sample.

[0008] In one of the above technical solutions, "obtaining system status information of a device" includes: saving the system status information in the form of a data structure of a ring buffer.

[0009] In one of the above-mentioned technical solutions, "inputting the jamming event information sample or the smooth event information sample into the system optimization neural network model to obtain a prediction result" includes: reducing the dimension of the jamming event information sample or the smooth event information sample to obtain a first event feature vector; inputting the first event feature vector into the system optimization neural network model to obtain a prediction result.

[0010] In one of the above technical solutions, the system optimizes the neural network model using the XLNet model.

[0011] In one of the above technical solutions, the system status information includes at least one of: hardware usage status information, application process information or user interaction information.

[0012] In one of the above-mentioned technical solutions, the hardware usage status information includes at least one of: cache occupancy information, CPU status information, memory status information or I / O status information, and the application process information includes at least one of: foreground process information, application event information or application running information.

[0013] In one of the above-mentioned technical solutions, "based on the system status information, obtaining a jamming event information sample or a smooth event information sample" includes: obtaining an abnormal event pool; when it is detected that an event in the abnormal event pool occurs but no jamming occurs, obtaining the system status information of this time window as a smooth event information sample.

[0014] In one of the above technical solutions, the number of the acquired smooth event information samples is greater than the number of the acquired jamming event information samples.

[0015] In a second aspect, the present invention provides an optimization method for improving system fluency, the method comprising: obtaining current system status information of a device; inputting the system status information into a trained system optimization neural network model according to the first aspect to obtain a prediction result; if the prediction result meets a preset condition, optimizing the device.

[0016] In one of the above-mentioned technical solutions, "if the prediction result meets the preset conditions, optimizing the device" includes: obtaining a freeze threshold; comparing the prediction result with the freeze threshold to determine whether the device is about to freeze; if it is determined that the device is about to freeze, optimizing the device.

[0017] In a third aspect, a control device is provided, which includes a processor and a storage device, wherein the storage device is suitable for storing multiple computer programs, and the computer programs are suitable for being loaded and run by the processor to execute the training method of the neural network model for predicting the occurrence of lag described in any of the technical solutions of the training method of the neural network model for predicting the occurrence of lag or the optimization method for improving the fluency of the system described in any of the technical solutions of the optimization method for improving the fluency of the system.

[0018] In a fourth aspect, a computer-readable storage medium is provided, which stores a plurality of computer programs, wherein the computer programs are suitable for being loaded and run by a processor to execute the training method for a neural network model for predicting the occurrence of lag described in any one of the technical solutions of the training method for a neural network model for predicting the occurrence of lag or the optimization method for improving the fluency of the system described in any one of the technical solutions of the optimization method for improving the fluency of the system.

[0019] The above one or more technical solutions of the present invention have at least one or more of the following Beneficial effects:

[0020] In the technical solution of the present invention, by obtaining the system status information of the device and generating corresponding event information samples, and then using the system optimization neural network model for prediction and training, the dynamic and adaptive optimization of the performance of the smart device is successfully achieved. This technology can significantly improve the response speed of the device, significantly enhance the user experience, and achieve continuous and comprehensive improvement of the device performance. In general, this technology brings a smoother, more efficient and adaptive operating experience to smart devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0022] Figure 1 It is a flowchart of main steps of a training method of a neural network model for predicting the occurrence of jamming according to an embodiment of the present invention;

[0023] Figure 2 is a flowchart of main steps of an optimization method for improving system fluency according to an embodiment of the present invention;

[0024] Figure 3is a flowchart of the secondary steps of a training method for a neural network model for predicting the occurrence of jamming according to an embodiment of the present invention;

[0025] Figure 4 is a schematic diagram of system status information within a time window of a training method for a neural network model for predicting occurrence of jamming according to an embodiment of the present invention;

[0026] Figure 5 The diagram is a schematic diagram of an execution framework of an optimization method for improving system fluency according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0028] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as computer programs, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store computer programs, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0029] The directional terms used in this document, such as "front", "front side", "front part", "rear", "rear side" and "rear part", are based on the front-to-back direction of the vehicle after the component is installed on the vehicle. The "longitudinal", "longitudinal" and "longitudinal section" mentioned in this document are based on the front-to-back direction after the component is installed on the vehicle, while "transverse", "transverse" and "cross section" refer to the direction perpendicular to the longitudinal direction.

[0030] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a method for training a neural network model for predicting the occurrence of jamming according to an embodiment of the present invention. Figure 1As shown, the training method of the neural network model for predicting the occurrence of lag in the embodiment of the present invention mainly includes the following steps S10 to S40.

[0031] Step S10: Obtain system status information of the device.

[0032] In this embodiment, the system status information is a parameter reflecting the current status of the system.

[0033] In one embodiment, various key state parameters of the device at a specific point in time are captured. These parameters, or features, form the basis for training a neural network model, enabling it to learn and predict future system behavior from these features. To make this clear, these parameters can be considered as a sample that represents the performance status of the device at a specific moment. By periodically collecting these samples, a dataset can be established that reflects the behavior of the device under different workloads and conditions.

[0034] In this embodiment, the system status information is a multi-dimensional collection, which includes at least one of hardware usage status information, application process information and user interaction information. Preferably, in one embodiment, the system status information is as follows: Figure 4 As shown, the system status information includes: hardware usage status information, application process information or user interaction information.

[0035] In this embodiment, the hardware usage status information is a record of the physical resource usage of the device. The hardware usage status information includes at least one of: cache occupancy information, CPU status information, memory status information or I / O status information. CPU status information may include current utilization, temperature and frequency, while I / O status information may involve disk read and write speed, network bandwidth usage, etc. This information provides in-depth insights into the current operating status of the device, making it easier to identify signs of lag later.

[0036] Application process information focuses on running applications. Application process information includes at least one of the following: foreground process information, application event information, or application running information. Foreground process information may include currently activated applications, their response time, and resource usage. Application event information may involve events such as application startup, shutdown, and update. This type of information enables the neural network model to understand the behavior of applications and their possible impact on system performance.

[0037] Preferably, in this embodiment, the system state information is stored in the form of a data structure of a ring buffer. The ring buffer provides an efficient way to store the system state information. Since the system state information is generated continuously and in real time, and only the most recent data may be considered for performance analysis and optimization, the ring buffer becomes an ideal choice. It can continuously collect and store the latest system state information, while automatically discarding outdated data.

[0038] Using a ring buffer data structure for storage means that the system can quickly access and retrieve recent status data without having to process large amounts of historical data. This approach provides high-speed access to data while ensuring the continuity and real-time nature of data storage, providing effective data support for subsequent performance analysis and optimization.

[0039] It should be noted that the device in this embodiment can be a mobile terminal, a computer terminal, a tablet computer, or other smart device that depends on an operating system.

[0040] Step S20: Based on the system status information, obtain a jam event information sample or a smooth event information sample.

[0041] In this embodiment, the jam event information sample refers to the information sample obtained in the last time window before the jam occurs. The smooth event information sample refers to other samples other than the jam event information sample. However, in order to achieve better results in the training of the neural network, other methods can be used to select the smooth event information sample.

[0042] In one implementation, the jam event information sample or the smooth event information sample is obtained through steps S201-S203, such as Figure 3 shown.

[0043] Step S201: When a freeze is detected, the system status information of the previous time window before the freeze is obtained as a freeze event information sample.

[0044] In this embodiment, the time window is a preset time period.

[0045] In one implementation, when a system freeze occurs, it is often caused by excessive demand or conflict for certain resources in a short period of time. Signs of these resource shortages or conflicts often begin to appear some time before the freeze. Therefore, capturing system status information before the freeze is the key to retrospectively analyzing these signs.

[0046] In this embodiment, the time window is a predefined, fixed-length time period that is used to normalize data collection. For example, if the time window is defined as 10 seconds, then at the moment when the freeze is detected, the system will immediately backtrack and capture all system status information for the previous 10 seconds. This information may include the usage of CPU, memory, I / O and other key hardware, as well as the activities of foreground and background processes, application events and user interactions.

[0047] Taking the CPU as an example, if the CPU usage rate continues to be higher than 90% in the time window before the lag, this may mean that a process or application has taken up a large amount of computing resources, resulting in other tasks not getting enough CPU time, thus causing lag.

[0048] In this implementation, the system status information is stored in the form of a data structure of a ring buffer as an example. Specifically, the system status information is collected regularly according to a time window and then sent to the ring buffer for storage. Each time window may include key information such as CPU usage, memory status, I / O status, etc., which provides an instant snapshot of the current state of the system.

[0049] In this implementation, when a certain system status information is detected as freeze information, the system status information of the previous time window is regarded as a freeze event information sample.

[0050] For example, if the system collects data every 10 seconds, and the ring buffer has the capacity of 5 time windows of data, when a freeze is detected in the sixth collected data, the data of the first time window in the ring buffer (that is, the fifth collected data) will be regarded as the freeze event information sample, and the data collected from the first to the fourth time will be overwritten by the subsequent new data.

[0051] In this embodiment, a method for detecting jamming is provided. For example, a typical smooth screen frame rate is 60 frames per second, which means that each frame should be rendered within about 16.67 milliseconds. In order to detect jamming, a threshold value, such as 20 milliseconds, can be set. When the rendering time of a frame exceeds this threshold value, it can be considered that jamming has occurred.

[0052] In summary, the core of step S201 is to quickly and accurately capture key system status information before the freeze when the freeze is detected. This information provides valuable data support for subsequent troubleshooting and system optimization.

[0053] Step S202: Acquire an abnormal event pool.

[0054] In this embodiment, the abnormal event pool is a collection of abnormal events.

[0055] In one implementation, the abnormal event pool is defined as a collection that includes various events that may cause system freezes, but not necessarily every time. These events usually include: excessive CPU utilization, memory leaks, I / O blocking, a large number of disk operations, or certain specific APPs being opened.

[0056] The main purpose of setting up an abnormal event pool is to provide a reference framework so that these events can be monitored and recorded in a real-time environment. When these events occur, the system does not always freeze, but these events provide important reference information for possible freezes.

[0057] Once the abnormal event pool is determined, these events can be continuously monitored while the application is running. When events in the abnormal event pool occur but there is no system freeze, the relevant system status information of these events will be marked as smooth event information samples. Such samples are extremely important for subsequent neural network training because they provide the network with examples that may not necessarily freeze when abnormal events occur, which helps to improve the network's judgment accuracy.

[0058] Step S203: when it is detected that an event in the abnormal event pool occurs but no freeze occurs, the system state information of this time window is obtained as a smooth event information sample.

[0059] In one implementation, when an event in the abnormal event pool is found to have occurred, but the application still runs smoothly without the expected lag, the system status information at this time becomes particularly important. This status information contains the key factors that did not cause the lag, which will be of great help to the subsequent neural network training.

[0060] The abnormal event pool ensures that the captured events are representative, and step S203 ensures that the situations in which the application remains smooth under these events are accurately recorded. This method effectively avoids a single, biased data source and enhances the generalization ability and prediction accuracy of the subsequent neural network model.

[0061] In this embodiment, preferably, the number of obtained smooth event information samples is greater than the number of obtained jam event information samples. As mentioned above, the number of smooth event information samples is significantly greater than the jam event information samples. Preferably, in this embodiment, the number of smooth event information samples is 3 to 5 times the number of jam event information samples.

[0062] During subsequent training, a subset of these smooth event information samples is randomly selected and combined with the jam event information samples to form an approximate 1:1 ratio. This strategy is called "downsampling" and its purpose is to ensure that both types of events are fully represented during training.

[0063] In this embodiment, the number of smooth event information samples obtained is greater than the number of jam event information samples to ensure diversity, and the diversity of smooth events can be captured, thereby ensuring that the selected subset can represent all possible smooth event scenarios. If only a small number of smooth event samples are collected, some key smooth event features or patterns may be missed.

[0064] Step S30: Input the jam event information sample or the smooth event information sample into the system optimization neural network model to obtain a prediction result.

[0065] In this embodiment, the prediction result is a probability value, which reflects the possibility of the system freezing in the future.

[0066] In one embodiment, the system optimization neural network model is a multi-layer structure, generally including an input layer, multiple hidden layers, and an output layer. Each layer is composed of multiple neurons, which are connected by weights. The model continuously adjusts these weights during the training process to learn how to predict the future performance of the system based on the input system state information. In this embodiment, the system optimization neural network model uses the XLNet model.

[0067] In this embodiment, both the jam event information samples and the smooth event information samples are converted into a form acceptable to the neural network, such as a vector, in which each element represents a specific state or feature of the system. After these samples are input into the input layer of the model, they are propagated layer by layer, and after the activation function and weight adjustment, a prediction result is finally generated at the output layer.

[0068] In this embodiment, the prediction result is a probability value between 0 and 1. When the value is close to 1, it means that the possibility of the system freezing in the future is high; and when the value is close to 0, it means that the possibility of freezing is low.

[0069] In this embodiment, the prediction result here reflects the possibility of the system freezing in the future. The length of the period of time is directly related to the selected positive sample. For example, if the selected freezing event information sample is a sample of the time window before the freezing occurs, then the theoretical system optimization neural network model predicts the probability of freezing in the next time window. If the selected freezing event information sample is a sample of the second time window before the freezing occurs, then the theoretical system optimization neural network model predicts the probability of freezing in the next time window. So in general, the period of time is a short time in the future, or it can be said that the period of time here is a preset time, and the preset time length here is related to the selected positive sample (freezing event information sample).

[0070] Preferably, in this embodiment, the prediction result is obtained through steps S301-S302, which are as follows:

[0071] Step S301: reducing the dimension of the jamming event information sample or the smooth event information sample to obtain a first event feature vector.

[0072] In this embodiment, the first event feature vector is one-dimensional data after dimensionality reduction.

[0073] In one implementation, Word2Vec is used for dimensionality reduction. Word2Vec is designed to convert text into a vector of a fixed size.

[0074] In this embodiment, firstly, the jam event information samples or smooth event information samples are appropriately preprocessed, and the state information in each sample is regarded as a "word", thereby obtaining a "sentence". Next, these "sentences" are trained using the Word2Vec model to learn the embedding vector of each state information.

[0075] After training, the embedding vectors of all state information in each sample can be averaged or weighted to obtain a vector representing the entire sample, namely the first event feature vector. This feature vector captures the main information in the original sample, but its dimension is much smaller than the original sample, making it more suitable for subsequent neural network processing.

[0076] In addition, another advantage of using Word2Vec for dimensionality reduction is that it can capture the relationship and similarity between state information. For example, two similar system states may be mapped to close positions in the space, which helps neural networks better understand and process this data.

[0077] In general, step S301 utilizes the powerful function of Word2Vec to convert high-dimensional system state information into low-dimensional feature vectors, thereby providing effective input data for subsequent neural network analysis.

[0078] Step S302: Input the first event feature vector into the system optimization neural network model to obtain a prediction result.

[0079] In this embodiment, in each iteration, the model receives a batch of feature vectors and tries to predict their corresponding labels. The label here is the actual performance of the system, for example, a sample of jamming event information is a jamming label, and a sample of smooth event information is a smooth label. Then, the output of the model is compared with the true label to calculate the value of the loss function. Common loss functions include mean square error, cross entropy loss, etc.

[0080] Step S40: Optimizing the neural network model based on the prediction result back propagation algorithm training system.

[0081] In this embodiment, the back propagation algorithm automatically adjusts the weights according to the prediction error, so that it can predict the system state more accurately.

[0082] In one embodiment, when the neural network model performs forward propagation and generates a prediction result, there is a certain error between the model output and the true label. This error is measured by a loss function, which describes the difference between the predicted value and the actual value.

[0083] In order to optimize this error, the model needs to know how to adjust its internal weights. The backpropagation algorithm is the tool for this purpose. It starts at the output layer, propagates the error back to each layer, and calculates the gradients of the weights of each layer. These gradients indicate how the weights should be adjusted to reduce the error in the next iteration.

[0084] Specifically, each neuron receives an input value and produces an output value. The relationship between these inputs and outputs is defined by an activation function such as ReLU, Sigmoid, or Tanh. The input of a neuron is related to the product of its weight and bias. Therefore, by adjusting the weights and biases, the output of the neuron can be changed.

[0085] During the backpropagation phase, the chain rule is used to calculate the partial derivative of the loss function with respect to each weight. This partial derivative (or gradient) indicates how the weights should be adjusted. Then, an optimizer such as stochastic gradient descent, Adam, or RMSProp is used to update the weights and biases based on these gradients.

[0086] This process is repeated multiple times, i.e. multiple iterations, until the model's loss drops to an acceptable level or other stopping criteria are met.

[0087] Figure 2 FIG. 1 is a flow chart of the main steps of an optimization method for improving system fluency according to an embodiment of the present invention. Figure 2 As shown, the optimization method for improving system fluency in the embodiment of the present invention mainly includes the following steps S50 to S70.

[0088] Step S50: Obtain the current system status information of the device.

[0089] In one implementation, the system status information generally includes a series of parameters and indicators to reflect the operating status of the device. Such information may include, but is not limited to, hardware usage status information, application process information, and user interaction information.

[0090] Preferably, in this embodiment, Figure 5As shown, in order to ensure data privacy and processing speed, the system optimization neural network model can be pre-loaded into the local storage of the device, and the system optimization neural network model becomes a local model. This means that when the device needs to obtain system status information for analysis, all data processing is done locally without sending the data to an external server. This localized processing method not only protects user privacy, but also reduces network latency, thereby providing users with feedback and optimization suggestions more quickly.

[0091] Step S60: input the system status information into the trained system optimization neural network model to obtain a prediction result.

[0092] In this embodiment, this step is similar to step S30 and will not be described again here.

[0093] Step S70: selectively optimizing the equipment based on the prediction results.

[0094] In one implementation, based on these prediction results, it can be determined whether a jam will occur. The optimization strategies can be divided into the following categories:

[0095] Resource management: If the prediction indicates that certain resources (such as CPU or RAM) are being used too high, you can take action to free up those resources. This might include killing unnecessary background processes, adjusting the priority of running applications, or cleaning up cached data.

[0096] Network optimization: Based on the prediction results, if the network connection is unstable or the speed is limited, you can try to reconnect to the network, switch to a different network band, or optimize the network settings.

[0097] Storage management: If the storage is close to full or there are redundant files, users can be reminded to clean up unnecessary files or recommended to defragment the disk.

[0098] Software updates and repairs: The prediction results may indicate that some applications or system components need to be updated or repaired. In this case, the system will recommend users to perform the corresponding updates or repairs to ensure smooth operation of the device.

[0099] In this implementation, a specific optimization measurement is provided, such as steps S701-S703, which are specifically as follows:

[0100] Step S701: Obtain the freeze threshold.

[0101] In one embodiment, the jamming threshold is a pre-set standard for determining when a device may reach its performance limit and may cause jamming.

[0102] Step S702: Compare the prediction result with the freeze threshold to determine whether the device is about to freeze.

[0103] In one embodiment, after receiving the prediction result of the neural network model, the system compares the result with the jamming threshold. If the prediction result exceeds the jamming threshold, it means that the device has a high probability of jamming or performance degradation in future operations.

[0104] For example, if the jamming threshold is 0.8 and the prediction result is 0.85, this means that the device is more likely to experience jamming in future operations, so optimization measures need to be taken.

[0105] Step S703: If it is determined that the device is about to freeze, the device is optimized.

[0106] In one implementation, once the system determines that the device may be stuck, it will take a series of optimization measures to prevent potential performance issues.

[0107] The system will detect abnormal status and background load of the current device. Abnormal status can be any abnormal operation, such as application crash or excessive memory usage. In addition, background load, such as background network access, download tasks, disk I / O operations, etc., will also be detected by the system.

[0108] After identifying these possible optimization targets, the system will take targeted actions. For example, based on the predicted probability of jamming, the system will proportionally limit the anomalies and load in the background. For example, if the current jamming probability is 80%, the resource usage of applications running in the background, especially those with anomalies or excessive load, will be reduced by 80%.

[0109] Specifically for applications with abnormal CPU, the system can adjust the CPU scheduling policy, reduce the scheduling priority of these applications or limit their CPU clock speed, so as to ensure that their CPU usage is effectively controlled. This can not only prevent device lag, but also improve the device's response speed and overall performance.

[0110] In addition to directly limiting resource usage, the system can also adopt other strategies, such as delaying or suspending certain non-critical tasks until the device status returns to normal. In this way, the user experience can be maximized while ensuring that the device runs in the best state.

[0111] In this embodiment, considering the daily usage pattern of the device and the user's habits, the system optimization neural network model is retrained when the device is idle, such as late at night. This strategy is intended to reduce interference with the user's daily activities while ensuring that the device is updated in a highly efficient state.

[0112] After retraining is complete, the system automatically replaces the old model with the new one, ensuring that the device always uses the latest and most optimized model in daily use. This dynamic adaptation method allows the model to continuously learn and adapt to changes in the device, thereby providing a more accurate and efficient optimization strategy to ensure that users have a smooth and fast device experience.

[0113] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of each of the above method embodiments when executed by the processor. Among them, the computer program includes a computer program, and the computer program can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electric carrier signals and telecommunication signals.

[0114] Furthermore, the present invention also provides a control device. In one embodiment of the control device according to the present invention, the control device includes a processor and a storage device. The storage device can be configured to store a program for executing the training method of the neural network model for predicting the occurrence of jamming or the optimization method for improving the system fluency of the above-mentioned method embodiment. The processor can be configured to execute the program in the storage device, which includes but is not limited to executing the training method of the neural network model for predicting the occurrence of jamming or the optimization method for improving the system fluency of the above-mentioned method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0115] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the training method of the neural network model for predicting the occurrence of jamming or the optimization method for improving the system fluency of the above-mentioned method embodiment, and the program can be loaded and run by the processor to implement the above-mentioned training method of the neural network model for predicting the occurrence of jamming or the optimization method for improving the system fluency. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0116] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0117] Those skilled in the art will appreciate that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0118] The relevant user personal information that may be involved in the various embodiments of this application is strictly in accordance with the requirements of laws and regulations, following the principles of legality, legitimacy and necessity, based on the reasonable purposes of business scenarios, to process the personal information that users actively provide during the use of products / services or generated due to the use of products / services, as well as the personal information obtained with the user's authorization.

[0119] The user personal information processed by the applicant will vary depending on the specific product / service scenario, and shall be based on the specific scenario in which the user uses the product / service, which may involve the user's account information, device information, driving information, vehicle information or other related information. The applicant will treat the user's personal information and its processing with a high degree of diligence.

[0120] The Applicant attaches great importance to the security of user personal information and has adopted reasonable and feasible security protection measures that comply with industry standards to protect user information and prevent personal information from being accessed, disclosed, used, modified, damaged or lost without authorization.

[0121] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A training method for a neural network model for predicting the occurrence of jamming, characterized in that: include: Get the system status information of the device; Based on the system status information, obtain a jam event information sample or a smooth event information sample; Input the jam event information sample or the smooth event information sample into the system optimization neural network model to obtain a prediction result; wherein the prediction result is a probability value, and the probability value reflects the possibility of jamming of the system in the future period of time; The system optimization neural network model is trained based on the prediction results by back propagation algorithm.

2. The training method of a neural network model for predicting the occurrence of jamming according to claim 1, characterized in that: “Based on the system status information, obtaining a jam event information sample or a smooth event information sample” includes: When a jam is detected, the system status information of the previous time window before the jam is obtained as a jam event information sample.

3. The training method of a neural network model for predicting the occurrence of jamming according to claim 1, characterized in that: "Get device system status information", including: The system status information is stored in the form of a data structure of a ring buffer.

4. The training method of a neural network model for predicting the occurrence of jamming according to any one of claims 1 to 3, characterized in that: "Inputting the jam event information sample or the smooth event information sample into the system optimization neural network model to obtain a prediction result", including: Performing dimension reduction on the jamming event information sample or the smooth event information sample to obtain a first event feature vector; The first event feature vector is input into the system optimization neural network model to obtain a prediction result.

5. The training method of a neural network model for predicting the occurrence of jamming according to any one of claims 1 to 3, characterized in that: The system status information includes at least one of: hardware usage status information, application process information or user interaction information.

6. The training method of a neural network model for predicting the occurrence of jamming according to claim 5, characterized in that: The hardware usage status information includes at least one of cache occupancy information, CPU status information, memory status information or I / O status information, and the application process information includes at least one of foreground process information, application event information or application running information.

7. The training method of a neural network model for predicting the occurrence of jamming according to claim 1, characterized in that: “Based on the system status information, obtaining a jam event information sample or a smooth event information sample” includes: Get the exception event pool; When it is detected that an event in the abnormal event pool occurs but no lag occurs, the system status information of this time window is obtained as a smooth event information sample.

8. The training method of a neural network model for predicting the occurrence of jamming according to claim 1, characterized in that: The number of the acquired smooth event information samples is greater than the number of the acquired jam event information samples.

9. An optimization method for improving system fluency, characterized in that: include: Get the current system status information of the device; Inputting the system status information into the trained system optimization neural network model according to any one of claims 1 to 8 to obtain a prediction result; If the prediction result meets the preset conditions, the device is optimized.

10. The optimization method for improving system fluency according to claim 9, characterized in that: “If the prediction result meets the preset conditions, optimizing the device” includes: Get the jamming threshold; Compare the prediction result with the freeze threshold to determine whether the device is about to freeze; If it is determined that the device is about to freeze, the device is optimized.

Images