Multi-case capacitor decision factor processing method based on big data and mathematical statistics

Abstract

The present invention relates to the field of integrated circuit data processing and discloses a multi-case capacitor decision factor processing method based on big data and mathematical statistics. The method comprises: collecting capacitor parameter and environmental variable data to generate a joint data set grouped by environmental parameter combinations; based on the environmental parameter combinations and groups, removing noise and repairing missing values ​​to generate a multi-dimensional feature matrix for environmental grouping; performing principal component analysis on the multi-dimensional feature matrix to construct a regression model to screen key factors; defining the environmental parameter combinations as state spaces and dynamically adjusting weight coefficients; loading the weight mapping table into a simulation platform to simulate capacitor characteristics under different environmental parameter combinations, and deploying the weight mapping table into a control module after verification.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated circuit software development data processing technology, and in particular to a multi-case capacitance decision factor processing method based on big data and mathematical statistics. Background Art

[0002] In the field of integrated circuit software development, the dynamic optimization of capacitor parameters directly affects the stability of circuit performance and environmental adaptability. In the existing technology, the extraction and optimization methods of capacitor decision factors under multiple environmental conditions usually rely on fixed thresholds or single environmental models, which makes it difficult to effectively quantify the complex nonlinear correlation between different environmental variables and capacitor characteristics. Especially when environmental parameters change dynamically, the existing methods lack the ability to mine the statistical characteristics of multi-dimensional capacitor data, resulting in rigid distribution of decision factor weights and insufficient model adaptability. For example, in scenarios with temperature fluctuations or unstable voltage, the existing methods cannot accurately identify the key influencing factors of capacitor parameters, resulting in reduced operating efficiency or abnormal response of integrated circuits. The above problems are caused by the insufficient ability of the data processing link to integrate multi-source heterogeneous capacitor features, and the lack of generalization ability of statistical models for dynamic environmental variables. It is urgent to build a unified cross-environment optimization framework through data-driven methods. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a multi-situation capacitance decision factor processing method based on big data and mathematical statistics. The present invention solves the problems of the capacitance decision factor extraction method based on fixed thresholds or single environmental models in the existing technology, which is difficult to quantify the capacitance characteristics and complex nonlinear correlation when multiple environmental variables change dynamically, resulting in rigid factor weight distribution and insufficient model environmental adaptability.

[0004] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0005] The present invention provides a multi-case capacitance decision factor processing method based on big data and mathematical statistics, including:

[0006] Step 1: Collecting a capacitance parameter data set through a distributed sensor network, synchronously correlating it with environmental variable data, formatting the data in a unified manner, and generating a capacitance and environment joint data set grouped by environmental parameter combinations. Collecting a capacitance parameter data set in real time through a distributed sensor network, synchronously correlating it with dynamically changing environmental variable data, formatting the capacitance parameter data set and the environmental variable data in a unified manner, and generating a time-series correlated capacitance and environment joint data set.

[0007] Step 2: Group the data set based on the combination of environmental parameters. The data set is grouped into groups according to temperature range, voltage level, and frequency range. Noise data is removed based on the outlier detection algorithm, and missing values ​​are repaired through intra-group interpolation and filtering techniques to generate a multi-dimensional feature matrix for environmental grouping.

[0008] Step 3: Perform principal component analysis on the multidimensional feature matrix, calculate the eigenvectors through the covariance matrix, and extract the characteristic principal components whose cumulative variance contribution rate is higher than the preset threshold; construct a multivariate regression model based on the characteristic principal components, establish a nonlinear mapping relationship between capacitance parameters and environmental variables, screen key decision factors and generate an initial factor weight distribution table.

[0009] Step 4: Define the environmental parameter combination as a discrete state space, use the reinforcement learning algorithm to dynamically adjust the weight coefficients in the initial factor weight distribution table, and generate an environmental state and factor weight mapping table.

[0010] Step 5: Load the environmental state and factor weight mapping table into the integrated circuit simulation platform to simulate the capacitor response characteristics under different environmental parameter combinations. After verifying the model stability by comparing the test data, deploy the optimized weight parameters into the control module.

[0011] Furthermore, in the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 1 includes:

[0012] Collecting capacitance parameters and dynamically changing environmental variable data through a distributed sensor network, adding a unified timestamp to the capacitance parameters and the environmental variable data, and achieving time axis alignment across source data;

[0013] Converting the analog signal in the time-axis aligned capacitance parameter data into a digital signal, and performing standardization processing on the digital signal to generate a structured data table including capacitance value, environmental parameter and timestamp fields;

[0014] The structured data table is stored as a data grid of the capacitance and environment joint data set.

[0015] Furthermore, in the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 2 includes:

[0016] The generated capacitance and environment joint data set is segmented according to a preset time window, and the window size of the time window is dynamically adjusted according to the frequency of change of the environmental parameters;

[0017] Noise detection is performed on the segmented data set. Noise data is removed based on the outlier detection algorithm. Missing values ​​are repaired through linear interpolation and smoothing filtering techniques of adjacent windows to generate a continuous time series data stream.

[0018] The continuous time series data stream is multi-dimensionally aggregated according to the capacitance type and the environmental parameter category to form a multi-dimensional feature matrix with a row and column structure.

[0019] Furthermore, in the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 3 includes:

[0020] Calculate the covariance matrix of the multidimensional feature matrix to generate eigenvectors and eigenvalues;

[0021] Based on the eigenvectors and eigenvalues, extracting principal components whose cumulative variance contribution rate exceeds a preset threshold;

[0022] Input the principal components, capacitance parameter data, and environmental variable data into a multiple regression model to calculate the influence weight of each principal component on capacitance performance;

[0023] The top N key decision factors are screened in descending order according to the impact weights to generate an initial factor weight distribution table.

[0024] Furthermore, in the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 4 includes:

[0025] Dividing the environmental parameters collected in real time into a state space according to a preset discretization interval, and associating the state space with the weight coefficients in the initial factor weight distribution table;

[0026] Based on the Q-learning strategy, the weight coefficients in the initial factor weight distribution table are dynamically adjusted according to the state space corresponding to the current environmental parameters;

[0027] The deviation value between the capacitance parameter data collected in real time and the prediction result of the multivariate regression model is calculated through sliding window statistics, and the weight parameters in the environmental state and factor weight mapping table are updated according to the deviation value.

[0028] Furthermore, in the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 5 includes:

[0029] Loading the environmental state and factor weight mapping table into the integrated circuit simulation platform;

[0030] Generate a random environmental fluctuation sequence based on the Monte Carlo method to simulate the extreme scenarios corresponding to the dynamically changing environmental variable data and output a capacitance characteristic curve;

[0031] Comparing the error rate of the capacitance characteristic curve with the actual test data to evaluate the stability of the environmental state and factor weight mapping table;

[0032] The dynamically adjusted weight parameters are deployed to the integrated circuit software control module through a lightweight API interface to complete the configuration of the dynamically adjusted weight parameters.

[0033] Furthermore, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention further includes:

[0034] The system uses a version control mechanism to manage the environment status and factor weight mapping table, retaining historical weight configurations and supporting incremental updates;

[0035] After the optimized weight parameters are deployed to the integrated circuit software control module, when a new capacitor parameter data set is subsequently collected, the system triggers a process including data cleaning, modeling, and dynamic adjustment to generate an updated environmental status and factor weight mapping table, and performs version control management on the updated mapping table to complete the continuous optimization deployment of the weight parameters.

[0036] Beneficial effects of the present invention:

[0037] The technical solution provided by the present invention constructs a multidimensional feature matrix through an environmental parameter combination and grouping mechanism, and combines principal component analysis to screen key decision factors, thus breaking through the limitations of the fixed threshold model. A reinforcement learning algorithm is used to dynamically adjust factor weight coefficients in a discrete state space to achieve real-time matching of environmental variable fluctuations and weight distribution. An extreme scenario verification mechanism based on Monte Carlo simulation is used to improve the generalization capability of capacitor characteristic prediction in dynamic environments. A closed-loop optimization system is formed through version control and incremental updates, allowing the weight mapping table to continuously adapt to changes in environmental parameters. This effectively solves the problem of factor weight rigidity caused by a single environmental model in the background technology, and improves the prediction accuracy and control stability of the capacitor characteristics of integrated circuits in multivariable coupling scenarios of temperature, voltage, and frequency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0039] Figure 1 This is a flowchart of a multi-case capacitance decision factor processing method based on big data and mathematical statistics provided by an embodiment of the present invention.

[0040] Figure 2 This is a high-precision stability verification diagram provided by an embodiment of the present invention.

[0041] Figure 3 A graph showing the capacitance-voltage characteristic prediction accuracy provided by an embodiment of the present invention.

[0042] Figure 4 A comparison chart of model error rates under different environmental stresses provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0044] See also Figures 1 to 4 The present invention provides a multi-case capacitance decision factor processing method based on big data and mathematical statistics, including:

[0045] Step 1: Collecting a capacitance parameter data set through a distributed sensor network, synchronously correlating it with environmental variable data, formatting the data in a unified manner, and generating a capacitance and environment joint data set grouped by environmental parameter combinations. Collecting a capacitance parameter data set in real time through a distributed sensor network, synchronously correlating it with dynamically changing environmental variable data, formatting the capacitance parameter data set and the environmental variable data in a unified manner, and generating a time-series correlated capacitance and environment joint data set.

[0046] Step 2: Group the data set based on the combination of environmental parameters. The data set is grouped into groups according to temperature range, voltage level, and frequency range. Noise data is removed based on the outlier detection algorithm, and missing values ​​are repaired through intra-group interpolation and filtering techniques to generate a multi-dimensional feature matrix for environmental grouping.

[0047] Step 3: Perform principal component analysis on the multidimensional feature matrix, calculate the eigenvectors through the covariance matrix, and extract the characteristic principal components whose cumulative variance contribution rate is higher than the preset threshold; construct a multivariate regression model based on the characteristic principal components, establish a nonlinear mapping relationship between capacitance parameters and environmental variables, screen key decision factors and generate an initial factor weight distribution table.

[0048] Step 4: Define the environmental parameter combination as a discrete state space, use the reinforcement learning algorithm to dynamically adjust the weight coefficients in the initial factor weight distribution table, and generate an environmental state and factor weight mapping table.

[0049] Step 5: Load the environmental state and factor weight mapping table into the integrated circuit simulation platform to simulate the capacitor response characteristics under different environmental parameter combinations. After verifying the model stability by comparing the test data, deploy the optimized weight parameters into the control module.

[0050] Step 1: The distributed sensor network collects capacitor parameter datasets in real time, including physical quantities such as capacitance value and dielectric constant, and synchronously associates dynamically changing environmental variable data, including temperature, voltage, and frequency parameters. A unified timestamp is added to the multi-source data to solve the problem of sampling frequency differences across devices and achieve time axis alignment. The capacitor analog signal is converted into a digital signal through the analog-to-digital conversion module, and all digital signals are standardized to eliminate dimensional differences. The output is a structured data table containing timestamps, capacitance values, and environmental parameters, which are grouped by environmental parameter combinations and stored as a joint capacitor and environmental dataset, where the grouping is based on the combination of temperature range, voltage level, and frequency range.

[0051] Step 2: Based on the combined environmental parameter grouping results, an event-triggered mechanism is used to dynamically segment the joint dataset: the time window is reset when the change in an environmental parameter exceeds a preset threshold. Outlier noise data is identified and removed using the isolation forest algorithm. Missing values ​​are filled using Lagrangian interpolation of adjacent data windows. This data is then smoothed using a Butterworth low-pass filter to generate a continuous time series data stream. Multidimensional aggregation operations are performed based on capacitor type and environmental parameter category, forming a multidimensional feature matrix for environmental grouping, with time series as rows and environmental features as columns.

[0052] Step 3: Calculate the covariance matrix of the multidimensional feature matrix and obtain the eigenvectors and eigenvalues ​​through eigendecomposition. Select the principal component vector whose cumulative variance contribution rate exceeds the preset threshold. The principal component represents the interaction effect between environmental variables. Input the principal component vector, capacitance parameters, and environmental variables into the multivariate regression model to establish a nonlinear mapping relationship between the capacitance value and the principal component. Filter the top N key decision factors based on the absolute value of the regression coefficient and generate an initial factor weight distribution table containing the factor name and weight coefficient.

[0053] Step 4: Discretize the environmental parameter combinations into a finite state space, with each state corresponding to a specific temperature, voltage, and frequency range combination. A state-action value function is constructed based on the Q-learning algorithm, using the mean absolute error between the measured capacitance value and the regression model prediction within a sliding window as the reward signal. A greedy strategy is used to select weight adjustment actions, dynamically updating the coefficients in the initial factor weight distribution table to generate a dynamic mapping table between environmental states and factor weights.

[0054] Step 5: Load the environmental state and factor weight mapping table into the integrated circuit simulation platform. Using the Monte Carlo method, a random environmental fluctuation sequence is generated to simulate extreme scenarios. Simulation curves of capacitance characteristics for different environmental parameter combinations are output, and the mean relative error between the simulation results and the actual test data is calculated. When the error rate falls below the preset tolerance threshold, the optimized weight parameters are written to the control module's configurable registers via a lightweight API, completing parameter deployment.

[0055] The system uses a version control mechanism to manage historical versions of the environment state-weight mapping table and supports incremental updates. After the weight parameters are deployed, the newly collected capacitance parameter dataset triggers the data cleaning, modeling, and dynamic adjustment process. The updated mapping table is compared with the historical version and, after verification through regression testing, deployed to the control module via wireless update, forming a closed loop for continuous iterative optimization of the weight parameters.

[0056] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 1 includes:

[0057] Collecting capacitance parameters and dynamically changing environmental variable data through a distributed sensor network, adding a unified timestamp to the capacitance parameters and the environmental variable data, and achieving time axis alignment across source data;

[0058] Converting the analog signal in the time-axis aligned capacitance parameter data into a digital signal, and performing standardization processing on the digital signal to generate a structured data table including capacitance value, environmental parameter and timestamp fields;

[0059] The structured data table is stored as a data grid of the capacitance and environment joint data set.

[0060] A distributed sensor network synchronously collects capacitor parameter data and environmental variable data. Capacitor parameter data includes physical quantities such as capacitance and dielectric constant, while environmental variable data includes temperature, voltage, and frequency parameters. To eliminate sampling timing deviations among multiple source devices, a unified timestamp based on the NTP protocol is added to both types of data, achieving timeline alignment with millisecond-level accuracy. After timeline alignment, the capacitor analog signal is converted into a digital signal via the ADC conversion module, while the environmental variable data undergoes direct digital normalization, employing a unified Z-score method to eliminate dimensional differences. The normalized data is reorganized by field type into a structured data table containing five core fields: timestamp, capacitance value, temperature, voltage, and frequency. Finally, the structured data table is stored in a distributed database as key-value pairs, forming a data grid for the joint capacitor and environmental dataset. The data grid is indexed and partitioned based on the three-dimensional combination of temperature range, voltage level, and frequency range.

[0061] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 2 includes:

[0062] The generated capacitance and environment joint data set is segmented according to a preset time window, and the window size of the time window is dynamically adjusted according to the frequency of change of the environmental parameters;

[0063] Noise detection is performed on the segmented data set. Noise data is removed based on the outlier detection algorithm. Missing values ​​are repaired through linear interpolation and smoothing filtering techniques of adjacent windows to generate a continuous time series data stream.

[0064] The continuous time series data stream is multi-dimensionally aggregated according to the capacitance type and the environmental parameter category to form a multi-dimensional feature matrix with a row and column structure.

[0065] The combined capacitor and environmental dataset uses a dynamic segmentation mechanism triggered by environmental parameter change events. A new time window is generated when the change in temperature, voltage, or frequency exceeds a preset threshold. The window size is adaptively adjusted based on the frequency of environmental parameter fluctuations: in high-frequency fluctuations, the window is reduced to capture transient features, while in low-frequency fluctuations, the window is expanded to enhance data continuity. An isolation forest algorithm is used to detect outliers in the segmented dataset, identifying and removing noisy data points exceeding three standard deviations. Missing values ​​are filled using Lagrangian interpolation of adjacent window data. Time series smoothing is then performed using a Butterworth low-pass filter, resulting in a continuous and uninterrupted time series data stream. This data stream is aggregated and reorganized based on capacitor material type and the three environmental parameter dimensions of temperature, voltage, and frequency, forming a multidimensional feature matrix with timestamps as row indices and combinations of capacitor type and environmental parameters as column features.

[0066] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 3 includes:

[0067] Calculate the covariance matrix of the multidimensional feature matrix to generate eigenvectors and eigenvalues;

[0068] Based on the eigenvectors and eigenvalues, extracting principal components whose cumulative variance contribution rate exceeds a preset threshold;

[0069] Input the principal components, capacitance parameter data, and environmental variable data into a multiple regression model to calculate the influence weight of each principal component on capacitance performance;

[0070] The top N key decision factors are screened in descending order according to the impact weights to generate an initial factor weight distribution table.

[0071] The covariance matrix of the multidimensional feature matrix is ​​calculated, and the Jacobi iteration method is used to solve the eigenvectors and eigenvalues. Based on the eigenvalue size sorting, the top K principal component vectors whose cumulative variance contribution rate exceeds the preset threshold are selected. The principal component represents the interaction pattern between environmental variables. The principal component vector, the original capacitor parameter data, and the environmental variable data are input into the multivariate regression model together, and the least squares method is used to fit the mapping relationship between the principal component and the capacitor performance index. The standardized regression coefficient of each principal component is calculated in the regression model as its influence weight value on the capacitor performance. Sort in descending order according to the influence weight value, and select the top N principal components with the highest weight value corresponding to the environmental variables as the key decision factors. Output a set of key-value pairs containing factor names and weight coefficients to generate an initial factor weight distribution table.

[0072] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 4 includes:

[0073] Dividing the environmental parameters collected in real time into a state space according to a preset discretization interval, and associating the state space with the weight coefficients in the initial factor weight distribution table;

[0074] Based on the Q-learning strategy, the weight coefficients in the initial factor weight distribution table are dynamically adjusted according to the state space corresponding to the current environmental parameters;

[0075] The deviation value between the capacitance parameter data collected in real time and the prediction result of the multivariate regression model is calculated through sliding window statistics, and the weight parameters in the environmental state and factor weight mapping table are updated according to the deviation value.

[0076] Real-time environmental parameters collected are divided into a state space according to a preset discretization rule. Temperature is discretized into Celsius intervals, voltage into volt levels, and frequency into Hertz bands. Each discrete unit is defined as an independent state. A mapping relationship is established between each state in the state space and the weight coefficient in the initial factor weight distribution table, forming an initial state-weight association table. A state-action value function is constructed based on the Q-learning algorithm, where the action space is defined as operations that increase, decrease, or maintain the weight coefficient. Based on the discrete state corresponding to the current environmental parameter, an ε-greedy strategy is used to select weight adjustment actions and output a set of dynamically adjusted weight coefficients. Using a sliding statistical mechanism within a fixed time window, the mean absolute deviation between the measured capacitance parameter values ​​and the predicted values ​​of the multivariate regression model within the window is calculated. This deviation is used as the immediate reward signal for Q-learning to update the weight parameter values ​​in the environmental state-factor weight mapping table.

[0077] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention, step 5 includes:

[0078] Loading the environmental state and factor weight mapping table into the integrated circuit simulation platform;

[0079] Generate a random environmental fluctuation sequence based on the Monte Carlo method to simulate the extreme scenarios corresponding to the dynamically changing environmental variable data and output a capacitance characteristic curve;

[0080] Comparing the error rate of the capacitance characteristic curve with the actual test data to evaluate the stability of the environmental state and factor weight mapping table;

[0081] The dynamically adjusted weight parameters are deployed to the integrated circuit software control module through a lightweight API interface to complete the configuration of the dynamically adjusted weight parameters.

[0082] The environmental state and factor weight mapping table is loaded into the memory workspace of the SPICE integrated circuit simulation platform to establish a library of mapping relationships between environmental parameters and capacitor response characteristics. A Monte Carlo random sampling method is used to generate an environmental fluctuation sequence, where temperature fluctuations are simulated using a random walk model, voltage fluctuations using a Poisson jump model, and frequency fluctuations using a Gaussian distribution model. This generates a simulation input sequence that includes extreme scenarios. The simulation platform performs circuit netlist calculations and outputs capacitance-frequency characteristic curves for different environmental parameter combinations. The mean relative error between this characteristic curve and the actual test data set is calculated, and the stability of the mapping table is evaluated based on whether the error rate is below a preset tolerance threshold. A lightweight RESTful API is used to call the control module's configuration registers, writing the dynamically adjusted weight parameters in JSON format to the non-volatile memory of the FPGA programmable logic unit, completing the configuration update of the dynamically adjusted weight parameters.

[0083] Specifically, the multi-case capacitance decision factor processing method based on big data and mathematical statistics of the present invention also includes:

[0084] The system uses a version control mechanism to manage the environment status and factor weight mapping table, retaining historical weight configurations and supporting incremental updates;

[0085] After the optimized weight parameters are deployed to the integrated circuit software control module, when a new capacitor parameter data set is subsequently collected, the system triggers a process including data cleaning, modeling, and dynamic adjustment to generate an updated environmental status and factor weight mapping table, and performs version control management on the updated mapping table to complete the continuous optimization deployment of the weight parameters.

[0086] The system uses a version control system based on branch management to manage the environmental status and factor weight mapping table. Each time the weight parameters are updated, an independent version snapshot is generated to retain a complete traceable record of the historical weight configuration. It supports an incremental update mechanism, and uses a difference comparison algorithm to identify the area where the weight coefficients of the new and old mapping tables have changed, and only performs version storage operations on the changed parts. After the optimized weight parameters are deployed to the integrated circuit software control module, the system starts the data acquisition event listener. When the subsequent distributed sensor network collects a new capacitor parameter data set, it automatically triggers the data cleaning, feature matrix construction, principal component analysis and reinforcement learning dynamic adjustment process. The updated mapping table output by the new process is merged with the current effective version. After the regression test verifies the environmental adaptability, the incremental update package is deployed to the firmware storage area of ​​the control module via wireless OTA to complete the continuous iterative optimization of the weight parameters.

[0087] The specific embodiment of the present invention collects capacitance parameter data sets and environmental variable data through a distributed sensor network, wherein the environmental variable data includes temperature, voltage, and frequency parameters, and the capacitance parameter data includes capacitance value and dielectric constant physical quantity. A unified timestamp based on the NTP protocol is added to the multi-source data to achieve time axis alignment with millisecond precision. The capacitance analog signal after time axis alignment is processed into a digital signal by the ADC conversion module, and the environmental variable data is standardized using Z-score to eliminate dimensional differences, and a structured data table containing timestamps, capacitance values, and environmental parameters is output. The data table is indexed and partitioned according to the three-dimensional combination of temperature range, voltage level, and frequency range, and stored as a data grid of the joint capacitance and environment data set.

[0088] A dynamic segmentation mechanism is triggered by environmental parameter change events. The time window is reset when the temperature change or voltage fluctuation exceeds a preset threshold. An isolation forest algorithm is used to identify outlier noise outside of three standard deviations. Missing values ​​are filled through Lagrangian interpolation of adjacent windows, and a Butterworth low-pass filter is used to generate a continuous time series data stream. Capacitor material type is classified and aggregated according to environmental parameter dimensions to form a multidimensional feature matrix with rows corresponding to time series and columns corresponding to feature dimensions. The covariance matrix of the feature matrix is ​​calculated, and the Jacobi iteration method is used to extract principal components whose cumulative variance contribution exceeds the threshold. These components are then input into a multivariate regression model to calculate standardized regression coefficients as influence weights. Key decision factors are then screened in descending order of weight to generate an initial weight distribution table.

[0089] Real-time environmental parameters are discretized into a state space based on Celsius ranges, volt levels, and Hertz frequency bands. Each state is associated with an initial weight distribution table. A state-action value function is constructed based on the Q-learning algorithm, defining the action space as operations that increase or decrease weight coefficients. An ε-greedy strategy is used to select actions. The mean absolute deviation between the measured capacitance values ​​and the model's predicted values ​​is calculated over a fixed time window as a reward signal, dynamically updating the mapping table between environmental states and factor weights. This mapping table is loaded into a SPICE simulation platform, where a series of extreme temperature fluctuations is generated based on a Monte Carlo random walk model, outputting a capacitance-frequency characteristic curve. The mean relative error between the curve and the actual test data is calculated. When the error rate falls below a preset tolerance threshold, the weight parameters are written to the non-volatile memory of the FPGA programmable logic unit (FPGA) via a RESTful API in JSON format.

[0090] The system uses a branch management mechanism to version the mapping table, generating independent snapshots for each update. Newly collected capacitor parameter datasets trigger the entire data cleaning and dynamic adjustment process. After outputting the updated mapping table, a difference comparison algorithm is executed to identify changed areas. After regression testing verifies environmental adaptability, incremental update packages are deployed to the control module firmware storage area via over-the-air (OTA) updates, forming a continuous closed-loop optimization mechanism for weight parameters, addressing the environmental adaptability deficiencies of static models in prior art.

[0091] In the technical solution of the present invention, the environmental parameter combination grouping mechanism divides the data set through a three-dimensional combination of temperature range, voltage level, and frequency range, providing a structured input basis for subsequent noise removal and feature extraction. The principal component analysis algorithm performs covariance calculation and eigenvalue decomposition on the multidimensional feature matrix, extracts the principal component vector whose cumulative variance contribution rate exceeds the threshold, eliminates multicollinearity interference between environmental variables, and screens out the key influencing factors that characterize capacitor performance. The multivariate regression model establishes a nonlinear mapping between the principal component vector and the original capacitor parameters and environmental variables, calculates the standardized regression coefficient through least squares fitting, and generates an initial factor weight distribution table.

[0092] The reinforcement learning algorithm utilizes a discretized state space design, dividing real-time environmental parameters into discrete state units according to preset intervals. Each state is associated with an initial weight distribution table. The Q-learning strategy constructs a state-action value function, defining the action space as a function of increasing, decreasing, or maintaining the weight coefficient. Actions are selected based on an ε-greedy strategy. A sliding window calculates the mean absolute deviation between the measured capacitance and the model's predicted value as an immediate reward signal, driving dynamic updates of the weight parameters and ensuring real-time matching of environmental state changes with weight distribution.

[0093] Monte Carlo simulation uses a random walk model to generate temperature fluctuation sequences, a Poisson jump model to simulate voltage fluctuations, and a Gaussian distribution model to construct frequency perturbations, covering scenarios with extreme variations in environmental parameters. The error rate between the simulated capacitance characteristic curve and actual test data provides a quantitative basis for model stability. The version control mechanism utilizes branch management and difference comparison algorithms, storing only incremental updates for changed areas of the mapping table. This system, combined with regression testing and OTA wireless deployment, forms a closed-loop optimization system. These algorithms work together to achieve a transition from static weight allocation to environmentally adaptive decision-making, addressing the environmental generalization limitations of fixed threshold models.

[0094] Existing methods for extracting capacitance decision factors rely on fixed thresholds or single environmental models, and their defects stem from two aspects: first, the fixed threshold cannot adapt to the dynamic coupling effects of multiple environmental parameters such as temperature, voltage, and frequency, resulting in inaccurate quantification of capacitance characteristics and nonlinear correlations; second, the static weight allocation mechanism ignores the state space changes of environmental parameter combinations. When environmental variables fluctuate continuously, the factor weights cannot respond to changes in real time, resulting in insufficient environmental adaptability of the model.

[0095] The present invention solves the above problems through the state space discretization of environmental parameter combinations and the dynamic weight adjustment mechanism of reinforcement learning. First, the environmental parameter combination is defined as a discrete state space, and the initial mapping relationship between the environmental state and the factor weight is established; secondly, a state-action value function is constructed based on the Q-learning strategy, and the deviation between the actual capacitance value and the model prediction value is used as the reward signal to dynamically adjust the coefficients in the initial factor weight distribution table; finally, a sequence of extreme environmental fluctuations is generated through Monte Carlo simulation to verify the stability of the weight mapping table in dynamic scenarios. This mechanism makes the factor weight distribution adaptive to the environmental state.

[0096] The multidimensional feature matrix of the grouped environments provides input for principal component analysis, extracting key decision factors through covariance matrix calculation, thus overcoming the information limitations of a single environmental model. A dynamic weight adjustment process driven by reinforcement learning transforms environmental parameter changes into state transitions, enabling real-time matching of weight coefficients with environmental fluctuations. A version control mechanism supports incremental updates of mapping tables, triggering a closed-loop optimization process based on new data collection, resulting in continuously enhanced environmental generalization capabilities. Finally, dynamic weight parameters are deployed in the integrated circuit software control module to improve the accuracy of capacitor characteristic prediction.

Claims

1. A multi-case capacitance decision factor processing method based on big data and mathematical statistics, characterized in that: include: Step 1: Collecting a capacitance parameter data set through a distributed sensor network, synchronously correlating it with environmental variable data, formatting the data in a unified manner, and generating a capacitance and environment joint data set grouped by environmental parameter combinations. Collecting a capacitance parameter data set in real time through a distributed sensor network, synchronously correlating it with dynamically changing environmental variable data, formatting the capacitance parameter data set and the environmental variable data in a unified manner, and generating a time-series correlated capacitance and environment joint data set. Step 2: Group the dataset based on the combination of environmental parameters. The dataset is grouped by temperature range, voltage level, and frequency range. Noise data is removed using an outlier detection algorithm, and missing values ​​are repaired through intra-group interpolation and filtering techniques to generate a multi-dimensional feature matrix for environmental groupings. Step 3: Perform principal component analysis on the multidimensional feature matrix, calculate the eigenvectors through the covariance matrix, and extract the principal components with cumulative variance contribution rates higher than the preset threshold; A multivariate regression model is constructed based on the characteristic principal components, a nonlinear mapping relationship between capacitance parameters and environmental variables is established, key decision factors are screened, and an initial factor weight distribution table is generated; Step 4: Define the environmental parameter combination as a discrete state space, use the reinforcement learning algorithm to dynamically adjust the weight coefficients in the initial factor weight distribution table, and generate a mapping table between environmental state and factor weight; Step 5: Load the environmental state and factor weight mapping table into the integrated circuit simulation platform to simulate the capacitor response characteristics under different environmental parameter combinations. After verifying the model stability by comparing the test data, deploy the optimized weight parameters into the control module.

2. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 1 is characterized in that: The step 1 comprises: Collecting capacitance parameters and dynamically changing environmental variable data through a distributed sensor network, adding a unified timestamp to the capacitance parameters and the environmental variable data, and achieving time axis alignment across source data; Converting the analog signal in the time-axis aligned capacitance parameter data into a digital signal, and performing standardization processing on the digital signal to generate a structured data table including capacitance value, environmental parameter and timestamp fields; The structured data table is stored as a data grid of the capacitance and environment joint data set.

3. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 2 is characterized in that: The step 2 includes: The generated capacitance and environment joint data set is segmented according to a preset time window, and the window size of the time window is dynamically adjusted according to the frequency of change of the environmental parameters; Noise detection is performed on the segmented data set. Noise data is removed based on the outlier detection algorithm. Missing values ​​are repaired through linear interpolation and smoothing filtering techniques of adjacent windows to generate a continuous time series data stream. The continuous time series data stream is multi-dimensionally aggregated according to the capacitance type and the environmental parameter category to form a multi-dimensional feature matrix with a row and column structure.

4. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 3 is characterized in that: The step 3 comprises: Calculate the covariance matrix of the multidimensional feature matrix to generate eigenvectors and eigenvalues; Based on the eigenvectors and eigenvalues, extracting principal components whose cumulative variance contribution rate exceeds a preset threshold; Input the principal components, capacitance parameter data, and environmental variable data into a multiple regression model to calculate the influence weight of each principal component on capacitance performance; The top N key decision factors are screened in descending order according to the impact weights to generate an initial factor weight distribution table.

5. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 4 is characterized in that: The step 4 comprises: Dividing the environmental parameters collected in real time into a state space according to a preset discretization interval, and associating the state space with the weight coefficients in the initial factor weight distribution table; Based on the Q-learning strategy, the weight coefficients in the initial factor weight distribution table are dynamically adjusted according to the state space corresponding to the current environmental parameters; The deviation value between the capacitance parameter data collected in real time and the prediction result of the multivariate regression model is calculated through sliding window statistics, and the weight parameters in the environmental state and factor weight mapping table are updated according to the deviation value.

6. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 5 is characterized in that: The step 5 comprises: Loading the environmental state and factor weight mapping table into the integrated circuit simulation platform; Generate a random environmental fluctuation sequence based on the Monte Carlo method to simulate the extreme scenarios corresponding to the dynamically changing environmental variable data and output a capacitance characteristic curve; Comparing the error rate of the capacitance characteristic curve with the actual test data to evaluate the stability of the environmental state and factor weight mapping table; The dynamically adjusted weight parameters are deployed to the integrated circuit software control module through a lightweight API interface to complete the configuration of the dynamically adjusted weight parameters.

7. The multi-case capacitance decision factor processing method based on big data and mathematical statistics according to claim 6 is characterized in that: Also includes: The system uses a version control mechanism to manage the environment status and factor weight mapping table, retaining historical weight configurations and supporting incremental updates; After the optimized weight parameters are deployed to the integrated circuit software control module, when a new capacitor parameter data set is subsequently collected, the system triggers a process including data cleaning, modeling, and dynamic adjustment to generate an updated environmental status and factor weight mapping table, and performs version control management on the updated mapping table to complete the continuous optimization deployment of the weight parameters.

Images