Industrial asset indoor wireless positioning method based on ensemble learning and RNN neural network

Abstract

The invention discloses an industrial asset indoor wireless positioning method based on integrated learning and an RNN neural network, and relates to the technical field of mobile industrial asset target indoor positioning. WIFI positioning and UWB positioning are combined, firstly, a large-range indoor positioning area is divided into unit areas, and the unit areas are divided into unit areas; wIFI positioning adopts an integrated learning algorithm aggregated by various machine learning algorithm models and is used for resolving a specific area where a moving target is located; a UWB positioning RNN model in the area is used to determine the accurate position or moving path of the moving target in the area; and finally, according to the coordinate back propagation error of accurate positioning, updating the weight of various machine learning to correct the model, and gradually optimizing the current model through multiple iterations to achieve the optimal positioning model in the scene. According to the invention, a moving target indoor positioning method with clear grading is established, a reliable target positioning method and process are provided, and the disadvantages of unclear grading, limited adaptability, incapability of full-process tracking and the like of the existing positioning mode are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of indoor positioning of mobile targets, and in particular to an indoor wireless positioning method for industrial assets based on integrated learning and RNN neural network. Background Art

[0002] With the rapid development of my country's manufacturing industry, the processing flow of factory products is gradually becoming diversified. From the procurement and transportation of raw materials, processing and testing, to product assembly, packaging and warehousing, various processes will inevitably go through various transportation processes. In order to avoid economic losses caused by the loss of important industrial production materials during transportation, determining the precise location of industrial assets during transportation has become an urgent problem to be solved.

[0003] The positioning of industrial assets usually has higher requirements than general indoor positioning: including real-time, accuracy, economic benefits and stability: during the transportation of industrial assets, it is necessary to monitor their location data information in real time, which is usually achieved by deploying continuous sensors or data collectors to achieve continuity and real-time data transmission; the learning model trained with historical data is closer to the specific location of the target than the general location calculation rules and has higher accuracy; since the transportation range of industrial assets is generally set in large places such as factories, docks, and workshops, the existing positioning method will lead to high positioning and maintenance costs due to the large number of sensors deployed, resulting in low economic benefits of positioning, and is easily affected by electromagnetic waves, space magnetic fields, and electric fields, resulting in unstable positioning accuracy. Summary of the invention

[0004] In order to solve the above problems, the present invention provides an indoor wireless positioning method for industrial assets based on ensemble learning and RNN neural network, which has the advantages of saving positioning equipment cost, clear positioning classification, improving positioning accuracy through self-iteration, and tracking positioning information throughout the entire process. It proposes an effective solution to the problems of resource waste caused by the deployment of a large number of beacon points for the transportation positioning of important means of production in the current manufacturing industry, algorithm failure caused by a single positioning method or a single positioning algorithm in a complex indoor positioning environment, low positioning accuracy, and large errors.

[0005] To achieve the above object, the present invention provides an indoor wireless positioning method for industrial assets based on ensemble learning and RNN neural network, which specifically includes the following steps:

[0006] Step S1: Data set construction, deployment of indoor AP access points and beacon points, collection of signal strength identifier RSSI fingerprint library to build positioning correspondence;

[0007] Step S2: WIFI positioning prediction: Based on the signal strength value obtained by measuring the AP access point, the real-time position solution is performed through the integrated learning algorithm obtained by integrating the machine learning model to obtain the location area with the maximum probability of the moving target;

[0008] Step S3: UWB positioning prediction, combined with the specific location area determined by the WIFI positioning prediction in step S2, the server retrieves the original data obtained by the UWB beacon points in the location area, calls the RNN model to solve the precise location of the mobile target, and returns the location coordinate value;

[0009] Step S4: Reversely update the aggregation weights, calculate the mean square error of the result points inferred by the integrated learning according to the precisely located coordinate points, and correct the weights of the weighted aggregation of the machine learning model in the WIFI positioning process;

[0010] Step S5: Return to step S2 for iteration, and calculate and output new positioning coordinates according to the position changes of the mobile target between various indoor areas and the weights of the machine learning model updated in step S4, so as to gradually reduce the error.

[0011] Preferably, in step S2, the following steps are specifically included:

[0012] S21: Use the test set data measured in step S1 to test the accuracy of the positioning model in different time periods of the same map under a specific scenario to obtain a set of positioning errors, and set the error threshold to

[0013]

[0014] S22: Initialize the machine learning model and initialize the weighted fusion value The RSSI and location coordinate information are passed to the machine learning model as a positioning reference;

[0015] S23: Use T machine learning models to respectively calculate the real-time RSSI collected by the AP points, and calculate the coordinate point L predicted by the T machine learning models corresponding to a set of signal strength values i (i=1,2,...,T)=(x i ,y i );

[0016] S24: According to the weighted fusion value, the coordinate points of T kinds of predictions are fused into heterogeneous data positioning coordinates to form a certain coordinate. The fusion adopts the enhanced weighted K nearest neighbor algorithm. According to the error threshold Remove the reference points that are larger than the threshold, and merge the remaining coordinate points into the WIFI area positioning coordinates as follows According to the specific regional distribution under WIFI positioning, the positioning coordinates are located in the specific UWB area to complete the preliminary positioning:

[0017]

[0018] Preferably, in step S3, the original data collected by the beacon point is specifically included (D i ,P i ) Train the RNN model to obtain the precise location information of the moving target in the determined area;

[0019] Where D i Indicates the distance D between the beacon point and the moving target i =[d i,1 ,d i,2 ,...,d i,n ], measured by n beacon points deployed in the area, P i Represents the actual coordinate point corresponding to the n-dimensional distance vector.

[0020] Preferably, in step S3, the UWB positioning method includes a method of mapping the local positioning of the mobile target to the global positioning, which is specifically expressed as:

[0021] Mapping the relative coordinate points of the UWB positioning in a specific room to the absolute coordinate points of the entire map includes the following steps:

[0022] Step 1: Use the RNN model to solve the real-time distance vector collected by the beacon point to obtain the point in the internal coordinate system of the area solved by the UWB positioning model (X i ,Y i );

[0023] Step 2: Combine the specific coordinates of the deployed UWB positioning beacon points and the positional relationship between the mobile target and the beacon points to map the relative coordinate points to the overall positioning area to obtain the absolute coordinates of the relative coordinate points in the global positioning area.

[0024] Preferably, in step S4, the weight updating method includes a reverse updating process based on the positioning result, which specifically includes the following steps:

[0025] S41: Based on the T predicted coordinate points determined by the WIFI positioning prediction in step S2 and the absolute coordinates obtained in step S3 The mean square error σ i 2 , update the weight V of the weighted fusion of the machine learning model i size, get V i ' is the updated weight;

[0026]

[0027] S42: Use the updated weight V i 'Calculate the error threshold of all test points under the prediction of the existing ensemble learning model results, recorded as

[0028]

[0029] Preferably, in step S5, the specific implementation process is as follows:

[0030] S51: for the machine learning model after the weights are updated, re-determine the error and calculate the error threshold;

[0031] S52: Solve the real-time RSSI data using the machine learning model with updated weights to obtain coordinate points corresponding to T types of predictions of the machine learning model;

[0032] S53: According to the updated error threshold, the coordinates below the threshold are screened out, and the coordinates below the threshold are subjected to K-nearest neighbor fusion according to the updated weights to obtain fused coordinate points, and enter the UWB positioning stage;

[0033] S54: Continue to execute the UWB positioning prediction process of step S3, and use the precise coordinate values ​​calculated by the RNN algorithm to update the machine learning model weights again, and continue to iterate.

[0034] Therefore, the present invention adopts the above-mentioned industrial asset indoor wireless positioning method based on ensemble learning and RNN neural network, which has the following beneficial effects:

[0035] (1) The present invention avoids the waste of resources caused by deploying a large number of UWB positioning beacon points indoors in real scenarios with high positioning accuracy requirements, and saves indoor positioning costs by combining low-cost WIFI positioning with high-cost UWB positioning.

[0036] (2) The present invention updates the aggregation weights of various machine learning models in WIFI regional positioning through UWB precise positioning information. After multiple iterations, the most suitable regional positioning algorithm models in the positioning scenario are determined and assigned high weights. The predicted values ​​of these models with higher accuracy account for a higher proportion in the final result than other models, making the positioning algorithm adaptive. Through the continuous updating of weights, it can be applied to different positioning scenarios and adapt to different positioning needs.

[0037] (3) The present invention combines regional positioning with precise positioning within the region to gradually narrow the possible range of the mobile target, reduce errors, and improve accuracy. Compared with the traditional indoor positioning method that uses a single positioning method and a single model, it has more accurate prediction results.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a structural flow chart of the WIFI positioning machine learning model of the present invention;

[0040] Figure 2 A diagram showing the area division in an application scenario in an embodiment of the present invention;

[0041] Figure 3 It is the UWB positioning recurrent neural network RNN ​​model structure in the present invention;

[0042] Figure 4 It is a flowchart of the present invention for updating various machine learning weights of WIFI according to the UWB positioning results;

[0043] Figure 5 It is a graph showing the change of positioning error with the number of iterations drawn by the weight updating method of the present invention;

[0044] Figure 6 This is a simulation result diagram of the actual collection of WIFI fingerprint database and UWB beacon point value experiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0046] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0047] The words "include" or "comprises" and the like used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms "inside", "outside", "upper", "lower", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly specified and limited, the terms "attachment" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] like Figure 1 As shown, an indoor wireless positioning method for industrial assets based on ensemble learning and RNN neural network specifically includes the following steps:

[0049] Step S1: Data set construction, deployment of indoor AP access points and beacon points, collection of signal strength identifier RSSI fingerprint library to build positioning correspondence;

[0050] The AP points for WIFI positioning are laid out in the factory, and the signal is required to cover the entire area. UWB beacon points are deployed in the indoor area according to the rooms or channels that need to be positioned, and the overall WIFI positioning area is divided into several UWB positioning sub-areas. Figure 2 As shown, this is the actual floor plan of a factory and the specific division of the positioning method.

[0051] Step S2: WIFI positioning prediction, according to the signal strength value obtained by measuring the AP access point, the real-time position solution is performed through the integrated learning algorithm obtained by integrating the machine learning model to obtain the location area with the maximum probability of the moving target; six machine learning algorithm models are selected, including the nearest neighbor algorithm (NN), K nearest neighbor algorithm (KNN), K weighted nearest neighbor method (WKNN), Bayesian probability algorithm (Prob), Gaussian kernel density algorithm (Gk) and KNN method improved algorithm (Stg):

[0052] The nearest neighbor algorithm (NN) calculates the Euclidean distance L between the measured RSS vector and the RSS vector of each record in the database according to the following formula: i Then select the position coordinates of the recorded point corresponding to the minimum distance as the result output. As shown in the following formula, RSSI i is the RSS vector of the point to be measured, RSSI j is the RSS vector recorded for each reference point in the database, j is the number of reference points:

[0053]

[0054] The main idea of ​​the K-nearest neighbor algorithm (KNN) is to find the K (K>2) RPs (database vectors) with the highest WiFi fingerprint similarity to the TP (positioning measurement vector), and then average the positions corresponding to these K RPs to finally get the position of the TP. As shown in the following formula, is the positioning estimation result, (x i ,y i ) is the coordinate corresponding to the i-th selected fingerprint information:

[0055]

[0056] The difference between the K-weighted nearest neighbor method (WKNN) and the KNN algorithm is that after selecting the nearest K (K>2) reference points, a weighting coefficient is multiplied on the coordinates of each database vector. The weighting method uses the inverse of the Euclidean distance for weighting. As shown in the following formula, is the positioning estimation result, (x i ,y i ) is the coordinate corresponding to the i-th selected fingerprint information:

[0057]

[0058] The main idea of ​​the Bayesian probability algorithm (Prob) is to calculate the posterior probability of TP appearing on each RP through the Bayesian formula, take the k RPs with the largest probability, and then average the positions corresponding to these K RPs to finally get the position of TP. As shown in the following formula, is the positioning estimation result, (x i ,y i ) is the coordinate corresponding to the i-th selected fingerprint information:

[0059]

[0060] The main idea of ​​the Gaussian kernel density algorithm (Gk) is to calculate the likelihood of the relative supersaturation of each fingerprint position through the Gaussian kernel density estimator (GaussianKernelDensityEstimator, GKDE), and determine the position of the TP by averaging the position corresponding to the highest likelihood value. As shown in the following formula, x and y are input samples, ||xy|| 2 represents the square of the Euclidean distance, and δ represents a hyperparameter of the Gaussian kernel function:

[0061]

[0062] The KNN method improved algorithm (Stg) is an improved method of the KNN method. It screens RPs through the k strongest signal access points of TPs and then applies the KNN method.

[0063] In step S2, the following steps are specifically included:

[0064] S21: Using the test set data measured in step S1, based on a specific scenario, the six models are used to solve the RSSI of the test set data respectively, and the predicted point is compared with the real coordinate point, and the median error of each positioning method is recorded. Take the average of the six error medians Get the threshold

[0065]

[0066] S22: Initialize the machine learning model and initialize the weighted fusion value The RSSI and location coordinate information are passed to the machine learning model as a positioning reference;

[0067] S23: Use T machine learning models to respectively calculate the real-time RSSI collected by the AP points, and calculate the coordinate point L predicted by the T machine learning models corresponding to a set of signal strength values i (i=1,2,...,T)=(x i ,y i );

[0068] S24: According to the weighted fusion value, the coordinate points of T kinds of predictions are fused into heterogeneous data positioning coordinates to form a certain coordinate. The fusion adopts the enhanced weighted K nearest neighbor algorithm. According to the error threshold Remove the reference points that are larger than the threshold, and merge the remaining coordinate points into the WIFI area positioning coordinates as follows According to the specific regional distribution under WIFI positioning, the positioning coordinates are located in the specific UWB area to complete the preliminary positioning:

[0069]

[0070] In step S3, the original data collected by the beacon point is specifically included (D i ,P i ) Train the RNN model to obtain the precise location information of the moving target in the determined area;

[0071] Where D i Indicates the distance D between the beacon point and the moving target i =[d i,1 ,d i,2 ,...,d i,n ], measured by n beacon points deployed in the area, P i Represents the actual coordinate point corresponding to the n-dimensional distance vector.

[0072] Step S3: UWB positioning prediction, combined with the specific location area determined by the WIFI positioning prediction in step S2, the server retrieves the original data obtained by the UWB beacon points in the location area, calls the RNN model to solve the precise location of the mobile target, and returns the location coordinate value; in step S3, the UWB positioning method includes mapping the local positioning of the mobile target to the global positioning method, using the pre-measured beacon point collection data and the actual coordinate position to train the positioning model, and record the network structure and node parameters. Its basic structure is as follows Figure 3As shown in the figure, assuming that n UWB beacon points are deployed indoors, the input of the model is an n-dimensional vector, which represents the distance between each beacon point and the moving target. The input data passes through the recurrent neural network layer RNN, the linear layer Linear, the normalization layer BatchNormalization, the activation function layer Relu, and the linear layer Linear to obtain the output result.

[0073] The input of the RNN layer is an n-dimensional vector, which contains 256 hidden layers; the output enters the first linear layer and is converted into 512 outputs with a bias; the 512 feature dimensions are normalized using the normalization layer to satisfy the distribution law of mean 0 and variance 1; the normalized features are then input to the activation function layer for classification, and the results are passed through the second linear layer and output as two-dimensional data, which represents the predicted horizontal and vertical coordinate values.

[0074] Specifically expressed as:

[0075] Mapping the relative coordinate points of the UWB positioning in a specific room to the absolute coordinate points of the entire map includes the following steps:

[0076] Step 1: Use the RNN model to solve the real-time distance vector collected by the beacon point to obtain the point in the internal coordinate system of the area solved by the UWB positioning model (X i ,Y i );

[0077] Step 2: Combine the specific coordinates of the deployed UWB positioning beacon points and the positional relationship between the mobile target and the beacon points to map the relative coordinate points to the overall positioning area to obtain the absolute coordinates of the relative coordinate points in the global positioning area.

[0078] Step S4: Reversely update the aggregation weights, calculate the mean square error of the result points inferred by the integrated learning according to the precisely located coordinate points, and correct the weights of the weighted aggregation of the machine learning model in the WIFI positioning process;

[0079] In step S4, the weight updating method includes a reverse updating process based on the positioning result, which specifically includes the following steps:

[0080] S41: Figure 4 As shown, the Euclidean distance calculation formula is used to calculate the coordinate points of the T predicted points determined by the WIFI positioning prediction in step S2 and the absolute coordinates obtained in step S3 The mean square error σ i 2 , update the weight V of the weighted fusion of the machine learning model i size, get V i' is the updated weight, and the median prediction error is updated using the updated ensemble learning method with weights;

[0081]

[0082] S42: Use the updated weight V i 'Calculate the error threshold of all test points under the prediction of the existing ensemble learning model results, recorded as

[0083]

[0084] Step S5: Return to step S2 for iteration, and calculate and output new positioning coordinates according to the position changes of the mobile target between various indoor areas and the weights of the machine learning model updated in step S4, so as to gradually reduce the error.

[0085] like Figure 5 As shown, in step S5, the specific implementation process is as follows:

[0086] S51: for the machine learning model after the weights are updated, re-determine the error and calculate the error threshold;

[0087] S52: Solve the real-time RSSI data using the machine learning model with updated weights to obtain coordinate points corresponding to T types of predictions of the machine learning model;

[0088] S53: According to the updated error threshold, the coordinates below the threshold are screened out, and the coordinates below the threshold are subjected to K-nearest neighbor fusion according to the updated weights to obtain fused coordinate points, and enter the UWB positioning stage;

[0089] S54: Continue to execute the UWB positioning prediction process of step S3, and use the precise coordinate values ​​calculated by the RNN algorithm to update the machine learning model weights again, and continue to iterate.

[0090] Experimental simulation

[0091] An experimental simulation was conducted, and the experimental data was based on the real collected WIFI fingerprint database and UWB beacon point values. Figure 6As shown in the figure, the horizontal axis represents different training weeks. The prediction data of each week can be regarded as an iteration. The vertical axis represents the difference between the predicted coordinate point and the true coordinate point. The error changes of all six types of algorithms in various machine learning are shown. In the current positioning environment, the solution errors of the three algorithms based on probability Prob, K nearest neighbor improved algorithm Stg, and nearest neighbor algorithm NN are significantly greater than those of the K nearest neighbor algorithm KNN, K weighted nearest neighbor algorithm WKNN, and Gaussian kernel density algorithm GK. After improving the weighted fusion value of the small error algorithm, it can be observed that the "Gathered" curve (fused coordinates) gradually converges to the small error algorithm. Compared with positioning using a single algorithm, the regional positioning method with updated weights reduces the error by 20.12% to 37.58%. Under stable environmental conditions, the error can be kept within 2.5 meters to 3.5 meters, which meets the indoor positioning error standard.

[0092] Therefore, the present invention adopts the above-mentioned indoor wireless positioning method of industrial assets based on integrated learning and RNN neural network, combines WIFI positioning with UWB positioning, first divides the large-scale indoor positioning area into unit areas, and WIFI positioning adopts an integrated learning algorithm aggregated by various machine learning algorithm models to solve the specific area where the mobile target is located; then use the UWB positioning RNN model in the area to determine the precise position or moving path of the mobile target in the area; finally, according to the precise positioning coordinates, backpropagate the error, update the weights of various machine learning to correct the model, and gradually optimize the current model after multiple iterations to achieve the best positioning model in this scenario. It has the advantages of saving positioning equipment cost, clear positioning classification, improving positioning accuracy through self-iteration, and tracking positioning information throughout the process. It proposes an effective solution to the problems of resource waste caused by the deployment of a large number of beacon points for the transportation positioning of important production materials in the current manufacturing industry, algorithm failure caused by a single positioning method or a single positioning algorithm in a complex indoor positioning environment, low positioning accuracy, and large errors.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network, Features: The specific steps include: Step S1: Data set construction, deployment of indoor AP access points and beacon points, collection of signal strength identifier RSSI fingerprint library to build positioning correspondence; Step S2: WIFI positioning prediction: Based on the signal strength value obtained by measuring the AP access point, the real-time position solution is performed through the integrated learning algorithm obtained by integrating the machine learning model to obtain the location area with the maximum probability of the moving target; Step S3: UWB positioning prediction, combined with the specific location area determined by the WIFI positioning prediction in step S2, the server retrieves the original data obtained by the UWB beacon points in the location area, calls the RNN model to solve the precise location of the mobile target, and returns the location coordinate value; Step S4: Reversely update the aggregation weights, calculate the mean square error of the result points inferred by the integrated learning according to the precisely located coordinate points, and correct the weights of the weighted aggregation of the machine learning model in the WIFI positioning process; Step S5: Return to step S2 for iteration, and calculate and output new positioning coordinates according to the position changes of the mobile target between various indoor areas and the weights of the machine learning model updated in step S4, so as to gradually reduce the error.

2. According to the method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network described in claim 1, Features: In step S2, the following steps are specifically included: S21: Use the test set data measured in step S1 to test the accuracy of the positioning model in different time periods of the same map under a specific scenario to obtain a set of positioning errors, and set the error threshold to S22: Initialize the machine learning model and initialize the weighted fusion value The RSSI and location coordinate information are passed to the machine learning model as a positioning reference; S23: Use T machine learning models to respectively calculate the real-time RSSI collected by the AP points, and calculate the coordinate point L predicted by the T machine learning models corresponding to a set of signal strength values i (i=1,2,...,T)=(x i ,y i ); S24: According to the weighted fusion value, the T predicted coordinate points are fused into a certain coordinate using the enhanced weighted K nearest neighbor algorithm. Remove the reference points larger than the threshold, and merge the remaining coordinate points into the WIFI area positioning coordinates as follows According to the specific regional distribution under WIFI positioning, the positioning coordinates are located in the specific UWB area to complete the preliminary positioning:

3. According to the method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network described in claim 1, Features: In step S3, the original data collected by the beacon point is specifically included (D i ,P i ) Train the RNN model to obtain the precise location information of the moving target in the determined area; Where D i Indicates the distance D between the beacon point and the moving target i =[d i,1 ,d i,2 ,...,d i,n ], measured by n beacon points deployed in the area, P i Represents the actual coordinate point corresponding to the n-dimensional distance vector.

4. According to the method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network as described in claim 3, Features: UWB positioning methods include mapping the local positioning of mobile targets to global positioning, which is specifically expressed as: Mapping the relative coordinate points of the UWB positioning in a specific room to the absolute coordinate points of the entire map includes the following steps: Step 1: Use the RNN model to solve the real-time distance vector collected by the beacon point to obtain the point in the regional internal coordinate system (X i ,Y i ); Step 2: Combine the specific coordinates of the deployed UWB positioning beacon points and the positional relationship between the mobile target and the beacon points to map the relative coordinate points to the overall positioning area to obtain the absolute coordinates of the relative coordinate points in the global positioning area.

5. According to the method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network as described in claim 4, Features: In step S4, the weight updating method includes a reverse updating process based on the positioning result, which specifically includes the following steps: S41: Based on the T predicted coordinate points determined by the WIFI positioning prediction in step S2 and the absolute coordinates obtained in step S3 The mean square error σ i 2 , update the weight V of the weighted fusion of the machine learning model i size, get V i ' is the updated weight; S42: Use the updated weight V i 'Calculate the error threshold of all test points under the prediction of the existing ensemble learning model results, recorded as 6. According to the method for indoor wireless positioning of industrial assets based on ensemble learning and RNN neural network as described in claim 5, Features: In step S5, the specific implementation process is as follows: S51: for the machine learning model after the weights are updated, re-determine the error and calculate the error threshold; S52: Solve the real-time RSSI data using the machine learning model with updated weights to obtain coordinate points corresponding to T types of predictions of the machine learning model; S53: According to the updated error threshold, the coordinates below the threshold are screened out, and the coordinates below the threshold are subjected to K-nearest neighbor fusion according to the updated weights to obtain fused coordinate points, and enter the UWB positioning stage; S54: Continue to execute the UWB positioning prediction process of step S3, and use the precise coordinate values ​​calculated by the RNN algorithm to update the machine learning model weights again, and continue to iterate.

Images