Adaptive pervasive edge computing task unloading method based on imitation learning

Abstract

The invention relates to the technical field of edge computing, and discloses a self-adaptive pervasive edge computing task unloading method based on imitation learning, and the method comprises the steps: firstly constructing a task unloading strategy imitation model and an environment adaptability evaluation model, the former is used for learning and simulating a historical task unloading strategy, and the latter is used for evaluating the influence of environment change on the strategy; a deep reinforcement learning algorithm is introduced to train a simulation model, and real-time feedback of an environment evaluation model is combined to continuously optimize and adjust an unloading strategy so as to generate a highly adaptive task unloading decision. According to the method, the task unloading instruction can be intelligently generated and executed according to the real-time edge computing environment state and the task characteristics, the dynamic change of the environment is effectively coped with, and efficient utilization of resources and rapid execution of tasks are achieved. Compared with a traditional method depending on a static rule or a predefined strategy, the task unloading flexibility is remarkably improved, and powerful support is provided for intelligent application in the edge computing environment.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and specifically to an adaptive pervasive edge computing task offloading method based on imitation learning. Background Art

[0002] With the rapid development of IoT technology and the widespread adoption of smart devices, edge computing, as an emerging computing model, is becoming a key technology supporting large-scale, distributed intelligent applications. Edge computing effectively reduces data transmission latency, improves data processing efficiency, and increases system responsiveness by deploying computing and data storage resources at the edge of the network, close to the device or data source. However, task offloading decisions in edge computing environments face numerous challenges. Intelligently selecting the optimal offloading strategy based on real-time environmental conditions and task characteristics to achieve efficient resource utilization and rapid task execution remains a key research topic.

[0003] Current task offloading methods often rely on static rules or predefined policies, which are insufficient in the dynamically changing edge computing environment. Edge computing environment conditions, such as network bandwidth, computing resources, and energy consumption, are constantly changing, while task characteristics, such as computational complexity, data size, and real-time requirements, also vary. Therefore, fixed offloading policies struggle to adapt to these changing scenarios, easily leading to problems such as irrational resource allocation, increased task execution delays, and excessive energy consumption. Summary of the Invention

[0004] The purpose of the present invention is to provide an adaptive ubiquitous edge computing task offloading method based on imitation learning to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive ubiquitous edge computing task offloading method based on imitation learning, the method comprising:

[0006] Construct a task offloading strategy simulation model and an environmental adaptability evaluation model. The task offloading strategy simulation model is used to imitate and learn historical task offloading strategies, and the environmental adaptability evaluation model is used to evaluate the current state of the edge computing environment and the impact of its changes on the task offloading strategy.

[0007] A deep reinforcement learning algorithm is used to train the task offloading strategy imitation model, and the strategy is adjusted based on feedback from the environmental adaptability evaluation model to generate adaptive task offloading decisions.

[0008] Based on the trained model, combined with the real-time edge computing environment status and task characteristics, task offloading instructions are generated and executed.

[0009] Preferably, a long short-term memory network (LSTM) is used to construct a task offloading strategy imitation model, which simulates and predicts future task offloading strategies by learning historical task offloading sequences.

[0010] Preferably, the training step of the task offloading strategy imitation model includes:

[0011] Data preparation: Collect and organize historical task offloading records, including task type, task size, offloading time, offloading destination, and corresponding system status;

[0012] Build an LSTM network structure: Design an LSTM model, consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives preprocessed historical task offloading records, the LSTM layer captures long-term dependencies in sequence data, the fully connected layer maps the LSTM layer output to the task offloading policy space, and the output layer outputs the predicted task offloading policy.

[0013] Model training: The LSTM model is trained using the backpropagation algorithm and the gradient descent optimizer to optimize the model parameters by minimizing the difference between the predicted policy and the actual policy. The calculation formula of the gradient descent optimizer is:

[0014]

[0015] in, Represents the model parameters, including all weights and biases that the model needs to learn; Represents the learning rate, which is used to control the step size of parameter update; Represents the loss function About parameters The gradient, is a vector pointing to the direction where the loss function grows fastest; Indicates assignment operation, i.e. updating parameters The value of

[0016] Model validation: Evaluate the predictive performance of the model on the validation set and adjust the model structure and hyperparameters based on the evaluation results.

[0017] Preferably, a convolutional neural network (CNN) is used to construct an environmental adaptability evaluation model, which extracts the characteristics of the edge computing environment and evaluates the impact of the environmental state on the task offloading strategy.

[0018] Preferably, the training step of the environmental adaptability assessment model includes:

[0019] Data collection: Collects real-time status data of the edge computing environment, including computing resources, network bandwidth, latency, and energy consumption;

[0020] Feature extraction: Use the CNN model to extract features from the collected environmental status data. The CNN model is defined as including convolutional layers, pooling layers, fully connected layers, and output layers. The convolutional layers are used to extract local features, the pooling layers are used to reduce feature dimensions, the fully connected layers are used to map features to the evaluation value space, and the output layer outputs the environmental adaptability evaluation results.

[0021] Model training: The CNN model is trained using supervised learning methods to optimize model parameters by minimizing the difference between the evaluation results and the actual environment state. The mean square error is set as the loss function to measure the difference between the model prediction value and the actual value. The MSE calculation formula is: ,in, is the sample size, is the actual environmental state of the i-th sample, is the model's predicted environmental state for the i-th sample;

[0022] Model update: Regularly update the CNN model based on real-time environment status data.

[0023] Preferably, the steps of adjusting the strategy in combination with the deep reinforcement learning algorithm include:

[0024] Define a reward function: Based on the success rate, latency, and energy consumption indicators of task offloading, define a reward function to evaluate the quality of the task offloading strategy;

[0025] Strategy Exploration: Use the ε-greedy exploration strategy in deep reinforcement learning algorithms to explore new strategies in the task offloading strategy space;

[0026] Strategy update: Based on the feedback and reward function of the environmental adaptability evaluation model, the task offloading strategy imitation model is updated using the update strategy of the deep reinforcement learning algorithm Q-learning to maximize long-term rewards;

[0027] Strategy execution: Generate and execute task offloading instructions based on the updated task offloading strategy simulation model.

[0028] Preferably, the method for collecting model training data includes:

[0029] Task offloading record collection: Extract historical task offloading records from the edge computing system's log files, including task type, task size, offloading time, offloading destination, system status, and task offloading results. The system status includes CPU usage, memory usage, and network bandwidth.

[0030] Environmental status data collection: Through monitoring tools deployed on edge computing nodes, environmental status data is collected in real time, including computing resource usage, network status, energy consumption, and environmental parameters.

[0031] Preferably, the method further comprises: performing data cleaning and standardization processing on the collected model training data.

[0032] Preferably, the method of cleaning the data includes:

[0033] For missing values, linear interpolation algorithm is used to fill them;

[0034] For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

[0035] Preferably, the method of normalizing the data includes: using the Z-score normalization method to normalize the continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for normalization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, The data are standardized.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] By constructing a task offloading strategy simulation model, this paper can imitate and learn from the successful experiences of historical task offloading strategies, avoiding repeated trial and error and wasting resources. Furthermore, the environmental adaptability assessment model can evaluate the current state of the edge computing environment and its changes in real time, ensuring that the offloading strategy matches the current environment, thereby achieving efficient resource utilization.

[0038] Current static rules or predefined strategies are often difficult to adapt to the dynamically changing edge computing environment, resulting in increased task execution delays. The present invention trains the task offloading strategy imitation model through a deep reinforcement learning algorithm, and adjusts the strategy based on the feedback of the environmental adaptability evaluation model. It can quickly respond to environmental changes, generate the optimal offloading strategy, and significantly reduce task execution delays. In an edge computing environment, energy consumption is an important consideration. The present invention avoids unnecessary energy waste by intelligently selecting offloading strategies. The task offloading strategy imitation model can learn which tasks are suitable for local execution and which tasks are suitable for offloading to the edge server for execution, thereby minimizing energy consumption while ensuring task execution efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Schematic diagram of the working principle of the adaptive ubiquitous edge computing task offloading method based on imitation learning according to the present invention;

[0040] Figure 2 Flowchart for building a model for task offloading strategy imitation;

[0041] Figure 3 A diagram showing the steps for adjusting the strategy using a deep reinforcement learning algorithm. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] See also Figure 1-3 The present invention provides a technical solution: an adaptive ubiquitous edge computing task offloading method based on imitation learning, the method comprising:

[0044] A task offloading policy imitation model is constructed to simulate and learn from historical task offloading policies. The model inputs include historical task offloading records, such as task type, size, offloading time, and offloading destination, as well as corresponding offloading results, such as execution latency and energy consumption. By learning from historical data using a deep neural network (such as an LSTM), the model can capture the correlation between task offloading policies and environmental and task characteristics, thereby generating a preliminary offloading policy.

[0045] At the same time, an environmental adaptability assessment model is constructed to evaluate the current edge computing environment state and the impact of changes on task offloading strategies. The model inputs include real-time environmental status data, such as network bandwidth, computing resource utilization, and energy consumption, as well as task characteristic data, such as computational complexity, data size, and real-time requirements. By integrating and evaluating environmental status and task characteristics using a convolutional neural network (CNN), the model outputs a score of the environment's adaptability to the offloading strategy, providing a basis for subsequent policy adjustments.

[0046] After constructing the task offloading strategy imitation model, it is trained using a deep reinforcement learning algorithm. During training, feedback from the environmental adaptability assessment model serves as a reward signal to guide the task offloading strategy imitation model in adjusting its strategy. Specifically, for each offloading decision, the environmental adaptability assessment model evaluates its performance in the actual environment and assigns a corresponding reward or penalty. Based on this feedback signal, the task offloading strategy imitation model adjusts model parameters using optimization algorithms such as gradient descent to maximize long-term cumulative rewards.

[0047] During training and actual application, the task offloading strategy simulation model continuously generates new offloading strategies and evaluates them using the environmental adaptability evaluation model. Based on the evaluation results, the offloading strategy is dynamically adjusted. If a strategy performs poorly in the actual environment (e.g., excessive execution latency or high energy consumption), the model parameters are adjusted using a deep reinforcement learning algorithm to reduce the probability of its selection. Conversely, if a strategy performs well, its selection probability is increased. In this way, the task offloading strategy can continuously adapt to changes in the environment and tasks, achieving adaptive offloading decisions.

[0048] Based on the trained task offloading strategy simulation model and environmental adaptability assessment model, combined with the real-time edge computing environment status and task characteristics, specific task offloading instructions are generated. The instructions include detailed information such as the offloading destination, offloading time, and transmission method. These instructions are sent to the corresponding edge computing node or cloud server for execution, ensuring fast and efficient task processing.

[0049] The present invention will be further described below in conjunction with Examples 1 to 3:

[0050] Example 1:

[0051] This embodiment uses a long short-term memory (LSTM) network to build a task offloading strategy simulation model. By learning historical task offloading sequences, it simulates and predicts future task offloading strategies. The following is a specific implementation method for building and training the task offloading strategy simulation model in this invention:

[0052] Collect and organize historical task offloading records as training data. These records should contain rich information so that the LSTM model can learn the complex relationship between task offloading strategies, environment states, and task characteristics. Specifically, each record includes:

[0053] Task type: Indicates the nature or category of the task, such as image processing, speech recognition, etc.

[0054] Task size: Indicates the amount of data or computing required by the task.

[0055] Uninstallation time: records the time when the task is uninstalled.

[0056] Offload destination: Indicates the edge computing node or cloud server to which the task is offloaded.

[0057] System status: includes environmental parameters such as network bandwidth, computing resource utilization, and energy consumption during offloading.

[0058] These data need to be preprocessed, such as normalization and missing value processing, to ensure the stability and effectiveness of model training.

[0059] Constructing LSTM network structure:

[0060] Design an LSTM model whose structure includes input layer, LSTM layer, fully connected layer and output layer.

[0061] Input layer: Receives preprocessed historical task offloading records and converts them into a format acceptable to the model.

[0062] LSTM layer: Consists of multiple LSTM units and is used to capture long-term dependencies in sequence data, that is, to learn the impact of historical task offloading strategies on current decisions.

[0063] Fully connected layer: maps the output of the LSTM layer to the task offloading strategy space, which usually includes the selection of the offloading destination and the determination of the offloading time.

[0064] Output layer: Generates a predicted task offloading strategy based on the output of the fully connected layer. For classification problems (such as selecting an offloading destination), the output layer can be normalized using the softmax function to obtain a probability distribution for each option.

[0065] The LSTM model is trained using the backpropagation algorithm and a gradient descent optimizer. The training objective is to minimize the difference between the predicted policy and the actual policy, that is, to optimize the loss function. The choice of loss function should be task-specific. For example, cross-entropy loss can be used for classification problems.

[0066] The calculation formula of the gradient descent optimizer is:

[0067]

[0068] in, Represents the model parameters, including all weights and biases that the model needs to learn; Represents the learning rate, which is used to control the step size of parameter update; Represents the loss function About parameters The gradient, is a vector pointing to the direction where the loss function grows fastest; Indicates assignment operation, i.e. updating parameters During the training process, the model gradually converges to the optimal solution by iteratively calculating the gradient and updating the parameters.

[0069] After training is complete, the model's predictive performance needs to be evaluated on a validation set. The validation set should contain data that is independent and identically distributed with the training set to ensure the reliability of the evaluation results. Based on the evaluation results, the model structure and hyperparameters can be adjusted. For example, if the model performs poorly on the validation set, consider increasing the number of LSTM layers, adjusting the learning rate, or modifying the loss function. Through continuous iterations of training, validation, and adjustment, a high-performance task offloading strategy mimicking model can be obtained.

[0070] Taking the image processing task in the edge computing environment as an example, it is assumed that the historical task offloading record contains the task type (image processing), task size (expressed in the number of pixels), offloading time (timestamp), offloading destination (edge node number) and system status (network bandwidth, CPU utilization, energy consumption, etc.).

[0071] Following the above steps, the data is first collected and preprocessed. An LSTM model is then constructed. The input layer receives the preprocessed records, the LSTM layer captures long-term dependencies in the sequence, and the fully connected layer maps the LSTM layer output to the selection space of offload destinations. The output layer uses a softmax function to output the selection probability of each edge node. The model is trained using a backpropagation algorithm and a gradient descent optimizer, optimizing model parameters by minimizing the difference between the predicted and actual offload destinations. During training, the changes in the loss function and the accuracy on the validation set can be monitored to evaluate the model's training progress and performance. Finally, after obtaining a high-performance task offloading policy imitation model, it can be applied to actual edge computing environments to generate and execute task offloading instructions based on real-time environmental conditions and task characteristics, achieving intelligent and adaptive task offloading decisions.

[0072] Example 2:

[0073] This example describes a method for building an environmental adaptability assessment model. This method extracts the characteristics of the edge computing environment and evaluates the impact of the environmental state on the task offloading strategy. The following is a specific implementation method for building and training the environmental adaptability assessment model in this invention:

[0074] Collect real-time status data of the edge computing environment. This data should fully reflect all aspects of the environment so that the CNN model can accurately evaluate the impact of the environment status on the task offloading strategy. Specifically, the collected data includes:

[0075] Computing resources: Such as CPU utilization, memory usage, and GPU utilization, reflecting the computing power of edge computing nodes.

[0076] Network bandwidth: refers to the data transmission rate between edge computing nodes, which affects the transmission time of task offloading.

[0077] Latency: This includes network latency and computational latency, and is an important indicator for evaluating task offloading efficiency.

[0078] Energy consumption: reflects the energy consumption of edge computing nodes when performing tasks, which is particularly important for resource-constrained edge devices.

[0079] These data need to be collected in real time through sensors or monitoring systems and stored in a database for subsequent processing.

[0080] The CNN model is used to extract features from the collected environmental state data. The CNN model consists of convolutional layers, pooling layers, fully connected layers, and output layers.

[0081] Convolutional layer: Consists of multiple convolution kernels, used to extract local features from environmental state data. The convolution kernel slides over the data, extracting the data's feature map through convolution operations.

[0082] Pooling layer: This follows the convolutional layer and is used to reduce the dimensionality of the feature map, reducing computational effort while retaining important features. Common pooling operations include max pooling and average pooling.

[0083] Fully connected layer: Flattens the feature map output by the pooling layer into a one-dimensional vector and maps it to the evaluation value space through a fully connected neural network.

[0084] Output layer: Generates environmental adaptability assessment results based on the output of the fully connected layer. For regression problems (such as evaluating the impact of environmental conditions on task offloading strategies), the output layer is typically a neuron that outputs continuous evaluation values.

[0085] The CNN model is trained using supervised learning. The training objective is to minimize the difference between the evaluation results and the actual environment state, that is, to optimize the loss function. Here, the mean squared error (MSE) is set as the loss function to measure the difference between the model's predicted value and the actual value. The MSE is calculated as: ,in, is the sample size, is the actual environment state (or processed label value) of the i-th sample, is the model's predicted environmental state for the i-th sample.

[0086] Because edge computing environments are constantly changing, CNN models need to be regularly updated to adapt to new environmental conditions. This update process includes collecting new environmental data, retraining the model, or fine-tuning model parameters. The specific update frequency and method should be determined based on actual conditions and can be triggered based on factors such as time intervals, the degree of environmental change, or the degree of model performance degradation.

[0087] Example 3:

[0088] The following is a specific implementation of strategy adjustment combined with a deep reinforcement learning algorithm. This method is applied to the optimization scenario of task offloading strategy.

[0089] First, define the reward function. In this embodiment, the design of the reward function comprehensively considers three key indicators: the success rate of task offloading, latency, and energy consumption. Specifically, the reward function R is defined as:

[0090] R = α×Success_Rate - β×Delay - γ×Energy_Consumption,

[0091] Among them, Success_Rate represents the success rate of task offloading, Delay represents the delay in the task offloading process, and Energy_Consumption represents the energy consumption in the task offloading process. α, β, and γ are the weight coefficients of these three indicators, which are adjusted according to the needs of the actual application scenario to ensure that the reward function can accurately reflect the comprehensive performance of the task offloading strategy.

[0092] This example uses the ε-greedy exploration strategy from a deep reinforcement learning algorithm to explore new strategies in the task offloading strategy space. In practice, an exploration probability ε is set. At each decision, a task offloading strategy is randomly selected as the current strategy with a probability of ε, and the currently known optimal strategy is selected with a probability of 1-ε. This approach ensures extensive strategy exploration while leveraging existing empirical knowledge, improving exploration efficiency.

[0093] Based on the feedback from the environmental adaptability evaluation model and the reward function, the task offloading policy imitation model is updated using the Q-learning deep reinforcement learning algorithm's update strategy. Specifically, a Q-value table is maintained to store the expected reward for each state-action pair. After each task offloading, the actual reward for the current state-action pair is calculated based on the feedback from the environmental adaptability evaluation model, and the corresponding value in the Q-value table is updated according to the Q-learning update formula. Through continuous iterative updates, the Q-value table gradually converges, and the task offloading policy imitation model is optimized accordingly, aiming to maximize long-term reward.

[0094] Finally, the policy is executed. Based on the updated task offloading policy simulation model, task offloading instructions are generated and executed. In specific implementation, the state of each task to be offloaded is input into the task offloading policy simulation model. Based on the current state and the optimized policy, the model outputs the corresponding task offloading instruction. The task offloading operation is then carried out according to this instruction, achieving efficient, low-latency, and low-energy offloading of tasks.

[0095] The following is a specific implementation of collecting and processing model training data:

[0096] Collect task offloading records. Specifically, extract historical task offloading records from the edge computing system's log files. These records detail the task type (e.g., compute-intensive, data-intensive), task size (measured in data volume or compute), offloading time (the timestamp when the task offloading began), offloading destination (i.e., the edge computing node or cloud server to which the task was offloaded), system status (including key metrics such as CPU usage, memory utilization, and network bandwidth), and the result of the task offloading (success or failure). These records provide the model with rich historical experience data, helping it learn system performance and results under different task offloading strategies.

[0097] Collecting environmental status data. To achieve real-time environmental awareness, monitoring tools deployed on edge computing nodes are used to collect real-time environmental status data. This data includes, but is not limited to, computing resource usage (such as real-time CPU, memory, and storage utilization), network conditions (such as bandwidth, latency, and packet loss), energy consumption (such as node energy consumption levels), and environmental parameters (such as temperature and humidity, which may affect device performance). This data provides the model with a comprehensive view of the current environment, enabling it to make more intelligent and adaptable task offloading decisions.

[0098] After collecting the raw data, data cleaning is performed. Missing values are filled using a linear interpolation algorithm. Specifically, for missing points in time series data, linear interpolation is performed using the values of the adjacent points before and after to estimate the missing value, thereby maintaining data continuity and integrity. For outliers, a threshold judgment method is used to handle them. First, a reasonable threshold range is set based on historical data and experience. Data points outside this range are then considered outliers. These outliers are then replaced with the mean of their neighboring points to eliminate any negative impact they may have on model training.

[0099] Finally, the data is normalized. To eliminate the differences in dimensions and value ranges between different features and improve the training efficiency and accuracy of the model, the Z-score normalization method is used to normalize continuous features. In specific implementation, for each numerical data point, the mean of the feature is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for normalization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, This is the standardized data. In this way, all features are ensured to have the same weight and influence during model training.

[0100] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0101] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive pervasive edge computing task offloading method based on imitation learning, characterized in that: The method comprises: Construct a task offloading strategy simulation model and an environmental adaptability evaluation model. The task offloading strategy simulation model is used to imitate and learn historical task offloading strategies, and the environmental adaptability evaluation model is used to evaluate the current state of the edge computing environment and the impact of its changes on the task offloading strategy. A deep reinforcement learning algorithm is used to train the task offloading strategy imitation model, and the strategy is adjusted based on feedback from the environmental adaptability evaluation model to generate adaptive task offloading decisions. Based on the trained model, combined with the real-time edge computing environment status and task characteristics, task offloading instructions are generated and executed.

2. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 1 is characterized in that: A long short-term memory (LSTM) network is used to construct a task offloading strategy imitation model. By learning historical task offloading sequences, the future task offloading strategies can be simulated and predicted.

3. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 2 is characterized in that: The training steps of the task offloading strategy imitation model include: Data preparation: Collect and organize historical task offloading records, including task type, task size, offloading time, offloading destination, and corresponding system status; Build an LSTM network structure: Design an LSTM model, consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives preprocessed historical task offloading records, the LSTM layer captures long-term dependencies in sequence data, the fully connected layer maps the LSTM layer output to the task offloading policy space, and the output layer outputs the predicted task offloading policy. Model training: The LSTM model is trained using the backpropagation algorithm and the gradient descent optimizer to optimize the model parameters by minimizing the difference between the predicted policy and the actual policy. The calculation formula of the gradient descent optimizer is: ， in, Represents the model parameters, including all weights and biases that the model needs to learn; Represents the learning rate, which is used to control the step size of parameter update; Represents the loss function About parameters The gradient, is a vector pointing to the direction where the loss function grows fastest; Indicates assignment operation, i.e. updating parameters The value of Model validation: Evaluate the predictive performance of the model on the validation set and adjust the model structure and hyperparameters based on the evaluation results.

4. The method for adaptive ubiquitous edge computing task offloading based on imitation learning according to claim 1 is characterized in that: The convolutional neural network (CNN) is used to build an environmental adaptability evaluation model. By extracting the characteristics of the edge computing environment, the impact of the environmental status on the task offloading strategy is evaluated.

5. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 4 is characterized in that: The training steps of the environmental adaptability assessment model include: Data collection: Collects real-time status data of the edge computing environment, including computing resources, network bandwidth, latency, and energy consumption; Feature extraction: Use the CNN model to extract features from the collected environmental status data. The CNN model is defined as including convolutional layers, pooling layers, fully connected layers, and output layers. The convolutional layers are used to extract local features, the pooling layers are used to reduce feature dimensions, the fully connected layers are used to map features to the evaluation value space, and the output layer outputs the environmental adaptability evaluation results. Model training: The CNN model is trained using supervised learning methods to optimize model parameters by minimizing the difference between the evaluation results and the actual environment state. The mean square error is set as the loss function to measure the difference between the model prediction value and the actual value. The MSE calculation formula is: ,in, is the sample size, is the actual environmental state of the i-th sample, is the model's predicted environmental state for the i-th sample; Model update: Regularly update the CNN model based on real-time environment status data.

6. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 1 is characterized in that: The steps for adjusting the strategy by combining deep reinforcement learning algorithms include: Define a reward function: Based on the success rate, latency, and energy consumption indicators of task offloading, define a reward function to evaluate the quality of the task offloading strategy; Strategy Exploration: Use the ε-greedy exploration strategy in deep reinforcement learning algorithms to explore new strategies in the task offloading strategy space; Strategy update: Based on the feedback and reward function of the environmental adaptability evaluation model, the task offloading strategy imitation model is updated using the update strategy of the deep reinforcement learning algorithm Q-learning to maximize long-term rewards; Strategy execution: Generate and execute task offloading instructions based on the updated task offloading strategy simulation model.

7. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 1 is characterized in that ,Methods for collecting model training data include: Task offloading record collection: Extract historical task offloading records from the edge computing system's log files, including task type, task size, offloading time, offloading destination, system status, and task offloading results. The system status includes CPU usage, memory usage, and network bandwidth. Environmental status data collection: Through monitoring tools deployed on edge computing nodes, environmental status data is collected in real time, including computing resource usage, network status, energy consumption, and environmental parameters.

8. The adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 7 is characterized in that: The method also includes: performing data cleaning and standardization processing on the collected model training data.

9. The adaptive pervasive edge computing task offloading method based on imitation learning according to claim 8 is characterized in that ,The methods for cleaning data include: For missing values, linear interpolation algorithm is used to fill them; For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

10. The method for using the adaptive ubiquitous edge computing task offloading method based on imitation learning according to claim 8, characterized in that: The method of standardizing data includes: using the Z-score standardization method to standardize continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, The data are standardized.

Images