Wireless mouse low-power-consumption intelligent optimization method based on deep learning

Abstract

The invention provides a wireless mouse low-power-consumption intelligent optimization method based on deep learning, and the method comprises the steps: achieving the real-time collection of the operation state data of a wireless mouse through a multi-sensor fusion technology, building a power consumption prediction model based on a long and short-term memory network, so as to learn the time sequence relation between the user behavior and the mouse power consumption change, based on a prediction result, an intelligent power consumption control strategy optimization model is constructed, an optimal power consumption control strategy is generated through a strategy iterative optimization method in combination with parameter configuration of power management, clock gating, sensor sampling and the like, power consumption control parameters are analyzed, and a control instruction is generated. The CPU dominant frequency of the mouse, the voltage of a power management unit, the sampling rate of a sensor and a peripheral working mode are dynamically adjusted, and a power consumption control strategy is optimized by adopting a negative feedback mechanism. According to the wireless mouse, the energy consumption of the wireless mouse can be effectively reduced, meanwhile, the operation smoothness is guaranteed, and the cruising ability and the user experience are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer peripherals, and in particular relates to a low-power intelligent optimization method for a wireless mouse based on deep learning. Background Art

[0002] As an important component of computer peripherals, wireless mice are widely used in scenarios such as office work, gaming, and industrial control. Since wireless mice rely on battery power, their battery life directly affects the user experience. As wireless mice become increasingly versatile, such as high-precision sensors, multi-mode connections, programmable buttons, and RGB lighting effects, these additional features improve the user experience while also bringing higher energy consumption, making the battery life of wireless mice a major issue restricting their use. Existing wireless mouse power management methods mainly rely on fixed low-power modes, such as timed sleep, low refresh rate mode, and passive power management. These methods can reduce the mouse's energy consumption to a certain extent, but they have at least the following shortcomings:

[0003] (1) Most of them use fixed thresholds and preset rules. Due to different user operating habits and different mouse application scenarios, the mouse may be put into sleep mode or maintain high power consumption mode for a long time, resulting in power waste or degraded user experience.

[0004] (2) When predicting power consumption, it is only based on a limited number of states, which makes it difficult to capture complex user behavior patterns, resulting in limited prediction accuracy. In addition, the model relies on manually set state transition probabilities, making it difficult to perform adaptive optimization.

[0005] (3) Existing mouse energy-saving technologies mainly rely on simple sensor trigger mechanisms, such as mechanical switches and optical sensors, and are not combined with intelligent algorithms to optimize power consumption. This results in a relatively rigid power consumption management method that is difficult to adapt to different user habits.

[0006] Therefore, we need to develop a deep learning-based intelligent low-power optimization method for wireless mice, which can realize intelligent power consumption prediction, dynamically adjust the power consumption control strategy according to real-time data, improve the battery life of wireless mice and reduce unnecessary energy loss. Summary of the Invention

[0007] The purpose of the present invention is to provide a low-power intelligent optimization method for wireless mice based on deep learning, so as to solve the problems mentioned in the above background technology of existing wireless mice, such as fixed threshold power consumption management, limited prediction accuracy, and lack of intelligent optimization power consumption strategy.

[0008] To achieve the above objectives, the present invention provides a low-power intelligent optimization method for wireless mice based on deep learning, which is as follows:

[0009] Step S1: Realize real-time collection of the operating status data of the wireless mouse through multi-sensor fusion technology;

[0010] Step S2: Extract and fuse the features of the running status data to obtain a unified feature data frame stream. ,in Represents feature data frame, each frame Both include the current power consumption and motion characteristics of the mouse;

[0011] Step S3: constructing a power consumption prediction model for the wireless mouse based on the long short-term memory network, wherein the power consumption prediction model takes the feature data frame stream as input and accurately predicts the future power consumption of the mouse by learning the temporal relationship between user behavior and changes in mouse power consumption;

[0012] Step S4: Based on the power consumption prediction model, an intelligent power consumption control strategy optimization model is constructed. The strategy optimization model uses the power consumption prediction sequence output by the power consumption prediction model as the environmental state, and uses a configuration combination of adjustable parameters including power management, clock gating, and sensor sampling as the control behavior. The optimal power consumption control strategy is generated through strategy iterative optimization.

[0013] Step S5: Based on the optimal power consumption control strategy, the power consumption control parameters are analyzed, and control instructions are generated to adjust the CPU main frequency, the power management unit voltage, the sensor sampling rate and the peripheral working mode. Based on the adjusted power consumption control parameters, the mouse power consumption status is monitored in real time, the power consumption error between the actual power consumption and the predicted power consumption is calculated, and a negative feedback mechanism is used to optimize the optimal power consumption control strategy so that the power consumption control parameters dynamically converge to the optimal power consumption configuration; if power consumption abnormality is detected, switch to the safe mode and execute the corresponding power consumption management strategy.

[0014] Based on the above solution, the multi-sensor includes a voltage sensor, a current sensor, an acceleration sensor and a gyroscope sensor, and the collected operating status data are the power supply voltage, working current, motion acceleration and angular velocity of the mouse;

[0015] The feature data frame ,in is the timestamp of the feature data frame, based on the voltage and current sampling time. They are the instantaneous power, synthetic acceleration, velocity change rate, angular velocity energy and total motion energy of the mouse at the mth frame, and the generation frequency of the feature data frame is , which is consistent with the sampling frequency of the gyroscope sensor and the interpolated acceleration sensor.

[0016] Based on the above scheme, the continuous feature data frame Constitute a sample point, each frame corresponds to a time step of the power consumption prediction model, let the sample point sequence be ,but Can be expressed as a A matrix with 6 rows and columns, a matrix Each row corresponds to the power consumption prediction model The input data of time steps, the output of LSTM is set to the future time steps The power consumption prediction value of the mouse is expressed as a dimensional vector : ,in Indicates the Sample future The power consumption prediction value for each time step.

[0017] Based on the above solution, the power consumption prediction model is expressed as follows:

[0018]

[0019] in: Indicates initializing the hidden state and unit state to zero vector, Indicates the The state output of the hidden layer in time steps, R represents the real number domain, is the hidden layer dimension; Indicates the The cell state at time steps, Indicates the The cell state at each time step; It is The input of a time step corresponds to a frame of data in the feature data frame stream; Indicates connecting the hidden state of the previous time step with the current input; Represent the forget gate, input gate, and output gate respectively; is the candidate unit status; is the model parameter matrix; is the bias term; Represents element-wise multiplication (Hadamard product); and Respectively Function and hyperbolic tangent function, as the activation function of the gate unit and candidate state, the final output The power consumption prediction value corresponding to the i-th sample at the (j+N-1)-th moment in the future.

[0020] Based on the above scheme, after the power consumption prediction model is trained, the power consumption prediction model is used to perform real-time online wireless mouse power consumption prediction; for any new input sample point sequence , output by the power consumption prediction model Power consumption prediction sequence for the next M steps , the expression is: , Elements in For the future The power consumption prediction value of each time step takes into account the previous The user behavior characteristics of each time step reflect the impact of user usage patterns on power consumption.

[0021] Based on the above solution, the power consumption prediction sequence output by the power consumption prediction model is used as the environmental state, and the configuration combination of adjustable parameters including power management, clock gating, and sensor sampling is used as the control behavior, specifically:

[0022] Let the environment state space be , contains the power consumption prediction sequence within a time window, then at any time The mouse's environmental state It can be expressed as: , where R represents the field of real numbers, Indicates from time From now on, the future The power consumption prediction value of time steps, let the agent behavior space be , each control behavior is defined as a set of control parameter combinations, ,in Indicates the CPU main frequency, Indicates the supply voltage, represents the sensor sampling rate, is the total number of tunable parameters.

[0023] Based on the above scheme, at time , the agent is based on the current state of the environment Take a controlling action , then enter the next environment state , and feed back a scalar reward to the agent , Taking into account the actions The power consumption gain and performance loss after , are defined as: ,in express The actual power consumption of the mouse at this moment, and Respectively The response speed and sampling quality of the moment, and They are and The minimum required threshold, is the weight coefficient;

[0024] Find an optimal power consumption control strategy to maximize the long-term cumulative reward of the agent, is the discount factor, the cumulative reward is expressed as: ,in Indicates that in the strategy Under expectations, is the time step The reward value at the moment, the optimal power consumption control strategy for low power consumption optimization of the mouse Defined as: .

[0025] Based on the above solution, the power consumption control parameters are analyzed and the control instructions are generated as follows:

[0026] The current mouse environment state , through the optimal power consumption control strategy The mapping obtains the corresponding optimal control behavior: ,in , which represents the optimal configuration combination of various power consumption control parameters of the mouse, and the optimal control behavior is converted into Converted into a set of specific control instructions, including: adjusting the CPU main frequency to ; Adjust the output voltage of the power management unit to ; Adjust the sensor sampling rate to ;Adjust the working mode of peripherals;

[0027] The control instructions are sent to the execution unit through the driver layer interface, instructing the execution unit to adjust the corresponding mouse parameters, thereby switching the working state of the mouse to the optimal power consumption mode.

[0028] Based on the above solution, the real-time monitoring of the mouse power consumption status is specifically as follows:

[0029] The closed-loop control module monitors the power consumption level of the mouse in real time through a dedicated detection circuit. Switch to controlling behavior at the moment , and at the moment Read the current and voltage , then t to The average power consumption of the mouse during the time window is estimated to be:

[0030]

[0031] in is the number of sampling points, is the sampling interval, and Respectively represent The instantaneous values ​​of voltage and current at each sampling point.

[0032] Based on the above solution, calculating the power consumption error between the actual power consumption and the predicted power consumption includes:

[0033] The average power consumption and power consumption expectations Compare and calculate the power consumption error of the mouse : ,like Greater than the preset tolerance threshold , it means that the current optimal power consumption control strategy has deviations in actual effect, and a The optimal control behavior is corrected by the negative feedback correction amount, and the corrected optimal control behavior is output: ,in is the feedback gain matrix; It is adopted as the new optimal control behavior and drives the execution unit to adjust the mouse parameters accordingly.

[0034] The present invention has the following advantages and effects compared to the prior art:

[0035] (1) By learning the user's historical operation behavior through LSTM, the power consumption requirements of the mouse can be predicted more accurately, so that the power consumption management of the mouse can intelligently adapt to the user's habits instead of relying on a fixed time threshold, thus avoiding false sleep and ineffective power consumption;

[0036] (2) Based on the real-time data of the mouse, the power consumption control strategy of the mouse is adjusted through a reinforcement learning algorithm, so that it can intelligently optimize the power consumption distribution according to different usage scenarios, rather than adjusting it based only on fixed parameters;

[0037] (3) Through multi-sensor data fusion, combined with LSTM to predict the power consumption requirements of mouse users, and combined with reinforcement learning to optimize power consumption management, the mouse can intelligently adjust the sensor sampling rate, wireless communication mode and power consumption control strategy to achieve more accurate power consumption optimization. Compared with traditional sensor triggering methods, it improves the battery life of the mouse and reduces unnecessary power consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0039] Figure 1 This is a flowchart of a low-power intelligent optimization method for a wireless mouse based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to more clearly illustrate the purpose, technical solutions and advantages of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The example implementation methods can be implemented in various forms and should not be understood as being limited to the examples described herein. On the contrary, these implementation methods are provided to make the present invention more comprehensive and complete, and to fully convey the concepts of the example implementation methods to those skilled in the art.

[0041] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, it will be appreciated by those skilled in the art that the technical solutions of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0042] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0043] The step diagrams shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor do they necessarily need to be performed in the order described. For example, some operations / steps can be decomposed, while others can be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0044] The present invention will be described in detail below with reference to specific embodiments:

[0045] As attached Figure 1As shown, an embodiment of the present invention provides a low-power intelligent optimization method for a wireless mouse based on deep learning. The working steps of the method include: steps S1 to S5, which are specifically as follows:

[0046] Step S1: Real-time collection of wireless mouse operating status data is achieved through multi-sensor fusion technology.

[0047] Specifically, step S1 integrates a voltage sensor, a current sensor, an acceleration sensor, and a gyroscope sensor, which are respectively used to collect key parameters of the mouse, such as the power supply voltage, working current, motion acceleration, and angular velocity.

[0048] For example, the voltage sensor is connected to the power management unit of the mouse to monitor the output voltage of the battery or external power supply in real time. The sampling period is , then Sampling time . Define the voltage sampling value ,in represents the voltage sampling function, , and They are the minimum and maximum ranges of the voltage sensor respectively.

[0049] For example, the current sensor is connected in series in the main power circuit of the mouse to measure the instantaneous working current of the mouse circuit. Similar to voltage sampling, the current sampling value is defined as ,in represents the current sampling function, , and They are the minimum and maximum ranges of the current sensor respectively.

[0050] For example, the accelerometer and gyroscope are integrated into the mouse to detect the linear motion and angular motion of the mouse in three-dimensional space. These two motions correspond to the translation and rotation operations performed by the user holding the mouse. Assume that the sampling frequencies of the accelerometer and gyroscope are the same, both , then subsampling time .Will The instantaneous acceleration and angular velocity components collected at each moment are expressed as: Represents the acceleration component in the positive direction of the x-axis, in m / s²; Represents the acceleration component in the positive direction of the y-axis, in m / s²; Represents the acceleration component in the positive direction of the z-axis, in m / s²; Represents the angular velocity component in the positive direction of the x-axis, in rad / s; Represents the angular velocity component in the positive direction of the y-axis, in rad / s; Represents the angular velocity component in the positive direction of the z-axis, in rad / s.

[0051] It should be noted that the selection of the sensor coordinate system needs to be compatible with the mechanical structure of the mouse. For example, the x-axis and y-axis are aligned with the horizontal and vertical directions of the mouse, respectively, and the z-axis is perpendicular to the mouse surface and upward, but the specific coordinate system definition can be adjusted according to design needs. In addition, the layout of the sensor will also affect the quality of the collected data. To reflect the movement state of the mouse as realistically as possible, the accelerometer and gyroscope sensor should be close to the geometric center of the mouse and parallel to its main movement plane. The voltage sensor and current sensor should be as close as possible to the power input terminal to reduce the measurement error caused by line loss.

[0052] In summary, through the voltage sensor, current sensor, acceleration sensor and gyroscope sensor, at the moment and Collect the power supply voltage of the mouse separately , working current , instantaneous acceleration and instantaneous angular velocity .

[0053] Step S2, based on the voltage, current, acceleration and angular velocity data of the wireless mouse collected in real time, feature extraction and fusion of multi-source heterogeneous mouse operation status data are performed, including time alignment, data synchronization, feature calculation, data fusion and other steps.

[0054] Specifically, since the sampling frequencies and timestamps of the sensors may not be completely consistent, it is necessary to time-align the data streams they collect. The sensor with the highest sampling frequency is selected as the reference sensor, and the data of other sensors are interpolated or downsampled so that they correspond one-to-one with the data of the reference sensor in the time dimension. Assume that the sampling frequency of the gyroscope sensor is Higher than the sampling frequency of the accelerometer sensor , and both satisfy ,in If it is an integer greater than 1, the data of the acceleration sensor is Order linear interpolation is performed to increase the sampling frequency to be consistent with the gyroscope sensor; the original sampling sequence of the acceleration sensor is ,in , the new sequence obtained after interpolation , then: , in Through interpolation, the timestamps of the acceleration data are aligned with the gyroscope data, and the sampling frequency is also unified to The voltage and current data are processed in a similar way. After synchronization, all sensor data will form a unified time base.

[0055] It should be noted that refers to the moment when the original sensor data is sampled, refers to the data moment after interpolation;

[0056] Furthermore, based on time alignment, some important features are extracted from each sensor data to characterize the instantaneous power consumption and motion state of the wireless mouse. The following are the feature definitions of important features: Instantaneous power Indicates that the mouse is at the sampling time The electric power is calculated as follows: ,in and Respectively The voltage and current values ​​collected at all times, The unit is Watt (W). Indicates that the mouse is at the sampling time The modulus of the three-axis acceleration vector is calculated as follows: ,in Represent the interpolated Axis acceleration components, The unit is , Reflects the mouse The amplitude or intensity of movement at a moment. Rate of change of speed Characterization of mouse The sharpness of the momentary motion is approximately calculated by the difference of the resultant acceleration over time: ,in is the sampling interval, , The unit is The larger the value, the more dramatic the change in mouse speed. Angular velocity energy Defined as the mouse pointer Half of the sum of the squares of the three-axis angular velocity components at the moment, that is: ,in Respectively represent the gyroscope sensor in Collected at all times Axis angular velocity components, The unit is . The physical meaning is the rotational kinetic energy of the mouse rotation motion, the total kinetic energy Combining translational and rotational energies, the total motion intensity of the mouse is approximately evaluated, which is defined as: ,in and is the weight factor of translation-rotation energy, which can be adjusted according to the mechanical characteristics of the mouse and the actual application scenario. , , The unit is , which comprehensively judges the range of motion of each degree of freedom of the mouse.

[0057] Furthermore, the extracted important features are packaged into a unified feature vector and necessary metadata information is added to form a feature data frame that is easy to store and transmit. Frame data frame is recorded as , then: ,in is the timestamp of the frame data, take , that is, the voltage and current sampling time is used as the standard. The frequency of data frame generation is , which is consistent with the sampling frequency of the gyroscope sensor and the interpolated acceleration sensor.

[0058] Furthermore, all data frames are fused to form a unified feature data frame stream ,Each frame contains a comprehensive description of the current power consumption and motion state of the mouse.

[0059] In step S3, a power consumption prediction model for wireless mice is constructed based on a long short-term memory (LSTM) network. The power consumption prediction model takes the fused feature data frame stream as input and accurately predicts the future power consumption of the mouse by mining the temporal correlation between user behavior and mouse power consumption changes.

[0060] Specifically, the continuous A feature data frame constitutes a sample point, and each frame corresponds to a time step of LSTM. Let the sample point sequence be ,but Can be expressed as a A matrix with 6 rows and 6 columns:

[0061]

[0062] The subscript Indicates the number of the sample point, j=0,1,2,...,N-1, Respectively represent The first sample Frame data timestamp, instantaneous power, synthetic acceleration, velocity change rate, angular velocity energy and total motion energy. Each row of corresponds to the LSTM Considering that there may be a certain lag effect in the correlation between actual power consumption and behavior, the output of LSTM is set to the future time steps The power consumption prediction value of the mouse is expressed as a dimensional vector : ,in Indicates the Sample future The power consumption prediction value for each time step.

[0063] Furthermore, based on the above input and output definitions, the mathematical expression of the LSTM-based power consumption prediction model is as follows, where each feature data frame Corresponding to one time step input of the model:

[0064]

[0065] in: Indicates initializing the hidden state and unit state to zero vector, , indicating the The state output of the hidden layer in time steps, R represents the real number domain, is the hidden layer dimension; , indicating the The cell state at time steps, Indicates the The cell state at each time step; It is The input of a time step corresponds to a frame of data in the feature data frame stream; Indicates connecting the hidden state of the previous time step with the current input; Represent the forget gate, input gate, and output gate respectively; is the candidate unit status; is the model parameter matrix; is the bias term; Represents element-wise multiplication (Hadamard product); and Respectively Function and hyperbolic tangent function as activation functions of gate units and candidate states; the final output The power consumption prediction value corresponding to the i-th sample at the (j+N-1)-th moment in the future.

[0066] Furthermore, in view of the sleep and wake-up mechanism of wireless mice and the discontinuous working characteristics, a mask vector is introduced to process variable-length sequences when training and applying the LSTM model. The details are as follows: For each sample point , generate a corresponding mask vector ,in express No. The frame data is valid, otherwise it is padding data. In the forward calculation, the mask vector is used to mask the output and hidden state of each gate unit, so: ; ; ; ; and Replace the original power consumption prediction model , to achieve effective processing of indefinite length sequences.

[0067] Furthermore, the power consumption prediction model is trained and the mean square error (MSE) is used as the loss function, namely: ,in represents all parameters of the model, and Respectively represent The actual power consumption sequence and predicted value sequence of samples, The model training process uses optimization algorithms such as stochastic gradient descent (SGD) to minimize the loss function and iteratively update the parameters. Until the preset stop condition is reached.

[0068] Furthermore, after the model training is completed, the power consumption prediction model is used to perform real-time online wireless mouse power consumption prediction. , the power consumption prediction model gives its power consumption prediction sequence for the next M steps , the expression is: , Elements in The future The power consumption prediction value of each time step takes into account the previous The behavioral characteristics of each time step reflect the impact of user usage patterns on power consumption.

[0069] It should be noted that the mouse usage data (such as current, voltage, acceleration, etc.) collected by the sensor is used as input to predict future power consumption requirements through LSTM. Compared with traditional methods such as HMM, LSTM can learn more complex user behavior patterns and improve prediction accuracy.

[0070] Step S4: Based on the power consumption prediction model, an intelligent power consumption control strategy optimization model is constructed. The strategy optimization model uses the power consumption prediction sequence output by the power consumption prediction model as the environmental state, and the configuration combination of adjustable parameters including power management, clock gating, and sensor sampling as the intelligent agent behavior. The optimal power consumption control strategy for mouse power consumption optimization is obtained through strategy iterative optimization, thereby achieving a dynamic trade-off between mouse energy saving and efficiency improvement and performance experience.

[0071] Specifically, let the environment state space be , contains the power consumption prediction sequence within a time window, then at any time The mouse's environmental state It can be expressed as: , where R represents the field of real numbers, Indicates from time From now on, the future The power consumption prediction value of time steps. Let the agent behavior space be , each behavior It is defined as a set of control parameter combinations, namely: ,in Indicates the CPU main frequency, Indicates the supply voltage, represents the sensor sampling rate, is the total number of adjustable parameters. Different parameter combinations correspond to different energy-saving modes, for example, high performance mode corresponds to higher 、 and , while low power mode does the opposite.

[0072] Furthermore, at the moment , the agent is based on the current state of the environment Take an action , then enter the next environment state , and feed back a scalar reward to the agent , the reward takes into account the behavior The power consumption gain and performance loss after , are defined as: ,in express The actual power consumption of the mouse at this moment, and Respectively indicate time Response speed and sampling quality, and They are and The minimum required threshold, is the weight coefficient. The first term represents the negative utility of power consumption, and the last two terms represent the penalty for performance violation. Further find an optimal power consumption control strategy , so that under this strategy, the long-term cumulative reward of the agent is maximized, let is the discount factor, the cumulative reward is expressed as: ,in Indicates that in the strategy Under expectations, is the time step The reward value at the moment, the optimal power consumption control strategy for low power consumption optimization of the mouse Defined as: .

[0073] Solving the optimal power consumption control strategy based on policy gradient reinforcement learning algorithm , including REINFORCE and Actor-Critic. Specifically, taking the Actor-Critic algorithm as an example, let the policy function (Actor) be , the value function (Critic) is ,in and is the parameter to be learned. At each time step , according to the strategy Generate a behavior , and observe the state of the environment at the next moment and instant rewards Evaluate the current strategy through temporal difference (TD) error Pros and cons: ,pass Update the parameters of the policy function and the value function. For the policy function, the parameter update formula is: ,in is the learning rate of the policy function. For the value function, the parameter update formula is: ,in is the learning rate of the value function. The Actor-Critic algorithm optimizes the policy function and the value function through bidirectional iteration, gradually improving the long-term effectiveness of the policy.

[0074] It should be noted that in this embodiment, in order to further balance exploration and exploitation, the ϵ-greedy exploration mechanism is introduced to The probability of adopting the current optimal power consumption control strategy is ϵ, and the probability of randomly taking exploratory behavior is ϵ. As the training progresses, According to a certain schedule, from the initial value Decrease to 0, and gradually transition from exploration to utilization.

[0075] It should be noted that based on the data obtained in step S1, the mouse power consumption prediction through LSTM, the temporal relationship between user behavior and mouse power consumption changes, and the Actor-Critic reinforcement learning model adjusts the mouse working mode in real time, achieving the following technical effects: when it is predicted that the user will not use the mouse for a long time, the system reduces the sensor sampling rate in advance and adjusts the wireless communication mode to reduce power consumption. Compared with the traditional fixed threshold method, this embodiment provides a more intelligent and accurate power consumption optimization strategy.

[0076] After multiple rounds of iterative optimization, an optimal power consumption control strategy for the mouse power consumption is obtained. In the reasoning phase, for any input environment state s, the agent takes action , achieving real-time power consumption optimization control. After the control instructions are sent to the underlying execution unit, feedback is obtained through mechanisms such as power consumption detection to form supervised learning data for continuous optimization of the strategy optimization model.

[0077] It should be noted that the above optimal power consumption control strategy The optimization process can be performed in the cloud or locally, using either online or offline learning. During online learning, each wireless mouse acts as an independent agent, optimizing its power consumption control strategy based on its own usage data, which is stored locally. During offline learning, the terminal uploads its state-behavior-reward data to the cloud, which aggregates data from multiple users to train a common strategy, which is then distributed to the terminal. These two modes can be combined to fine-tune the common strategy.

[0078] Step S5: The optimal power consumption control strategy output by the strategy optimization model It converts the power consumption into specific execution instructions and monitors the system status of the mouse in real time. It continuously optimizes and corrects the power consumption control strategy through the negative feedback mechanism to cope with the complex and changeable mouse usage environment.

[0079] Specifically, for the current state of the environment , through the optimal power consumption control strategy The mapping obtains the corresponding optimal control behavior: ,in , which represents the optimal configuration combination of various mouse power consumption control parameters.

[0080] Furthermore, the closed-loop control module converts the optimal control behavior into a set of specific control instructions, including: adjusting the CPU main frequency to ;Regulate the output voltage of the power management unit (PMU) to ; Adjust the sensor sampling rate to ; Adjust the working mode of other peripherals such as LEDs, motors, wireless communications, etc. Control instructions are sent to the functional units and peripherals of the execution unit through the driver layer interface, guiding them to make corresponding parameter adjustments, thereby switching the working state of the mouse to the optimal power consumption mode.

[0081] At the same time, the closed-loop control module also monitors the power consumption level of the mouse in real time through a dedicated detection circuit. Switch to Control Behavior , and at the moment Read the current and voltage , then t to The average power consumption of the mouse in the time window can be estimated as:

[0082]

[0083] in is the number of sampling points, is the sampling interval, and Respectively represent The instantaneous values ​​of voltage and current at each sampling point.

[0084] Furthermore, the predicted average power consumption and power consumption expectations For comparison, calculate the power consumption error of the mouse: ,like Greater than the preset tolerance threshold , it means that the current strategy has deviations in actual effect, and a The negative feedback correction amount is used to correct the optimal control behavior. The corrected output is: ,in is the feedback gain matrix; This is adopted as the new optimal control behavior and drives the execution unit to adjust the mouse parameters accordingly. Through the negative feedback mechanism, the system can adaptively mitigate policy deviations, allowing the actual power consumption level of the mouse to gradually converge to the optimal power consumption configuration.

[0085] It should be noted that compared with traditional fixed thresholds or simple trigger mechanisms, this embodiment not only uses sensors to detect the mouse status, but also intelligently analyzes sensor data and optimizes power consumption management strategies, enabling the mouse to adaptively adjust its power consumption mode according to different user usage patterns, thereby improving energy efficiency.

[0086] In addition to power consumption feedback, the closed-loop control module is also responsible for monitoring the working status of the mouse, including performance indicators (such as response delay, data throughput) and abnormal events (such as overheating, undervoltage, etc.). Once a serious performance degradation or abnormal event is detected, the system will automatically switch to emergency energy-saving mode and adopt predetermined safety strategies (such as forced frequency reduction, shutting down non-core peripherals) to ensure basic functions, while also feeding back alarm information. Based on this information, the operating system and intelligent applications can take corresponding energy-saving measures (such as postponing low-priority tasks, reducing the interface refresh rate), or remind users to charge or replace batteries in time. Feedback information such as power consumption errors, performance warnings, and abnormal events is further incorporated into consideration, and the policy model is dynamically adjusted through mechanisms such as online learning to continuously improve the accuracy and robustness of energy efficiency control. This forms a self-optimizing and adaptive closed-loop control process.

[0087] The entire closed-loop control process requires no human intervention and is applicable to a wide range of application scenarios. Thanks to negative feedback correction and online learning capabilities, the control system can proactively perceive changes in the external environment and its own state, autonomously adjusting its optimization strategy and intelligently balancing energy efficiency, performance, safety, and other dimensions. This adaptive closed-loop control mechanism significantly improves the wireless mouse's energy efficiency and user experience while reducing the user's operational burden.

[0088] It should be noted that the mathematical model and algorithm used in the closed-loop control mechanism described above can be appropriately simplified or improved based on specific hardware configurations and application requirements. The core concept is to empower devices with intelligent "self-management and self-optimization" capabilities through self-monitoring, self-analysis, self-decision-making, and self-execution technologies, enabling them to achieve or approach optimal comprehensive energy efficiency in complex application environments.

[0089] This embodiment covers multiple steps, including data acquisition, feature extraction, power consumption prediction, intelligent control strategy optimization, and closed-loop feedback adjustment. The wireless mouse's operating status data is acquired through voltage sensors, current sensors, accelerometers, and gyroscope sensors, and key feature information is extracted using time alignment and data fusion techniques. A power consumption prediction model is constructed based on a long short-term memory network to learn the temporal relationship between user behavior and mouse power consumption changes, enabling accurate prediction of the mouse's future power consumption. Based on the prediction results, a policy gradient optimization algorithm is used to construct an intelligent power consumption control strategy optimization model. The optimal power consumption control strategy is generated by adjusting the CPU main frequency, power management, power management unit voltage, sensor sampling rate, and peripheral operating mode. A closed-loop control mechanism monitors the mouse's power consumption in real time, calculates the error between actual and predicted power consumption, and dynamically adjusts the control strategy using a negative feedback mechanism to ensure the stability and adaptability of the mouse's energy consumption optimization. This approach effectively reduces the wireless mouse's energy consumption, extends its battery life, and maintains a balanced user experience, making it suitable for a variety of application scenarios.

[0090] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present invention are indicated by the claims. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and that various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A low-power intelligent optimization method for wireless mouse based on deep learning, characterized in that: include: Step S1: Realize real-time collection of the operating status data of the wireless mouse through multi-sensor fusion technology; Step S2: Extract and fuse the features of the running status data to obtain a unified feature data frame stream. ,in Represents feature data frame, each frame Both include the current power consumption and motion characteristics of the mouse; Step S3: constructing a power consumption prediction model for the wireless mouse based on the long short-term memory network, wherein the power consumption prediction model takes the feature data frame stream as input and accurately predicts the future power consumption of the mouse by learning the temporal relationship between user behavior and changes in mouse power consumption; Step S4: Based on the power consumption prediction model, an intelligent power consumption control strategy optimization model is constructed. The strategy optimization model uses the power consumption prediction sequence output by the power consumption prediction model as the environmental state, and uses a configuration combination of adjustable parameters including power management, clock gating, and sensor sampling as the control behavior. The optimal power consumption control strategy is generated through strategy iterative optimization. Step S5: Based on the optimal power consumption control strategy, the power consumption control parameters are analyzed, and control instructions are generated to adjust the CPU main frequency, the power management unit voltage, the sensor sampling rate and the peripheral working mode. Based on the adjusted power consumption control parameters, the mouse power consumption status is monitored in real time, the power consumption error between the actual power consumption and the predicted power consumption is calculated, and a negative feedback mechanism is used to optimize the optimal power consumption control strategy so that the power consumption control parameters dynamically converge to the optimal power consumption configuration; if power consumption abnormality is detected, switch to the safe mode and execute the corresponding power consumption management strategy.

2. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 1, characterized in that: The multi-sensor includes a voltage sensor, a current sensor, an acceleration sensor and a gyroscope sensor, and the collected operating status data are the power supply voltage, working current, motion acceleration and angular velocity of the mouse respectively; The feature data frame ,in is the timestamp of the feature data frame, based on the voltage and current sampling time. They are the instantaneous power, synthetic acceleration, velocity change rate, angular velocity energy and total motion energy of the mouse at the mth frame, and the generation frequency of the feature data frame is , which is consistent with the sampling frequency of the gyroscope sensor and the interpolated acceleration sensor.

3. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 2, characterized in that: The continuous feature data frame Constitute a sample point, each frame corresponds to a time step of the power consumption prediction model, let the sample point sequence be ,but Can be expressed as a A matrix with 6 rows and columns, a matrix Each row corresponds to the power consumption prediction model The input data of time steps, the output of LSTM is set to the future time steps The power consumption prediction value of the mouse is expressed as a dimensional vector : ,in Indicates the Sample future The power consumption prediction value for each time step.

4. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 3, characterized in that: The power consumption prediction model is expressed as follows: ,in: Indicates initializing the hidden state and unit state to zero vector, Indicates the The state output of the hidden layer in time steps, R represents the real number domain, is the hidden layer dimension; Indicates the The cell state at time steps, Indicates the The cell state at each time step; It is The input of a time step corresponds to a frame of data in the feature data frame stream; Indicates connecting the hidden state of the previous time step with the current input; Represent the forget gate, input gate, and output gate respectively; is the candidate unit status; is the model parameter matrix; is the bias term; Represents element-wise multiplication (Hadamard product); and Respectively Function and hyperbolic tangent function, as the activation function of the gate unit and candidate state, the final output The power consumption prediction value corresponding to the i-th sample at the (j+N-1)-th moment in the future.

5. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 3, characterized in that: After the power consumption prediction model is trained, the power consumption prediction model is used to predict the power consumption of the wireless mouse in real time. , output by the power consumption prediction model Power consumption prediction sequence for the next M steps , the expression is: , Elements in For the future The power consumption prediction value of each time step takes into account the previous The user behavior characteristics of each time step reflect the impact of user usage patterns on power consumption.

6. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 1, characterized in that: The power consumption prediction sequence output by the power consumption prediction model is used as the environmental state, and the configuration combination of adjustable parameters including power management, clock gating, and sensor sampling is used as the control behavior, specifically: Let the environment state space be , contains the power consumption prediction sequence within a time window, then at any time The mouse's environmental state It can be expressed as: , where R represents the field of real numbers, Indicates from time From now on, the future The power consumption prediction value of time steps, let the agent behavior space be , each control behavior is defined as a set of control parameter combinations, ,in Indicates the CPU main frequency, Indicates the supply voltage, represents the sensor sampling rate, is the total number of tunable parameters.

7. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 6, characterized in that: At the moment , the agent is based on the current state of the environment Take a controlling action , then enter the next environment state , and feed back a scalar reward to the agent , Taking into account the actions The power consumption gain and performance loss after , are defined as: ,in express The actual power consumption of the mouse at this moment, and Respectively The response speed and sampling quality of the moment, and They are and The minimum required threshold, is the weight coefficient; Find an optimal power consumption control strategy to maximize the long-term cumulative reward of the agent, is the discount factor, the cumulative reward is expressed as: ,in Indicates that in the strategy Under expectations, is the time step The reward value at the moment, the optimal power consumption control strategy for low power consumption optimization of the mouse Defined as: .

8. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 5, characterized in that: The specific steps of analyzing the power consumption control parameters and generating the control instructions are as follows: The current mouse environment state , through the optimal power consumption control strategy The mapping obtains the corresponding optimal control behavior: ,in , which represents the optimal configuration combination of various power consumption control parameters of the mouse, and the optimal control behavior is converted into Converted into a set of specific control instructions, including: adjusting the CPU main frequency to ; Adjust the output voltage of the power management unit to ; Adjust the sensor sampling rate to ;Adjust the working mode of peripherals; The control instructions are sent to the execution unit through the driver layer interface, instructing the execution unit to adjust the corresponding mouse parameters, thereby switching the working state of the mouse to the optimal power consumption mode.

9. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 8, characterized in that: The real-time monitoring of the mouse power consumption state is specifically as follows: The closed-loop control module monitors the power consumption level of the mouse in real time through a dedicated detection circuit. Switch to controlling behavior at the moment , and at the moment Read the current and voltage , then t to The average power consumption of the mouse during the time window is estimated to be: ,in is the number of sampling points, is the sampling interval, and Respectively represent The instantaneous values ​​of voltage and current at each sampling point.

10. The method for low-power intelligent optimization of wireless mouse based on deep learning according to claim 9, characterized in that: Calculating the power consumption error between the actual power consumption and the predicted power consumption includes: The average power consumption and power consumption expectations Compare and calculate the power consumption error of the mouse : ,like Greater than the preset tolerance threshold , it means that the current optimal power consumption control strategy has deviations in actual effect, and a The negative feedback correction amount is used to correct the optimal control behavior and output the corrected optimal control behavior: ,in is the feedback gain matrix; It is adopted as the new optimal control behavior and drives the execution unit to adjust the mouse parameters accordingly.

Images