Intelligent self-adapting environment perception sensor

Abstract

The application relates to an intelligent self-adaptive environment sensing sensor, which detects and reads preliminary environment parameters of a current sensor; analyzes the working state of the sensor according to the preliminary environment parameters; adjusts the sensor detection parameters according to the working state to obtain corrected environment parameters. The application solves the problem that the prior art cannot adapt to extreme environments, thereby leading to low detection precision. A Bayesian network model of sensor environment parameters and working states is constructed to quantify the relationship between extreme environments and abnormalities, so that accurate analysis of detection abnormalities in extreme environments is realized, and a foundation is laid for correction of detection results. A generalized additive model between the probability of the working state being abnormal and detection deviation is established, and the accuracy of detection deviation correction is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of sensors and relates to an intelligent self-adaptive environment sensing sensor. BACKGROUND

[0002] Sensor technology originates from people's unremitting pursuit of accurate sensing and measuring the surrounding environment. In the early stage, people realized the need for equipment to obtain basic data such as temperature and pressure in simple industrial production and scientific research activities to ensure stable production and accurate experiments. With the progress of science and technology, traditional mechanical or simple electronic means cannot meet the requirements of precision and function. The development of electronic components such as electronic tubes and transistors provides a hardware basis for sensor technology; the rise of integrated circuit technology enables sensors to be miniaturized and intelligentized. In daily life, from automatic temperature-controlled home appliances to automobile safety systems, the demand for sensors in various fields has increased dramatically, prompting sensor technology to continue to innovate to adapt to diverse scenarios.

[0003] However, existing sensors directly contact the measured environment during work, and thus are also affected by the environment. When the sensor is in an extreme environmental condition, it often cannot adapt and leads to low precision. Therefore, the existing technology requires the sensor to be able to sense its working state in real time and automatically adjust the detection strategy and parameters according to environmental changes, so as to maintain high detection precision in various extreme environments and meet the needs of different fields for reliable operation of sensors in complex and harsh environments. SUMMARY

[0004] To solve the above problems in the prior art, the application provides an intelligent self-adaptive environment sensing sensor, which aims to analyze the detection deviation of the sensor in the historical period under extreme environmental conditions to solve the problem of low precision caused by the inability of the prior art to adapt to extreme environmental conditions.

[0005] The object of the application can be achieved by the following technical solutions:

[0006] The application provides an intelligent self-adaptive environment sensing sensor, which comprises an environment detection module, an environment analysis module and a parameter adjustment module, and the environment detection module, the environment analysis module and the parameter adjustment module are communicatively connected, wherein:

[0007] The environment detection module is configured to detect and read the preliminary environmental parameters of the current sensor.

[0008] The environment analysis module is configured to analyze the working state of the sensor according to the preliminary environmental parameters.

[0009] The parameter adjustment module is configured to adjust the detection parameters of the sensor according to the working state to obtain corrected environmental parameters.

[0010] Further, the environment detection module comprises a main detection unit and a plurality of redundant detection units, and the specific work is as follows:

[0011] When the preliminary environment parameter detected does not exceed the set environment bearing threshold, the detection result of the main detection unit is adopted;

[0012] When the preliminary environment parameter detected exceeds the set environment bearing threshold, all the redundant detection units are used for measurement, and the optimal detection result is analyzed through a multi-layer perception machine model.

[0013] Further, the optimal detection result analyzed through the multi-layer perception machine model comprises the following steps:

[0014] S1, determining initial weights: comprehensively considering the hardware characteristics, past detection data accuracy and stability under different environmental conditions of the redundant detection units, determining initial weights for each redundant detection unit;

[0015] S2, feature extraction and data preprocessing: using Fourier transform and wavelet transform methods, extracting key features from the multi-parameter environment detection data obtained by the redundant detection units; normalizing the data to unify the data of different parameter ranges to the same scale;

[0016] S3, constructing a neural network model: constructing a multi-layer perception machine model, so that the number of input layer nodes corresponds to the number of extracted features; adjusting the number of layers and nodes of the hidden layer according to the complexity of the data and the calculation resources, and using a nonlinear activation function to transform the input;

[0017] S4, model training and optimization: taking the data features of each redundant detection unit after preprocessing as input and the known accurate environment parameters as labels, inputting them into the neural network model for training; calculating the error between the model prediction result and the label through the back propagation algorithm, and adjusting the weights of each layer of the network according to the error; using optimization algorithms such as stochastic gradient descent, and constantly iterating until the error converges to a preset value;

[0018] S5, determining the optimal detection result: inputting the detection data features of the current redundant detection unit into the neural network model after training and optimization, and obtaining the result output by the model.

[0019] Further, the analysis of the working state of the sensor according to the preliminary environment parameter comprises the following steps:

[0020] Obtaining historical period environment parameter detection data set samples and corresponding sensor working states, the working states being divided into normal and abnormal;

[0021] Using a k-means clustering method to divide the data set into a plurality of environment levels;

[0022] An abnormal probability analysis model is constructed with the environmental grade as an explanatory variable and the corresponding working state as a response variable.

[0023] The environmental grade of the preliminary environmental parameter is obtained, input into the abnormal probability analysis model, and the probability of the working state of the sensor being abnormal is analyzed.

[0024] Further, the k-means clustering method comprises the following steps:

[0025] K1, determining the number of clusters: according to the characteristics of the data, an empirical rule is used or an elbow method is used to assist in determining the number of clusters k;

[0026] K2, randomly initializing k cluster centers: randomly selecting k data points from the environmental parameter detection data set as initial cluster centers as the starting point of clustering;

[0027] K3, assigning data points to the nearest cluster center: calculating the distance between each data point in the data set and the k cluster centers, and assigning the data points to the corresponding cluster according to the principle of nearest distance, to form a preliminary clustering result;

[0028] K4, updating the cluster center: calculating the mean of all data points in each cluster, and taking it as the new cluster center;

[0029] K5, repeating step K3 and step K4 until convergence: constantly repeating the data point assignment and cluster center updating steps until the cluster center changes less than a threshold value or reaches a maximum iteration number, completing the clustering division of the data set.

[0030] Further, the abnormal probability analysis model is configured as a Bayesian network model, comprising the following steps:

[0031] Defining nodes and variables: defining the nodes of the Bayesian network model, i.e. the environmental grade and the working state of the sensor, and assigning corresponding variables to each node to determine the model factors;

[0032] Determining the network structure: according to the causal relationship between the environmental grade and the working state of the sensor, the network structure is constructed to show the logical relationship between the nodes by directed edges;

[0033] Estimating the conditional probability table: for each node, the probability distribution is estimated according to the historical data, the prior probability of the environmental grade node is determined, and the conditional probability of the working state of the sensor under the given environmental grade is determined;

[0034] Model verification and optimization: verify the model with part of the historical data, compare the predicted and actual working states, and adjust the probability values or network structure according to the error index;

[0035] Input and reasoning: input the current environment level into the optimized model, and use the Bayesian reasoning algorithm combined with the conditional probability table to calculate the probability that the sensor working state is abnormal;

[0036] Further, the adjustment of the sensor detection parameter according to the working state to obtain the corrected environmental parameter comprises the following steps:

[0037] Obtain the probability that the working state is abnormal and the corresponding detection bias data sample in the historical period;

[0038] Establish a relationship conversion model between the probability that the working state is abnormal and the detection bias;

[0039] Input the probability that the current sensor working state is abnormal into the relationship conversion model, and output the detection bias;

[0040] According to the output detection bias, the detection parameter of the sensor is adjusted, so as to obtain the corrected environmental parameter.

[0041] Further, the relationship conversion model is configured as a generalized additive model, comprising the following steps:

[0042] Data preparation: collect abnormal probability and detection bias historical data, clean and remove noise, outliers and missing values, and standardize or normalize;

[0043] Model structure setting: construct the relationship between abnormal probability and detection bias with a generalized additive model, use a spline function as a smoothing function, determine the spline type, node number and position;

[0044] Model fitting: use an iterative algorithm to fit the model, adjust the smoothing function parameters to minimize the prediction error, select a suitable penalty parameter to balance the fitting and complexity, and determine the optimal value through cross-validation;

[0045] Model evaluation: evaluate the model with an independent validation set, and calculate the determination coefficient to judge the prediction and generalization ability of the model.

[0046] Further, the generalized additive model has the following formula:

[0047]

[0048] In the formula, g(u y ) is an identity connection function with additive property; u y is a response variable, i.e. detection bias; s(x i , k i ) is a smoothing function, the smoothing function is configured as a cubic spline function, x i is an explanatory variable, i.e. abnormal probability, k iThe number of nodes of the cubic spline function controls the number of intervals in the numerical range of the explanatory variable, each interval has a curve fitted by a cubic spline function, and n is the number of explanatory variables.

[0049] The beneficial effects of the present application are:

[0050] (1) By detecting and reading the preliminary environmental parameters of the current sensor, analyzing the working state of the sensor according to the preliminary environmental parameters, adjusting the sensor detection parameters according to the working state, and obtaining the corrected environmental parameters. The present application solves the problem that the prior art cannot adapt to extreme environment, resulting in low detection accuracy.

[0051] (2) Constructing a Bayesian network model of sensor environmental parameters and working state to quantify the relationship between extreme environment and abnormality, thereby realizing accurate analysis of detection abnormalities in extreme environment and laying a foundation for correcting detection results.

[0052] (3) Establishing a generalized additive model between the probability of abnormal working state and detection bias to improve the accuracy of detection bias correction. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the accompanying drawings.

[0054] Figure 1 It is a structure diagram of an intelligent self-adaptive environment perception sensor in the present application.

[0055] Figure 2 It is a flowchart of analyzing the working state of the sensor according to the preliminary environmental parameters in an embodiment of the present application.

[0056] Figure 3 It is a flowchart of adjusting the sensor detection parameters according to the working state to obtain the corrected environmental parameters in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific embodiments, structures, features and effects according to the present application will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0058] Please refer to Figure 1 The present application provides an intelligent self-adaptive environment perception sensor, which comprises an environment detection module, an environment analysis module and a parameter adjustment module, and the environment detection module, the environment analysis module and the parameter adjustment module are communicatively connected, wherein:

[0059] The environment detection module is used for detecting and reading the preliminary environmental parameters of the current sensor;

[0060] The environment analysis module is configured to analyze the working state of the sensor according to the preliminary environment parameter.

[0061] The parameter adjustment module is configured to adjust the sensor detection parameter according to the working state to obtain the corrected environment parameter.

[0062] Further, the environment detection module includes a main detection unit and a plurality of redundant detection units, and the specific working process is as follows:

[0063] When the detected preliminary environment parameter does not exceed the set environment bearing threshold, the detection result of the main detection unit is adopted;

[0064] When the detected preliminary environment parameter exceeds the set environment bearing threshold, all the redundant detection units are used for measurement, and the optimal detection result is analyzed through a multi-layer perception machine model.

[0065] In the embodiment, in the intelligent adaptive environment sensing sensor, the environment detection module is a key component for obtaining environment information, which is composed of a main detection unit and a plurality of redundant detection units. This design aims to comprehensively guarantee the accuracy and reliability of the detection result.

[0066] When the environment is relatively stable, that is, the detected preliminary environment parameter does not exceed the set environment bearing threshold, the main detection unit can fully exert its high-efficiency and accurate detection advantage, and the sensor directly adopts the detection result thereof. This is because the main detection unit is carefully calibrated and optimized under normal environment, and can output reliable detection data with extremely high efficiency, so as to ensure that the sensor can rapidly and accurately perceive environment information in most cases.

[0067] However, once the detected preliminary environment parameter exceeds the set environment bearing threshold, it means that the sensor encounters extreme or complex environmental conditions, and the detection accuracy of the main detection unit may be affected. To cope with this challenge, all the redundant detection units are immediately started to work for measurement. These redundant detection units are not simply repeated settings, and they may use different detection principles, technical means or have unique performance characteristics, and collect environment data from multiple dimensions in extreme environment, providing rich and diverse information sources for subsequent analysis.

[0068] On this basis, for the large amount of data obtained after the determination of the redundancy detection unit, a multi-layer perception (MLP) model is used to analyze and obtain the optimal detection result. The MLP model is a powerful artificial neural network model, which includes an input layer, multiple hidden layers and an output layer. The input layer receives detection data from the redundancy detection unit, the hidden layer extracts and processes deep features of the data through a series of nonlinear transformations, and mines potential complex patterns and relationships in the data. In this process, the model automatically learns the association between different environmental parameters and detection results. The output layer outputs an optimal detection result based on the processing result of the hidden layer, which considers all the information of the redundancy detection unit. Through the analysis of the MLP model, the data of multiple redundancy detection units can be effectively integrated, the limitations of a single detection unit in extreme environments can be overcome, and the accuracy and reliability of the detection result in harsh environments can be improved, providing more accurate environmental parameter information for intelligent adaptive environmental perception sensors.

[0069] Further, the analysis of the optimal detection result by the multi-layer perception model comprises the following steps:

[0070] S1, determining initial weights: when analyzing the optimal detection result by using a neural network model, the initial weights of each redundancy detection unit need to be determined first. This process needs to consider the hardware characteristics of the redundancy detection unit, the accuracy of the past detection data, and the stability under different environmental conditions. For example, for a redundancy detection unit that has always maintained high precision detection in high temperature environment, a relatively high initial weight is given; while for the unit with large fluctuation in detection accuracy, a lower initial weight is given. These weights will reflect the contribution of each redundancy detection unit to the final result in the subsequent calculation process.

[0071] S2, feature extraction and data preprocessing: after the redundancy detection unit obtains the detection data, the data needs to be extracted and preprocessed. For environmental detection data, it may involve temperature, humidity, gas concentration and other parameters. Through mathematical methods such as Fourier transform and wavelet transform, key features such as trend and periodicity are extracted from the data. At the same time, the data is normalized to the same scale, such as mapping the temperature range from 0-100℃, the humidity range from 0-100% and other different range data to the interval of 0-1, to eliminate the influence of data dimension difference on model calculation, and ensure that the neural network can process data more effectively.

[0072] S3, build a neural network model: build a neural network model suitable for this task, this embodiment is a multilayer perceptron (MLP). The number of input layer nodes of the model corresponds to the number of extracted features, and the number of layers and nodes of the hidden layer can be adjusted according to the complexity of the data and the computing resources. The hidden layer transforms the input through a nonlinear activation function (such as the ReLU function) to learn the complex patterns in the data. The output layer outputs a result representing the optimal estimate of the environmental parameter according to the task requirements. In the construction process, the hyperparameters of the network should be set reasonably, such as learning rate, iteration times, etc., to ensure that the model can quickly and accurately converge.

[0073] S4, model training and optimization: use the collected preprocessed data to train the neural network model. In the training process, the data features of each redundancy detection unit are taken as input, and the known accurate environmental parameters are taken as labels. Through the back propagation algorithm, the error between the model prediction result and the label is calculated, and the weights of each layer in the network are adjusted according to the error, so that the error gradually decreases. This process is iterated until the error of the model converges to an acceptable range. In the training process, optimization algorithms such as stochastic gradient descent can be used to improve training efficiency and avoid the model falling into a local optimal solution.

[0074] S5, determine the optimal detection result: after training and optimization of the neural network model, input the detection data features of the current redundancy detection unit, and the output result of the model is the optimal detection result after considering all redundancy detection units. This result integrates the information of each redundancy detection unit, and through the learning and processing of the neural network, it can more accurately reflect the true parameters of the current environment. Compared with traditional methods, the neural network model can automatically learn the complex relationships in the data, and when processing multi-source detection data in extreme environments, it shows higher adaptability and accuracy, providing more reliable environmental parameter detection results for intelligent adaptive environmental perception sensors.

[0075] Further, the working state of the preliminary environmental parameter analysis sensor according to the working state of the preliminary environmental parameter analysis sensor includes the following steps:

[0076] Obtain historical environmental parameter detection data set samples and corresponding sensor working states, the working states are divided into normal and abnormal;

[0077] This step lays the data foundation for subsequent analysis. By collecting environmental parameter data detected by the sensor over a period of time, these data cover various environmental conditions such as temperature, humidity, light intensity, etc. At the same time, record the corresponding working state of the sensor at each detection, which is simply divided into "normal" and "abnormal" two kinds. These historical data samples will become the training material of the analysis model, helping us find the potential relationship between environmental parameters and sensor working state.

[0078] Divide the data set into several environmental levels using k-means clustering method;

[0079] K-means clustering is an unsupervised learning algorithm that divides data points into different groups (clusters). Here, the historical environmental parameter detection data set is taken as input, and through the k-means algorithm, the entire data set is divided into several different groups according to the similarity between data points, each group representing an environmental level. For example, environmental data with similar temperature, humidity, etc. Parameters can be classified into one level. In this way, the complex and diverse environmental data is classified and arranged, making subsequent analysis easier on different environmental level levels to find the rules of sensor working state under different environmental conditions.

[0080] Build an abnormal probability analysis model with environmental level as explanatory variable and corresponding working state as response variable;

[0081] After completing the environmental level division, take each environmental level as the explanatory variable (i.e. independent variable), and the corresponding working state (normal or abnormal) of the sensor under this environmental level as the response variable (i.e. dependent variable). Use statistical or machine learning methods to build a model, the purpose of this model is to predict the probability of the sensor being in abnormal working state through the given environmental level. For example, by analyzing the frequency of sensor abnormal working state under different environmental levels, etc. Information, build a model that can reflect the relationship between environmental level and sensor abnormal working probability. This model will become an important tool for judging the working state of the current sensor.

[0082] Get the environmental level of the preliminary environmental parameters, input the abnormal probability analysis model, and analyze the probability of the sensor's working state being abnormal.

[0083] For the preliminary environmental parameters detected by the current sensor, first of all, according to the previously determined k-means clustering rule, the environmental grade to which it belongs is determined. Then, this environmental grade information is input into the abnormal probability analysis model which has been constructed. The model calculates and outputs the probability that the sensor working state is abnormal under the current environmental condition according to the relationship between the environmental grade and the abnormal working probability learned before. Through this probability value, whether the current working state of the sensor is normal can be quantitatively judged, which provides a basis for whether the sensor needs to be adjusted or maintained subsequently.

[0084] Further, the k-means clustering method comprises the following steps:

[0085] K1, determine the number of clusters k: when using the k-means clustering method to process the environmental parameter detection data set, the primary task is to determine the number of clusters k, that is, how many environmental grades the data set is divided into. This decision usually needs to be made in combination with specific business requirements and data characteristics. Some empirical rules can be used, for example, according to the understanding of the environment parameter in the past, the number of typical environmental states that may exist is estimated. Some algorithms can also be used to assist in determination, such as elbow method. The elbow method calculates the sum of distances (i.e. error sum of squares) from different k values of data points to their respective cluster centers. As k increases, the error sum of squares will gradually decrease. When k increases to a certain extent, the decline of the error sum of squares will become very small. At this point, just like the elbow bending point, the k value corresponding to this point is usually the more appropriate number of clusters.

[0086] K2, randomly initialize k cluster centers: after determining the number of clusters k, k data points are randomly selected from the environmental parameter detection data set as initial cluster centers. These cluster centers will be updated in the subsequent iteration process and finally represent the typical characteristics of each environmental grade. The purpose of random initialization is to provide a starting point for the clustering process, but due to random selection, different initial cluster centers may produce different clustering results. However, through subsequent iteration optimization, the influence of this initial selection will gradually decrease.

[0087] K3, assign data points to the nearest cluster center: for each environmental parameter data point in the data set, calculate the distance between it and the k cluster centers. The distance calculation here can use common distance measurement methods such as Euclidean distance. According to the calculated distance, the data point is assigned to the cluster represented by the nearest cluster center. Through this step, each data point is divided into a specific environmental grade, forming a preliminary clustering result.

[0088] K4, Update cluster centers: After the assignment of all data points is completed, the center of each cluster needs to be updated. For each cluster, calculate the mean of all data points in this cluster (if it is multi-dimensional data, then the mean of each dimension), and take this mean as the new cluster center. The purpose of updating the cluster center is to make the cluster center more representative of the characteristics of all data points in the cluster, thereby improving the accuracy of clustering.

[0089] K5, Repeat step K3 and step K4 until convergence: constantly repeat step K3 (assign data points to the nearest cluster center) and step K4 (update cluster center), until the convergence condition is met. The convergence condition can generally be that the cluster center no longer changes significantly, i.e. the distance between the new cluster center and the cluster center of the last iteration is less than a certain pre-set threshold; or the maximum number of iterations is reached. Through multiple iterations, the clustering result will gradually stabilize, and finally obtain the result of dividing the environmental parameter detection data set into k environmental levels, which provides an effective data classification basis for subsequent construction of anomaly probability analysis model.

[0090] Further, the anomaly probability analysis model is configured as a Bayesian network model, comprising the following steps:

[0091] Define nodes and variables: determine the nodes in the Bayesian network, mainly including environmental level nodes and sensor working state nodes. The environmental level nodes take different environmental levels obtained by k-means clustering as their state values, such as low, medium and high environmental levels, etc. The sensor working state nodes take "normal" and "abnormal" as their state values. These nodes represent the key factors involved in the model. At the same time, each node is assigned a corresponding variable for subsequent probability calculation and reasoning.

[0092] Determine network structure: build the Bayesian network structure according to the potential causal relationship between environmental level and sensor working state. Generally speaking, the change of environmental level will affect the working state of the sensor, so the environmental level node is the parent node and the sensor working state node is the child node. A directed edge is established from the environmental level node to the sensor working state node. This structure reflects our preliminary understanding of the logical relationship between the two factors and provides a framework for subsequent probability calculation.

[0093] Estimating Conditional Probability Tables: For each node, the conditional probability table (CPT) needs to be estimated. For the environmental level node, the prior probability distribution is estimated based on the frequency of each environmental level appearing in the historical period environmental parameter detection dataset. That is, the probability of each environmental level appearing. For the sensor working state node, under the condition of different environmental levels, the conditional probability is determined according to the proportion of the sensor working state being "normal" or "abnormal" under the environmental level in the historical data. For example, under the "high" environmental level, the probability of the sensor working state being "abnormal" is 0.3, and the probability of being "normal" is 0.7. These conditional probability tables quantify the dependencies between nodes.

[0094] Model Verification and Optimization: Use part of the historical data as a validation set, take the environmental level as input, calculate the probability of the sensor working state being "abnormal" through the Bayesian network model, and compare it with the actual working state. Calculate the error indicators between the predicted results and the actual results, such as accuracy, recall rate, etc. According to the error indicators, optimize the model. You can adjust the probability values in the conditional probability table, or fine-tune the network structure, such as adding or deleting some edges, to improve the prediction accuracy of the model.

[0095] Input and Inference: Obtain the environmental level of the current preliminary environmental parameter, input it into the optimized Bayesian network model. Use Bayesian inference algorithms such as variable elimination or joint tree algorithm to calculate the probability of the sensor working state being "abnormal" under the environmental level, combined with the determined conditional probability table. In this way, based on historical data and the constructed model, the working state abnormal probability of the current sensor is effectively evaluated.

[0096] Further, the adjusting of the sensor detection parameter according to the working state to obtain the corrected environmental parameter comprises the following steps:

[0097] Collecting the probability of the working state being abnormal and the corresponding detection bias data sample in the historical period: This step aims to collect historical data as the basis for subsequent analysis. By recording the past sensor operation, the probability data of the working state being determined as abnormal is sorted out. At the same time, record the deviation data between the sensor detection result and the true environmental parameter when the working state is abnormal each time. These deviation data reflect the degree of inaccuracy of the sensor detection under abnormal state. Collect a large number of such sample data to fully understand the correlation pattern between the working state abnormal probability and the detection bias.

[0098] Establishing a relationship transformation model between the probability of the working state being abnormal and the detection bias: Based on the data samples obtained in the previous step, a model is constructed using mathematical or statistical methods that can describe the internal relationship between the probability of the working state being abnormal and the detection bias. For example, regression analysis, machine learning algorithms (such as generalized additive models, linear regression, decision tree regression, etc.) can be used to determine this relationship. The process of establishing the model is to mine the rules from a large amount of data and find a function or algorithm that can predict the detection bias based on the abnormal probability. This model will become the key basis for adjusting the sensor detection parameters.

[0099] Inputting the current probability of the sensor working state being abnormal into the relationship transformation model and outputting the detection bias: After knowing the current probability of the sensor working state being abnormal, this probability value is substituted into the established relationship transformation model. The model processes the input abnormal probability according to the previously learned rules and outputs a corresponding detection bias value. This detection bias value is the predicted size of the possible detection error of the current sensor based on historical data and the model.

[0100] Adjusting the detection parameters of the sensor according to the output detection bias to obtain the corrected environmental parameters: After obtaining the detection bias, according to the working principle and structural characteristics of the sensor, the detection parameters of the current sensor are adjusted accordingly using this bias value. For example, if the detection bias indicates that the sensor detection value is too high, the sensitivity of the sensor can be appropriately reduced or the detection result can be appropriately subtracted, etc. Through such adjustment, the detection result of the sensor is closer to the true environmental parameters, thereby obtaining the corrected environmental parameters, improving the accuracy and reliability of the sensor detection.

[0101] Further, the relationship transformation model is configured as a generalized additive model, comprising the following steps:

[0102] Data preparation: Collect the probability of the working state being abnormal and the corresponding detection bias data samples in the historical period, and clean and preprocess these data. The cleaning process includes removing noise, outliers and missing values in the data. For missing values, mean filling, median filling or interpolation filling according to the distribution characteristics of the data can be used. Preprocessing also includes standardizing or normalizing the data, mapping the abnormal probability and detection bias data to a suitable scale range to ensure that the model can better process the data and improve the stability and convergence speed of the model.

[0103] Model structure setting: The generalized additive model (GAM) assumes that the relationship between the response variable (detection bias) and the explanatory variable (probability of working state being abnormal) can be represented as the sum of multiple smooth functions. In this scenario, the model structure is set such that the detection bias is equal to a constant term plus a smooth function of the probability of working state being abnormal. The smooth function is typically approximated using a spline function, which has good fitting ability and can flexibly capture the non-linear relationship between variables. The appropriate spline type, such as natural spline or cubic spline, is selected, and the number and location of spline nodes are determined, which will affect the complexity and fitting effect of the model.

[0104] Model fitting: The generalized additive model is fitted using the prepared data. An iterative algorithm, such as the locally weighted regression scatterplot smoothing method (LOWESS) or penalized likelihood estimation method, is used to continuously adjust the parameters of the smooth function, so that the error between the predicted value of the model and the actual detection bias is minimized. During the fitting process, a suitable penalty parameter needs to be selected to balance the goodness of fit and complexity of the model. If the penalty parameter is too large, the model will be too smooth and may not capture the subtle changes in the data; if the penalty parameter is too small, the model will be too complex and prone to overfitting. The optimal penalty parameter is selected through cross-validation and other methods.

[0105] Model evaluation: The fitted generalized additive model is evaluated using an independent validation data set. The evaluation indicators of the model, such as mean squared error (MSE), mean absolute error (MAE), or coefficient of determination (R 2 ), are calculated. The mean squared error measures the average squared deviation between the predicted value of the model and the actual value, the mean absolute error measures the average absolute deviation, and the coefficient of determination represents the degree of explanation of the data by the model. Through these indicators, the prediction performance and generalization ability of the model are evaluated to determine whether the model can accurately predict the detection bias corresponding to the probability of working state being abnormal.

[0106] Prediction and application: The probability of the current sensor working state being abnormal is input into the evaluated and optimized generalized additive model, and the model calculates and outputs the corresponding detection bias prediction value according to the fitted smooth function. According to this predicted detection bias, the detection parameters of the sensor are adjusted to obtain the corrected environmental parameters. In actual application, the prediction effect of the model is continuously monitored, and the model is updated and optimized according to new data to ensure that the model can adapt to changes in the environment and drifts in sensor performance.

[0107] Further, the generalized additive model has a calculation formula as follows:

[0108] The generalized additive model has an expression as follows:

[0109]

[0110] wherein g(u y ) is an identity connection function with additive property; u y is a response variable, i.e. detection bias; s(x i , k i ) is a smoothing function, which is configured as a cubic spline function, x i is an explanatory variable, i.e. abnormal probability, k i is the number of nodes of the cubic spline function, which controls the number of intervals in the numerical range of the explanatory variable, each interval has a curve fitted by the cubic spline function, and n is the number of explanatory variables (1 in this embodiment).

[0111] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed with the preferred embodiments as above, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution of the present application. Any modification, change, and modification of the above embodiments, which is based on the technical essence of the present application, still belongs to the scope of the technical solution of the present application.

Claims

1. An intelligent self-adapting environment-aware sensor, characterized by: The application relates to an environment parameter adjustment method and device, and belongs to the field of sensor parameter adjustment. The application comprises an environment detection module, an environment analysis module and a parameter adjustment module, which are communicatively connected, wherein: The environment detection module is used for detecting and reading preliminary environment parameters of a current sensor; The environment detection module comprises a main detection unit and a plurality of redundant detection units, and the specific operation is as follows: When the detected preliminary environment parameters do not exceed a set environment bearing threshold, the detection result of the main detection unit is adopted; When the detected preliminary environment parameters exceed the set environment bearing threshold, all the redundant detection units are used for measurement, and the optimal detection result is analyzed through a multilayer perception machine model; The optimal detection result is analyzed through the multilayer perception machine model, and the method comprises the following steps: S1, determining initial weights: the initial weights of each redundant detection unit are determined by comprehensively considering the hardware characteristics, past detection data accuracy and stability under different environment conditions of the redundant detection units; S2, feature extraction and data preprocessing: key features are extracted from the multi-parameter environment detection data obtained by the redundant detection units by using Fourier transform and wavelet transform methods; the data are normalized to unify the data in different parameter ranges to the same scale; S3, constructing a neural network model: a multilayer perception machine model is constructed, the number of input layer nodes is corresponding to the number of extracted features; the number of layers and nodes of the hidden layer are adjusted according to the data complexity and calculation resources, and a nonlinear activation function is used to transform the input; S4, model training and optimization: the data features of each redundant detection unit after preprocessing are input as the input, and the known accurate environment parameters are input as the labels, and then the neural network model is trained; the error between the model prediction result and the label is calculated through a back propagation algorithm, and the weights of each layer of the network are adjusted according to the error; a stochastic gradient descent optimization algorithm is used for iteration until the error converges to a preset value; S5, determining the optimal detection result: the detection data features of the current redundant detection unit are input into the neural network model after training and optimization, and the result output by the model is obtained; The environment analysis module is used for analyzing the working state of the sensor according to the preliminary environment parameters; 2. The intelligent self-adapting environment-aware sensor of claim 1, wherein: The parameter adjustment module is used for adjusting the sensor detection parameters according to the working state to obtain corrected environment parameters. The working state of the sensor is analyzed according to the preliminary environment parameters, and the method comprises the following steps: A historical period environment parameter detection data set sample and the working state of the corresponding sensor are obtained, and the working state is divided into normal and abnormal; A k-means clustering method is used to divide the data set into a plurality of environment levels; An abnormal probability analysis model is constructed by taking the environment level as an explanatory variable and the corresponding working state as a response variable; 3. The intelligent self-adapting environment-aware sensor of claim 2, wherein: The environment level of the preliminary environment parameters is input into the abnormal probability analysis model, and the probability that the working state of the sensor is abnormal is analyzed. The k-means clustering method comprises the following steps: K1, determining the number of clusters: the number of clusters k is determined according to the data characteristics by using an empirical rule or with the aid of an elbow method; K2, randomly initialize k cluster centers: randomly select k data points from the environmental parameter detection dataset as initial cluster centers, as the starting point of clustering; K3, assign data points to the nearest cluster center: calculate the distance between each data point in the dataset and the k cluster centers, and assign the data points to the corresponding cluster according to the principle of nearest distance, forming a preliminary clustering result; K4, update the cluster center: calculate the mean of all data points in each cluster, and take it as the new cluster center; K5, repeat step K3 and step K4 until convergence: constantly repeat the data point assignment and cluster center update steps until the cluster center changes less than a threshold value or reaches a maximum iteration number, completing the clustering division of the dataset.

4. The intelligent self-adapting environment-aware sensor of claim 2, wherein: The anomaly probability analysis model is configured as a Bayesian network model, including the following steps: Define nodes and variables: define the nodes of the Bayesian network model, i.e. the environmental level and the sensor working state, and assign corresponding variables to each node to determine the model factors; Determine the network structure: according to the causal relationship between the environmental level and the sensor working state, construct the network structure to show the logical relationship between nodes with directed edges; Estimate the conditional probability table: for each node, estimate the probability distribution based on historical data, determine the prior probability for the environmental level node, and determine the conditional probability for the sensor working state node under a given environmental level; Model verification and optimization: verify the model with part of the historical data, compare the predicted and actual working states, and adjust the probability values or network structure according to the error index; Input and reasoning: input the current environmental level into the optimized model, use the Bayesian reasoning algorithm combined with the conditional probability table to calculate the probability of the sensor working state being abnormal.

5. The intelligent self-adapting environment-aware sensor of claim 4, wherein: The adjustment of the sensor detection parameters according to the working state to obtain the corrected environmental parameters includes the following steps: Obtain the probability of the working state being abnormal and the corresponding detection bias data sample in the historical period; Establish a relationship conversion model between the probability of the working state being abnormal and the detection bias; Input the probability of the current sensor working state being abnormal into the relationship conversion model to output the detection bias; Adjust the detection parameters of the sensor according to the output detection bias to obtain the corrected environmental parameters.

6. The intelligent self-adapting environment-aware sensor of claim 5, wherein: The relationship conversion model is configured as a generalized additive model, including the following steps: Data preparation: collect abnormal probability and detection bias historical data, clean up noise, outliers and missing values, and standardize or normalize; Model structure setting: build an abnormal probability and detection bias relationship with a generalized additive model, use a spline function as a smoothing function, determine the spline type, node number and position; Model fitting: use an iterative algorithm to fit the model, adjust the smoothing function parameters to minimize the prediction error, select an appropriate penalty parameter to balance the fitting and complexity, and determine the optimal value through cross-validation; Model evaluation: evaluate the model with an independent validation set, calculate the determination coefficient to judge the prediction and generalization ability of the model.

7. The intelligent self-adapting environment-aware sensor of claim 6, wherein: The generalized additive model has the following formula: ， wherein is an identity link function with additive properties; u y is a response variable, i.e. a detection bias; is a smoothing function, configured as a cubic spline function, x i is an explanatory variable, i.e. an anomaly probability, k i is a number of knots of the cubic spline function, controlling a number of intervals in a range of values of the explanatory variable, each interval having a curve fitted by the cubic spline function, n is a number of explanatory variables.

Images