Multi-sensor nonlinear error compensation method and device and storage medium
By using a hybrid SSA-DBN-LSTM model for multi-sensor nonlinear error compensation, the problems of low sensor calibration accuracy and insufficient feature extraction are solved, achieving higher calibration accuracy and feature extraction capability, and improving the accuracy of health status prediction.
Patent Information
- Application Number
- CN202510938078.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-10-28
AI Technical Summary
Existing technologies suffer from low calibration accuracy and poor performance in sensor nonlinear error compensation, insufficient feature extraction, and difficulty in optimizing model parameters, resulting in insufficient accuracy and reliability of measurement data.
A hybrid SSA-DBN-LSTM model is used for multi-sensor nonlinear error compensation. Data is preprocessed by moving average filtering, outlier removal and time synchronization. Feature extraction and parameter optimization are performed by combining a deep belief network (DBN) and an optimized long short-term memory network (LSTM). The hyperparameters of the LSTM model are optimized by using the Sparrow Search algorithm (SSA).
It improves the calibration accuracy and feature extraction capability of sensor data, enhances the accuracy of health status prediction, and overcomes the problems of insufficient feature representation capability and low parameter optimization efficiency of traditional methods.
Smart Images
Figure CN120849947A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of measurement technology, and in particular to a multi-sensor nonlinear error compensation method, device, and storage medium. Background Art
[0002] With the development of sensor technology, sensors have been widely developed and applied in various fields, including but not limited to industrial production, logistics and transportation, environmental monitoring, and aerospace.
[0003] The nonlinear compensation problem of sensors refers to the nonlinear relationship between the sensor's output and input, which leads to errors between the sensor's output value and the actual value. Compensation is needed to improve the accuracy and reliability of the measurement data. Sensor output values are typically affected by various factors, such as environmental conditions like temperature, humidity, and pressure, as well as the sensor's own characteristics and operating state. Due to the complexity and nonlinearity of these factors, accurate correction of sensor output data cannot often be achieved simply through linear models.
[0004] To address this problem, extensive research has been conducted both domestically and internationally. Currently, commonly used methods include hardware compensation, software compensation, and hybrid compensation. Hardware compensation uses electronic circuits and components for correction; however, the compensation circuits are often complex, influenced by numerous factors, and have limited accuracy. Furthermore, issues such as the potential for interference between zero-point and full-point adjustments exist. Common software compensation methods include: lookup table method, Newton's iteration method, curve fitting method, neural network method, particle swarm optimization (PSO) algorithm, and support vector machine (SVM) method. Different methods have their own suitable application scenarios. The lookup table method is highly sensitive to segmented intervals, significantly impacting its efficiency. Newton's iteration method is classic and accurate, but its efficiency is low. Curve fitting and linear regression methods are heavily influenced by the parameters of the fitted model, leading to model accuracy issues. The PSO algorithm also suffers from drawbacks such as poor local search capability, susceptibility to local optima, and low search accuracy. With large training samples, machine learning methods like SVM exhibit slow convergence speeds and sensitivity to data. Hardware calibration methods require identifying the causes of errors based on the actual application. They necessitate superior component design, excellent manufacturing, and precise installation to minimize the introduction of error sources. This method is expensive and limited by current limitations. Summary of the Invention
[0005] The purpose of this invention is to provide a multi-sensor nonlinear error compensation method, device, and storage medium to solve the problems of low calibration accuracy, poor calibration effect, insufficient feature extraction, and difficulty in optimizing model parameters in multi-sensor nonlinear error compensation.
[0006] To achieve the above objectives, the present invention provides a multi-sensor nonlinear error compensation method, the specific steps of which are as follows:
[0007] Step S1: Acquire data collected by multiple sensors and perform preprocessing;
[0008] Step S2: Input the preprocessed data into the trained and optimized SSA-DBN-LSTM hybrid model. The SSA-DBN-LSTM hybrid model includes a DBN model (deep belief network model) and an optimized LSTM model (long short-term memory network model) connected in sequence. The optimized LSTM model is the model obtained by optimizing the parameters of the initial LSTM model using the SSA algorithm (sparrow search algorithm).
[0009] Step S3: The SSA-DBN-LSTM hybrid model outputs multi-sensor data after error compensation.
[0010] Preferably, in step S1, the preprocessing process includes a moving average filtering method, outlier removal, missing value processing, and time synchronization.
[0011] The preferred method, the moving average filtering method, has the following specific steps:
[0012] Obtain the dataset: Denote the monitored dataset as h. i i = 1, 2, ..., N, where N is the number of monitoring points;
[0013] The denoising result is obtained by summing and averaging the data: The denoising result is obtained by summing and averaging the data of each monitoring point within the sliding window. The calculation formula is as follows:
[0014]
[0015] Where, x i The initial denoising result obtained by accumulation, n L and n R Let represent the number of points contained to the left and right of point i, respectively. For window n L +n R The sum of the datasets of all points within +1, h i+n This represents the data for the (i+n)th monitoring point, where n is the index of the point within the window.
[0016] Preferably, outlier removal uses the data density method. When a point in the monitoring point group deviates from the local density by a set value, the point is considered an outlier; otherwise, it is considered a normal value. The local density is the absolute value of the difference between the point data and the average value of the monitoring point group data, and the set value is k times the standard deviation of the monitoring point group data.
[0017] The preferred optimization training process for the SSA-DBN-LSTM hybrid model is as follows:
[0018] Step 1: Acquire known sensor data and preprocess the known sensor data to obtain sample data. Divide the sample data into training set and test set according to the set data.
[0019] Step 2: Extract features from the training set using the DBN model to obtain data feature values;
[0020] Step 3: Input the data feature values into the LSTM model to obtain the compensated sensor data, and optimize the parameters of the LSTM model using the SSA algorithm to obtain the parameter-optimized LSTM model;
[0021] Step 4: Determine whether the iteration has ended based on whether the LSTM model output has reached the set requirements or the number of iterations. If the iteration has not ended, return to Step 2. After the iteration ends, you will get the trained DBN model and the trained LSTM model with optimized parameters based on the SSA algorithm.
[0022] Preferably, the DBN model comprises several stacked layers of restricted Boltzmann machines, and is trained and optimized through a combination of unsupervised forward training and backward fine-tuning;
[0023] The specific steps are as follows:
[0024] Initialize the DBN model and set the number of hidden layers, the number of neurons in each layer, and the maximum number of iterations;
[0025] Unsupervised pre-training and supervised adjustment training are performed sequentially. The supervised adjustment training uses gradient descent to adjust the parameters of the DBN model in reverse.
[0026] Preferably, in step three, the hyperparameters of the LSTM model are optimized using the SSA algorithm. The hyperparameters of the LSTM model include the learning rate, the number of hidden layer nodes, the number of iterations, and the dropout rate.
[0027] 1) Initialize the sparrow population: Use the hyperparameters of the LSTM model as individual sparrows in the SSA algorithm, and use the accuracy of the LSTM model output as the fitness function of individual sparrows;
[0028] 2) Calculate its fitness value and update the population;
[0029] 3) Calculate the fitness value of each sparrow individual based on the fitness function, and sort the population accordingly;
[0030] 4) Repeat steps 2) and 3) until the iteration stops, and obtain the LSTM model with optimized parameters.
[0031] An apparatus for performing the above-described multi-sensor nonlinear error compensation method includes:
[0032] The data acquisition module includes several different types of sensors for real-time data acquisition.
[0033] The data processing module preprocesses the collected real-time data;
[0034] The compensation module contains an SSA-DBN-LSTM hybrid model, which performs nonlinear error compensation on the preprocessed data to obtain the compensated sensor data.
[0035] A storage medium storing a computer program, which, when executed by a processor, implements the various steps of a multi-sensor nonlinear error compensation method.
[0036] Therefore, the present invention employs the aforementioned multi-sensor nonlinear error compensation method, device, and storage medium, which has the following beneficial effects: It utilizes a hybrid SSA-DBN-LSTM model. The DBN model extracts features from the preprocessed data, effectively extracting deep features from high-dimensional data and overcoming the problem of insufficient feature representation capability in traditional methods. The combination of the DBN model and the parameter-optimized LSTM model leverages the advantages of both deep feature extraction and temporal modeling, improving the accuracy of health status prediction. The SSA algorithm is used to optimize the parameters of the LSTM model, resulting in high parameter optimization efficiency.
[0037] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0038] Figure 1 This is a flowchart of a multi-sensor nonlinear error compensation method according to the present invention;
[0039] Figure 2 This is a flowchart of the DBN model training process of the present invention. Detailed Implementation
[0040] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0041] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0042] like Figure 1 As shown, a multi-sensor nonlinear error compensation method includes the following specific steps:
[0043] Step S1: Acquire data from multiple sensors and perform preprocessing. The multiple sensors include sensors for temperature, heat flow, strain, and humidity, etc. The type and number of sensors are set according to actual needs.
[0044] The preprocessing process includes moving average filtering, outlier removal, missing value handling, and time synchronization.
[0045] The moving average filtering method has the following specific steps:
[0046] Obtain the dataset: Denote the monitored dataset as h. i i = 1, 2, ..., N, where N is the number of monitoring points;
[0047] The denoising result is obtained by summing and averaging the data: The denoising result is obtained by summing and averaging the data of each monitoring point within the sliding window. The calculation formula is as follows:
[0048]
[0049] Where, x i The initial denoising result obtained by accumulation, n L and n R Let represent the number of points contained to the left and right of point i, respectively. For window n L +n R The sum of the datasets of all points within +1, h i+nThis represents the data for the (i+n)th monitoring point, where n is the index of the point within the window.
[0050] Outlier removal uses the data density method. When a point in the monitoring point group deviates from the local density by a set value, the point is considered an outlier; otherwise, it is considered normal. The local density is the absolute value of the difference between the point data and the average value of the monitoring point group data. The set value is three times the standard deviation of the monitoring point group data.
[0051] Missing value handling and time synchronization will be handled using existing methods, which will not be described in detail here.
[0052] Step S2: Input the preprocessed data into the trained and optimized SSA-DBN-LSTM hybrid model. The SSA-DBN-LSTM hybrid model includes a DBN model and an optimized LSTM model connected in series. The optimized LSTM model is the model obtained by optimizing the parameters of the initial LSTM model using the SSA algorithm.
[0053] The optimization training process of the SSA-DBN-LSTM hybrid model is as follows:
[0054] Step 1: Acquire known sensor data and preprocess it to obtain sample data. Divide the sample data into training set and test set according to the set data.
[0055] Step 2: Extract features from the training set using the DBN model to obtain data feature values.
[0056] Step 3: Input the data feature values into the LSTM model to obtain the compensated sensor data, and optimize the parameters of the LSTM model using the SSA algorithm to obtain the parameter-optimized LSTM model.
[0057] Step 4: Determine whether the iteration has ended based on whether the LSTM model output has reached the set requirements or the number of iterations. If the iteration has not ended, return to Step 2. After the iteration ends, you will get the trained DBN model and the trained LSTM model with optimized parameters based on the SSA algorithm.
[0058] The hyperparameters of the LSTM model are optimized using the SSA algorithm. The hyperparameters of the LSTM model include the learning rate, the number of hidden layer nodes, the number of iterations, and the dropout rate.
[0059] 1) Initialize the sparrow population: Use the hyperparameters of the LSTM model as individual sparrows in the SSA algorithm, and use the accuracy of the LSTM model output as the fitness function of individual sparrows;
[0060] 2) Calculate its fitness value and update the population;
[0061] 3) Calculate the fitness value of each sparrow individual based on the fitness function, and sort the population accordingly;
[0062] 4) Repeat steps 2) and 3) until the iteration stops (the maximum number of iterations is reached or the difference between the model output value and the true value is less than the set value), and obtain the LSTM model with optimized parameters.
[0063] The attribute values corresponding to the maximum fitness function are the optimized hyperparameters of the LSTM model.
[0064]
[0065] Where F represents the fitness value, n 诊断标签＝真实标签 This indicates that the number of diagnostic labels is the same as the number of actual labels.
[0066] Based on the optimized hyperparameters, an LSTM model is built. The model has 20 nodes in the first hidden layer, 14 nodes in the second hidden layer, a learning rate of 0.002, and 50 training iterations.
[0067] The DBN model consists of several stacked layers of restricted Boltzmann machines, and is trained and optimized by a combination of unsupervised forward training and backward fine-tuning.
[0068] The specific steps are as follows:
[0069] Initialize the DBN model and set the number of hidden layers, the number of neurons in each layer, and the maximum number of iterations;
[0070] Unsupervised pre-training and supervised adjustment training are performed sequentially. The supervised adjustment training uses gradient descent to adjust the parameters of the DBN model in reverse.
[0071] like Figure 2 As shown, the specific steps are as follows:
[0072] Initialize the DBN model and set the number of hidden layers, the number of neurons in each layer, and the maximum number of iterations;
[0073] Unsupervised pre-training and supervised tuning training are performed sequentially. Supervised tuning training uses gradient descent to adjust the parameters of the DBN model.
[0074] The model was then validated using a test set to evaluate its performance.
[0075] Step S3: The SSA-DBN-LSTM hybrid model outputs multi-sensor data after error compensation.
[0076] An apparatus for performing the above-described multi-sensor nonlinear error compensation method includes:
[0077] The data acquisition module includes several different types of sensors for real-time data acquisition.
[0078] The data processing module preprocesses the collected real-time data;
[0079] The compensation module contains an SSA-DBN-LSTM hybrid model, which performs nonlinear error compensation on the preprocessed data to obtain the compensated sensor data.
[0080] A storage medium storing a computer program, which, when executed by a processor, implements the various steps of the aforementioned multi-sensor nonlinear error compensation method.
[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A multi-sensor nonlinear error compensation method, characterized in that, The specific steps are as follows: Step S1: Acquire data collected by multiple sensors and perform preprocessing; Step S2: Input the preprocessed data into the trained and optimized SSA-DBN-LSTM hybrid model. The SSA-DBN-LSTM hybrid model includes a DBN model and an optimized LSTM model connected in series. The optimized LSTM model is the model obtained by optimizing the parameters of the initial LSTM model using the SSA algorithm. Step S3: The SSA-DBN-LSTM hybrid model outputs multi-sensor data after error compensation.
2. The multi-sensor nonlinear error compensation method according to claim 1, characterized in that: In step S1, the preprocessing process includes moving average filtering, outlier removal, missing value handling, and time synchronization.
3. The multi-sensor nonlinear error compensation method according to claim 2, characterized in that: The moving average filtering method has the following specific steps: Obtain the dataset: Denote the monitored dataset as h. i i = 1, 2, ..., N, where N is the number of monitoring points; The denoising result is obtained by summing and averaging the data: The denoising result is obtained by summing and averaging the data of each monitoring point within the sliding window. The calculation formula is as follows: Among them, x i The initial denoising result obtained by accumulation, n L and n R Let represent the number of points contained to the left and right of point i, respectively. For window n L +n R The sum of the datasets of all points within +1, h i+n This represents the data for the (i+n)th monitoring point, where n is the index of the point within the window.
4. The multi-sensor nonlinear error compensation method according to claim 3, characterized in that: Outlier removal uses the data density method. When a point in the monitoring point group deviates from the local density by a set value, the point is considered an outlier; otherwise, it is considered normal. The local density is the absolute value of the difference between the point data and the average value of the monitoring point group data. The set value is k times the standard deviation of the monitoring point group data.
5. The multi-sensor nonlinear error compensation method according to claim 4, characterized in that: The optimization training process of the SSA-DBN-LSTM hybrid model is as follows: Step 1: Acquire known sensor data and preprocess the known sensor data to obtain sample data. Divide the sample data into training set and test set according to the set data. Step 2: Extract features from the training set using the DBN model to obtain data feature values; Step 3: Input the data feature values into the LSTM model to obtain the compensated sensor data, and optimize the parameters of the LSTM model using the SSA algorithm to obtain the parameter-optimized LSTM model; Step 4: Determine whether the iteration has ended based on whether the LSTM model output has reached the set requirements or the number of iterations. If the iteration has not ended, return to Step 2. After the iteration ends, you will get the trained DBN model and the trained LSTM model with optimized parameters based on the SSA algorithm.
6. The multi-sensor nonlinear error compensation method according to claim 5, characterized in that: The DBN model consists of several stacked layers of restricted Boltzmann machines, and is trained and optimized by a combination of unsupervised forward training and backward fine-tuning. The specific steps are as follows: Initialize the DBN model and set the number of hidden layers, the number of neurons in each layer, and the maximum number of iterations; Unsupervised pre-training and supervised adjustment training are performed sequentially. The supervised adjustment training uses gradient descent to adjust the parameters of the DBN model in reverse.
7. The multi-sensor nonlinear error compensation method according to claim 6, characterized in that: In step three, the hyperparameters of the LSTM model are optimized using the SSA algorithm. The hyperparameters of the LSTM model include the learning rate, the number of hidden layer nodes, the number of iterations, and the dropout rate. 1) Initialize the sparrow population: Use the hyperparameters of the LSTM model as individual sparrows in the SSA algorithm, and use the accuracy of the LSTM model output as the fitness function of individual sparrows. 2) Calculate its fitness value and update the population; 3) Calculate the fitness value of each sparrow individual based on the fitness function, and sort the population accordingly; 4) Repeat steps 2) and 3) until the iteration stops, and obtain the LSTM model with optimized parameters.
8. An apparatus for implementing the multi-sensor nonlinear error compensation method according to any one of claims 1-7, characterized in that, include: The data acquisition module includes several different types of sensors for real-time data acquisition. The data processing module preprocesses the collected real-time data; The compensation module contains an SSA-DBN-LSTM hybrid model, which performs nonlinear error compensation on the preprocessed data to obtain the compensated sensor data.
9. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the various steps of the multi-sensor nonlinear error compensation method according to any one of claims 1-5.