A push button switch adaptive optimization method and storage medium

Abstract

The present invention discloses a push button switch adaptive optimization method and storage medium. This method acquires tactile signals through high-density piezoelectric sensors and combines spatiotemporal convolution kernels with a dynamic baseline calibration algorithm to extract pressure distribution, gradient change rate, and energy concentration characteristics. It integrates time-frequency domain analysis of historical operation sequences with neural differential equation modeling to predict operation probabilities and flag anomalies. It also integrates environmental parameters to generate anti-interference control instructions and uses a gated attention mechanism to dynamically weight multi-source data. The system establishes a multi-level reliability protection mechanism, achieving fault immunity through feature drift detection, digital twin residual verification, and hybrid control switching. It responds to real-time needs, dynamically optimizes heterogeneous computing resource allocation, performs three-dimensional storage reconstruction, and performs precision bit width switching, forming a closed-loop cognitive evolutionary chain.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a button switch adaptive optimization method and a storage medium. Background Art

[0002] Traditional pushbutton switches face significant technical challenges in complex industrial environments. Existing pushbutton control systems suffer from high response latency, making them difficult to meet real-time requirements, especially when handling high-frequency operations. The complexity of indicator light logic creates a high cognitive load for users and makes maintenance difficult. Furthermore, they lack environmental adaptability, making them susceptible to malfunctions caused by temperature fluctuations, mechanical vibration, and electromagnetic interference. In other words, they have poor environmental adaptability and are susceptible to interference-induced malfunctions.

[0003] Furthermore, existing methods often rely on a single signal modality and lack the ability to fuse multi-source data, making it difficult to accurately capture user operational intent. Regarding fault handling, traditional systems typically employ fixed threshold detection and are unable to dynamically adapt to feature drift, resulting in inefficient fault recovery, insufficient reliability, and insufficient fault handling capabilities. While some research has attempted to incorporate machine learning algorithms, these efforts have not effectively combined heterogeneous computing resource optimization with dynamic reconstruction of three-dimensional storage, making it difficult to balance real-time performance with energy efficiency. Summary of the Invention

[0004] (1) Technical issues to be resolved

[0005] To solve the above problems, the present invention proposes a button switch adaptive optimization method and storage medium, aiming to solve the problem in the existing technology that it relies on a single signal modality, lacks the ability to fuse multi-source data, and is difficult to accurately capture user operation intentions.

[0006] (2) Technical solution

[0007] A push button switch adaptive optimization method of the present invention,

[0008] include:

[0009] Acquire a tactile signal of the push button switch, and obtain a spatiotemporal feature vector based on the tactile signal, wherein the spatiotemporal feature vector includes pressure distribution, pressure gradient change rate, and energy concentration;

[0010] Acquire a historical operation event sequence of the push button switch, and obtain a time series prediction result based on the spatiotemporal feature vector and the operation event sequence, wherein the time series prediction result includes an operation probability distribution and an abnormality mark;

[0011] Acquiring environmental state parameters of the push button switch, and generating control instructions based on the environmental state parameters, the time series prediction results, and the spatiotemporal feature vectors, for converting the abstract decision of the neural network into executable control instructions;

[0012] Based on the execution status of the control instructions and the response delay of the physical equipment, a multi-level reliability protection mechanism is constructed to achieve fault immunity through feature drift detection, physical-digital state residual analysis and hybrid control loop switching;

[0013] In response to the feature drift detection results and fault immunity requirements, the heterogeneous computing resource allocation strategy is dynamically optimized, and three-dimensional storage space reconstruction and dynamic precision bit width switching are performed based on real-time task priorities and energy consumption constraints, forming a closed-loop cognitive evolution link.

[0014] In the present invention, the method for extracting the spatiotemporal feature vector of the tactile signal includes:

[0015] Construct a three-dimensional pressure field to collect the pressure distribution matrix, and apply the spatiotemporal convolution kernel to smooth the pressure distribution matrix;

[0016] A dynamic baseline calibration algorithm is used to eliminate ambient temperature drift. The dynamic baseline algorithm includes a smoothing factor that is adaptively adjusted with the operating frequency.

[0017] In the present invention, the spatiotemporal convolution kernel smoothes the pressure distribution matrix, further comprising:

[0018] Normalizing the pressure gradient change rate to generate a standardized gradient vector;

[0019] The energy concentration of the pressure distribution matrix is ​​calculated.

[0020] In the present invention, the method for generating the time series prediction result includes:

[0021] Perform multi-scale wavelet scattering transform on the historical operation event sequence, extract the time-frequency domain feature tensor, and use neural differential equations to model the implicit dynamic process;

[0022] The probability distribution of future action intentions is predicted through a bidirectional gated recurrent unit, and when the prediction residual exceeds a statistical threshold, it is marked as an abnormal action.

[0023] In the present invention, the environmental state parameters include temperature, mechanical vibration main frequency, electromagnetic interference intensity and power supply ripple coefficient. The method for fusing the environmental state parameters with the adaptive control instructions is:

[0024] Build a preset environmental condition template library containing legal state transition paths for multiple typical environmental parameter combinations. Use the attention mechanism to dynamically weight the impact of environmental parameters on the control strategy and generate anti-interference optimization instructions.

[0025] When the electromagnetic interference intensity exceeds the degradation threshold, the degradation mode is forcibly enabled to limit high-frequency state switching operations.

[0026] In the present invention, the multi-level reliability protection mechanism includes:

[0027] Online monitoring of feature drift, detecting abnormal inputs through Mahalanobis distance, and triggering an alarm when the Mahalanobis distance value is greater than 5.99;

[0028] Monitoring the ReLU activation rate of a hidden layer of a neural network, and freezing network parameters when the ReLU activation rate exceeds a freezing threshold;

[0029] Compare sensor data with the digital twin's predicted value and switch to the backup PID controller when the residual exceeds three times the standard deviation.

[0030] In the present invention, the heterogeneous computing resource allocation strategy includes:

[0031] Build a heterogeneous computing resource partitioning mechanism to decouple heterogeneous task flows into four parallel execution units based on functional modes: perception signal analysis, decision logic generation, safety constraint verification, and system interaction and coordination;

[0032] A dynamic numerical representation mode switching strategy is designed to adaptively adjust the data precision topology structure according to the criticality level of the real-time task, realizing dynamic migration between low-entropy computing state and high-fidelity analysis state.

[0033] Another computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the button switch adaptive optimization method described in any one of the above technical solutions.

[0034] (3) Beneficial effects

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention greatly reduces the delay of state switching and improves the response speed through the spatiotemporal convolution kernel and dynamic baseline calibration algorithm, and adopts the combined entropy compression and adaptive light language system to reduce the logic complexity of the indicator light.

[0037] The present invention greatly shortens the fault recovery time through multi-level protection mechanism and hybrid control loop switching. Dynamic precision bit width switching and three-dimensional storage reconstruction technology greatly improve the reasoning ability and meet the strict power consumption constraints of industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 Schematic diagram of the process structure of the adaptive optimization method;

[0040] Figure 2 This is the pressure thermal diagram of the push button switch;

[0041] Figure 3 Schematic diagram of the process structure of the closed-loop cognitive evolution link. DETAILED DESCRIPTION

[0042] This paper constructs a tactile intelligent agent that achieves the autonomous evolution of industrial interaction interfaces through the deep integration of spatiotemporal perception and dynamic decision-making topology. The system core constructs and analyzes a multidimensional tactile field, upscaling the spatiotemporal gradient of the pressure matrix into a continuous manifold of operational semantics, and reconstructing the topological space of human-machine dialogue at the level of differential geometry. The fractal decision-making architecture transforms mechanical impulses into a semantically dense control language through a cognitive transition from microscopic to mesoscopic to macroscopic, forming an embodied cognitive closed loop.

[0043] Example 1

[0044] like Figure 1-Figure 3 A push button switch adaptive optimization method shown includes the following steps:

[0045] S100 : Acquire a tactile signal of a push button switch, and obtain a spatiotemporal feature vector based on the tactile signal, where the spatiotemporal feature vector includes pressure distribution, pressure gradient change rate, and energy concentration.

[0046] During the tactile signal acquisition phase, a high-density piezoelectric sensor array is used to construct a three-dimensional dynamic pressure field. The original signal flows through the baseline drift compensation module to construct a sliding window energy entropy model:

[0047] ,in Represents the energy entropy value within the window, which is used to measure the degree of disorder of pressure distribution. For the window Normalized probability of pressure values For the window Pressure sampling values, is the sum of all pressure values ​​within the window, is the number of samples in the sliding window. The sliding window energy entropy model quantifies the degree of disorder in the pressure distribution and dynamically detects and compensates for baseline drift. This addresses tactile signal distortion caused by environmental interference or hardware aging. It is a core preprocessing module that ensures the accuracy of spatiotemporal feature vectors.

[0048] When the window entropy value exceeds the threshold, dynamic compensation is triggered, where the threshold is 2.5. Dynamic compensation is triggered when , and the compensation amount is generated by weighted fusion of the historical baseline sliding mean and the current window minimum: ,in is the historical baseline sliding mean, calculated by exponential moving average (EMA), , , reflecting the degree of retention of historical data, is the current baseline estimate. The historical baseline sliding mean Its purpose is to capture the long-term baseline trend of the sensor signal and suppress low-frequency drift such as signal offset caused by temperature change. For the time window Internal, tactile signals The function of the voltage regulator is to detect instantaneous undershoot or abnormal low voltage of the signal in real time, reflecting the current environmental interference, such as sudden mechanical vibration.

[0049] Noise suppression uses a joint filtering strategy—the anisotropic diffusion equation is applied in the spatial dimension: ,in Represents the space-time pressure field function, describing the button surface at position ,time The pressing intensity, in which the spatiotemporal pressure field function is used as the raw tactile signal input to be processed, contains noise and effective pressure distribution information; represents the spatial gradient operator of the pressure field, Used to quantify the local change rate of the pressure field and to identify the edge and texture features of the pressed area; It represents the modulus of the gradient, which is used to measure the severity of local pressure changes. High modulus values ​​correspond to edges or mutation areas. is the diffusion coefficient, , is the gradient threshold parameter, in this disclosure The gradient threshold parameter is mainly used to dynamically adjust the diffusion intensity. When the modulus of the gradient is much larger than the gradient threshold parameter, Strong diffusion is allowed to suppress noise; when the modulus of the gradient is much smaller than the gradient threshold parameter, diffusion is suppressed to retain details. Used to represent the divergence operator acting on the diffusion flux , which is used to describe the anisotropic diffusion process of the pressure field and adaptively smooth along the gradient direction.

[0050] The time dimension uses an improved Kalman filter, and the process noise covariance matrix Adaptive adjustment according to the pressure change rate, ,in Indicates the Kalman filter The process noise covariance matrix at the moment is used to characterize the uncertainty of the system model. The larger the value, the lower the trust in the predicted value. In tactile signal processing, Dynamically adjust the filter's response speed and stability to balance the two. In the above expression, 0.1 represents the basic noise covariance, which is used to ensure the minimum noise tolerance under static or slowly changing pressure, and avoid excessive filtering sensitivity leading to signal distortion; 0.05 represents the pressure change rate weight coefficient, which is used to adjust the contribution of the pressure change rate to the noise covariance and control the sensitivity of the dynamic response; Indicates the amplitude of the rate of change of pressure over time.

[0051] Specifically, the pressure distribution field is extracted through multi-scale morphological analysis: constructing structural elements , representing pixel disks with diameters of 3, 5, and 6, respectively, and performing opening operations and closed operation , calculate the morphological gradient: ,in Indicates location The morphological gradient value at Indicated in the structural element The upper bound is the maximum value in the defined neighborhood. Indicated in the structural element Take the lower bound, which is the minimum value, within the defined neighborhood. Represent the original signal In position The value at .

[0052] The final feature vector consists of three parts: center of mass positioning, shape description and complexity evaluation. The center of mass positioning calculates the center of gravity coordinates of the pressure field through the zero-order moment and the first-order moment to identify the core pressing area; the shape description quantifies the overall shape of the pressure distribution, such as symmetry and ductility, by extracting the first three-order features of the Hu invariant moment; and the complexity evaluation characterizes the degree of randomness of the tactile signal by calculating the normalized entropy value of the pressure distribution.

[0053] The pressure gradient change rate is determined through a combined spatiotemporal analysis. The Sobel operator is used in the spatial dimension to detect the lateral and longitudinal gradients of the pressure field, while the five-point central difference method is used in the temporal dimension to calculate the instantaneous rate of change. The system sorts the gradient data by significance, extracts the top 5% of high-gradient regions, generates a histogram, and then calculates the spatial distribution frequency in 20 bins, effectively capturing the spatiotemporal characteristics of sudden operating behaviors, such as rapid key presses.

[0054] Energy concentration analysis integrates time- and frequency-domain information. Spatially, the energy integral of each sensor unit within a 50ms window is calculated, and principal component analysis is used to extract the energy field's concentration direction. Temporally, a short-time Fourier transform is used to monitor the center-of-gravity frequency of the energy spectrum. When high-frequency energy events are detected, the system automatically triggers a transient response mode, such as increasing filtering strength in environments with strong mechanical vibration. In this embodiment, high-frequency energy refers to energy frequencies greater than 150Hz.

[0055] Spatiotemporal feature fusion utilizes an attention-guided mechanism. Three encoder branches process the raw pressure, gradient, and energy data, respectively, and dynamically assign feature weights using cross-attention. For example, when oil contamination causes localized sensor failure, the system automatically reduces the attention weight of the abnormal region and instead enhances the contribution of spatiotemporal gradient features. The fused feature vectors are optimized through contrastive learning, forcing features of similar operations to cluster closely in the vector space and significantly separating heterogeneous operations.

[0056] S200 , obtaining a historical operation event sequence of a push button switch, and obtaining a time series prediction result based on the spatiotemporal feature vector and the operation event sequence, wherein the time series prediction result includes an operation probability distribution and an abnormality mark.

[0057] In the processing of historical operation event sequences, a multi-scale sliding window mechanism is used to capture operation characteristics at different time granularities. The short-term window focuses on transient pressure changes and calculates the maximum and variance of the pressure gradient within the window; the medium-term window counts the operation frequency and pressure distribution entropy to quantify the randomness of the operation.

[0058] The long-term window extracts the first five components of the Fourier coefficients and identifies periodic patterns. The three-level features are concatenated into a 17-dimensional time series tensor. ,in is the number of time steps, forming a hierarchical temporal representation.

[0059] The fusion of spatiotemporal features and operation sequences utilizes a dynamic attention mechanism. The spatiotemporal feature vector undergoes a linear transformation to generate a query vector, and the temporal features of the operation sequence are convolutionally encoded to form key-value pairs. By calculating attention weights, historical operation segments that are strongly correlated with the current spatiotemporal state are selected. For example, when a specific pressure distribution pattern is detected, the model focuses on the regularity of past operation intervals that exhibit similar patterns. The fused features are input into a bidirectional gated recurrent network, whose hidden state simultaneously predicts future operation probabilities and anomaly risks.

[0060] The operation probability distribution is dynamically generated through a mixed Gaussian model. The model outputs three Gaussian distribution parameters (mean, variance, and weight), which correspond to three typical modes: quick response, normal operation, and delayed triggering. For example, if the current pressure is concentrated in the edge area and changes slowly, the model will increase the weight of the delayed triggering mode and predict that the next operation interval may be extended to more than 800ms. In real-time detection, if the actual interval continues to deviate from the predicted value by more than three times the standard deviation, the primary anomaly mark is triggered. In order to distinguish occasional fluctuations from real faults, a secondary verification is introduced: after three consecutive primary anomalies, the Mahalanobis distance of the operation characteristics in the window is calculated. If it exceeds the statistical threshold, it is confirmed as a final anomaly.

[0061] Anomaly detection combines gradient analysis and time localization. Backpropagation is used to calculate the contribution of the neural network's hidden layer to anomaly flagging, pinpointing the critical time period that triggered the anomaly. For example, an anomaly flag might be caused by three abnormal pressure spikes within the past two seconds. When flagging an anomaly, the system generates an event report with a timestamp and influencing factors, helping maintenance personnel quickly locate the root cause.

[0062] Model training utilizes progressive multi-task learning. The primary task minimizes the error in predicting the operation interval, while the secondary task optimizes anomaly classification accuracy. Initial training focuses on normal operation data, gradually introducing 5% to 30% of abnormal samples to enhance the model's sensitivity to rare events. To prevent overfitting, a temporal smoothing constraint is added to enforce a smooth transition between hidden states at adjacent moments. A dynamic parameter update strategy is employed during deployment. After every 1,000 operations, the model is fine-tuned using the latest data, retaining the core weights unchanged and updating only the output layer parameters. Furthermore, a library of anomaly patterns is maintained, automatically triggering incremental model learning when new anomalies are detected.

[0063] S300, obtaining environmental state parameters of a push button switch, and generating control instructions based on the environmental state parameters, the timing prediction results, and the spatiotemporal feature vectors, for converting the abstract decision of the neural network into executable control instructions.

[0064] In the environmental state parameter collection stage, a three-dimensional perception network integrating temperature, humidity and electromagnetic interference is used: the temperature sensor uses an NTC thermistor and is calibrated using a second-order polynomial equation. Convert the voltage signal to an accurate temperature value, where is the calibrated temperature value, which represents the actual temperature calculated by the sensor voltage. The original voltage input of the sensor is the unprocessed electrical signal output by the thermistor. is the quadratic coefficient, which characterizes the curvature of voltage changing with temperature. is the linear coefficient, which represents the linear proportional relationship between voltage and temperature. is a constant term, which represents the temperature offset when the voltage is zero, such as baseline compensation; humidity measurement uses a capacitive sensing unit, whose output Adaptive baseline compensation Eliminate the influence of temperature drift; the electromagnetic interference intensity is captured by the wide-band detection circuit, and the energy integral value in the 10kHz-1GHz frequency band is extracted , is the electromagnetic interference intensity, which represents the sum of electromagnetic energy in a specified frequency band. They represent the lower and upper limits of the integral frequency, respectively, and define the monitoring frequency band of electromagnetic interference. It is the frequency domain representation of the electromagnetic interference signal, obtained by Fourier transform (FFT) of the time domain signal; the energy integral value is converted into a standardized signal with a dynamic range of 0-5V through logarithmic compression. Environmental parameters and spatiotemporal feature vectors The fusion uses a gated attention mechanism: , Represents a comprehensive index of environmental stress, which is determined by temperature and humidity Fusion generation, and are the normalized temperature and humidity values; The time series prediction state of the LSTM network characterizes the time evolution pattern of button operations; represents the learnable weight matrix, used to generate queries and keys respectively; is the query vector, generated based on spatiotemporal features, is the key vector, which is generated by concatenating the environment indicator and the LSTM state; represents the attention weight matrix, which represents the correlation strength between spatiotemporal features and environmental / temporal features, where Indicates calculating the similarity between the query and the key. is the probability distribution normalized by Softmax after scaling.

[0065] is a comprehensive indicator of environmental stress, and the fused features Input control strategy generator, which consists of two parallel fully connected layers: the main network outputs the original control quantity ,in is the weight matrix of the main network, used to integrate the fusion features Mapped to the control volume space, is the fused feature vector, including spatiotemporal features , environmental parameters and time series forecast status The fused feature vector encodes the multi-dimensional information of the current system status, such as button pressure distribution, ambient temperature and humidity, historical operation mode, etc. is the bias term, which is used to adjust the reference output of the control quantity; is the activation function, which is the Sigmoid function in this embodiment, used to compress the linear transformation result to Range. The original control amount is obtained through the weight matrix Fusion features Perform linear transformation and superimpose bias After activation function Generate normalized raw control volume For example, when the ambient temperature rises and the pressure gradient changes dramatically, The weight of the corresponding dimension in increases, resulting in The output tends to 1, triggering a fast state switching instruction; when the confidence level is low (such as severe electromagnetic interference), the safety constraint network is reduce The weight of Ensure system security.

[0066] Safety Constraint Network Generation Confidence , is the weight matrix of the security constraint network, which is used to integrate the fusion feature vector The linear transformation parameters mapped to the confidence space are learned through training and represent the weight of the impact of different feature dimensions on security; is the bias term; confidence It is a dynamic safety gate that achieves a balance between intelligent decision-making and robustness. The final control instructions are fused through dynamic weighting: , The final generated control instruction is used to directly drive a button switch or indicator light. Its main function is to integrate the output of the main network decision and safety constraints, balancing real-time response and system safety. It is a preset safety instruction, generated based on historical data or expert rules. Adjust the weight of the main network and security policy in real time, such as when the environmental pressure Increase (such as a sudden temperature rise) Reduce, The proportion increases to avoid overheating risks; when electromagnetic interference reduce, Increase, making full use of the optimization capabilities of neural networks.

[0067] Preset safety instructions Generated based on historical operating modes, such as forcing the switch trigger sensitivity to be reduced in a high temperature environment. Instruction discretization uses an adaptive threshold mechanism: defining the critical value of state switching ,in For the last 100 times The mean of is the standard deviation, when Generates high level command , otherwise maintain In order to deal with electromagnetic interference scenarios, an instruction verification link is added: a 3-bit parity check code is added to each control instruction. The receiving end corrects single-bit errors through a majority voting mechanism. If the verification fails, the timing is backtracked and the valid instructions of the previous two cycles are re-executed. In extreme working conditions, for example , start degradation mode: turn off the neural network inference engine and switch to a conservative strategy based on the sliding average filter to ensure the minimum functional availability of the system.

[0068] S400. Based on the execution status of the control instruction and the response delay of the physical device, a multi-level reliability protection mechanism is constructed to achieve fault immunity through feature drift detection, physical-digital state residual analysis and hybrid control loop switching.

[0069] During the feature drift detection phase, the system builds a dynamic protection mechanism by real-time monitoring of the execution status of control instructions, such as response delay, execution success rate, and resource utilization. Based on the sliding window statistical method, the system continuously calculates the degree of deviation between the current state and the historical normal mode. When the state indicators, such as a sudden increase in delay or a sudden drop in success rate, differ from the long-term statistical mean by more than three times the standard deviation, a security response is immediately triggered: freeze the parameter update of the online learning module, roll back to the stable weight snapshot of the previous 5 minutes, and start the deep diagnosis process. This mechanism can effectively deal with feature drift problems caused by sudden environmental changes or model degradation. For example, in industrial robot arm control, when sudden load changes cause abnormal delays, the system can complete the state freeze within 2 milliseconds to prevent the issuance of erroneous instructions.

[0070] Physical-digital state residual analysis is achieved through bidirectional verification between the digital twin and the actual device. The system incorporates a high-precision device dynamics model, generating real-time predictions for key parameters such as motor speed and temperature, which are then compared with physical sensor data in milliseconds. Residual assessment utilizes a dynamic threshold strategy: This strategy comprehensively considers historical fluctuation patterns and instantaneous change rates. Hardware anomalies are identified when residual amplitudes consistently exceed limits and are accompanied by a surge in high-frequency noise energy, such as when energy in the 2-4kHz band exceeds the baseline by 10dB.

[0071] The hybrid control loop switching adopts a gradual migration strategy to ensure control continuity in fault scenarios. In the initial stage, the neural network controller and the preset PID controller output instructions in parallel. By comparing the differences, for example, if the output torque deviation exceeds the rated value by 10%, it is determined whether switching is necessary. During the transition period, exponential decay weighted fusion is used to gradually reduce the weight of the neural network output. At the same time, Lyapunov stability constraints are injected to prevent sudden changes in speed or position signals. During the stable period, it completely switches to PID control and starts offline retraining of the neural network. This design performs well in the injection molding machine pressure control test. When a sudden change in material fluidity causes control instability, the system quickly completes the disturbance-free switching, and the control accuracy of the output fluctuation amplitude is greatly improved.

[0072] A multi-level protection mechanism enhances system resilience through layered collaboration. Feature drift detection serves as the first line of defense, resolving soft faults; residual analysis focuses on diagnosing hardware anomalies; and hybrid control provides the ultimate safeguard. These three elements work together through an event bus and priority scheduler, forming a closed-loop protection system from anomaly perception to decision execution.

[0073] S500. In response to the feature drift detection results and fault immunity requirements, dynamically optimize the heterogeneous computing resource allocation strategy, perform three-dimensional storage space reconstruction and dynamic precision bit width switching based on real-time task priorities and energy consumption constraints, and form a closed-loop cognitive evolution link.

[0074] In the feature drift detection phase, the system adopts a real-time monitoring mechanism that integrates multiple indicators. By comparing the probability differences between the real-time data stream and the historical benchmark distribution, such as the KL divergence to measure the change in distribution shape, and the statistical shift in the multi-dimensional feature space, such as the Mahalanobis distance to evaluate the overall distribution shift, a dynamic drift assessment model is constructed. When moderate drift is detected, such as when the indicator exceeds 2 times the standard deviation of the historical fluctuation range, the adaptive sampling strategy is automatically triggered - the data collection frequency is increased from the conventional 100Hz to 200Hz, and the incremental learning mode is activated to quickly fine-tune the parameters of the last layer of the model at a high learning rate of 0.01. If the drift continues to intensify, for example, if the indicator does not fall back for three consecutive periods, the emergency response protocol is initiated: a snapshot of the benchmark model stored in the safe zone is loaded, and the latest feature patterns are aggregated from the edge nodes through the federated learning framework to achieve online reconstruction of the model structure.

[0075] Dynamic scheduling of heterogeneous computing resources utilizes a "time-space dual-drive" optimization strategy. In the temporal dimension, a dynamic task priority assessment model is designed, comprehensively considering task urgency, processing deadlines, and resource utilization. For example, fault diagnosis tasks are given the highest priority, tasks closer to the deadline receive higher weights, and non-critical tasks are downgraded when GPU memory utilization exceeds 80%. In the spatial dimension, based on hardware topology awareness technology, the operating status of computing units such as the CPU, GPU, and FPGA, including temperature, load, and energy consumption, is analyzed in real time. Graph neural networks are used to predict the system benefits of different task mapping schemes. When the FPGA temperature is detected to exceed a safe threshold, convolution operations are immediately migrated to an idle GPU, and matrix sparsification compression technology is enabled, significantly reducing power consumption while ensuring computational accuracy.

[0076] like Figure 2 As shown, the reconstructed 3D storage architecture uses pressure sensors to generate a pressure heat map of the pushbutton switches. The reconstruction of the 3D storage architecture focuses on intelligent tiering and resilience design. First, a data heat map analysis module is constructed. By tracking the access frequency, data relevance, and lifecycle of storage cells, high-frequency access parameters, such as the weights of the last layer of a neural network, are migrated to the near-computing core (HBM2e) of the 3D stacked storage, significantly reducing data access latency. Second, an energy-aware storage model is established. While strictly meeting the constraints of 10ns access latency and 5W peak power consumption, dynamic voltage and frequency scaling (DVFS) is used to reduce the L3 cache voltage from 1.2V to 0.9V based on real-time load. Finally, a cross-layer redundancy protection mechanism is deployed, implementing multi-copy erasure coding storage for key parameters. Four parity blocks are added to every eight data blocks, allowing for simultaneous damage of three storage cells while still allowing for complete data recovery. Hash tree parity is also used to ensure data integrity.

[0077] like Figure 3 As shown, the dynamic precision control system utilizes a closed-loop control chain of "perception-decision-execution." The perception layer monitors task characteristics, hardware status, and environmental constraints in real time. For example, image recognition tasks require a high dynamic range, energy-saving mode is activated when the remaining battery power drops below 20%, and frequency reduction is enforced in high-temperature environments. The decision layer builds a reinforcement learning model to transform precision selection into a multi-objective optimization problem: dynamically switching between FP32 full-precision mode and INT8 quantization mode, achieving a Pareto optimal balance between precision loss and energy savings through a deep Q-network. When a hardware anomaly is detected, the execution layer immediately switches to FP32 mode and initiates error recovery, ensuring output reliability through a recalculation mechanism.

[0078] The closed-loop cognitive evolution system achieves continuous optimization through a five-level adaptive mechanism. The perception network uses multiple embedded probes to collect fine-grained chip-level metrics, such as transistor-level leakage current and system-level heat dissipation efficiency, to build a digital twin that maps the physical system state in real time. The analysis engine uses a temporal graph convolutional network (T-GCN) to mine the spatiotemporal correlations of multi-dimensional metrics and predict performance inflection points within the next five seconds, such as the critical moment when the cache hit rate drops to 80%. The decision center uses a mixed integer programming algorithm to generate candidate configurations and select the optimal strategy based on a multi-dimensional utility function, such as prioritizing resource allocation for fault recovery links. The execution unit converts control instructions into micro-ops through the hardware abstraction layer, ensuring extremely fast computational resource reallocation, storage topology reconstruction, and precision mode switching.

[0079] Example 2

[0080] An embodiment of the present invention provides a computer-readable storage medium.

[0081] The computer-readable storage medium provided in the embodiment of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned push button switch adaptive optimization methods can be implemented.

[0082] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0083] For an introduction to the computer-readable storage medium provided in an embodiment of the present invention, please refer to the above method embodiment, and the present invention will not elaborate on it here.

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0085] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0086] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0087] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Any modifications and improvements made to the technical solution of the present invention by a person of ordinary skill in the art without departing from the design concept of the present invention shall fall within the scope of protection of the present invention. The technical content for which protection is sought in the present invention is fully set forth in the claims.

Claims

1. A push button switch adaptive optimization method, characterized in that: include: Acquire a tactile signal of the push button switch, and obtain a spatiotemporal feature vector based on the tactile signal, wherein the spatiotemporal feature vector includes pressure distribution, pressure gradient change rate, and energy concentration; Acquire a historical operation event sequence of the push button switch, and obtain a time series prediction result based on the spatiotemporal feature vector and the operation event sequence, wherein the time series prediction result includes an operation probability distribution and an abnormality mark; Acquiring environmental state parameters of the push button switch, and generating control instructions based on the environmental state parameters, the time series prediction results, and the spatiotemporal feature vectors, for converting the abstract decision of the neural network into executable control instructions; Based on the execution status of the control instructions and the response delay of the physical equipment, a multi-level reliability protection mechanism is constructed to achieve fault immunity through feature drift detection, physical-digital state residual analysis and hybrid control loop switching; In response to the feature drift detection results and fault immunity requirements, the system dynamically optimizes the heterogeneous computing resource allocation strategy, performs three-dimensional storage space reconstruction and dynamic precision bit width switching based on real-time task priorities and energy consumption constraints, and forms a closed-loop cognitive evolution chain. The physical-digital state residual analysis includes: The residual amplitude is obtained by comparing the residual analysis of the sensor data with the predicted value of the digital twin, and the hardware abnormality is determined based on the residual amplitude.

2. The push button switch adaptive optimization method according to claim 1, characterized in that: The method for extracting the spatiotemporal feature vector of the tactile signal includes: Construct a three-dimensional pressure field to collect the pressure distribution matrix, and apply the spatiotemporal convolution kernel to smooth the pressure distribution matrix; A dynamic baseline calibration algorithm is used to eliminate ambient temperature drift. The dynamic baseline algorithm includes a smoothing factor that is adaptively adjusted with the operating frequency.

3. The push button switch adaptive optimization method according to claim 2, characterized in that: The spatiotemporal convolution kernel performs smoothing processing on the pressure distribution matrix, further comprising: Normalizing the pressure gradient change rate to generate a standardized gradient vector; The energy concentration of the pressure distribution matrix is ​​calculated.

4. The push button switch adaptive optimization method according to claim 3, characterized in that: The method for generating the time series prediction result includes: Perform multi-scale wavelet scattering transform on the historical operation event sequence, extract the time-frequency domain feature tensor, and use neural differential equations to model the implicit dynamic process; The probability distribution of future action intentions is predicted through a bidirectional gated recurrent unit, and when the prediction residual exceeds a statistical threshold, it is marked as an abnormal action.

5. The push button switch adaptive optimization method according to claim 1 or 4, characterized in that: The environmental state parameters include temperature, mechanical vibration main frequency, electromagnetic interference intensity and power supply ripple coefficient. The method for fusing the environmental state parameters with the adaptive control instructions is as follows: Build a preset environmental condition template library containing legal state transition paths for multiple typical environmental parameter combinations. Use the attention mechanism to dynamically weight the impact of environmental parameters on the control strategy and generate anti-interference optimization instructions. When the electromagnetic interference intensity exceeds the degradation threshold, the degradation mode is forcibly enabled to limit high-frequency state switching operations.

6. The push button switch adaptive optimization method according to claim 5, characterized in that: The multi-level reliability protection mechanism includes: Online monitoring of feature drift, detecting abnormal inputs through Mahalanobis distance, and triggering an alarm when the Mahalanobis distance value is greater than 5.99; Monitoring the ReLU activation rate of a hidden layer of a neural network, and freezing network parameters when the ReLU activation rate exceeds a freezing threshold; Compare sensor data with the digital twin's predicted value and switch to the backup PID controller when the residual exceeds three times the standard deviation.

7. The push button switch adaptive optimization method according to claim 6, characterized in that: The heterogeneous computing resource allocation strategy includes: Build a heterogeneous computing resource partitioning mechanism to decouple heterogeneous task flows into four parallel execution units based on functional modes: perception signal analysis, decision logic generation, safety constraint verification, and system interaction and coordination; A dynamic numerical representation mode switching strategy is designed to adaptively adjust the data precision topology structure according to the criticality level of the real-time task, realizing dynamic migration between low-entropy computing state and high-fidelity analysis state.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the push button switch adaptive optimization method according to any one of claims 1 to 7 is implemented.

Images