UWB abnormal signal identification method based on genetic algorithm optimization random forest

Abstract

The invention discloses a UWB abnormal signal identification method based on a genetic algorithm optimization random forest, and relates to the technical field of wireless communication and positioning navigation. On the premise of not considering the important characteristic of channel impulse response (CIR), the measurement distance and the received signal strength are used as training characteristics to be input into a random forest (RF) model; according to the UWB signal identification method based on the genetic algorithm, the genetic algorithm (GA) is used for optimizing training parameters, and identification of a UWB line of sight (LOS) signal, a non-line of sight (NLOS) signal and a multi-path (MP) signal can be achieved. The UWB signal identification method based on the genetic algorithm can be used for identifying the UWB line of sight (LOS) signal, the NLOS signal and the multi-path (MP) signal, and can be used for identifying the UWB line of sight (LOS) signal and the multi-path (MP) signal. The problem of abnormal signal identification when low-cost equipment cannot acquire CIR information is solved.

Description

Technical Field

[0001] The invention relates to the technical field of wireless communication and positioning navigation, and is a UWB abnormal signal recognition method based on genetic algorithm optimization of random forest. Background Art

[0002] Since its introduction, ultra-wideband (UWB) positioning technology has been widely used in indoor positioning systems due to its signal stability and low power consumption; it provides good technical support for applications such as smart life interconnection, Internet of Things, industrial automation, medical tracking, and supermarket assistance. The more common UWB positioning method currently uses the signal transmission and reception time difference between the target and the base station to provide accurate distance estimation, and then determines the three-dimensional coordinates of the target through the geometric relationship between multiple base stations. However, in indoor positioning, due to the influence of obstacles such as walls and furniture, the signal propagation between the target and the base station often forms a non-line-of-sight situation with obstruction and a multipath situation with reflection and refraction, resulting in a large distance estimation deviation, which in turn affects the accuracy of target positioning.

[0003] At present, domestic and foreign scholars have done a lot of research on the problem of NLOS signal recognition, and proposed NLOS recognition methods from different angles. The research schemes are mainly divided into two categories: one is based on the distance measurement results, and the NLOS distance measurement value is regarded as an abnormal value. The residual weighting method is used to smooth the influence of abnormal signals on the positioning results to improve the positioning accuracy. The other is based on the characteristics of the wireless channel, and the typical features of the wireless signal are selected for training using machine learning theory to achieve the effect of autonomous classification of NLOS signals. After eliminating the identified abnormal signals, the LOS signal is used to solve the three-dimensional position coordinates of the tag.

[0004] However, most scholars in the research on UWB signal recognition such as the above-mentioned ones are often limited to the binary classification of LOS and NLOS, and often ignore MP signals or identify NLOS and MP signals together. Another noteworthy issue is that in the process of signal recognition using machine learning methods, important features such as channel impulse response and first path signal harmonic amplitude can often be easily obtained, which is obviously difficult for signal recognition of low-cost devices. Summary of the invention

[0005] In view of this, the present invention provides a UWB abnormal signal recognition method based on genetic algorithm optimized random forest.

[0006] This method regards the MP signal as an independent signal type, and uses the estimated distance, total received signal strength and first path received signal strength as the training feature input model, realizing the three-value classification of LOS, NLOS and MP signals, providing a new design idea for the UWB indoor positioning method of low-cost equipment.

[0007] On the one hand, an embodiment of the present application provides a UWB abnormal signal recognition method based on genetic algorithm optimized random forest, comprising:

[0008] a. Preparation of UWB signal dataset

[0009] Furthermore, the step of preparing the UWB signal data set includes data collection, calibration and preprocessing. The collection specifically includes the arrangement of UWB base stations, the arrangement of signal receivers, the placement of obstacles, and the collection of data sets. The data set used in the present invention is a UWB signal collected by a company, with a total of 61,930 groups.

[0010] Furthermore, the calibration of the UWB signal data set is to calibrate the collected data set. Specifically, it includes manually verifying the obstacles between the signal generator and the receiver. The signal collected in the space with obstacles is marked as NLOS signal, the signal collected in the space without obstacles but narrow space and multipath propagation effect cannot be ignored is marked as MP signal, and the signal collected in the space without obstacles is marked as LOS signal. The marked file is saved in csv format for subsequent model training and testing.

[0011] Furthermore, the preprocessing of the UWB signal data set includes balancing the data set and dividing it into a training set and a test set. Balancing means that the initial UWB data set obtained by measurement is unbalanced for the three required categories (LOS, NLOS and MP), so it is necessary to balance the data by a random resampling method so that the number of observations of the three signals NLOS, LOS and MP is consistent. Dividing the training set and the test set means that the model is trained using all the data, and one-third of the total sample volume is extracted from each of the three categories of data by random sampling to form a test data set.

[0012] b. Establish a random forest recognition model optimized by genetic algorithm

[0013] Furthermore, a random forest model for identifying abnormal UWB signals is established, which is an integrated learning classifier based on a decision tree as a base learner. First, a sample with a capacity of N is subjected to N replacement extractions, and 1 sample is extracted each time to form N samples. Secondly, when each node of the decision tree needs to be split, n attributes are randomly selected from the m attributes of each sample. Then, an attribute is selected from the n attributes as the split attribute of the node through an optimization strategy. Repeat the cycle until it cannot be split. Finally, a large number of decision trees are established according to the above method to form a random forest. In an embodiment of the present application, N is a preprocessed resampled data set, m=3, n<3.

[0014] Furthermore, a genetic algorithm optimization model is established. Genetic algorithm is an efficient, parallel, and global search method that can automatically acquire and accumulate knowledge about the search space during the search process, and adaptively control the search process to obtain the optimal solution. First, the encoding method is selected, and then the initial population is determined; then the fitness function is calculated to determine whether it meets the termination condition; then the genetic process including replication, crossover, and mutation is started; finally, the result is output to determine the optimal solution. Among them, the reproduction generation, population size, mating probability, and mutation probability are the four parameters that determine the performance of the genetic algorithm. The reproduction generation, that is, the number of replications, is generally selected in the range of 100-500; if the generation is too small, the population is not mature enough to effectively search for the optimal solution; if the generation is too large, the population is mature and no longer has evolutionary significance, which greatly increases the invalid running time of the model and reduces efficiency. The population size is generally selected in the range of 20-100; if the number is too small, it cannot provide enough sampling points, causing the population to mate with relatives and then reproduce pathological genes, affecting the generation of competitive groups; if the number is too large, it will increase the amount of calculation, and it is not easy to converge, thereby reducing robustness. The mating probability is generally taken as 0.4-0.99 empirically; if the probability is too low, the search will be sluggish and the population cannot be effectively updated; if the probability is too high, the optimal individual structure of the population reproduction will be destroyed. The mutation probability is generally taken as 0.0001-0.1; if the probability is too low, the diversity of the population will decrease, new genes will be difficult to produce, and high-quality genes will be easily lost, eventually leading to premature population; if the probability is too high, it will destroy high-fitness individuals and change the model from optimal search to random search. In the embodiment of the present application, the values ​​of the four parameters are 100, 100, 0.6, and 0.01, respectively.

[0015] Furthermore, a genetic algorithm optimization model is used to search for the optimal values ​​of the two parameters that most affect the recognition performance of the random forest recognition model, namely the number of decision trees and the depth of decision trees, and finally a GA-RF recognition model is formed.

[0016] c. Use UWB signal dataset to train GA-RF model

[0017] Furthermore, a large-scale UWB signal data set is used to train the GA-RF model, so that the model can identify and distinguish three types of signals: LOS, NLOS and MP. The UWB signal data set is a resampled data set obtained after data preprocessing.

[0018] d. Test the performance of the GA-RF model using a randomly sampled UWB signal dataset

[0019] In order to test the application effect of the method in a real scenario, the performance of the GA-RF model is tested using the randomly sampled UWB signal data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. As shown in the accompanying drawings, the above and other purposes, features and advantages of the present invention will be clearer. The same reference numerals in all the drawings indicate the same parts. The drawings are not deliberately scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0021] Figure 1 Schematic diagram of the results of evaluating the recognition performance of the GA-RF method using the confusion matrix.

[0022] Figure 2 The figure is a flowchart of a UWB abnormal signal recognition method based on genetic algorithm optimized random forest. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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.

[0024] like Figure 1 As shown, it is a schematic diagram of the results of using the confusion matrix to evaluate the recognition performance of the GA-RF method. In order to verify the performance of the method described in the embodiment of the present invention, the GA-RF model is tested using a randomly sampled UWB signal data set, and the confusion matrix is ​​used as a performance evaluation indicator. At the same time, the present invention considers that the relevant definitions and interpretations of the confusion matrix are basic industry knowledge and are not specifically explained. Analysis of the confusion matrix shows that the GA-RF model can achieve an accuracy of 92.94% for NLOS signals and 91.35% for MP signals. At the same time, the recall rate for NLOS signals can reach 94.26%, and the recall rate for MP signals can reach 94.27%. Comprehensive analysis of the confusion matrix results shows that the GA-RF model can achieve an accuracy of 92.52% for UWB abnormal signals. It can be considered that the method provided by the present invention can provide good signal recognition technical support for low-cost UWB devices.

[0025] See also Figure 2 , is a flow chart of a method for identifying abnormal UWB signals based on a genetic algorithm to optimize a random forest, which is exemplified in an embodiment of the present application. The method comprises the following steps:

[0026] S1. Preparation of UWB signal dataset

[0027] Specifically include: data collection, calibration and preprocessing. The collection specifically includes the layout of UWB base stations, the layout of signal receivers, the placement of obstacles, and the collection of data sets. The data set used in the present invention is the UWB signal collected by a company, with a total of 61930 groups.

[0028] S2. Establish a genetic algorithm optimized random forest (GA-RF) recognition model

[0029] Specifically, it includes: establishing a random forest model for identifying abnormal UWB signals, which is an integrated learning classifier based on decision trees. Furthermore, a genetic algorithm optimization model is established and used to search for the optimal values ​​of the two parameters that have the greatest impact on the recognition performance of the random forest recognition model, namely the number of decision trees and the depth of the decision trees, and finally forming a GA-RF recognition model.

[0030] S3. Use UWB signal dataset to train GA-RF model

[0031] Specifically, it includes: using a large-scale UWB signal data set to train the GA-RF model so that the model can identify and distinguish three types of signals: LOS, NLOS and MP. The UWB signal data set is a resampled data set obtained after data preprocessing.

[0032] S4. Testing the performance of the GA-RF model using a randomly sampled UWB signal dataset

[0033] Specifically, it includes: using the randomly sampled UWB signal data set to test the performance of the GA-RF model, and the evaluation indicator is the confusion matrix.

[0034] Compared with the prior art, the above method has the following advantages:

[0035] 1. Avoid using channel impulse response as a model training feature, solving the problem that low-cost devices cannot obtain this information.

[0036] 2. The genetic algorithm is used to search for the optimal solution of the parameters required for the random forest model. Compared with the traditional empirical method, the time required is greatly reduced and the acquisition of the optimal solution of the parameters also indicates that the performance of the recognition model can be maximized.

[0037] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0038] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A UWB abnormal signal recognition method based on genetic algorithm optimized random forest, the method comprising: Preparation of UWB signal dataset Establishing Genetic Algorithm Optimized Random Forest (GA-RF) Recognition Model Using UWB signal dataset to train GA-RF model The performance of the GA-RF model is tested using a randomly sampled UWB signal dataset.

2. According to claim 1, a UWB abnormal signal recognition method based on genetic algorithm optimized random forest is characterized in that: Preparation of UWB signal dataset, including: The data set is a data set made for training the GA-RF model. It is sent by the UWB base station and received by the signal collection point. The signal is calibrated into three types: NLOS signal, LOS signal and MP signal according to the signal characteristics. At the same time, three signal parameters containing distance and signal strength information (estimated distance (Dist), total received signal strength (RSS_RX) and first path received signal strength (RSS_FP)) are extracted as training features. Directly input data.

3. The UWB abnormal signal recognition method based on genetic algorithm optimized random forest according to claim 1 is characterized in that: Establish a random forest model to identify abnormal UWB signals, including: This model is an ensemble learning classifier based on decision tree. First, N samples with capacity N are extracted by permutation, and 1 sample is extracted each time to form N samples. Secondly, when each node of the decision tree needs to be split, n attributes much smaller than m are randomly selected from the m attributes of each sample. Then, an attribute is selected from the n attributes as the split attribute of the node through the optimization strategy. The cycle is repeated until it cannot be split. Finally, a large number of decision trees are established according to the above method to form a random forest.

4. The UWB abnormal signal recognition method based on genetic algorithm optimized random forest according to claim 1 is characterized in that: Establish a genetic algorithm to optimize the model of random forest influencing parameters, including: Genetic algorithm is an efficient, parallel, global search method that can automatically acquire and accumulate knowledge about the search space during the search process, and adaptively control the search process to obtain the optimal solution. First, the encoding method is selected, and then the initial population is determined; then the fitness function is calculated to determine whether it meets the termination condition; then the genetic process including replication, crossover, and mutation is started; finally, the results are output to determine the optimal solution. This model uses genetic algorithms to search for the optimal values ​​of the two parameters in the random forest model that have the greatest impact on the model recognition performance, namely the number of decision trees and the depth of decision trees.

5. The UWB abnormal signal recognition method based on genetic algorithm optimized random forest according to claim 1 is characterized in that: Perform data preprocessing on the initial data set, including: In the initial UWB data set, the number of the three types of observations measured is not the same, so the random resampling method is used to preprocess the data to equally balance the three types of observations.

6. The UWB abnormal signal recognition method based on genetic algorithm optimized random forest according to claim 1 is characterized in that: The GA-RF model is trained using the UWB signal dataset, including: A large-scale UWB signal dataset is used to train the GA-RF model, enabling the model to identify and distinguish three types of signals: LOS, NLOS and MP.

7. The UWB abnormal signal recognition method based on genetic algorithm optimized random forest according to claim 1 is characterized in that: The performance of the GA-RF model was tested using a randomly sampled UWB signal dataset, including: In order to test the application effect of the method in a real scenario, the performance of the GA-RF model is tested using the randomly sampled UWB signal data set.

8. The UWB signal data set is a data set obtained after data preprocessing; the randomly sampled UWB signal data set is a test data set formed by randomly sampling the preprocessed data set and extracting one third of the total number of samples from each of the three types of data; the GA-RF model consists of two parts: a random forest recognition model and a genetic algorithm optimization model. The use of the UWB signal data set to train the GA-RF model includes the training of the recognition model and the adjustment of the genetic algorithm parameters; the use of the public UWB signal data set to test the performance of GA-RF is intended to test the application effect of the method in real scenarios.

Images