Self-adaptive positioning method and system for intelligent factory

Abstract

The invention relates to the technical field of image data processing, in particular to a self-adaptive positioning method and system for an intelligent factory, and the method comprises the steps: obtaining a positioning environment image signal of a to-be-detected target through a CCD image shooting device, and carrying out the preprocessing of the image signal of the to-be-detected target, network training is carried out on the processed image signals through deep learning, and image features are extracted; the method comprises the following steps: collecting physical data of a to-be-detected target through various sensors, fusing the physical data, classifying the physical data of the to-be-detected target by using a convolutional neural network, and establishing a training model; performing big data analysis on the image features and the training model, and identifying an actual positioning application scene and a positioning demand of the intelligent factory; and intelligently matching a positioning mode according to an actual positioning application scene and a positioning demand. The positioning mode can be adaptively matched according to the actual application scene and the positioning requirement of the intelligent factory, and the system cost is reduced while the positioning precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular relates to an adaptive positioning method and system for smart factories. Background Art

[0002] With the deep integration of informatization and industrialization, the manufacturing model is further developing towards integration, networking and intelligence. Intelligent manufacturing technology has emerged and attracted much attention. Intelligent manufacturing emphasizes the interconnection between equipment and improves production efficiency, flexibility and resource utilization through data analysis and automatic decision-making. As a carrier for promoting industrial transformation and realizing intelligent manufacturing, the intelligent production process of smart factories cannot be separated from accurate location identification technology. The use of positioning technology can provide accurate location information for personnel, materials, equipment, etc., ensure personnel and production safety, and improve production efficiency, so as to realize the automated production, intelligent management and efficient cost control of smart factories.

[0003] The existing positioning methods are mainly based on non-range measurement and ranging-based positioning methods. For example, CN117499869A discloses a smart factory positioning method for indoor terminals based on the weighted nearest neighbor method. The wireless nodes of the method are distributed in a nine-square grid format and can be deployed according to the size of the specific factory area in the factory. As long as the distance between two adjacent wireless nodes is equidistant, the wireless nodes are distributed and named according to the longitudinal and transverse directions, so as to achieve the functions of adjusting the number and position of signal nodes according to actual usage requirements. However, the positioning accuracy of this method is low and it is not suitable for fine-grained positioning scenarios in smart factories. CN116582818A proposes a non-line-of-sight effect compensation indoor positioning method based on UWB ranging. By analyzing the geometric characteristics of signal propagation, it identifies whether the target node is affected by non-line-of-sight during the positioning process and determines the affected anchor nodes. The non-line-of-sight measurement information is reversely compensated so that the parameter perception after compensation is the same as line-of-sight transmission, so that it can also be approximated to free space propagation in complex indoor environments, achieving more accurate positioning. However, when studying how to reduce the impact of non-line-of-sight paths on positioning accuracy, the positioning-related patent mainly focuses on traditional methods such as mathematical calculations, and the combination with cutting-edge technologies such as artificial intelligence is relatively weak, which limits the positioning accuracy. CN116709202A discloses an indoor 3D fingerprint positioning method and system based on millimeter wave beam pattern trajectory. By establishing beam pattern trajectory information for specific positions of both communicating parties, a directional fingerprint is established, and the known spacing information of two millimeter wave receiving devices is used to calculate the 3D position of a specific sending device, thereby achieving high-precision indoor 3D positioning. Although this method has high positioning accuracy, the deployment cost is also high.

[0004] In summary, the existing positioning methods are relatively simple and cannot accurately match the positioning requirements of smart factories with different granularities. They also lack intelligent processing and analysis of data, which limits the applicability, robustness and economy of positioning methods in practical applications. Summary of the invention

[0005] The present invention provides an adaptive positioning method and system for smart factories, aiming to solve the problem in the prior art that both positioning cost and positioning accuracy cannot be taken into account.

[0006] A first aspect: The present invention provides an adaptive positioning method for a smart factory, the method comprising the following steps: Step S1: acquiring a positioning environment image signal of the target to be measured through a CCD image capture device, preprocessing the image signal of the target to be measured, and performing network training on the processed image signal through deep learning to extract image features; Step S2: collecting physical data of the target to be measured through a variety of sensors, wherein the sensors include a direction sensor, an acceleration sensor, a gyroscope, and an air pressure sensor; Step S3: fusing the physical data of the target to be measured, and using a convolutional neural network to classify the physical data of the target to be measured and establish a training model; Step S4: performing big data analysis on the image features and training models to identify actual positioning application scenarios and positioning requirements of smart factories; Step S5: Adaptively match the positioning method according to the actual positioning application scenario and positioning requirements.

[0007] Furthermore, before executing step S2, the method includes: detecting an image signal by a CCD image pickup device to determine whether there is an obstruction between the target to be detected and the base station, and if so, enhancing the base station signal strength.

[0008] Furthermore, the preprocessing of the image signal of the target to be measured specifically includes: gray-scaling, filtering, noise reduction and image enhancement of the image signal of the target to be measured.

[0009] Furthermore, the specific method for fusing the physical data of the target to be measured is any one of a weighted average method, a Kalman filter method, and a Bayesian estimation method.

[0010] Furthermore, using a convolutional neural network to classify the physical data of the target to be measured and establish a training model includes: evaluating the posture, direction, and moving step of the target to be measured, using CNN to classify the physical data of the target to be measured, generating various sets, and performing network training respectively.

[0011] Furthermore, the adaptive matching of the positioning mode according to the actual positioning application scenario and positioning requirements is specifically: querying the positioning information table according to the actual positioning scenario and positioning requirements to determine the positioning mode, and the positioning mode includes a non-ranging positioning mode and a ranging positioning mode.

[0012] Furthermore, the non-ranging positioning method adopts a PSO-optimized DV-Hop positioning algorithm, and the ranging positioning method is an RSS positioning algorithm based on deep learning optimization.

[0013] Furthermore, the PSO optimization DV-Hop positioning algorithm is specifically as follows: Step S500: Initializing the population according to the size of the wireless network, the size of the target area, and the desired positioning accuracy; Step S501: Calculate the fitness value of each particle according to the fitness function of the particle swarm algorithm PSO, and update the extreme value pBest of each particle according to the fitness value; Step S502: compare the extreme values ​​of all particles, select M extreme values ​​of particles with the smallest fitness values ​​and save them into the set Best, where M>1; Step S503: determine whether the number of iterations is greater than a preset number or the extreme value of each particle in the set Best is less than a certain accuracy threshold. If yes, execute step S504; if not, execute step S501; Step S504: Calculate the average value of all extreme values ​​in the set Best, take the position corresponding to the average value as the optimal position, and use the parameters associated with the optimal position for positioning.

[0014] Furthermore, the RSS positioning algorithm based on deep learning optimization is specifically: Step S510: Offline stage: detecting the basic characteristic information of the base station signal received by each grid node, processing the basic characteristic information into a signal fingerprint, and thereby constructing a signal fingerprint database; Step S511: pre-processing the fingerprint data in the fingerprint database, dividing the training set and the validation set, creating a cross entropy loss function, configuring the Adam optimizer and setting the learning rate parameter, and using the deep learning CNN model to carry out training until the Loss value is the lowest; Step S512: Online stage: collect the current fingerprint data of the target to be tested as a test set, use the CNN model to test, and screen out signal data with LOS characteristics by comparing with the basic feature information; Step S513: Use the sensor node transmission power and reference distance The path loss model and the path loss factor are optimized to compensate the signal data with LOS characteristics, and the optimized signal data is used to perform distance estimation.

[0015] In the second aspect, the present invention provides an adaptive positioning system for a smart factory, comprising a base station, a server, and a target to be measured, wherein the base station is arranged in the smart factory and provides a radio signal source, and the target to be measured is provided with a signal receiver that can receive the radio signal from the base station; When positioning begins, the base station transmits a radio signal, and the CCD image capture device obtains the positioning environment image signal of the target to be measured and transmits the image signal to the server; a variety of sensors collect the physical data of the target to be measured and transmit the physical data to the server, and the sensors include direction sensors, acceleration sensors, gyroscopes, and air pressure sensors; The server is used to preprocess the image signal of the target to be measured, and perform network training on the processed image signal through deep learning to extract image features; fuse the physical data of the target to be measured, and use a convolutional neural network to classify the physical data of the target to be measured and establish a training model; perform big data analysis on the image features and the training model to identify the actual positioning application scenarios and positioning requirements of the smart factory; and adaptively match the positioning method according to the actual positioning application scenarios and positioning requirements.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The CCD image acquisition device of the present invention is responsible for the image evaluation of the intelligent factory positioning environment, pre-processes the identified image signal, and performs neural network training, extracts the image feature signal and hands it over to the big data platform for positioning scene analysis; multiple types of sensors collect the physical data of the target to be measured, and use the convolutional neural network (CNN) to deeply process the physical data, which is also handed over to the big data platform for positioning demand analysis. Thus, adaptive positioning is achieved according to the actual positioning scene and positioning demand. At the same time, the non-range positioning method adopts the PSO optimized DV-Hop positioning algorithm, and the range positioning method adopts the RSS positioning algorithm based on deep learning optimization, which greatly saves the deployment cost, technical overhead, and energy consumption cost required for positioning. That is, the present invention can adaptively match the positioning method for different application scenarios and positioning requirements of the intelligent factory, adopt the ranging-based positioning method for the positioning scene with higher positioning requirements, and adopt the non-range positioning method for the positioning scene with lower positioning requirements, thereby realizing the adaptive matching of the positioning method, while improving the positioning accuracy and reducing the system cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0018] Figure 1 This is a flow chart of the adaptive positioning method for smart factories of the present invention; Figure 2 This is a schematic diagram of the adaptive positioning method for smart factories of the present invention; Figure 3 A network diagram of the adaptive positioning method for smart factories of the present invention; DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the specific embodiments described herein are only used to explain the present invention and are part of the embodiments of the present invention, rather than 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.

[0020] Example 1 In smart factories, the following scenarios have relatively high requirements for positioning: ① Logistics and warehousing: Automated guided vehicles (AGVs) and automatic storage and retrieval systems (AS / RS) require high-precision positioning to accurately transport, store and extract materials. For example, in large stereoscopic warehouses, AS / RS relies on precise positioning to accurately place goods in the corresponding cargo positions. Excessive errors will lead to chaotic storage of goods. For inventory management of raw materials and finished products, precise positioning helps to quickly find target goods and improve the efficiency of warehousing. ② Production equipment management: When maintaining and repairing some large, high-value production equipment, it is necessary to accurately locate faulty parts. For example, in semiconductor production equipment and precision assembly equipment, the internal structure is complex. Only by accurately locating the problem can efficient maintenance be carried out to reduce downtime. Real-time positioning of equipment can optimize equipment layout, facilitate monitoring of collaboration between equipment, and improve production efficiency. ③ Personnel management: In dangerous areas, such as areas with high temperature, high pressure, and toxic and harmful gases, it is necessary to accurately locate staff to ensure their safety. Once a dangerous situation occurs, timely rescue can be carried out. For some large factories across regions, accurate personnel positioning helps to reasonably allocate tasks and supervise work processes, making it easier for managers to quickly find the corresponding personnel. ④ Environmental monitoring: The structure of a smart factory is complex. Once a harmful gas leak occurs, the gas may spread between the complex equipment architecture, and the setting may spread outside the factory area, causing harm to the surrounding environment and residents. Therefore, it is necessary to accurately locate the location of the harmful gas leak to ensure that timely measures can be taken. However, like ordinary office areas: places where employees work daily, such as administrative offices, conference rooms, etc. The work in these places is mainly processing documents, holding meetings, etc., and high-precision positioning is not required.

[0021] According to whether ranging is performed, the existing smart factory positioning methods can be divided into two types: non-ranging-based positioning and ranging-based positioning. Non-ranging-based positioning does not require measuring distance, angle or signal strength. Its advantages are low hardware overhead, simple configuration, and high system scalability. Although the accuracy of non-ranging-based positioning can meet the requirements of general positioning occasions, this method is not suitable for working scenarios with high positioning accuracy requirements. Ranging-based positioning extracts measurement information based on distance, angle, etc. from different characteristics of radio signals. Specific indicators include received signal strength (RSS), arrival time (TOA), signal arrival time difference (TODA) and arrival angle (AOA). The ranging-based positioning method can obtain information such as the received signal strength, interval distance and azimuth of the target, which provides the necessary conditions for accurate positioning, but compared with non-ranging-based positioning, this type of method is more expensive. At the same time, factors such as the complex working conditions of manufacturing plants make it difficult for positioning technology to meet the multiple requirements of high accuracy, high coverage, high reliability, high scalability, low power consumption and cost-effectiveness. In addition, radio waves are blocked by obstacles, reflected, refracted or scattered, forming non-line-of-sight (NLOS) propagation. As the propagation distance increases, the measurement error will also increase accordingly, seriously affecting the positioning accuracy. Various positioning technologies are susceptible to such NLOS environments, but the existing patents related to indoor positioning focus on mathematical modeling and other calculation methods when studying how to reduce the impact of non-distance measurement, and are not sufficiently integrated with cutting-edge technologies such as artificial intelligence deep learning. To be precise, there is a lack of an adaptive positioning method for smart factories to achieve accurate matching of positioning requirements for complex application scenarios in smart factories.

[0022] like Figure 1-Figure 3 As shown, an embodiment of the present invention relates to an adaptive positioning method for a smart factory, the method comprising the following steps: Step S1: acquiring a positioning environment image signal of the target to be measured through a CCD image capture device, preprocessing the image signal of the target to be measured, and performing network training on the processed image signal through deep learning to extract image features; With the help of high-precision CCD image capture devices, the image signals of the complex positioning environment around the target to be measured are accurately acquired. These image signals may be interfered by various factors such as light and noise, so a series of preprocessing operations are required, such as grayscale processing, filtering and denoising, and image enhancement, to improve the clarity and quality of the image. Subsequently, the processed image signals are divided into training sets, validation sets, and test sets, and advanced deep learning algorithms are used to conduct in-depth network training on the processed image signals. By constructing a multi-layer neural network model, it can automatically learn and extract key image features related to positioning in the image, such as the shape, color, texture of the target to be measured, and its relative position in the environment, etc., to provide an accurate visual information basis for subsequent positioning analysis. More specifically, first, the server uses a high-precision CCD image capture device to accurately obtain a wide-view image of the coverage area and a complex positioning environment image around the target to be measured. Then, the server calculates the similarity between the global features of the received wide-view image and the global features of the positioning environment image, and selects the wide-view image with the global features closest to the positioning environment image. It should be noted that the server can not only calculate the similarity of global features, but also calculate the similarity of the data about the object, image description and segmentation of each image. By calculating the similarity between the positioning environment image and each wide-view image, and using a histogram to calculate the weight value of each panoramic image, further by counting the number of similar images between the positioning environment image generated from a specific wide-view image and the positioning environment image received during positioning, a higher weight is given to the wide-view image with a large number of similar images, and further according to the similarity and weight value of each parameter, the ranking of each wide-view image is calculated, and the wide-view image most similar to the positioning environment image is obtained. Then, the server calculates the local features of the selected panoramic image, such as the position coordinates and length information of the key points in the image, and calculates the local features of the positioning environment image. The server calculates the local features of the environment image and provides data support for further positioning.

[0023] Furthermore, in order to reduce the positioning error caused by NLOS, before obtaining the positioning environment image signal of the target to be measured through the CCD camera, it is determined whether there is an obstruction between the target to be detected and the base station, that is, whether non-line-of-sight (NLOS) propagation is formed, resulting in signal attenuation. If so, the base station signal strength will be enhanced to reduce the positioning error caused by the RSSI positioning technology being affected by NLOS. The signal base station is arranged in the smart factory to provide a radio signal source, and the target node to be measured is equipped with a signal receiver that can receive the radio signal from the base station. At the beginning of positioning, the signal base station transmits a radio signal. At the same time, the CCD image capture device obtains the positioning environment image signal of the target to be measured, and then transmits it to the server for storage and processing.

[0024] Step S2: collecting physical data of the target to be measured through a variety of sensors, wherein the sensors include a direction sensor, an acceleration sensor, a gyroscope, and an air pressure sensor; In order to obtain more comprehensive information about the target to be measured, the complementarity between various sensors is fully utilized, and the physical data of the target to be measured is collected through sensors with different functions. Among them, the direction sensor can accurately sense the direction angle of the target, and can accurately capture the direction changes in both plane and three-dimensional space; the acceleration sensor can monitor the acceleration of the target in real time, and accurately measure the changes in the target's motion state, such as acceleration, deceleration, stillness or uniform motion, which is of great value in judging the dynamic behavior characteristics of the target; the gyroscope focuses on detecting the rotation angle and angular velocity of the target, and provides key data support for the rotation of the target on its own axis; the pressure sensor can calculate the approximate height information of the target by measuring the changes in the air pressure value of the surrounding environment, which supplements the important vertical dimension data for positioning in the smart factory environment with a certain height difference. That is, when positioning begins, the signal base station transmits a radio signal. At the same time, various sensors such as the direction sensor, acceleration sensor, gyroscope, and pressure sensor collect the physical data of the target to be measured, and then transmit it to the server for storage and processing.

[0025] Step S3: fusing the physical data of the target to be measured, and using a convolutional neural network to classify the physical data of the target to be measured and establish a training model; After collecting the physical data, the server efficiently fuses the physical data of the target to be measured, integrates and calibrates the data from different sensors to eliminate errors and inconsistencies between the data, and forms a complete and accurate target physical information data set. Next, the posture, direction, moving step length, etc. of the target to be measured are comprehensively evaluated, and the data set is deeply processed using a convolutional neural network (CNN) with powerful feature extraction and classification capabilities. Through learning and training of a large amount of sample data, classification models and prediction models for different physical data patterns are constructed to more accurately identify and analyze the physical state and behavior of the target, and provide reliable physical data basis for positioning decisions. That is, the present invention proposes to give full play to the complementarity between sensors by intelligently integrating data from different types of sensors. Combined with advanced data fusion algorithms, such as neural networks and Kalman filters, the accuracy and reliability of positioning data are improved.

[0026] The specific method for fusing the physical data of the target to be measured is: any one of the weighted average method, the Kalman filter method, and the Bayesian estimation method. For example: if the physical data of the target to be measured is relatively stable, the fluctuation is small, and the sensor measurement error is also relatively stable, the weighted average method can be selected. If the data is dynamically changing and there is a certain system model (such as a linear system), the Kalman filter method is more suitable. When the uncertainty of the data is large and prior knowledge needs to be considered for fusion, the Bayesian estimation method is more advantageous; further, the present invention also includes verification and evaluation of the fused data: using known standard data or reference data obtained by other reliable measurement methods, the fused data is compared and verified, the accuracy and stability of the fusion algorithm are evaluated, and the parameters of the fusion algorithm are adjusted and optimized according to the results.

[0027] In order to obtain more accurate and comprehensive positioning data, the physical data of the target to be measured needs to be proofread before being fused. For example, sensor data acquisition: obtain raw data from each sensor (such as temperature, pressure, humidity, acceleration, etc.) and record the corresponding timestamp to ensure the synchronization and timeliness of the data. Filter the raw data to remove noise and outliers. At the same time, perform data conversion and normalization to map the data of different sensors to the same numerical range for subsequent processing. Use timestamp information to align data from different sensors at the same or similar times to handle data misalignment problems caused by possible differences in acquisition frequency. For sensors with spatial position information (such as multiple cameras or positioning sensors), coordinate transformation, rotation and translation and other operations are used to make their spatial reference systems consistent, ensuring that the measured physical quantities are described in the same spatial coordinate system.

[0028] Step S4: performing big data analysis on the image features and training models to identify actual positioning application scenarios and positioning requirements of smart factories; Based on the image features and training models extracted in the previous steps, big data analysis technology is used for comprehensive and in-depth mining and analysis. Through correlation analysis, cluster analysis, and pattern recognition of massive historical data and real-time data, the positioning application scenarios in different areas and production links in the smart factory can be accurately identified, such as the high-precision positioning requirements of short-distance transportation vehicles in the material handling area, and the real-time positioning monitoring requirements of the equipment positioning area in the production workshop, etc. At the same time, the specific positioning requirements in each scenario can be accurately grasped, including positioning accuracy requirements, positioning real-time requirements, positioning stability requirements, etc., so as to provide strong decision support for selecting the most appropriate positioning method.

[0029] Step S5: Adaptively match the positioning method according to the actual positioning application scenario and positioning requirements.

[0030] Finally, according to the precisely determined positioning application scenarios and detailed positioning requirements, screening and matching are performed from the pre-stored positioning information table (Table 1 is an example of the positioning information table) to determine the most suitable positioning method. For example, in the precision assembly area with extremely high positioning accuracy requirements, a high-precision positioning method based on ranging is selected; while in ordinary office areas with large-area, low-precision positioning requirements, a non-ranging-based positioning method may be used. This ensures intelligent classification of different application scenarios in different smart factories, thereby realizing adaptive matching of positioning methods, improving positioning accuracy while reducing system costs, and meeting the actual needs of automated production and management of smart factories.

[0031] The positioning information table is formulated by various departments (such as production department, logistics department, information management department, etc.) according to the special requirements for positioning in the actual application scenario. At the same time, combined with technical literature and previous project experience, the positioning method suitable for each scenario is selected. In order to ensure accuracy, the positioning information table is updated regularly (such as every quarter or every six months). For example: Since the production process of a smart factory may change with factors such as product upgrades and process improvements, the positioning requirements of the application scenario may also change accordingly. At this time, the positioning requirements in the application scenario should be re-evaluated and the positioning information table should be modified synchronously.

[0032] In order to quickly filter out the most suitable positioning method from the pre-stored positioning information table (Table 1 is an example of the positioning information table), you can use a database query language (such as MySQL) to write an intelligent query statement to traverse the positioning information table. For example, to find a suitable positioning method in the "production workshop scene" with "high" positioning accuracy, "high" positioning real-time performance, and "high" positioning stability, you can write the following query statement: SELECT positioning method FROM positioning information table WHERE application scenario = 'production workshop scene' AND positioning accuracy = 'high' AND positioning real-time performance = 'high' AND positioning stability = 'high'.

[0033] Table 1

[0034] Furthermore, in order to further improve the positioning accuracy and reduce the system cost, the non-ranging positioning method adopts the PSO optimized DV-Hop positioning algorithm, and the ranging positioning method is the RSS positioning algorithm based on deep learning optimization.

[0035] The PSO optimization DV-Hop positioning algorithm is specifically as follows: Step S500: Initializing the population according to the size of the wireless network, the size of the target area, and the desired positioning accuracy; The population is initialized based on the key factors of the size of the wireless sensor network (WSN), the size of the target area, and the desired positioning accuracy. This means that the characteristics of the actual application scenario should be comprehensively considered, and the number of particles in the particle swarm, the initial parameter range represented by each particle, etc. should be reasonably determined to prepare for the subsequent optimization search using the PSO algorithm.

[0036] Step S501: Calculate the fitness value of each particle according to the fitness function of the particle swarm algorithm PSO, and update the extreme value pBest of each particle according to the fitness value; Calculate fitness value and update extreme value: Calculate the fitness value corresponding to each particle according to the pre-set fitness function. Then update the extreme value of each particle according to this value, that is, record the optimal position (corresponding to the optimal fitness performance) found by each particle in the search process so far.

[0037] The fitness function is specifically:

[0038] in, , is the positive learning factor, which is usually taken according to experience ; rand() represents a random number between 0 and 1; d represents the dimension; k represents the number of calculations; represents the position of particle i in the dth dimension, represents the best position of particle i so far, represents the best position of all particles so far, is the inertia weight. A slightly larger inertia weight w will give the particle a stronger spatial exploration ability, which can prevent the algorithm from falling into a local extreme value too early. A slightly smaller inertia weight w will bring a stronger local optimization ability and accelerate the convergence speed of the algorithm. Therefore, the present invention adopts an adaptive strategy to control the change of the inertia weight w:

[0039] in, is the maximum number of updates, T is the number of updates currently made by the algorithm, , Represent the maximum and minimum weights respectively. It can be seen that as the algorithm iterates, the value of w will become smaller, which can enable the algorithm to have a stronger spatial exploration ability in the early stage. In the later stage of optimization, the smaller weight will also allow particles to conduct more detailed searches locally and accelerate the convergence speed.

[0040] Step S502: compare the extreme values ​​of all particles, select M extreme values ​​of particles with the smallest fitness values ​​and save them into the set Best, where M>1; Compare the extreme values ​​of all particles, select M (for example, M=5) extreme values ​​of particles with the smallest fitness values, and save them in the set Best. Since the smaller the fitness value, the better the positioning effect, the parameter combination corresponding to these five extreme values ​​is likely to be the best, and further processing will be done based on them later.

[0041] Step S503: determine whether the number of iterations is greater than a preset number or the extreme value of each particle in the set Best is less than a certain accuracy threshold. If yes, execute step S504; if not, execute step S501; To determine the termination condition of the iteration, two conditions are set here. One is to see whether the number of iterations is greater than a certain preset number. The preset number is ; The second is to check whether all values ​​in the set Best meet the established requirements (for example, the extreme values ​​are all less than a certain precision threshold, etc.). If any of these two conditions are met, it means that the optimization process of the algorithm can be ended and the next step can be executed; if not, it is necessary to jump back to step S501 to continue updating the extreme value pBest of each particle and continue the next round of search optimization.

[0042] Step S504: Calculate the average value of all extreme values ​​in the set Best, take the position corresponding to the average value as the optimal position, and use the parameters associated with the optimal position for positioning.

[0043] After determining that the termination condition is met, the average value of the five extreme values ​​in the set Best is calculated, and the position corresponding to this average value is regarded as the optimal position. Finally, the parameters associated with this optimal position are used to perform positioning operations, hoping to achieve better positioning effects based on the optimized parameters and meet the requirements of WSN for positioning accuracy in corresponding scenarios.

[0044] In general, this is a process that uses the particle swarm algorithm to optimize the search for positioning-related parameters and ultimately obtain better positioning parameters for actual positioning. Through continuous iteration and screening, it aims to improve the accuracy and effectiveness of positioning.

[0045] The RSS positioning algorithm based on deep learning optimization is specifically: Step S510: Offline stage: detecting the basic characteristic information of the base station signal received by each grid node, processing the basic characteristic information into a signal fingerprint, and thereby constructing a signal fingerprint database; Step S511: pre-processing the fingerprint data in the fingerprint database, dividing the training set and the validation set, creating a cross entropy loss function, configuring the Adam optimizer and setting the learning rate parameter, and using the deep learning CNN model to carry out training until the Loss value is the lowest; Step S512: Online stage: collect the current fingerprint data of the target to be tested as a test set, use the CNN model to test, and screen out signal data with LOS characteristics by comparing with the basic feature information; Step S513: Use the sensor node transmission power and reference distance The path loss model and the path loss factor are optimized to compensate for the signal data with LOS characteristics, and the distance is estimated after using the optimized signal data.

[0046] The specific calculation formula for distance estimation of the optimized signal data is:

[0047] in, Transmit power for sensor nodes, The reference distance is The path loss model is is the power received at a distance d from the sending node, is to indicate that it obeys the mean of 0 and the variance of is a normal random distribution, and r is the path loss factor.

[0048] This RSS positioning algorithm based on deep learning optimization automatically extracts the features of line-of-sight LOS path signals by using CNN. Through training and testing of the fingerprint data of the target to be measured, the LOS path signal is screened out before calculating the positioning coordinates. On this basis, the measurement values ​​containing NLOS errors are optimized through deep learning models to reduce the interference of obstacles on the signal, thereby significantly improving the positioning accuracy and enhancing the system's adaptability.

[0049] Of course, in many fields of positioning technology, the non-ranging positioning method involved in the present invention has a rich variety of implementation methods and expansion possibilities, and is by no means limited to the case of PSO optimizing the DV-Hop positioning algorithm. In fact, the non-ranging positioning method covers methods based on various principles such as signal strength difference and arrival time difference. Similarly, the ranging positioning method is also diverse, and is by no means limited to the RSS positioning algorithm based on deep learning optimization. The ranging positioning method also includes many categories such as laser ranging, ultrasonic ranging, and ranging based on arrival angle. Those skilled in the art can choose the algorithm category according to actual needs.

[0050] The present invention can intelligently classify different application scenarios of smart factories. According to the classification rules, a ranging-based positioning method is adopted for positioning scenarios with higher positioning requirements, and a non-ranging-based positioning method is adopted for positioning scenarios with lower positioning accuracy requirements, thereby realizing adaptive matching of positioning methods, improving positioning accuracy and reducing system costs.

[0051] Example 2 The present invention provides an adaptive positioning system for a smart factory, comprising a base station, a server, and a target to be measured. The base station is arranged in the smart factory and provides a radio signal source. The target to be measured is provided with a signal receiver and can receive the radio signal from the base station. When positioning begins, the base station transmits a radio signal, and the CCD image capture device obtains the positioning environment image signal of the target to be measured and transmits the image signal to the server; a variety of sensors collect the physical data of the target to be measured and transmit the physical data to the server, and the sensors include direction sensors, acceleration sensors, gyroscopes, and air pressure sensors; The server is used to preprocess the image signal of the target to be measured, and perform network training on the processed image signal through deep learning to extract image features; fuse the physical data of the target to be measured, and use a convolutional neural network to classify the physical data of the target to be measured and establish a training model; perform big data analysis on the image features and the training model to identify the actual positioning application scenarios and positioning requirements of the smart factory; and adaptively match the positioning method according to the actual positioning application scenarios and positioning requirements.

[0052] The present invention can adaptively match positioning methods according to different application scenarios and positioning requirements of smart factories. For positioning scenarios with higher positioning requirements, a positioning method based on ranging is adopted, and for positioning scenarios with lower positioning requirements, a positioning method based on non-ranging measurement is adopted, thereby realizing adaptive matching of positioning methods, improving positioning accuracy and reducing system costs.

[0053] The above embodiments only express the preferred implementation mode of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An adaptive positioning method for smart factories, characterized in that: The following steps are involved: Step S1: acquiring a positioning environment image signal of the target to be measured through a CCD image capture device, preprocessing the image signal of the target to be measured, and performing network training on the processed image signal through deep learning to extract image features; Step S2: collecting physical data of the target to be measured through a variety of sensors, wherein the sensors include a direction sensor, an acceleration sensor, a gyroscope, and an air pressure sensor; Step S3: fusing the physical data of the target to be measured, and using a convolutional neural network to classify the physical data of the target to be measured and establish a training model; Step S4: performing big data analysis on the image features and training models to identify actual positioning application scenarios and positioning requirements of smart factories; Step S5: Adaptively match the positioning method according to the actual positioning application scenario and positioning requirements.

2. The adaptive positioning method for smart factories according to claim 1 is characterized in that: Before executing the step S2, the method includes: detecting an image signal by a CCD image pickup device to determine whether there is an obstruction between the target to be detected and the base station, and if so, enhancing the base station signal strength.

3. The adaptive positioning method for smart factories according to claim 2 is characterized in that: The preprocessing of the image signal of the target to be measured specifically includes: graying, filtering, noise reduction and image enhancement of the image signal of the target to be measured.

4. The adaptive positioning method for smart factories according to claim 3 is characterized in that: The specific method for fusing the physical data of the target to be measured is: any one of the weighted average method, the Kalman filter method, and the Bayesian estimation method.

5. The adaptive positioning method for smart factories according to claim 3 is characterized in that: Using a convolutional neural network to classify the physical data of the target to be measured and establish a training model includes: evaluating the posture, direction, and moving step length of the target to be measured, using CNN to classify the physical data of the target to be measured, generating various sets, and performing network training respectively.

6. The adaptive positioning method for smart factories according to claim 1, characterized in that: The adaptive matching of the positioning mode according to the actual positioning application scenario and positioning requirements is specifically: querying the positioning information table according to the actual positioning scenario and positioning requirements, so as to determine the positioning mode, and the positioning mode includes a non-ranging positioning mode and a ranging positioning mode.

7. The adaptive positioning method for smart factories according to claim 6, characterized in that: The non-ranging positioning method adopts the PSO optimized DV-Hop positioning algorithm, and the ranging positioning method is the RSS positioning algorithm based on deep learning optimization.

8. The adaptive positioning method for smart factories according to claim 7, characterized in that: The PSO optimization DV-Hop positioning algorithm is specifically as follows: Step S500: Initializing the population according to the size of the wireless network, the size of the target area, and the desired positioning accuracy; Step S501: Calculate the fitness value of each particle according to the fitness function of the particle swarm algorithm PSO, and update the extreme value pBest of each particle according to the fitness value; Step S502: compare the extreme values ​​of all particles, select M extreme values ​​of particles with the smallest fitness values ​​and save them into the set Best, where M>1; Step S503: determine whether the number of iterations is greater than a preset number or the extreme value of each particle in the set Best is less than a certain accuracy threshold. If yes, execute step S504; if not, execute step S501; Step S504: Calculate the average value of all extreme values ​​in the set Best, take the position corresponding to the average value as the optimal position, and use the parameters associated with the optimal position for positioning.

9. The adaptive positioning method for smart factories according to claim 7, characterized in that: The RSS positioning algorithm based on deep learning optimization is specifically: Step S510: Offline stage: detecting the basic characteristic information of the base station signal received by each grid node, processing the basic characteristic information into a signal fingerprint, and thereby constructing a signal fingerprint database; Step S511: pre-processing the fingerprint data in the fingerprint database, dividing the training set and the validation set, creating a cross entropy loss function, configuring the Adam optimizer and setting the learning rate parameter, and using the deep learning CNN model to carry out training until the Loss value is the lowest; Step S512: Online stage: collect the current fingerprint data of the target to be tested as a test set, use the CNN model to test, and screen out signal data with LOS characteristics by comparing with the basic feature information; Step S513: Use the sensor node transmission power and reference distance The path loss model and the path loss factor are optimized to compensate the signal data with LOS characteristics, and the optimized signal data is used to perform distance estimation.

10. An adaptive positioning system for smart factories, comprising a base station, a server, and a target to be measured, characterized in that: The base station is arranged in the smart factory and provides a radio signal source. The target to be measured is provided with a signal receiver that can receive the radio signal from the base station. When positioning begins, the base station transmits a radio signal, and the CCD image capture device obtains the positioning environment image signal of the target to be measured and transmits the image signal to the server; a variety of sensors collect physical data of the target to be measured and transmit the physical data to the server, and the sensors include direction sensors, acceleration sensors, gyroscopes, and air pressure sensors; The server is used to pre-process the image signal of the target to be measured, and perform network training on the processed image signal through deep learning to extract image features; The physical data of the target to be measured are fused, and a convolutional neural network is used to classify the physical data of the target to be measured and establish a training model; big data analysis is performed on the image features and the training model to identify the actual positioning application scenarios and positioning requirements of the smart factory; and the positioning method is adaptively matched according to the actual positioning application scenarios and positioning requirements.

Images