A Method and System for Controlling the Power Consumption of an Edge Computing Unit System

Abstract

Embodiments of the present disclosure provide a method and system for controlling the power consumption of an edge computing unit system. Applied to the field of edge computing technology, it includes: collecting historical input data of edge computing devices, establishing an LSTM time series prediction model in the cloud, and deploying it on the edge computing devices; integrating the prediction model into the edge device firmware, and deploying a lightweight random forest classifier to determine the task output, dividing the device operation period, and completing the setting of the device operation state; constructing a device adjustment operation mechanism, adjusting the operation of the edge device, and collecting feedback data; and periodically updating the parameters of the LSTM time series task prediction model based on the feedback data and using an incremental learning method. In this way, the present invention can achieve accurate task prediction, intelligent scheduling decision-making and low-overhead model deployment on the edge side, and balance the extension of device life and the guarantee of task service quality by reducing the number of invalid wake-up events.

Description

Technical Field

[0001] The present disclosure relates to the field of edge computing technology, and particularly to a method and system for controlling the system power consumption of an edge computing unit. Background Art

[0002] In the Internet of Things (IoT) system and the field of edge computing, devices are usually deployed in resource-constrained environments and rely on battery power supply or low-power design to achieve long-term operation. To extend the device life cycle, traditional power management strategies generally adopt the "idle sleep" mechanism, that is, when the device has no task to process, the power supply voltage of the main chip is turned off or reduced to make it enter the low-power sleep state, thereby reducing the static power consumption. This method reduces the energy consumption baseline of the device to a certain extent and becomes the basic idea for extending the life of the IoT system.

[0003] However, the existing technologies have significant defects: frequent sleep-wake switching will introduce additional energy consumption overhead. Specifically, when the device wakes up from the sleep state, it needs to go through processes such as voltage recovery, clock stabilization, and peripheral initialization. The transient power consumption at this stage may be much higher than the steady-state operation power consumption. In scenarios with intensive or highly random task requests (such as intelligent monitoring, industrial sensor networks), the device may repeatedly switch states due to frequent responses to sudden tasks, resulting in increased energy consumption instead of decreased energy consumption. Research shows that for some low-power chips, the energy consumption of a single wake-up process is equivalent to the power consumption of continuous operation for several seconds to dozens of seconds. If the wake-up frequency per hour exceeds the threshold, the overall energy consumption is even higher than the continuous active state.

[0004] Existing solutions attempt to optimize the wake-up strategy through static sleep cycle adjustment or simple threshold judgment, but there are still the following limitations: Static scheduling inadaptability: Fixed sleep intervals cannot dynamically respond to changes in task load, resulting in over-wake-up during low load or response delay during high load. Task processing fragmentation: Scattered tasks trigger independent wake-up events, lacking a task merging and batch processing mechanism, resulting in repeated energy consumption overhead. Insufficient prediction ability: Traditional heuristic algorithms (such as the moving average method) are difficult to accurately predict non-periodic or sudden tasks, resulting in ineffective wake-up.

[0005] In recent years, although some studies have tried to introduce machine learning models to predict task arrival patterns, most solutions rely on cloud-based centralized training, do not fully consider the resource constraints of edge devices (such as memory and computing power limitations), and the accuracy loss after model quantization and deployment has not been effectively compensated. In addition, the existing technologies pay insufficient attention to the collaborative optimization of task classification and priority scheduling, and cannot guarantee the real-time performance of high-priority tasks while reducing the wake-up times.

[0006] Therefore, there is an urgent need for a power consumption control method that can achieve accurate task prediction, intelligent scheduling decision-making, and low-overhead model deployment on the edge side. Summary of the Invention

[0007] The present disclosure provides a method and system for power consumption control of an edge computing unit system, which achieves a balance between extending the device life and ensuring the quality of task services by reducing the number of invalid wake-up events, and at least solves the technical problems of difficult wake-up, insufficient collaboration, and insufficient prediction ability in the existing methods.

[0008] According to the first aspect of the present disclosure, a method for power consumption control of an edge computing unit system is provided, including the following steps:

[0009] Collect historical input data of the edge computing device, establish an LSTM time series prediction model in the cloud, and deploy the LSTM time series prediction model on the edge computing device;

[0010] Integrate the constructed prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine the task output, and divide the device operation period to complete the device operation state setting;

[0011] Construct a device adjustment operation mechanism based on the set device operation state, adjust the operation of the edge device, and collect feedback data;

[0012] Periodically update the parameters of the LSTM time series task prediction model based on the feedback data and using an incremental learning method.

[0013] In the above aspect and any possible implementation manner, a further implementation manner is provided, where the historical input data of the edge computing device includes: task arrival time, task data volume, and power consumption data.

[0014] In the above aspect and any possible implementation manner, a further implementation manner is provided, where the LSTM time series task prediction model includes an input layer, an LSTM layer, and an output layer;

[0015] The input layer inputs the historical input data of the edge computing device, the LSTM layer is used to capture the dynamic changes in the time series, and the output layer outputs the task prediction situation for the next time step.

[0016] In the above aspect and any possible implementation manner, a further implementation manner is provided, where the input layer performs z-score normalization processing on the three-dimensional tensor composed of the number of samples, the time step length, and the number of features, and generates a continuous time series sliding window in chronological order;

[0017] The LSTM layer includes a first LSTM layer and a second LSTM layer stacked on top of each other. The first LSTM layer is used to process the original input sequence and extract local temporal dependencies, and the second LSTM layer is used to further capture long-term dependencies and global patterns based on the output result of the first LSTM layer;

[0018] The output layer uses a linear activation function, and the output result of the output layer is the predicted task arrival time and density in the future time window.

[0019] In the above-described aspect and any possible implementation, a further implementation is provided. The LSTM time series task prediction model includes a loss function, and the construction process of the loss function is as follows:

[0020] Based on the true task arrival time and the task arrival time predicted by the prediction model, the mean squared error is used to measure and calculate the time prediction error loss function;

[0021] Based on the one-hot encoding of the true density category and the probability distribution of the density category predicted by the prediction model, the cross-entropy loss method is used to calculate the task density loss function;

[0022] The time prediction loss function and the task density loss function are weighted and combined to obtain a joint objective function;

[0023] A L2 regularization constraint is constructed for the model weights and combined with the joint objective function to obtain the loss function.

[0024] In the above-described aspect and any possible implementation, a further implementation is provided. The task output includes simple tasks and complex tasks. The simple tasks are low-load computing and high-tolerance delay tasks, and the complex tasks are high-computing load and low-latency requirement tasks.

[0025] In the above-described aspect and any possible implementation, a further implementation is provided. The division of the device operation period is based on the task arrival time and the density output by the LSTM time series task prediction model. The specific process is as follows:

[0026] When the predicted task density Dt in a certain period is greater than Dth, it is determined as the wake-up period. When the predicted task density Dt in a certain period is less than Dth, it is the sleep period, where Dth is the threshold.

[0027] In the above-described aspect and any possible implementation, a further implementation is provided.

[0028] The process of constructing a device adjustment operation mechanism based on the set device operation state and adjusting the operation of the edge device is as follows:

[0029] During complex task execution and wake-up periods, in combination with the predicted task density of the prediction model and by detecting the load condition of the model, a determination is made on whether to save power. If a set voltage-frequency pair is reached, the voltage is adjusted and the clock generator is synchronously configured to the set frequency, and the switching sequence of the voltage and frequency is ensured.

[0030] During periods of no task execution, simple task execution, and sleep, monitoring and processing of simple tasks are performed, and a complex task cache queue and complex task trigger conditions are set.

[0031] In the aspects and any possible implementation manners as described above, a further implementation manner is provided.

[0032] The process of periodically updating the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method is as follows:

[0033] Based on the local data samples and labels in the feedback data, the model gradient is calculated and gradient clipping is performed to obtain the clipped gradient.

[0034] Gaussian noise is added to the clipped gradient, and multi-device gradient aggregation is performed. Based on the aggregated gradient, the global model parameters are updated.

[0035] According to the second aspect of the present disclosure, an edge computing unit system power consumption control system is provided, including:

[0036] A data acquisition module, a task classification module, a dynamic scheduling module, and a model update module;

[0037] The data acquisition module is used to collect historical input data of the edge computing device, establish an LSTM time series prediction model in the cloud, and deploy the LSTM time series prediction model on the edge computing device.

[0038] The task classification module is used to integrate the constructed prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine the task output, and divide the device operation period to complete the setting of the device operation state.

[0039] The dynamic scheduling module is used to construct a device adjustment operation mechanism based on the set device operation state, perform operation adjustment on the edge device, and collect feedback data.

[0040] The model update module is used to periodically update the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method.

[0041] Compared with the prior art, the present invention has the following technical effects:

[0042] Through data-driven intelligent scheduling, the present invention achieves a balance between extending the device lifespan and ensuring the quality of task services by reducing the number of ineffective wake-up events. Moreover, through data-driven intelligent scheduling, with the goal of reducing the number of ineffective device wake-ups, the present invention improves energy efficiency, extends the device lifespan, and meets the power management requirements of edge computing devices.

[0043] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In conjunction with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present disclosure will become more apparent. The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0045] Figure 1 shows a schematic flowchart of a method for controlling the power consumption of an edge computing unit system according to an embodiment of the present disclosure;

[0046] Figure 2 shows a schematic structural diagram of a power consumption control system for an edge computing unit system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without making creative efforts fall within the scope of protection of the present disclosure.

[0048] To make the above objectives, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0049] Referring to Figure 1 as shown, this embodiment provides a method for controlling the power consumption of an edge computing unit system, including the following steps:

[0050] S101. Collect historical input data of the edge computing device, establish an LSTM time series prediction model in the cloud, and deploy the LSTM time series prediction model on the edge computing device.

[0051] In this embodiment, the historical input data received by the cloud from the edge device is used as a dataset for LSTM model training. The specific historical data includes: 1. Task arrival time, recording the generation timestamp of each task. 2. Task data volume, the data size of each task. 3. Power consumption data, the real-time power consumption during task processing. The dataset is divided into a training set and a test set in chronological order with a ratio of 7:3 to avoid future data leakage into the training set. The training set is used for model training, and the test set is used to evaluate the difference between the predicted values and the true values of the model to test its prediction performance and accuracy.

[0052] Specifically, the LSTM time series task prediction model includes: an input layer, an LSTM layer, and an output layer. Among them, the input layer inputs the historical task data of the device, the LSTM layer is used to capture the dynamic changes in the time series, and the output layer outputs the task prediction situation for the next time step.

[0053] In the input layer, the three-dimensional tensor composed of the number of samples, the time step, and the number of features undergoes data preprocessing. Each feature is normalized by z-score, and a continuous time series sliding window is generated in chronological order. The z-score normalization formula is:

[0054] (1)

[0055] where x is the original data value, μ is the mean of the dataset, and σ is the standard deviation of the dataset.

[0056] Specifically, in this embodiment, the mean μ and the standard deviation σ are respectively:

[0057] (2)

[0058] (3)

[0059] where n is the number of samples in the dataset, and xi is the value of the i-th sample.

[0060] The LSTM layer is stacked by two layers of LSTM, each layer having 128 hidden units respectively. Two independent LSTM layers, the output of the previous layer is used as the input of the next layer. Each LSTM layer has its own cell unit for extracting time series features at different levels. The first layer of LSTM: processes the original input sequence and extracts local time dependencies. The second layer of LSTM: further captures long-term dependencies and global patterns.

[0061] The output layer uses a linear activation function, and the output result gives the predicted task arrival time and density for the future time window. Time prediction: the task arrival time points {t1, t2, …, t for the next hour N}. Density classification: The task density for the next hour is P = [p1, p2, p3].

[0062] Model training and optimization will affect the final prediction error. The traditional LSTM model may not be able to effectively capture the complex relationships between different features when dealing with multivariate time series. Features such as task arrival time and density do not fully utilize the correlation between the two. Therefore, a joint optimization strategy is adopted, combining the continuous prediction of task arrival time and the discrete classification of density, while considering the correlation between the two. During the training process, the objective function (loss function) is defined, with minimizing the prediction error as the main optimization goal.

[0063] The objective function of the LSTM time series task prediction model ensures that the model can accurately predict the task arrival pattern and effectively support dynamic power consumption scheduling by jointly optimizing the mean square error of time prediction and the cross-entropy loss of density classification. In practical applications, α and λ are adjusted through experiments to achieve the best balance between prediction accuracy and generalization ability.

[0064] The loss function for task arrival time prediction uses the mean square error (MSE) to measure the error of time prediction:

[0065] (4)

[0066] where is the true arrival time of the i-th task, is the task arrival time predicted by the model, and N is the number of samples.

[0067] The prediction loss function for task density is divided into discrete categories (low, medium, and high density) and uses cross-entropy loss:

[0068] (5)

[0069] where is the one-hot encoding of the true density category, is the probability distribution of the density category predicted by the model, and C is the number of categories.

[0070] In this embodiment, one-hot encoding refers to converting categorical variables into a format suitable for machine learning algorithms. For example, here the task density categories are divided into low, medium, and high, and a binary feature vector is created for each category. The vector length is equal to the number of categories, and for each category, only the corresponding feature position is 1, and the rest are 0. Example: low density [1, 0, 0], medium density [0, 1, 0], high density [0, 0, 1].

[0071] The combined objective function combines the losses of time prediction and density classification with weighted combination to balance the optimization objectives of both:

[0072] (6)

[0073] where α is the weight coefficient, which adjusts the importance of time prediction and density classification (α = 0.7 indicates more attention to time accuracy). When α → 1, the model focuses on minimizing the time prediction error; when α → 0, the model focuses on improving the density classification accuracy. To prevent overfitting and ensure the efficient operation of the model on edge devices, L2 regularization is added to constrain the model weights:

[0074] (7)

[0075] where wk is the model weight parameter and λ is the regularization coefficient.

[0076] The final overall objective function (loss function) expression is:

[0077] (8)

[0078] After the model training is completed, quantization compression is performed to reduce the size and computational resource requirements of the model, making it suitable for running on resource-constrained edge devices. Use the TensorFlow Lite Converter to convert the floating-point model to the TFLite format, and compare the prediction errors of the model on the test set to ensure that the increase in MSE does not exceed 5%. Then integrate the quantized TFLite model into the edge device firmware.

[0079] S102. Integrate the constructed prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine the task output, and divide the device operation period to complete the device operation status setting.

[0080] In this embodiment, a lightweight random forest classifier is deployed in the coprocessor (MCU) of the edge computing device. Based on historical power consumption data and task characteristics (including computational complexity, real-time requirements, and data volume), the device functional tasks are divided into two types of outputs: 0 (simple tasks) and 1 (complex tasks). Simple tasks refer to tasks with low computational load and high tolerance for latency (sensor data filtering, heartbeat packet response), which are processed by the low-power coprocessor (MCU); complex tasks refer to tasks with high computational load and low latency requirements (AI inference, video encoding), which are processed by the main chip.

[0081] Deploy the TFLite quantization model of the LSTM time series task prediction model to the main chip of the edge computing system, and the main chip is responsible for performing computing, storage, and control tasks. Based on the task arrival density statistically calculated from historical data, that is, the number of tasks per minute, set a threshold Dth. The main chip divides the device operation cycle into a wake-up period and a sleep period by comparing the predicted task arrival time and density of the model. The specific discrimination method is that when the predicted task density Dt > Dth in a certain period, it is the wake-up period. Conversely, when the predicted task density Dt < Dth in a certain period, it is the sleep period. The main chip sends the divided periods to the MCU through the PCIe high-speed bus. The wake-up signal is sent by the MCU to the PMIC through the GPIO interface, and the PMIC restores the power supply of the main chip within 10 ms, starts the clock and peripheral devices, and the main chip switches from the deep sleep mode (power consumption < 5 mW) to the active mode (power consumption > 500 mW). The sleep signal sends a sleep instruction from the main chip to the PMIC through I²C and re-enters the deep sleep.

[0082] S103. Construct a device adjustment operation mechanism based on the set device operation status, perform operation adjustment on the edge device, and collect feedback data.

[0083] In this embodiment, the operation strategy is constructed through the divided wake-up period and sleep period, simple tasks, and complex tasks. Specifically:

[0084] During the period of dense task prediction, the main chip remains active to efficiently process tasks, while during the period of sparse task prediction, the main chip enters the deep sleep mode to reduce power consumption; at the same time, before the predicted task arrives, the system wakes up the main chip in advance to ensure that the device is ready, thereby reducing the response delay and optimizing the overall energy efficiency.

[0085] During the wake-up of the main chip for dense tasks, by combining the model-predicted task density and the real-time monitoring of its own load conditions, a decision is made on whether to reduce or increase performance to save power consumption. Specifically, the main chip selects a preset voltage-frequency pair: 1. Low load, voltage 0.8V, frequency locked at 800 MHz; 2. Medium load, voltage 1.0V, frequency 1.5 GHz; 3. High load, voltage 1.2V, frequency 2.0 GHz. Send a voltage regulation and frequency modulation instruction to the PMIC through the I²C interface. After receiving the instruction, the PMIC adjusts the output to the target voltage (such as from 1.2V to 0.8V) and synchronously configures the clock generator (PLL) to set the frequency (such as from 2.0 GHz to 800 MHz). At the same time, the PMIC will ensure the switching order of voltage and frequency (boost voltage first and then boost frequency, lower frequency first and then lower voltage) to prevent chip instability. In the above process, the PMIC cooperates with the main chip to achieve dynamic voltage and frequency scaling (DVFS), taking into account both performance and energy efficiency.

[0086] During the sleep of the main chip when the task is not predicted, the low-power coprocessor (MCU) monitors and processes simple tasks in real time; if a complex task arrives, the MCU caches it in the task queue. Only when one of the following wake-up conditions is triggered: the number of tasks in the queue reaches the preset threshold (50 tasks) or it enters the peak period of model prediction, the MCU sends a wake-up signal to the PMIC, and the PMIC restores the power supply of the main chip within 10 ms, starts the clock and peripheral devices.

[0087] S104. Periodically update the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method.

[0088] The cloud periodically updates the LSTM model parameters based on the feedback data and optimizes the prediction accuracy through incremental learning. The updated model is re-quantized and deployed to the edge device to form a closed-loop optimization system. Traditional incremental learning simply collects data in the cloud, retrains the model, and then pushes the update. Such a method has data privacy problems. Considering this problem, the present invention uses edge-side data desensitization during the upload process, first performs differential privacy, and then uploads it to the cloud. In this embodiment, the feedback data refers to two types of data, both generated by the edge device. 1. The difference between the prediction result of the lightweight model deployed on the local device and the actual result, which is used to guide the update of the cloud model parameters; 2. The original time series data such as the task arrival time, type, data volume, power consumption, etc. newly collected within a period of time, which is used to expand the training data set.

[0089] Specifically, the edge device uses local data to calculate the model gradient, clips the gradient to prevent the gradient amplitude from being too large, and limits its L2 norm not to exceed the threshold C. Add Gaussian noise N(0,σ²) to the gradient to satisfy ( ,δ)-differential privacy. Finally, the cloud aggregates the noisy gradients of multiple devices and updates the global model parameter w. The privacy budget consumption is further reduced by the random sampling method of only selecting some devices to participate in the update in each round. This method ensures the model prediction accuracy while protecting data privacy.

[0090] The related calculation formulas are as follows:

[0091] Gradient calculation formula:

[0092] (9)

[0093] Where is the loss function, is the model parameter, is the local data sample and label.

[0094] Gradient clipping formula:

[0095] (10)

[0096] Where C is the threshold value.

[0097] The calculation formula for the noise scale σ:

[0098] (11)

[0099] Where δ is the failure probability, is the privacy budget,

[0100] The gradient expression after adding noise:

[0101] (12)

[0102] The aggregation formula:

[0103] (13)

[0104] The model update formula:

[0105] (14)

[0106] Where is the learning rate.

[0107] As Figure 2 shown, this embodiment also provides a power consumption control system for an edge computing unit system, including: a data acquisition module 1, a task classification module 2, a dynamic scheduling module 3, and a model update module 4;

[0108] The data acquisition module 1 is used to collect historical input data of the edge computing device, establish an LSTM time series prediction model in the cloud, and deploy the LSTM time series prediction model on the edge computing device;

[0109] The task classification module 2 is used to integrate the constructed prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine the task output, and divide the device operation period to complete the device operation state setting;

[0110] The dynamic scheduling module 3 is used to construct a device adjustment operation mechanism based on the set device operation state, adjust the operation of the edge device, and collect feedback data;

[0111] The model update module 4 is used to perform periodic parameter updates on the LSTM time series task prediction model based on the feedback data and using the incremental learning method.

[0112] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.

[0113] It should be understood that various forms of processes shown above can be used, reordering, adding or deleting steps. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in the present application can be achieved, and no limitation is imposed herein.

[0114] The above specific implementation manners do not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A method for controlling the power consumption of an edge computing unit system, characterized in that, It includes the following steps: Collect historical input data of edge computing devices, establish an LSTM time series task prediction model in the cloud, and deploy the LSTM time series task prediction model on edge computing devices; Integrate the constructed LSTM time series task prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine task outputs, and divide the device operation period to complete the setting of the device operation state; Construct a device adjustment operation mechanism based on the set device operation state, perform operation adjustment on edge devices, and collect feedback data; Periodically update the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method; Among them, the LSTM time series task prediction model includes an input layer, an LSTM layer, and an output layer; the input layer inputs the historical input data of the edge computing device, the LSTM layer is used to capture dynamic changes in the time series, and the output layer outputs the task prediction situation at the next time step, including the predicted task arrival time and density in the future time window; The LSTM time series task prediction model includes a loss function, and the construction process of the loss function is as follows: Based on the true task arrival time and the task arrival time predicted by the LSTM time series task prediction model, use the mean square error measure to calculate the time prediction error loss function; Based on the one-hot encoding of the true density category and the density category probability distribution predicted by the LSTM time series task prediction model, use the cross-entropy loss method to calculate the task density loss function; Perform weighted combination of the time prediction error loss function and the task density loss function to obtain a joint objective function; Construct an L2 regularization constraint model weight and combine it with the joint objective function to obtain a loss function; Among them, the task outputs include simple tasks and complex tasks; based on the task arrival time and the density output by the LSTM time series task prediction model, divide the device operation period into a wake-up period and a sleep period; Construct a device adjustment operation mechanism based on the set device operation state, and perform operation adjustment on edge devices, including: During the operation of complex tasks and the wake-up period, combine the predicted task density of the LSTM time series task prediction model and detect the load situation of the model, judge whether to save power, if the set voltage-frequency pair is reached, adjust the voltage and synchronously configure the clock generator to set the frequency, and ensure the switching order of voltage and frequency; During the period of no task operation, simple task operation, and sleep stage, monitor and process simple tasks, and set a complex task cache queue and complex task trigger conditions.

2. The edge computing unit system power consumption control method according to claim 1, wherein The historical input data of the edge computing device includes: task arrival time, task data volume, and power consumption data.

3. The method for controlling the power consumption of the edge computing unit system according to claim 1, wherein The input layer performs z-score normalization processing on the three-dimensional tensor composed of the number of samples, time steps, and the number of features, and generates a continuous time series sliding window in chronological order; The LSTM layer includes a first LSTM layer and a second LSTM layer stacked on top of each other. The first LSTM layer is used to process the original input sequence and extract local temporal dependencies, and the second LSTM layer is used to further capture long-term dependencies and global patterns based on the output result of the first LSTM layer; The output layer uses a linear activation function.

4. The method for controlling the system power consumption of the edge computing unit according to claim 3, wherein The simple tasks are low-load computing and high-tolerance latency tasks, and the complex tasks are high-computation load and low-latency requirement tasks.

5. The method for controlling the power consumption of the edge computing unit system according to claim 4, wherein Based on the task arrival time and the density output by the LSTM time series task prediction model, the device operation period is divided into a wake-up period and a sleep period. The specific process is as follows: When the predicted task density Dt > Dth in a certain period, it is determined as the wake-up period. When the predicted task density Dt < Dth in a certain period, it is the sleep period, where Dth is the threshold.

6. The method for controlling the system power consumption of the edge computing unit according to claim 1, wherein The process of periodically updating the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method is as follows: Calculate the model gradient based on the local data samples and labels in the feedback data, and perform gradient clipping to obtain the clipped gradient; Add Gaussian noise to the clipped gradient, perform multi-device gradient aggregation, and update the global model parameters based on the aggregated gradient.

7. An edge computing unit system power consumption control system implemented by using the edge computing unit system power consumption control method according to any one of claims 1-6, characterized in that, Including: A data acquisition module (1), a task classification module (2), a dynamic scheduling module (3), and a model update module (4); The data acquisition module (1) is used to collect historical input data of the edge computing device, establish an LSTM time series task prediction model in the cloud, and deploy the LSTM time series task prediction model on the edge computing device; The task classification module (2) is used to integrate the constructed LSTM time series task prediction model into the edge device firmware, deploy a lightweight random forest classifier to determine the task output, and divide the device operation period to complete the setting of the device operation state; The dynamic scheduling module (3) is used to construct a device adjustment operation mechanism based on the set device operation state, adjust the operation of the edge device, and collect feedback data; The model update module (4) is used to periodically update the parameters of the LSTM time series task prediction model based on the feedback data and using the incremental learning method.

Images