A path tracking method, device and system

Abstract

The present application discloses a path tracking method, apparatus and system, relating to the technical field of the Internet of Things. The method includes: obtaining the measured positions of a target object determined by each reader at a first moment; inputting each of the measured positions into an adaptive target dynamics model based on a non-linear Kalman filtering algorithm that has been updated currently, to obtain an optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measured positions; and determining the path of the target object according to the optimal state vectors at different moments. The solution of the present application improves the accuracy of predicting the state vector of the target object, thereby improving the accuracy of tracking the motion trajectory / path of the target object.

Description

Technical Field

[0001] The present application relates to the technical field of the Internet of Things, and in particular, to a path tracking method, device, and system. Background Art

[0002] Introducing Internet of Things technology and the concept of "Industry 4.0" into the workshop production process can help managers track production process information comprehensively, automatically, real-time, and transparently, so as to quickly coordinate production plans and resources and improve the production management level of enterprises. In this context, the digital twin technology emerges as the times require, and virtual-real interaction has become one of the basic characteristics of an intelligent workshop or a digital twin workshop. In the actual application of a digital twin workshop, in order to map the flow of workpieces and materials in the workshop, most rely on radio frequency identification (RFID) technology for non-contact data communication, identify the tags on the workpieces, and transmit the data into the manufacturing execution system (MES). Using the event data source as the position information judgment condition, it is determined which workstation the workpiece is at and whether the workpiece is being processed. Judging that the workpiece leaves the workstation also originates from the event trigger mechanism. Therefore, the positioning and tracking accuracy of RFID will affect the accuracy of the synchronous mapping of the material transportation path, thereby affecting the decision-making of managers.

[0003] However, the widely used RFID technology at present generally extracts the distance between the tag and the reader based on the principle of received signal strength information (RSSI). This method will be interfered by surrounding noise sources, so accurate positioning cannot be performed. Summary of the Invention

[0004] The purpose of the present application is to provide a path tracking method, device, and system, so as to solve the problem in the prior art that the determined transportation path is inaccurate due to low positioning and tracking accuracy.

[0005] In a first aspect, in order to achieve the above object, the present application provides a path tracking method, including:

[0006] Obtain the measurement positions of the target object determined by each reader at the first moment;

[0007] Input each of the measurement positions into the currently updated adaptive target dynamics model based on the non-linear Kalman filter algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions;

[0008] Determine the path of the target object according to the optimal state vectors at different times.

[0009] Optionally, predicting the optimal state vector of the target object at the first time according to each of the measurement positions includes:

[0010] Obtain the predicted state vector of the target object at the first time predicted at the second time; wherein, the second time is the previous time of the first time;

[0011] Obtain the gain of the non-linear Kalman filter;

[0012] Predict the optimal state vector of the target object at the first time according to each of the measurement positions, the predicted state vector, and the gain.

[0013] Optionally, obtaining the predicted state vector of the target object at the first time predicted at the second time includes:

[0014] According to the formula x(t i ∣t i-1 )=A(t i-1 )x(t i-1 ∣t i-1 ), obtain the predicted state vector;

[0015] wherein, x(t i ∣t i-1 ) is the predicted state vector, A(t i-1 ) is the system state transition matrix, and x(t i-1 ∣t i-1 ) is the optimal state vector of the target object at the second time.

[0016] Optionally, the system state transition matrix is:

[0017]

[0018] wherein,

[0019] wherein, η = x, y, respectively representing the x and y coordinate axes; th i is the time interval between the first time and the second time; α is the maneuvering frequency parameter in the adaptive target dynamics model; α η is the maneuvering frequency on the x-axis or y-axis, is the Gaussian white noise variance.

[0020] Optionally, predicting the optimal state vector of the target object at the first time according to each of the measurement positions, the predicted state vector, and the gain includes:

[0021] According to the formula Predict the optimal state vector of the target object at the first moment;

[0022] Wherein, x(t i ∣t i ) is the optimal state vector of the target object at the first moment, x(t i ∣t i-1 ) is the predicted state vector of the target object at the first moment predicted at the second moment, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, h n [x n (t i )] is the measurement function corresponding to the nth reader, N(t i ) is the number of the readers.

[0023] Optionally, predicting the optimal state vector of the target object at the first moment according to each of the measured positions, the predicted state vector, and the gain includes:

[0024] According to the formula Predict the optimal state vector of the target object at the first moment;

[0025] Wherein, x(t i ∣t i ) is that of the target object at the first moment, x(t i ∣t i-1 ) is the predicted state vector, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, is the predicted position of the target object at the first moment predicted according to each of the measured positions.

[0026] Optionally, after obtaining the optimal state vector of the target object at the first moment, the method further includes:

[0027] Update the maneuvering frequency and / or the Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimated value in the optimal state vector of the target object at the first moment.

[0028] Optionally, updating the maneuver frequency and / or Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimated value in the optimal state vector of the target object at the first moment includes:

[0029] Determining the updated maneuver frequency and Gaussian white noise variance according to the following formula:

[0030]

[0031] Where is the acceleration estimated value of the target object at the first moment, is the acceleration estimated value of the target object at the second moment, and the second moment is the moment before the first moment, w a (t i-1 ) is a zero-mean white noise discrete sequence, β is the maneuver frequency of the acceleration discrete sequence, α is the maneuver frequency, th i is the time interval between the first moment and the second moment, is the variance of the zero-mean white noise discrete sequence, is the Gaussian white noise variance.

[0032] In a second aspect, to achieve the above object, an embodiment of the present application further provides a path tracking device, including:

[0033] An acquisition module, configured to acquire the measurement position of the target object at the first moment determined by each reader-writer;

[0034] A prediction module, configured to input each of the measurement positions into the currently updated adaptive target dynamics model based on the nonlinear Kalman filter algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions;

[0035] A determination module, configured to determine the path of the target object according to the optimal state vectors at different moments.

[0036] In a third aspect, to achieve the above object, an embodiment of the present application further provides a path tracking system, including: a processor, a memory, and a program stored on the memory and executable on the processor, and when the program is executed by the processor, the steps of the path tracking method described in the first aspect are implemented.

[0037] In a fourth aspect, to achieve the above object, an embodiment of the present application further provides a readable storage medium, on which a program is stored, and when the program is executed by a processor, the steps of the path tracking method described in the first aspect are implemented.

[0038] The above technical solution of this application has at least the following beneficial effects:

[0039] In the path tracking method of the embodiment of this application, first, obtain the measured positions of the target object determined by each reader at the first moment; secondly, input each of the measured positions into the currently updated adaptive target dynamics model based on the non-linear Kalman filtering algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measured positions; in this way, the positioning accuracy of the target object is improved. Finally, determine the path of the target object according to the optimal state vectors at different moments. In this way, on the basis of improving the positioning accuracy of the target object, the tracking accuracy of the path of the target object can be further improved, providing a reliable basis for the decision-making of the management personnel. Brief Description of the Drawings

[0040] Figure 1 is a schematic flowchart of the path tracking method of the embodiment of this application;

[0041] Figure 2 is a schematic diagram of the classification and common algorithms of data fusion;

[0042] Figure 3 is an example diagram showing the placement of RFID readers in the measurement range;

[0043] Figure 4 is a schematic diagram of the setting of the material transportation trajectory;

[0044] Figure 5 is an example diagram of each simulation data record;

[0045] Figure 6 is a curve graph of the estimated trajectory and the reference trajectory of the material based on EKF / UKF;

[0046] Figure 7 is a curve graph of the estimated trajectory and the reference trajectory of the horizontal and vertical axes based on EKF / UKF;

[0047] Figure 8 is a curve graph of the trajectory tracking error of the horizontal and vertical axes based on EKF / UKF;

[0048] Figure 9 is a schematic diagram of the characteristics of the distribution area of RFID readers;

[0049] Figure 10 is a curve graph of the placement positions of some RFID readers and the material transportation path in the repeated comparison test;

[0050] Figure 11It is a schematic structural diagram of the path tracking device according to the embodiment of the present application. Detailed implementation manners

[0051] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0052] The terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / " generally means that the related objects before and after are in an "or" relationship.

[0053] Before elaborating on the path tracking method, device, and system according to the embodiments of the present application, the relevant technical points will be explained first:

[0054] I. RFID

[0055] An Internet of Things technology that relies on wireless communication to achieve automatic identification of objects. The basic components of a typical RFID system include RFID tags, readers, antennas, and a back-end application system. Among them, the tags are attached to the objects to be identified and located, and when the reader uses the system antenna to receive the electromagnetic information generated when the object carrying the RFID tag enters the system identification range, it is then transmitted to the back-end application system for processing, so as to achieve various application effects required by the user.

[0056] In the field of positioning, outdoor positioning services based on the Global Positioning System (GPS) have been widely used. However, due to the lack of a line-of-sight wireless transmission channel between satellites and receivers, GPS cannot provide sufficiently accurate positioning services in indoor environments. In addition, many other problems make positioning and tracking in indoor environments more complex and challenging than in outdoor environments. For example, signal scattering caused by excessive obstacle distribution and multipath effects caused by signal reflection through walls and objects. With the development of technologies such as infrared, wireless local area network, ultra-wideband, and RFID, various channels have been provided for positioning problems in indoor environments. However, RFID can identify different people and devices, and due to its low cost, potential ability to accurately position objects, and ability to operate in occluded and non-line-of-sight environments, it has gradually been widely used in the field of indoor location awareness and target tracking, and has also become a widely used technology in wireless sensor networks in workshops.

[0057] II. Data Fusion:

[0058] Data fusion is a strategy and theoretical method for analyzing and processing multi-source data. By comprehensively analyzing and processing data from different times and spaces under certain criteria, it can accurately and timely obtain a consistent interpretation and description of the system, thereby realizing corresponding information processing, decision-making, and estimation tasks, and enabling the system to obtain more sufficient information. Currently, the data fusion field is mainly divided into three categories: sensor fusion, image fusion, and information fusion. Sensor fusion uses data from different sensors and integrates them into a single data through data measurement and analysis processes, enabling more accurate and specific inferences; image fusion improves the information content of images by reducing uncertainty and redundancy, including merging images in various forms into a single image; information fusion is a process for extracting information from incomplete information, processing multi-level processes such as the relevance and combination of information, and finally used to complete the required estimation and decision-making links.

[0059] Data fusion can also be classified into data-level fusion, feature-level fusion, and decision-level fusion according to the operation object. Among them, data-level fusion is also called pixel-level fusion in the image field. Each level performs fusion on the data itself, data features, and decision problems reflected by the data. The classification of each level and common algorithms are as Figure 2 shown. This application belongs to the data-level fusion level.

[0060] III. Kalman Filter:

[0061] An optimal autoregressive data processing algorithm that calculates the optimal estimated value by recursion based on the predicted value of the system state at the previous moment and the measured value at the current moment. In the fields of indoor positioning and target tracking, the Kalman filtering method and its various derivative algorithms usually can achieve excellent estimation effects.

[0062] In the actual system application, due to the interference of uncertain factors such as system state noise and measurement noise, there are often a large number of non-linear filtering problems in engineering practice. For example, the RFID measurement system is a typical non-linear system. However, the initially proposed Kalman filtering method is only applicable to linear systems, and its estimation effect on non-linear systems is very poor. Therefore, scholars have successively proposed non-linear Kalman filtering algorithms such as Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Cubature Kalman Filter (CKF), and Gauss-Hermite Kalman Filter (GHKF). Among them, this application mainly studies the RFID trajectory tracking data fusion based on EKF and UKF.

[0063] IV. EKF algorithm model

[0064] Consider the general discrete-time non-linear system model as:

[0065] x(k) = f[x(k - 1), w(k - 1)]

[0066] z(k) = h[x(k), v(k)]

[0067] Where, x(k) represents the system state vector to be estimated at time k; z(k) represents the system state measurement value at time k; w(k - 1) represents the system process noise, which is a Gaussian white noise sequence; v(k) represents the measurement noise, which is a Gaussian white noise independent of the system noise; f(·) and h(·) represent the non-linear process equation and measurement equation.

[0068] The EKF algorithm processes the non-linear system model into a linearized model after Taylor series expansion. The algorithm is mainly divided into the following parts:

[0069] (1) Predict x(k|k - 1):

[0070] x(k|k - 1) = f[x(k - 1|k - 1), 0]

[0071] (2) Update x(k|k):

[0072] x(k|k) = x(k|k - 1) + K(k)[z(k) - h(x(k|k - 1), 0)]

[0073] Where K(k) is the filter gain, representing the weights of the estimated value and the measured value on the true value.

[0074] (3) Filter gain K(k):

[0075] K(k) = P(k|k - 1)H T (k)[H T (k)P(k|k - 1)H(k) + M(k)R(k)M T (k)] -1

[0076] Where P(k|k - 1) represents the one-step-ahead prediction variance; R(k) represents the measurement noise covariance matrix, which is a known value; H(k) and M(k) are shown as follows.

[0077]

[0078]

[0079] (4) One-step-ahead prediction variance P(k|k - 1):

[0080] P(k|k - 1) = F(k - 1)P(k - 1|k - 1)F T (k - 1) + L(k - 1)Q(k - 1)L T (k - 1)

[0081] Where P(k - 1|k - 1) represents the estimation variance at time k - 1; Q(k - 1) represents the process noise variance, which is a known value; F(k) and L(k) are shown as follows.

[0082]

[0083]

[0084] (5) State estimation variance P(k|k):

[0085] P(k|k) = [I - K(k)H(k)]P(k|k - 1)

[0086] V. UKF algorithm model

[0087] The unscented transform refers to selecting a suitable number of points from many numerical points through a certain method, called sigma points, and using these sigma points to calculate the mean and variance, which can well approximate the true value. The specific selection method will not be elaborated here. The UKF algorithm uses the unscented transform to process the nonlinear system model for more accurate prediction and update. The algorithm is mainly divided into the following parts:

[0088] (1) Predict x(k|k - 1):

[0089] x (i) (k - 1|k - 1) = x(k - 1|k - 1) + x (i) , i = 1, …, 2n

[0090]

[0091]

[0092] x (i) (k|k - 1) = f[x (i) (k - 1|k - 1)]

[0093]

[0094] where x (i) (k - 1|k - 1) is the sigma point; represents the square root of the matrix after multiplying n by the estimated variance P, and selects the i-th row and transposes it to obtain an n-dimensional vector.

[0095] (2) Update x(k|k):

[0096] x (i) (k|k - 1) = x(k|k - 1) + x (i) , i = 1, …, 2n

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] (3) Filter gain K(k):

[0103]

[0104]

[0105]

[0106] (4) One-step-ahead prediction variance P(k|k - 1):

[0107]

[0108] (5) State estimation variance P(k + 1|k + 1):

[0109] P(k + 1|k + 1) = P(k|k + 1) - K(k)P z K T (k)

[0110] The following combines the accompanying drawings to detail the path tracking method, device, and system provided by the embodiments of the present application through specific embodiments and their application scenarios.

[0111] As Figure 1 shown, it is a schematic flowchart of the path tracking method of the embodiments of the present application. The method includes:

[0112] Step 101, obtain the measurement positions of the target object determined by each reader at the first moment;

[0113] It should be noted here that a tag is attached to the target object, and the reader determines the distance between the reader and the target object by identifying the tag on the target object, thereby further determining the measurement position of the target object;

[0114] Here, it also should be noted that the measurement position of the target object at the first moment can be represented by the distance vector between the reader and the target object.

[0115] Step 102, input each of the measurement positions into the currently updated adaptive target dynamics model based on the non - linear Kalman filter algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions;

[0116] Here, it should be noted that in this step, by inputting each measurement position into the adaptive target dynamics model based on the non - linear Kalman filter algorithm, the adaptive target dynamics model based on the non - linear Kalman filter algorithm can perform data fusion on these measurement positions to obtain the optimal state vector of the target object; wherein, the optimal state vector of the target object at least includes the displacement, velocity, acceleration, etc. of the target object.

[0117] Step 103, determine the path of the target object according to the optimal state vectors at different moments.

[0118] The path tracking method according to the embodiment of the present application, first, obtains the measured positions of the target object determined by each reader at the first moment; secondly, inputs each of the measured positions into the currently updated adaptive target dynamics model based on the non-linear Kalman filtering algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measured positions; thus, the positioning accuracy of the target object is improved. Finally, according to the optimal state vectors at different moments, the path of the target object is determined. In this way, on the basis of improving the positioning accuracy of the target object, the tracking accuracy of the path of the target object can be further improved, providing a reliable basis for the decision-making of the management personnel.

[0119] Here, the specific implementation process of step 101, obtaining the measured positions of the target object determined by each reader at the first moment, will be described:

[0120] The RFID measurement system obtains the distance z i from the detection target (target object) at time t n (t i ) through the RSSI method, as shown in the following formula:

[0121]

[0122] In the formula, d0 represents a fixed reference distance; P(d0) represents the received signal strength when the distance between the reader and the target object is d0; P(d) represents the received signal strength when the distance between the reader and the target object is d; n is the path loss exponent, which is determined by environmental factors.

[0123] In actual measurement, the measured distance z n (t i ) and the actual distance d n (t i ) are not the same, as shown in the following formula:

[0124] z n (t i ) = d n (t i ) + v n (t i )

[0125] wherein, v n (t i ) represents the measurement noise of the nth reader at time t i , and in practical applications, the measurement noise variance often satisfies a certain relationship with the actual distance.

[0126] As an alternative implementation, in step 102, predicting the optimal state vector of the target object at the first moment according to each of the measurement positions includes:

[0127] Obtaining the predicted state vector of the target object at the first moment predicted at the second moment; wherein, the second moment is the moment before the first moment;

[0128] Obtaining the gain of the non-linear Kalman filter;

[0129] Predicting the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector, and the gain.

[0130] In this alternative implementation, based on each of the measurement positions, the predicted state vector of the target object at the current moment (the first moment) predicted at the previous moment (the second moment), and the gain of the non-linear Kalman filter, the optimal state vector of the target object at the first moment is predicted, realizing that the optimal state vector at the current moment can be determined only based on the prediction result at the previous moment and the measurement result at the current moment, reducing the operation time in the prediction process and improving the prediction speed.

[0131] As a specific implementation, obtaining the predicted state vector of the target object at the first moment predicted at the second moment includes:

[0132] According to the formula x(t i ∣t i-1 ) = A(t i-1 )x(t i-1 ∣t i-1 ), obtaining the predicted state vector;

[0133] Wherein, x(t i ∣t i-1 ) is the predicted state vector, that is: the state vector of the target object predicted at time t i-1 at time t i ; A(t i-1 ) is the state transition matrix of the system at time t i-1 , wherein the system state matrix is updated in real time; x(t i-1 ∣t i-1 ) is the optimal state vector of the target object at the second moment (time t i-1 ).

[0134] Wherein, the state transition matrix of the system at time t i-1 is:

[0135]

[0136] Among them,

[0137] where η = x, y, representing the x and y coordinate axes respectively; th i is the time interval between the first moment (t i moment) and the second moment (t i-1 moment); α is the maneuver frequency parameter in the adaptive target dynamics model; α η is the component of α on the x-axis or y-axis, is the variance of Gaussian white noise.

[0138] As another alternative implementation, predicting the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector, and the gain includes:

[0139] According to the formula predict the optimal state vector of the target object at the first moment;

[0140] where x(t i ∣t i ) is the optimal state vector of the target object at the first moment, x(t i ∣t i-1 ) is the predicted state vector of the target object predicted at time t i-1 at time t i , K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measurement position measured by the nth reader-writer at the first moment, h n [x n (t i )] is the measurement function corresponding to the nth reader-writer, and N(t i ) is the number of the reader-writers.

[0141] Here, it should be noted that in this alternative implementation, the non-linear Kalman filter can specifically be an EKF; then the gain of the non-linear Kalman filter is specifically:

[0142]

[0143] where K n (t i ) is the gain of the Kalman filter corresponding to the nth reader-writer at the first moment (t i moment), P(t i ∣t i-1 ) is at the second moment (t i-1The state variance prediction matrix of the target object predicted at the first moment (t i moment); is the transpose matrix of the x-axis component of the measurement function corresponding to the nth reader at the first moment, is the inverse matrix of the covariance corresponding to the nth reader at the first moment (t i moment);

[0144]

[0145] where h nx (t i ) is the matrix of the x-axis component of the measurement function corresponding to the nth reader, P(t i |t i-1 ) is the state variance prediction matrix predicted at the second moment (t i-1 moment) at the first moment (t i moment); is the transpose matrix of the x-axis component of the measurement function corresponding to the nth reader, R(t i ) is the measurement noise covariance matrix at the first moment (t i moment);

[0146]

[0147] where P -1 (t i |t i ) is the inverse matrix of the optimal state variance matrix at the first moment (t i moment), P -1 (t i |t i-1 ) is the inverse matrix of the predicted state variance matrix predicted at the second moment (t i-1 moment) at the first moment (t i moment), R n (t i ) is the measurement noise covariance matrix corresponding to the nth reader at the first moment (t i moment), h nx (t i ) is the matrix of the x-axis component of the measurement function corresponding to the nth reader at the first moment (t i moment), is the inverse matrix of h nx (t i );

[0148]

[0149] where h n [ti , x(t i |t i-1 )] is the measurement function matrix corresponding to the nth reader, x i at the moment of t n (0) is the initial state vector of the x-axis corresponding to the nth reader set in advance, y n (0) is the initial state vector of the y-axis corresponding to the nth reader set in advance; x n (t i |t i-1 ) is the predicted state vector of the nth reader in the x-axis direction at the moment of t i-1 at the moment of t i at the moment of the nth reader; is the predicted state vector of the nth reader in the y-axis direction at the moment of t i-1 at the moment of t i at the moment of the nth reader;

[0150] h nx (t i )'s Jacobian matrix is:

[0151]

[0152] That is to say, when the nonlinear Kalman filter algorithm is the EKF algorithm, the process of determining the optimal state vector of the target object at the first moment based on the measured positions of the target object determined by each reader obtained is as follows:

[0153] (1), obtain each of the measured positions z n (t i )

[0154] (2), based on the formula x(t i |t i-1 ) = A(t i-1 )x(t i-1 |t i-1 ), obtain the predicted state vector of the target object at the first moment;

[0155] Among them, η = x, y;

[0156] (3) Based on the following formula, obtain the gain of the nonlinear Kalman filter:

[0157]

[0158] P(t i |t i-1 ) = A(t i-1 )P(t i-1 |t i-1 )AT (t i-1 ) + Q(t i-1 )

[0159]

[0160]

[0161]

[0162]

[0163]

[0164] where Q(t i-1 ) is the variance matrix of the process noise, T0 is the initial sampling period, is the component of the Gaussian white noise variance on the x-axis or y-axis, and other symbols are consistent with the descriptions of the same symbols above.

[0165] (4) Predict the optimal state vector of the target object at the first moment according to the formula

[0166] As another optional implementation, predicting the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector, and the gain includes:

[0167] Predict the optimal state vector according to the formula

[0168] where x(t i ∣t i ) is the optimal state vector of the target object at the first moment, x(t i ∣t i-1 ) is the predicted state vector of the target object at the first moment predicted at the second moment (t i-1 moment), K i (t n ) is the gain of the non-linear Kalman filter, z i (t n ) is the measurement position measured by the nth reader at the first moment (t i moment), i is the predicted position of the target object at the first moment predicted according to each of the measurement positions.

[0169] Here, it should be noted that this optional implementation is specifically applicable to the adaptive target dynamics model with the non-linear Kalman filter algorithm being the EKF algorithm; ​​​

[0170] Next, the process of determining the optimal state vector of the target object at the first moment based on the measured positions of the target object determined by each obtained reader when the non-linear Kalman filtering algorithm is the EKF algorithm will be described:

[0171] (1) Obtain each of the measured positions z n (t i );

[0172] (2) Based on the formula x(t i ∣t i-1 ) = A(t i-1 )x(t i-1 ∣t i-1 ), obtain the predicted state vector of the target object at the first moment;

[0173] Among them, η = x, y;

[0174] (3) Calculate the predicted position of the target object at the first moment

[0175] (a) According to the following formula, obtain the sigma points:

[0176]

[0177]

[0178]

[0179] Among them, W (0) is the mean weight; is the optimal state vector of the target object at the second moment (t i-1 moment); is a column vector with the same dimension as the process noise w(t i-1 ); is a column vector with the same dimension as the measurement noise v(t i-1 ); N x is the dimension of the system state; P(t i-1 ∣t i-1 ) is the estimated variance matrix at the second moment (t i-1 moment); is the i-th column principal square root of the matrix. Each point can be instantiated through the process model, and then the transformation set can be obtained.

[0180]

[0181] (b) According to the formula Predict the mean;

[0182] (c) According to the formula Predictive variance;

[0183] (d) Through the econometric model Respectively give the prediction points:

[0184] (e) According to the formula Determine the predicted position of the target object at the first moment

[0185] (4) According to the following formula, determine the gain of the non - linear Kalman filter:

[0186]

[0187] where, K n (t i ) is the non - linear Kalman filter gain, S n (t i ) is the covariance, P x,n (t i ) is the cross - covariance.

[0188] (5) According to the formula Predict the optimal state vector.

[0189] Here, it should also be noted that in the case where the first moment is the first moment, in the process of determining the optimal state vector of the target object at the first moment, the pre - set initial state vector / parameters can be used as the relevant state vector / parameters at the second moment mentioned above.

[0190] Furthermore, as an optional implementation manner, after obtaining the optimal state vector of the target object at the first moment, the method further includes:

[0191] Update the maneuvering frequency and / or Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimation value in the optimal state vector of the target object at the first moment.

[0192] In this optional implementation manner, by updating the maneuvering frequency and / or Gaussian white noise variance in the adaptive target dynamics model, the adaptive change of the adaptive target dynamics model is realized, thereby further improving the prediction accuracy of the adaptive target dynamics model.

[0193] As a specific implementation manner, updating the maneuvering frequency and / or Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimation value in the optimal state vector of the target object at the first moment includes:

[0194] Determine the updated maneuvering frequency and Gaussian white noise variance according to the following formula:

[0195]

[0196] where is the estimated acceleration value of the target object at the first moment, is the estimated acceleration value of the target object at the second moment, and the second moment is the moment before the first moment, w a (t i-1 ) is a zero-mean white noise discrete sequence, β is the maneuvering frequency of the acceleration discrete sequence, α is the maneuvering frequency, th i is the time interval between the first moment and the second moment, is the variance of the zero-mean white noise discrete sequence, is the Gaussian white noise variance.

[0197] Here, it should be noted that the formula indicates that the estimated acceleration satisfies a first-order discrete-time Markov process.

[0198] Here, it also should be noted that since then the autocorrelation relationship between β and can be expressed as:

[0199]

[0200] where l(0) and l(1) represent the autocorrelation function of the estimated acceleration, as shown in the following formula:

[0201]

[0202] Then the parameter values of β and can be obtained as:

[0203]

[0204] Therefore, the maneuvering frequency α and the Gaussian white noise variance can be obtained according to the relationship between the formula and Thus, the original adaptive target dynamics model can be updated using the obtained parameters, so as to realize the adaptive change of the target dynamics model.

[0205] In addition, the process of parameter update can also be updated in the following manner:

[0206] Suppose is η = x or y; the parameters updated in real time are: and α i,η ; where α i,η is the x-axis or y-axis component of the current updated maneuvering frequency α of the system, is the variance of the current updated Gaussian white noise of the x-axis or y-axis component; then:

[0207] When i ≤ K0, and α i,η are:

[0208]

[0209] When K0 is a positive number less than 10, the maneuvering frequency α i,η is positive, a M is a positive number, -a M is the opposite of a M .

[0210] When i > K0, and α i,η are obtained according to the following formula:

[0211]

[0212] Here, it should also be noted that the workshop data signal has big data characteristics such as massive scale, multi-source heterogeneity, multi-temporal and spatial scales, and multi-dimensions. In order to better combine the data fusion method with the digital twin system, a data transmission flow of the digital twin system based on data fusion is proposed. Among them, the soft sensor data includes data such as part performance parameter data, mechanism stroke planning, and working process simulation data that cannot be directly measured by sensors; the actual physical data of the sensors includes data such as physical attribute data and historical storage data that can be directly obtained by sensors. The path tracking method studied in the embodiments of the present application is also applicable to this data transmission flow.

[0213] Next, the experimental verification and analysis of the path tracking method applying the embodiments of the present application will be described with reference to the accompanying drawings:

[0214] First, a simulation data platform for the workshop RFID path tracking system is developed based on the improvement of the RFID measurement system based on the RSSI principle. It can simulate the placement of several RFID readers and can record the distance readings of each measurement point of the material transportation trajectory carrying RFID tags, and the transportation trajectory can be set and recorded by itself to provide a real value for comparison, such as Figure 3 、 Figure 4 and Figure 5 as shown.

[0215] After the simulation data is generated, the actual position of the material is estimated by the aforementioned EKF and UKF algorithms. At the same time, the motion process of the material is estimated using the workshop adaptive target dynamics model, and we can obtain the workshop RFID data fusion path tracking algorithm based on EKF / UKF.

[0216] Among them, the path tracking algorithm based on EKF includes parts such as state prediction, state fusion estimation (multi-RFID data fusion), and parameter update; while the path tracking algorithm based on UKF includes parts such as sigma point acquisition, predicted mean, predicted variance, covariance improvement, cross-covariance matrix, update, and parameter update. Among them, the update part uses the method of multi-RFID data fusion, which is different from the basic UKF method. The final experimental effect of the path tracking is as Figure 6 、 Figure 7 and Figure 8 shown.

[0217] From the experimental results Figures 6 to 8 it can be known that the estimation error of the EKF algorithm will increase to a large extent in some areas with fewer RFID readers, but it has a good estimation effect in most cases; while the UKF algorithm has a stable estimation effect during the overall trajectory tracking process, but the estimation effect is worse than that of the EKF algorithm in some areas with sufficient RFID readers. Among them, Figure 9 shows the characteristics of the distribution area of RFID readers in this simulation experiment.

[0218] Then, multiple repeated experiments were carried out to verify the obtained experimental results. By changing the number and placement position of RFID readers, and simulating various material transportation trajectories, and comparing experiments were carried out with the Gaussian-Hermite Kalman filter (GHKF) algorithm and the traditional method of particle filter. Some selected RFID placement positions and material trajectory paths are as Figure 10 shown.

[0219] After multiple repeated experiments, the average horizontal error of the EKF method is 4.7947, the average estimation covariance is 46.0732, the average vertical error is 2.5930, and the average estimation covariance is 28.9697; the average horizontal error of the UKF method is 3.8297, the average estimation covariance is 37.5835, the average vertical error is 2.3428, and the estimation covariance is 26.1844. Comparing the experimental results with the GHKF algorithm and the traditional method of particle filter, the results are shown in Table 1 below. This result shows that the workshop RFID data fusion path tracking algorithm based on EKF and UKF can effectively improve the workshop RFID tracking and positioning accuracy.

[0220] Algorithm Horizontal axis average error Horizontal axis estimated covariance Vertical axis average error Vertical axis estimated covariance EKF 4.7947 46.0732 2.5930 28.9697 UKF 3.8297 37.5835 2.3428 26.1844 GHKF 5.3467 52.1180 4.0976 38.6642 Particle filter method 5.2118 53.9901 3.7341 42.8906

[0221] Table 1

[0222] In addition, the EKF method has less estimation operation time, and the UKF method has stronger advantages in the overall estimation performance and can well adapt to various changes in the number of RFID and their placement. Moreover, when the number of RFID readers is sufficient, the accuracy of the two estimation methods is not much different. The reason is that after the materials with RFID tags are read by multiple readers simultaneously and the multiple groups of data go through the data fusion algorithm, the estimation performance of the algorithm has been greatly improved.

[0223] The path tracking method of the embodiments of the present application specifically relates to a data fusion algorithm model represented by the nonlinear Kalman filtering method and a workshop application method, and adopts a workshop RFID data fusion path tracking algorithm based on EKF and / or UKF to improve the accuracy of intelligent workshop material transportation trajectory tracking. The simulation results show that the algorithm proposed in the embodiments of the present application can effectively reduce the RFID positioning error. Among them, the method based on UKF can provide better overall estimation performance, but the method based on EKF has less operation time. At the same time, the research in this paper also plays a reference role in improving the measurement performance of other sensors in the workshop, and meets the requirements of intelligent workshops for high-precision sensors through methods such as data fusion.

[0224] In short, in the embodiments of the present application, first, according to the target motion characteristics of the workshop and the system model under irregular sampling intervals, a target dynamic model containing adaptive parameters (target maneuvering frequency α and noise variance ) is established; then, the acceleration model during the actual movement process of the target material is determined (predicting the optimal state vector at each moment based on EKF or UKF), so as to obtain the current acceleration estimate value of the target, and the system adaptive parameters are updated by virtue of the obtained estimate value; finally, the updated parameters are used to change the initial dynamic model of the target, and the acceleration estimation and model update processes are repeated until all measurement data are processed.

[0225] It should be noted that for the path tracking method provided in the embodiments of the present application, the execution subject may be a path tracking device, or a control module in the path tracking device for executing the loaded path tracking method. In the embodiments of the present application, the case where the path tracking device executes the loaded path tracking method is taken as an example to illustrate the path tracking method provided in the embodiments of the present application.

[0226] As Figure 11 shown, the embodiments of the present application also provide a path tracking device, including:

[0227] An acquisition module 1101, configured to acquire the measurement positions of the target object determined by each reader at the first moment;

[0228] A prediction module 1102, configured to input each of the measurement positions into an adaptive target dynamics model based on a non - linear Kalman filtering algorithm that has been currently updated, and obtain an optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions.

[0229] A determination module 1103, configured to determine the path of the target object according to the optimal state vectors at different moments.

[0230] For the path tracking device according to an embodiment of the present application, first, an acquisition module 1101 acquires the measurement positions of the target object determined by each reader at the first moment; secondly, a prediction module 1102 inputs each of the measurement positions into an adaptive target dynamics model based on a non - linear Kalman filtering algorithm that has been currently updated, and obtains an optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions. In this way, the positioning accuracy of the target object is improved. Finally, a determination module 1103 determines the path of the target object according to the optimal state vectors at different moments. In this way, on the basis of improving the positioning accuracy of the target object, the tracking accuracy of the path of the target object can be further improved, providing a reliable basis for the decision - making of management personnel.

[0231] Optionally, the prediction module 1102 includes:

[0232] A first acquisition sub - module, configured to acquire a predicted state vector of the target object at the first moment predicted at the second moment; wherein, the second moment is the moment before the first moment.

[0233] A second acquisition sub - module, configured to acquire the gain of the non - linear Kalman filter.

[0234] A prediction sub - module, configured to predict the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector, and the gain.

[0235] Optionally, the prediction sub - module is specifically configured to:

[0236] According to the formula \(x(t i \mid t i-1 ) = A(t i-1 )x(t i-1 \mid t i-1 ), obtain the predicted state vector;

[0237] wherein, \(x(t i \mid t i-1) is the predicted state vector, A(t i-1 ) is the system state transition matrix, x(t i-1 ∣t i-1 ) is the optimal state vector of the target object at the second moment.

[0238] Optionally, the system state transition matrix is:

[0239]

[0240] where,

[0241] where, η = x, y, respectively representing the x and y coordinate axes; th i is the time interval between the first moment and the second moment; α is the maneuvering frequency parameter in the adaptive target dynamics model; α η is the maneuvering frequency on the x-axis or y-axis, is the Gaussian white noise variance.

[0242] Optionally, the prediction sub-module is specifically configured to:

[0243] Predict the optimal state vector of the target object at the first moment according to the formula ;

[0244] where, x(t i ∣t i ) is the optimal state vector of the target object at the first moment, x(t i ∣t i-1 ) is the predicted state vector of the target object at the first moment predicted at the second moment, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, h n [x n (t i )] is the measurement function corresponding to the nth reader, N(t i ) is the number of the readers.

[0245] Optionally, the prediction sub-module is specifically configured to:

[0246] Predict the optimal state vector according to the formula ;

[0247] where, x(t i ∣t i ) is the optimal state vector of the target object at the first moment, x(ti ∣t i-1 ) is the predicted state vector, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, is the predicted position of the target object at the first moment predicted according to each of the measured positions.

[0248] Optionally, the device further includes:

[0249] An update module, configured to update the maneuvering frequency and / or the Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimation value in the optimal state vector of the target object at the first moment.

[0250] Optionally, the update module is specifically configured to:

[0251] Determine the updated maneuvering frequency and Gaussian white noise variance according to the following formula:

[0252]

[0253] where is the acceleration estimation value of the target object at the first moment, is the acceleration estimation value of the target object at the second moment, and the second moment is the moment before the first moment, w a (t i-1 ) is a zero-mean white noise discrete sequence, β is the maneuvering frequency of the acceleration discrete sequence, α is the maneuvering frequency, th i is the time interval between the first moment and the second moment, is the variance of the zero-mean white noise discrete sequence, is the Gaussian white noise variance.

[0254] The embodiment of the present application further provides a path tracking system, including: a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, it implements each process of the path tracking method embodiment as described above and can achieve the same technical effect. To avoid repetition, it will not be described here again.

[0255] The embodiments of the present application further provide a readable storage medium, on which a program is stored. When the program is executed by a processor, it implements each process of the path tracking method embodiment described above and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the readable storage medium can be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0256] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of another identical element in the process, method, article or device comprising the said element.

[0257] The above is the preferred embodiment of the present application. It should be pointed out that for those of ordinary skill in the art, without departing from the principle described in the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A path tracking method, characterized in that, Including: Obtain the measurement positions of the target object determined by each reader / writer at the first moment; Input each of the measurement positions into the currently updated adaptive target dynamics model based on the non-linear Kalman filtering algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions; Determine the path of the target object according to the optimal state vectors at different moments; Wherein, predicting the optimal state vector of the target object at the first moment according to each of the measurement positions includes: Obtain the predicted state vector of the target object at the first moment predicted at the second moment; wherein, the second moment is the moment before the first moment; Obtain the gain of the non-linear Kalman filter; Predict the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector and the gain; Wherein, predicting the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector and the gain includes: According to the formula predict the optimal state vector of the target object at the first moment; Among them, is the optimal state vector of the target object at the first moment, is the predicted state vector of the target object at the first moment predicted at the second moment, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, h n [x n (t i )] is the measurement function corresponding to the nth reader, N(t i ) is the number of the readers.

2. The method according to claim 1, wherein Obtaining the predicted state vector of the target object at the first moment predicted at the second moment includes: According to the formula Obtain the predicted state vector; Among them, is the predicted state vector, and A(t i-1 ) is the system state transition matrix, is the optimal state vector of the target object at the second moment.

3. The method according to claim 2, wherein The system state transition matrix is: Among them, where η = x, y, representing the x and y coordinate axes respectively; th i is the time interval between the first moment and the second moment; α is the maneuvering frequency parameter in the adaptive target dynamic model; α η is the maneuvering frequency on the x-axis or y-axis, is the Gaussian white noise variance.

4. The method according to claim 1, characterized in that Predicting the optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector and the gain includes: According to the formula Predict the optimal state vector of the target object at the first moment; Among them, is the optimal state vector of the target object at the first moment, is the predicted state vector, and K n (t i ) is the gain of the non-linear Kalman filter, and z n (t i ) is the measured position measured by the nth reader at the first moment, is the predicted position of the target object at the first moment predicted according to each of the measured positions.

5. The method according to claim 1, wherein After obtaining the optimal state vector of the target object at the first moment, the method further includes: Update the maneuvering frequency and / or the Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimated value in the optimal state vector of the target object at the first moment.

6. The method according to claim 1, wherein Updating the maneuvering frequency and / or the Gaussian white noise variance in the adaptive target dynamics model according to the acceleration estimated value in the optimal state vector of the target object at the first moment includes: Determine the updated maneuvering frequency and Gaussian white noise variance according to the following formula: Among them, is the estimated acceleration value of the target object at the first moment, is the estimated acceleration value of the target object at the second moment, where the second moment is the moment immediately preceding the first moment, w a (t i-1 ) is a discrete sequence of white noise with zero mean, β is the maneuvering frequency of the acceleration discrete sequence, α is the maneuvering frequency, th i is the time interval between the first moment and the second moment, is the variance of the discrete sequence of white noise with zero mean, is the Gaussian white noise variance.

7. A path tracking device, characterized in that, Including: An obtaining module, configured to obtain the measurement positions of the target object determined by each reader / writer at the first moment; A predicting module, configured to input each of the measurement positions into the currently updated adaptive target dynamics model based on the non-linear Kalman filtering algorithm to obtain the optimal state vector of the target object at the first moment; wherein, the adaptive target dynamics model is used to predict the optimal state vector of the target object at the first moment according to each of the measurement positions; A determining module, configured to determine the path of the target object according to the optimal state vectors at different moments; Wherein, the predicting module includes: A first obtaining sub-module, configured to obtain the predicted state vector of the target object at the first moment predicted at the second moment; wherein, the second moment is the moment before the first moment; A second obtaining sub-module, configured to obtain the gain of the non-linear Kalman filter; A predictor sub-module, configured to predict an optimal state vector of the target object at the first moment according to each of the measurement positions, the predicted state vector, and the gain; Specifically, the predictor sub-module is configured to: According to the formula Predict the optimal state vector of the target object at the first moment; Among them, is the optimal state vector of the target object at the first moment, is the predicted state vector of the target object at the first moment predicted at the second moment, K n (t i ) is the gain of the non-linear Kalman filter, z n (t i ) is the measured position measured by the nth reader at the first moment, h n [x n (t i )] is the measurement function corresponding to the nth reader, N(t i ) is the number of the readers.

8. A path tracking system, characterized in that, comprise: a processor, a memory, and a program stored on the memory and executable on the processor, where when the program is executed by the processor, the steps of the path tracking method according to any one of claims 1 to 6 are implemented.

9. A readable storage medium, characterized in that, A program is stored on the readable storage medium, and when the program is executed by the processor, the steps of the path tracking method according to any one of claims 1 to 6 are implemented.

Images