Robot parameter prediction method, device, storage medium and processor

By obtaining the correlation relationship and operating parameters between building robots and using neural network models to predict parameters, the problem of inaccurate prediction caused by data loss in the existing technology is solved, and the accuracy of building robot parameter prediction is improved.

CN114648148BActive Publication Date: 2024-11-05GUANGDONG BRIGHT DREAM ROBOTICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011507850.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-18
Publication Date
2024-11-05
Estimated Expiration
2040-12-18

AI Technical Summary

Technical Problem

In the prediction of building robot parameters, in the case of insufficient historical information or lack of covariate data, the prediction accuracy is low and effective prediction cannot be obtained.

Method used

By obtaining the correlation between the target robot and the auxiliary robot, using a preset neural network model to combine the operating parameters of the target robot and the auxiliary robot, parameter prediction is carried out, and the impact of the auxiliary robot on the target robot is introduced to improve the prediction accuracy.

Benefits of technology

It effectively improves the accuracy of building robot parameter prediction, solves the problem of prediction inaccurate due to data loss, and improves prediction accuracy and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114648148B_ABST
    Figure CN114648148B_ABST
Patent Text Reader

Abstract

The present invention discloses a parameter prediction method, device, storage medium and processor of a robot. The method comprises: obtaining the parameter prediction result of the target robot at the previous moment; obtaining the characteristic information of the target robot at the current moment, wherein the characteristic information is obtained according to the operating parameters of the target robot and the operating parameters of the auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; according to the parameter prediction result of the target robot at the previous moment and the characteristic information at the current moment, the parameters to be predicted of the target robot at the next moment are predicted by a preset neural network model. The present invention solves the technical problem of inaccurate parameter prediction of construction robots in the prior art.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of robots, and in particular to a method, device, storage medium and processor for predicting parameters of a robot. Background Art

[0002] In the field of smart buildings, a complex construction task usually requires the participation of multiple intelligent construction robots, that is, multiple machines work together. These construction robots often participate in the entire construction process in the form of group division of labor. For construction robots completing the same task, their working states are closely related. Therefore, during the construction process, it is very necessary to make real-time predictions of the key indicators of a certain important equipment.

[0003] Existing methods often use their own historical time series data and covariate information to predict the monitoring signal of the time series. However, when there is not enough historical information or the covariate data is missing due to some reasons, such methods often have the problem of low prediction accuracy, and sometimes even fail to obtain effective predictions.

[0004] With regard to the problem of inaccurate parameter prediction of construction robots in the prior art, no effective solution has been proposed yet. Summary of the invention

[0005] Embodiments of the present invention provide a robot parameter prediction method, device, storage medium and processor to at least solve the technical problem of inaccurate parameter prediction of a construction robot in the prior art.

[0006] According to one aspect of an embodiment of the present invention, a robot parameter prediction method is provided, comprising: obtaining a parameter prediction result of a target robot at a previous moment; obtaining feature information of the target robot at a current moment, wherein the feature information is obtained based on operating parameters of the target robot and operating parameters of an auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; and predicting parameters to be predicted of the target robot at the next moment through a preset neural network model based on the parameter prediction result of the target robot at a previous moment and the feature information at a current moment.

[0007] Furthermore, obtaining the parameter prediction result of the target robot at the previous moment includes: obtaining the output of the preset neural network model at the previous moment; and determining the output of the preset neural network model at the previous moment as the parameter prediction result of the target robot at the previous moment.

[0008] Furthermore, characteristic information of the target robot at the current moment is obtained, including: obtaining an adjacency matrix between the target robot and the auxiliary robot, wherein the adjacency matrix is ​​used to represent the association between the target robot and the auxiliary robot when performing the same set of tasks; determining the auxiliary parameters corresponding to each auxiliary robot according to the operating parameters of each auxiliary robot and the adjacency matrix; and weighting the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot to obtain characteristic information.

[0009] Furthermore, the auxiliary parameters corresponding to each auxiliary robot are determined according to the operating parameters and adjacency matrix of each auxiliary robot, including: obtaining the order of the auxiliary robot, wherein the order is used to indicate the closeness of the association between the auxiliary robot and the target robot when performing the same set of tasks; multiplying and transposing the order adjacency matrices to obtain an n-th order adjacency matrix, wherein n is used to indicate the number of multiplied adjacency matrices; and multiplying the operating parameters of each auxiliary robot with the n-th order adjacency matrix corresponding to each auxiliary robot to obtain the auxiliary parameters of each auxiliary robot.

[0010] Further, the n-order adjacency matrices are multiplied and transposed to obtain an n-order adjacency matrix, including: normalizing the adjacency matrix to obtain a row-normalized adjacency matrix; and multiplying and transposing the n-order normalized adjacency matrices to obtain an n-order adjacency matrix.

[0011] Furthermore, the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot are weighted to obtain characteristic information, including: obtaining weight values, and weighting the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot according to the weight values ​​to obtain characteristic information; wherein, the step of obtaining weight values ​​includes: obtaining constraints on the weight values, wherein the constraints include: the sum of the first weight value and all second weight values ​​is 1, the first weight value is greater than any second weight value, and the second weight value of a multi-order auxiliary robot is the nth power of the second weight value of a first-order auxiliary robot, where n represents the order of the auxiliary robot; solving the constraints to obtain the weight values ​​corresponding to the target robot and each auxiliary robot.

[0012] Furthermore, when the operating parameters of any robot are missing, the weight value of the robot missing the operating parameters is 0.

[0013] Furthermore, based on the parameter prediction results of the target robot at the previous moment and the feature information at the current moment, a preset neural network model is used to predict the parameters to be predicted of the target robot at the next moment, including: constructing a first relationship between the mean parameter and the parameter to be predicted in the normal distribution formula, and a second relationship between the variance parameter and the parameter to be predicted; solving the normal distribution formula according to the first relationship and the second relationship through the preset neural network model to obtain the prediction results of the parameters to be predicted.

[0014] Furthermore, the above method also includes: obtaining a preset neural network model, wherein the step of obtaining the preset neural network model includes: obtaining an initial neural network model; obtaining training data, wherein the training data includes parameter prediction results of the sample robot at the previous moment and feature information of the sample robot at the current moment, and the feature information of the sample robot at the current moment is determined according to the operating parameters of the sample robot and the operating parameters of the auxiliary robot corresponding to the sample robot; constructing a loss function, and training the initial neural network model through the training data, wherein the loss function is a logarithmic maximum likelihood function; when the function value of the loss function is less than a preset value, or the number of training iterations exceeds a preset number, the preset neural network model is obtained.

[0015] Furthermore, the preset neural network model is a gated recurrent network model.

[0016] Further, after the parameter prediction result of the target robot at the previous moment and the feature information at the current moment, the normal operating range is determined according to the distribution information of the parameter to be predicted, wherein the distribution information is obtained according to the model parameters of the preset neural network model; if the prediction result of the parameter to be predicted is not within the normal operating range, the prediction result is subjected to early warning processing;

[0017] According to one aspect of an embodiment of the present invention, a robot parameter prediction device is provided, comprising: a first acquisition module, used to obtain a parameter prediction result of a target robot at a previous moment; a second acquisition module, used to obtain feature information of the target robot at a current moment, wherein the feature information is obtained based on operating parameters of the target robot and operating parameters of an auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; a prediction module, used to predict the parameters to be predicted of the target robot at the next moment through a preset neural network model based on the parameter prediction result of the target robot at a previous moment and the feature information at a current moment.

[0018] According to one aspect of an embodiment of the present invention, a storage medium is provided, the storage medium including a stored program, wherein when the program is executed, a device where the storage medium is located is controlled to execute the above-mentioned parameter prediction of the robot.

[0019] According to one aspect of an embodiment of the present invention, a processor is provided, and the processor is used to run a program, wherein the program executes the above-mentioned parameter prediction of the robot when running.

[0020] In an embodiment of the present invention, the parameter prediction result of the target robot at the previous moment is obtained; the characteristic information of the target robot at the current moment is obtained, wherein the characteristic information is obtained according to the operating parameters of the target robot and the operating parameters of the auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; according to the parameter prediction result of the target robot at the previous moment and the characteristic information at the current moment, the parameters to be predicted of the target robot at the next moment are predicted by a preset neural network model. Since the target robot and the auxiliary robot are used to perform the same set of tasks, the target robot and the auxiliary robot have a certain influence on each other when performing tasks. When predicting the parameters to be predicted of the target robot, the above scheme not only predicts according to the detected operating parameters of the target robot itself, but also introduces the operating parameters of the auxiliary robot that performs the same set of tasks as the target robot, so that when predicting the parameters to be predicted of the target robot, the influence of the auxiliary robot on the target robot is introduced, and various types of relevant information are effectively used, thereby improving the prediction accuracy and solving the technical problem of inaccurate parameter prediction of construction robots in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0022] Figure 1 is a flow chart of a robot parameter prediction method according to an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of a robot parameter prediction method according to an embodiment of the present application;

[0024] Figure 3 is a flow chart of an optional robot parameter prediction method according to an embodiment of the present invention;

[0025] Figure 4 is a flowchart of another optional robot parameter prediction method according to an embodiment of the present invention;

[0026] Figure 5 It is a schematic diagram for determining the adjacency matrix through an undirected acyclic graph;

[0027] Figure 6 is a schematic diagram of a robot parameter prediction device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Example 1

[0031] According to an embodiment of the present invention, an embodiment of a parameter prediction method of a robot is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] Figure 1 is a flow chart of a robot parameter prediction method according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:

[0033] Step S102, obtaining the parameter prediction result of the target robot at the previous moment.

[0034] Specifically, the robot may be a construction robot, and the target robot may be the most important construction robot among multiple construction robots, that is, the robot whose parameters need to be predicted.

[0035] Step S104, obtaining characteristic information of the target robot at the current moment, wherein the characteristic information is obtained according to operating parameters of the target robot and operating parameters of the auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot.

[0036] Specifically, the above operating parameters may be multiple parameters obtained through detection. The auxiliary robot and the target robot are in the same building working scene, and when performing tasks in the building scene, there is a direct or indirect connection between the two. The above steps are used to extract features of the target robot, and the data used for feature extraction includes not only the data of the target robot itself, but also the data of the auxiliary robot in the same building environment as the target robot.

[0037] In an optional embodiment, the monitoring data of multiple robots in the same building work scene are obtained, including the monitoring index information of the target robot (i.e., the operating parameters of the target robot) and the covariate information of the auxiliary robot (i.e., the operating parameters of the auxiliary robot). After a period of recording, time series data can be obtained. The time points of signal collection are recorded as 1 to T time points. If there are n robots working together in the same work scene, the final data obtained is {Y 1:T , X 1:T}. Among them, {Y 1:T}=(Y 1,1:T , Y 2,1:T , ...Y n,1:T )T, 1≤t≤T, and Y 1:T ∈R T , X ij,t Represents the information of the jth covariate of the ith robot at time point t.

[0038] After obtaining the operating parameters of the target robot and the auxiliary robot, the characteristic information of the target robot at the current moment can be obtained by combining the association relationship between the target robot and the auxiliary robot.

[0039] Step S106, predicting the parameters to be predicted of the target robot at the next moment through a preset neural network model according to the parameter prediction results of the target robot at the previous moment and the feature information at the current moment.

[0040] The above-mentioned preset neural network model is a neural network model used to predict the parameters to be predicted. The neural network model is obtained by training sample data.

[0041] Figure 2 is a schematic diagram of a robot parameter prediction method according to an embodiment of the present application, combined with Figure 2As shown, the neural network model is a GRU (Gated Recurrent Unit) model. At time t-1, the GRU outputs the prediction result at time t-1 based on the GRU output information at time t-2 and the feature information at time t-1; at time t, the GRU outputs the prediction result at time t based on the GRU output information at time t-1 and the feature information at time t; at time t+1, the GRU outputs the prediction result at time t+1 based on the GRU output information at time t and the feature information at time t+1.

[0042] It can be seen that in the above embodiment of the present application, the parameter prediction result of the target robot at the previous moment is obtained; the characteristic information of the target robot at the current moment is obtained, wherein the characteristic information is obtained according to the operating parameters of the target robot and the operating parameters of the auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; according to the parameter prediction result of the target robot at the previous moment and the characteristic information at the current moment, the target robot and the auxiliary robot have a certain influence on each other when performing tasks. When predicting the parameters to be predicted of the target robot, the above scheme not only predicts according to the detected operating parameters of the target robot itself, but also introduces the operating parameters of the auxiliary robot that performs the same set of tasks as the target robot, so that when predicting the parameters to be predicted of the target robot, the influence of the auxiliary robot on the target robot is introduced, and various types of relevant information are effectively used, thereby improving the prediction accuracy and solving the technical problem of inaccurate parameter prediction of construction robots in the prior art.

[0043] Figure 3 is a flow chart of an optional robot parameter prediction method according to an embodiment of the present invention. As an optional embodiment, obtaining the parameter prediction result of the target robot at the previous moment includes:

[0044] Step S1021, obtaining the output of the preset neural network model at the previous moment;

[0045] Step S1023, determining that the output of the preset neural network model at the previous moment is the parameter prediction result of the target robot at the previous moment.

[0046] Combination Figure 2 As shown in the figure, the prediction result of the preset neural network model at the previous moment will be used as the input of the current moment. For example, the GRU input at time t-1 includes the output Y at the previous moment. I,t-2 , the GRU input at time t includes the output Y at the previous time I,t-1 , the GRU input at time t+1 includes the output Y at the previous timeI,t .

[0047] Figure 4 is a flowchart of another optional robot parameter prediction method according to an embodiment of the present invention. As an optional embodiment, obtaining feature information of the target robot at the current moment includes:

[0048] Step S1041, obtaining an adjacency matrix between the target robot and the auxiliary robot, wherein the adjacency matrix is ​​used to represent the association relationship between the target robot and the auxiliary robot when performing the same set of tasks;

[0049] Step S1043, determining the auxiliary parameters corresponding to each auxiliary robot according to the operating parameters of each auxiliary robot and the adjacency matrix;

[0050] Step S1045, weighting the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot to obtain feature information.

[0051] Specifically, the adjacency matrix is ​​used to represent the relationship between the target robot and the assistant robot. In an optional embodiment, the adjacency matrix between the target robot and the assistant robot can be determined by an undirected acyclic graph between the target robot and the assistant robot in the building scene. Figure 5 It is a schematic diagram of determining the adjacency matrix through an undirected acyclic graph, combined with Figure 5 As shown in FIG. 1 , each node in the graph is used to represent a robot in the same building environment. If two nodes are connected, it means that the robots represented by the two nodes have an association relationship when performing tasks. According to the association relationship embodied in the undirected acyclic graph, the adjacency matrix can be determined.

[0052] For example, the above adjacency matrix can be represented by an n*n dimensional matrix A. The element a at position (i, j) of the matrix A is ik =1, indicating that in the work scenario, device i and device k (1≤i, k≤n) are directly involved in the same task. ik = 0 means that there is no directly related task system between the two. In addition, since the robot cannot work with itself, all diagonal elements a of the matrix A are ii =0, (1≤i≤n).

[0053] After obtaining the adjacency matrix, the adjacency matrix can be calculated with the operating parameters of the auxiliary robot to obtain auxiliary parameters, and then the auxiliary parameters can be weighted with the operating parameters of the target robot, so that the collaborative relationship between the target robot and the auxiliary robot can be introduced into the calculation of feature information.

[0054] As an optional embodiment, the auxiliary parameters corresponding to each auxiliary robot are determined according to the operating parameters and adjacency matrix of each auxiliary robot, including: obtaining the order of the auxiliary robot, where the order is used to indicate the closeness of the relationship between the auxiliary robot and the target robot when performing the same set of tasks; multiplying and transposing the order adjacency matrices to obtain an n-th order adjacency matrix, where n is used to indicate the number of multiplied adjacency matrices; multiplying the operating parameters of each auxiliary robot with the n-th order adjacency matrix corresponding to each auxiliary robot to obtain the auxiliary parameters of each auxiliary robot.

[0055] Specifically, the above order is used to indicate the degree of correlation between the target robot and the auxiliary robot when performing a task. If the two are directly related, the order is 1; if the two are related with one other robot in between, the order is 2, and so on. Figure 5 As shown, if the node of the auxiliary robot is directly connected to the node of the target robot, the order of the auxiliary robot is 1st order; if the node of the auxiliary robot is connected to the node of the target robot via a node of another robot, the order of the auxiliary robot is 2nd order.

[0056] The above scheme determines the order of the auxiliary robot with respect to the target robot by the closeness of the association relationship between the auxiliary robot and the target robot, and applies the order to the feature information extraction of the target robot, thereby further improving the accuracy of the prediction.

[0057] As an optional embodiment, multiplying and transposing adjacency matrices of order to obtain an n-order adjacency matrix includes: normalizing the adjacency matrix to obtain a row-normalized adjacency matrix; multiplying and transposing the normalized adjacency matrices of order to obtain an n-order adjacency matrix.

[0058] Still in the above embodiment, the adjacency matrix A can be further processed by dividing each row element of the adjacency matrix A by the row sum to obtain It is the weighted adjacency matrix with weights after row normalization.

[0059] As an optional embodiment, the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot are weighted to obtain characteristic information, including: obtaining weight values, and weighting the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot according to the weight values ​​to obtain characteristic information; wherein, the step of obtaining weight values ​​includes: obtaining constraints on the weight values, wherein the constraints include: the sum of the first weight value and all second weight values ​​is 1, the first weight value is greater than any second weight value, and the second weight value of a multi-order auxiliary robot is the nth power of the second weight value of a first-order auxiliary robot, where n represents the order of the auxiliary robot; solving the constraints to obtain the weight values ​​corresponding to the target robot and each auxiliary robot.

[0060] In the above scheme, since the operating parameters of the target device itself should be the most important for predicting the future parameters of the target device, the first weight value of the target robot is greater than the second weight values ​​of all auxiliary robots. Since auxiliary robots with different orders have different effects on the target device, the second weight value of the multi-order auxiliary robot can be the nth power of the second weight value of the first-order auxiliary robot.

[0061] For example, the second weight value of the first-order auxiliary robot is ρ, and the second weight value of the second-order auxiliary robot is ρ 2 , and so on. Based on this, we can get the formula α+ρ+ρ 2 +…+ρ n =1, and α>ρ, so we can get Then get And 0<ρ<0.5. Since the first weight value is greater than any second weight value, that is, From this we can get the value range of ρ In addition, appropriate feature weight information can be selected according to the hyperparameter tuning method, and finally the weighted sum covariate information of each robot is obtained as the feature input of the target robot i.

[0062] It should be noted that if there are too many robots in the same scene and the task is too complicated, the adjacency matrix A can be directly used when extracting features instead of the row-normalized weighted adjacency matrix. This can avoid the diminishing effect of double weights in the above situation and retain more collaborative work information.

[0063] In an optional embodiment, an example is given to illustrate the extraction of the above feature information. First, it is assumed that the extracted feature information Z it For X t and The function of Among the characteristics of target device i, the monitored operating parameters of the target device i have the highest influence, followed by the operating parameters of the auxiliary robot directly related to it, and then the operating parameters of the auxiliary robot indirectly related to the target robot i, and so on. Therefore, based on the above, the characteristic Z of the target robot i at time t (1≤t≤T) is represented by it , can be expressed by the following formula:

[0064]

[0065] Where α>0 and ρ>0.

[0066] From the above, we can see that the feature information Z i,t It consists of two parts, namely the weighted values ​​of its own covariates and the covariates of the auxiliary robot. contains the sum of all the covariates of the first-order auxiliary robots, and their common coefficient is ρ. Furthermore, we can further obtain the second-order auxiliary robot information and the third-order or higher-order neighbor information based on the adjacency matrix. Specifically, It is used to represent the second-order adjacency information in the relationship between the target robot and the auxiliary robot. The non-zero elements in the i-th row of are the second-order neighbors of the target robot i. contains the sum of the covariate information of all second-order auxiliary robots, and its corresponding weight is ρ 2 . And so on for higher-level information inference for assistive robots.

[0067] As an optional embodiment, when the operating parameters of any robot are missing, the weight value of the robot missing the operating parameters is 0.

[0068] The robot that is missing operating parameters can be a target robot or an auxiliary robot. When a robot is missing operating parameters, the weight value of the robot can be set to 0, and the weight values ​​of other robots can be reset, so that even if the information of some robots is missing, the parameters to be predicted of the target robot can still be predicted.

[0069] Through the extraction of the above feature information, it can be seen that if the covariate information of the target robot is missing due to some factors (i.e., X i,t The above scheme can still use the covariate information of the auxiliary robot working with it to predict the signal, which is an effect that traditional prediction methods cannot achieve. It solves the problem of model failure caused by missing variables in conventional situations.

[0070] As an optional embodiment, based on the parameter prediction results of the target robot at the previous moment and the feature information at the current moment, the parameters to be predicted of the target robot at the next moment are predicted through a preset neural network model, including: constructing a first relationship between the mean parameter and the parameter to be predicted in the normal distribution formula, and a second relationship between the variance parameter and the parameter to be predicted; solving the normal distribution formula according to the first relationship and the second relationship through the preset neural network model to obtain the prediction result of the parameter to be predicted.

[0071] In an optional embodiment, it is assumed that the parameter to be predicted has a mean of μ and a variance of σ 2 A normally distributed random variable with the following distribution form: Modeling the two parameters mean and variance in the normal distribution separately, we get the first relationship mentioned above And the second relationship It is worth noting that the softplus function is used here to ensure that the estimate of the standard deviation is greater than 0.

[0072] Through the first relationship and the second relationship, the parameter to be predicted is substituted into the normal distribution formula and solved by the preset neural network model to obtain the prediction result.

[0073] As an optional embodiment, the method also includes: obtaining a preset neural network model, wherein the step of obtaining the preset neural network model includes: obtaining an initial neural network model; obtaining training data, wherein the training data includes parameter prediction results of the sample robot at the previous moment and feature information of the sample robot at the current moment, and the feature information of the sample robot at the current moment is determined according to the operating parameters of the sample robot and the operating parameters of the auxiliary robot corresponding to the sample robot; constructing a loss function, and training the initial neural network model through the training data, wherein the loss function is a logarithmic maximum likelihood function; when the function value of the loss function is less than a preset value, or the number of training iterations exceeds a preset number, the preset neural network model is obtained.

[0074] In the above scheme, the neural network model for predicting the parameters to be predicted of the target robot is obtained by training the initial neural network model. During the training process, at each time t, the feature Z extracted at the current time is i,t And the output h of the GRU unit at the previous moment i,t-1 Together as the input of the current GRU unit, so as to obtain the GRU unit output h at the current moment i,t .

[0075] In order to obtain the network parameters of the neural network model, the above loss function based on the logarithmic maximum likelihood function is used to ensure that the parameters in the distribution can be effectively estimated, that is, The parameters in the model are estimated according to the above loss function, and finally the training process of the entire model is completed.

[0076] The above scheme provides a method for estimating network parameters. The test process of the generated neural network model is similar to the training process, taking the output of the previous step and the feature information of the current time point as input for prediction. However, the difference is that during the test process, that is, when t>T, we can estimate the distribution of the parameters to be predicted. Then sample from this empirical distribution to get the sample And take the sample mean As the output of the current time node. For 1≤i≤n, t>T, we can As a point estimate of the forecast.

[0077] As an optional embodiment, the preset neural network model is a gated recurrent network model.

[0078] As an optional embodiment, after predicting the parameters of the target robot at the previous moment and the feature information at the current moment, the method further includes: determining the normal operating range based on the distribution information of the parameter to be predicted, wherein the distribution information is obtained based on the model parameters of a preset neural network model; if the prediction result of the parameter to be predicted is not within the normal operating range, performing early warning processing on the prediction result.

[0079] In the above scheme, an early warning is set for the parameter to be predicted. When the prediction result of the parameter to be predicted is not within the normal operating range, an early warning is immediately performed to prevent the target robot from operating abnormally.

[0080] In an optional embodiment, the distribution information of the parameter to be predicted can be obtained at each time point. Using this distribution information, we can construct the interval [μ 0.05 , μ 0.95 ] is used as the normal value interval prediction of the signal, where the upper and lower limits of the interval are 0.95 quantile and 0.05 quantile respectively. If the predicted result falls outside the interval, the predicted result is warned.

[0081] In another optional embodiment, if the parameter to be predicted is more sensitive to falling into the abnormal interval, a more stringent operation interval construction method can be adopted, and the 1-sigma principle is adopted, that is, the operation interval is constructed as A risk warning is issued for predicted values ​​that do not fall within this range.

[0082] At present, most of the existing prediction methods can only predict the predicted value of the signal point estimation in the future, but cannot warn the abnormal value of the predicted signal, which is not conducive to the prevention of abnormal signals and predictive maintenance. The above scheme can obtain the prediction of the signal and the normal operation range at the same time, which can solve this problem well.

[0083] The effects of the above embodiments of the present application are described below. In an optional embodiment, the covariate information of 5 robots is collected, each robot has covariate information in three dimensions, and the square root mean square error (RMSE) is used for evaluation. The final prediction results are shown in Table 1. Scheme 1 is the performance result of the scheme that only uses the operating parameters of the target device itself for prediction, and Scheme 2 is the performance result of the above scheme of this embodiment. It can be seen from the results in Table 1 that after introducing the covariate information of the auxiliary robots working in collaboration, the prediction accuracy has been effectively improved regardless of the number of iteration steps.

[0084] Table 1

[0085] Steps 10000 20000 30000 40000 50000 Solution 1 0.440 0.413 0.385 0.370 0.366 Solution 2 0.412 0.399 0.354 0.350 0.348

[0086] Example 2

[0087] According to an embodiment of the present invention, an embodiment of a parameter prediction device for a robot is provided. Figure 6 is a schematic diagram of a parameter prediction device for a robot according to an embodiment of the present invention, such as Figure 6 As shown, the device comprises the following steps:

[0088] A first acquisition module 60 is used to obtain the parameter prediction result of the target robot at the previous moment;

[0089] A second acquisition module 62 is used to acquire characteristic information of the target robot at a current moment, wherein the characteristic information is obtained according to operating parameters of the target robot and operating parameters of the auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot;

[0090] The prediction module 64 is used to predict the parameters to be predicted of the target robot at the next moment based on the parameter prediction results of the target robot at the previous moment and the feature information at the current moment, through a preset neural network model.

[0091] As an optional embodiment, the first acquisition module includes: a first acquisition submodule, used to obtain the output of the preset neural network model at the previous moment; and a first determination submodule, used to determine that the output of the preset neural network model at the previous moment is the parameter prediction result of the target robot at the previous moment.

[0092] As an optional embodiment, the second acquisition module includes: a second acquisition submodule, used to acquire the adjacency matrix between the target robot and the auxiliary robot, wherein the adjacency matrix is ​​used to represent the association between the target robot and the auxiliary robot when performing the same set of tasks; a second determination submodule, used to determine the auxiliary parameters corresponding to each auxiliary robot based on the operating parameters of each auxiliary robot and the adjacency matrix; a weighting submodule, used to weight the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot to obtain feature information.

[0093] As an optional embodiment, the second determination submodule includes: a first acquisition unit, used to acquire the order of the auxiliary robot, wherein the order is used to indicate the degree of association between the auxiliary robot and the target robot when performing the same set of tasks; a transposition unit, used to multiply and transpose order adjacency matrices to obtain an n-th order adjacency matrix, wherein n is used to indicate the number of multiplied adjacency matrices; a multiplication unit, used to multiply the operating parameters of each auxiliary robot with the n-th order adjacency matrix corresponding to each auxiliary robot to obtain the auxiliary parameters of each auxiliary robot.

[0094] As an optional embodiment, the transposition unit includes: a normalization subunit, used to perform row normalization on the adjacency matrix to obtain a row-normalized adjacency matrix; a transposition subunit, used to multiply and transpose the normalized adjacency matrices of order to obtain an n-th order adjacency matrix.

[0095] As an optional embodiment, the weighted submodule includes: a second acquisition unit, used to obtain weight values, and weight the operating parameters of the target robot and the auxiliary parameters of each auxiliary robot according to the weight values ​​to obtain feature information; wherein the second acquisition unit includes: a first acquisition subunit, and the step of obtaining weight values ​​includes: a second acquisition subunit, used to obtain constraint conditions for the weight values, wherein the constraint conditions include: the sum of the first weight value and all second weight values ​​is 1, the first weight value is greater than any second weight value, and the second weight value of a multi-order auxiliary robot is the nth power of the second weight value of a first-order auxiliary robot, where n represents the order of the auxiliary robot; a solving subunit, used to solve the constraint conditions to obtain the weight values ​​corresponding to the target robot and each auxiliary robot.

[0096] As an optional embodiment, when the operating parameters of any robot are missing, the weight value of the robot missing the operating parameters is 0.

[0097] As an optional embodiment, the prediction module includes: a construction submodule, which is used to construct a first relationship between a mean parameter and a parameter to be predicted in a normal distribution formula, and a second relationship between a variance parameter and the parameter to be predicted; and a solution submodule, which is used to solve the normal distribution formula according to the first relationship and the second relationship through a preset neural network model to obtain a prediction result of the parameter to be predicted.

[0098] As an optional embodiment, the above-mentioned device also includes: a third acquisition module, used to obtain a preset neural network model, wherein the third acquisition module includes: a third acquisition sub-module, used to obtain an initial neural network model; a fourth acquisition sub-module, used to obtain training data, wherein the training data includes the parameter prediction result of the sample robot at the previous moment and the feature information of the sample robot at the current moment, and the feature information of the sample robot at the current moment is determined according to the operating parameters of the sample robot and the operating parameters of the auxiliary robot corresponding to the sample robot; a training sub-module, used to construct a loss function and train the initial neural network model through the training data, wherein the loss function is a logarithmic maximum likelihood function; a stop training sub-module, used to obtain the preset neural network model when the function value of the loss function is less than a preset value, or the number of training iterations exceeds a preset number.

[0099] As an optional embodiment, the preset neural network model is a gated recurrent network model.

[0100] As an optional embodiment, the above-mentioned device also includes: a determination module, which is used to determine the normal operating range according to the distribution information of the parameters to be predicted based on the parameter prediction results of the target robot at the previous moment and the feature information at the current moment, wherein the distribution information is obtained according to the model parameters of the preset neural network model; an early warning module, which is used to perform early warning processing on the prediction results if the prediction results of the parameters to be predicted are not within the normal operating range.

[0101] Example 3

[0102] According to an embodiment of the present invention, a storage medium is provided, the storage medium including a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the parameter prediction of the robot described in Example 1.

[0103] Example 4

[0104] According to an embodiment of the present invention, a processor is provided, and the processor is used to run a program, wherein the program executes the parameter prediction of the robot described in Example 1 when running.

[0105] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0106] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0108] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0109] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0110] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0111] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A robot parameter prediction method, characterized in that: include: Get the parameter prediction results of the target robot at the previous moment; Acquire characteristic information of the target robot at the current moment, wherein the characteristic information is obtained according to operating parameters of the target robot and operating parameters of an auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; According to the parameter prediction result of the target robot at the previous moment and the feature information at the current moment, the parameters to be predicted of the target robot at the next moment are predicted by using a preset neural network model; Wherein, obtaining the characteristic information of the target robot at the current moment includes: obtaining an adjacency matrix between the target robot and the auxiliary robot, wherein the adjacency matrix is ​​used to represent the association relationship between the target robot and the auxiliary robot when performing the same set of tasks; determining the auxiliary parameters corresponding to each of the auxiliary robots according to the operating parameters of each of the auxiliary robots and the adjacency matrix; and weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots to obtain the characteristic information; Determining the auxiliary parameters corresponding to each of the auxiliary robots according to the operating parameters of each of the auxiliary robots and the adjacency matrix, including: obtaining the order of the auxiliary robot, wherein the order is used to indicate the closeness of the association between the auxiliary robot and the target robot when performing the same set of tasks; multiplying and transposing the adjacency matrices of the order to obtain an n-order adjacency matrix, wherein n is used to indicate the number of multiplied adjacency matrices; multiplying the operating parameters of each of the auxiliary robots with the n-order adjacency matrix corresponding to each of the auxiliary robots to obtain the auxiliary parameters of each of the auxiliary robots; The adjacency matrices of the order are multiplied and transposed to obtain an n-order adjacency matrix, including: performing row normalization on the adjacency matrix to obtain a row-normalized adjacency matrix; multiplying and transposing the normalized adjacency matrices of the order to obtain an n-order adjacency matrix; Weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots to obtain the characteristic information includes: obtaining weight values, and weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots according to the weight values ​​to obtain the characteristic information; wherein, the step of obtaining weight values ​​includes: obtaining constraints on the weight values, wherein the constraints include: the sum of the first weight value and all the second weight values ​​is 1, the first weight value is greater than any second weight value, and the second weight value of a multi-order auxiliary robot is the nth power of the second weight value of a first-order auxiliary robot, where n represents the order of the auxiliary robot; solving the constraints to obtain the weight values ​​corresponding to the target robot and each of the auxiliary robots.

2. The method according to claim 1, characterized in that Get the parameter prediction results of the target robot at the previous moment, including: Obtaining the output of the preset neural network model at the previous moment; Determine that the output of the preset neural network model at the previous moment is the parameter prediction result of the target robot at the previous moment.

3. The method according to claim 1, characterized in that When the operating parameters of any robot are missing, the weight value of the robot with the missing operating parameters is 0.

4. The method according to claim 1, characterized in that: According to the parameter prediction result of the target robot at the previous moment and the feature information at the current moment, the parameters to be predicted of the target robot at the next moment are predicted by a preset neural network model, including: Constructing a first relationship between a mean parameter and a parameter to be predicted in a normal distribution formula, and a second relationship between a variance parameter and the parameter to be predicted; The normal distribution formula is solved according to the first relationship and the second relationship by using a preset neural network model to obtain a prediction result of the parameter to be predicted.

5. The method according to claim 1, characterized in that The method further includes: obtaining the preset neural network model, wherein the step of obtaining the preset neural network model includes: Get the initial neural network model; Acquire training data, wherein the training data includes parameter prediction results of the sample robot at a previous moment and feature information of the sample robot at a current moment, wherein the feature information of the sample robot at the current moment is determined according to operating parameters of the sample robot and operating parameters of an auxiliary robot corresponding to the sample robot; Constructing a loss function, and training the initial neural network model using the training data, wherein the loss function is a logarithmic maximum likelihood function; When the function value of the loss function is less than a preset value, or the number of training iterations exceeds a preset number, the preset neural network model is obtained.

6. The method according to claim 1, characterized in that The preset neural network model is a gated recurrent network model.

7. The method according to claim 1, characterized in that After the parameter prediction result of the target robot at the previous moment and the feature information at the current moment, the method further includes: Determining a normal operating range according to distribution information of the parameter to be predicted, wherein the distribution information is obtained according to model parameters of the preset neural network model; If the prediction result of the parameter to be predicted is not within the normal operating range, early warning processing is performed on the prediction result.

8. A robot parameter prediction device, characterized in that: include: The first acquisition module is used to obtain the parameter prediction result of the target robot at the previous moment; a second acquisition module, configured to acquire characteristic information of the target robot at a current moment, wherein the characteristic information is obtained according to operating parameters of the target robot and operating parameters of an auxiliary robot, wherein the auxiliary robot is a robot that performs the same set of tasks as the target robot; A prediction module, used to predict the parameters to be predicted of the target robot at the next moment according to the parameter prediction results of the target robot at the previous moment and the feature information at the current moment, by using a preset neural network model; Wherein, obtaining the characteristic information of the target robot at the current moment includes: obtaining an adjacency matrix between the target robot and the auxiliary robot, wherein the adjacency matrix is ​​used to represent the association relationship between the target robot and the auxiliary robot when performing the same set of tasks; determining the auxiliary parameters corresponding to each of the auxiliary robots according to the operating parameters of each of the auxiliary robots and the adjacency matrix; and weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots to obtain the characteristic information; Determining the auxiliary parameters corresponding to each of the auxiliary robots according to the operating parameters of each of the auxiliary robots and the adjacency matrix, including: obtaining the order of the auxiliary robot, wherein the order is used to indicate the closeness of the association between the auxiliary robot and the target robot when performing the same set of tasks; multiplying and transposing the adjacency matrices of the order to obtain an n-order adjacency matrix, wherein n is used to indicate the number of multiplied adjacency matrices; multiplying the operating parameters of each of the auxiliary robots with the n-order adjacency matrix corresponding to each of the auxiliary robots to obtain the auxiliary parameters of each of the auxiliary robots; The adjacency matrices of the order are multiplied and transposed to obtain an n-order adjacency matrix, including: performing row normalization on the adjacency matrix to obtain a row-normalized adjacency matrix; multiplying and transposing the normalized adjacency matrices of the order to obtain an n-order adjacency matrix; Weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots to obtain the characteristic information includes: obtaining weight values, and weighting the operating parameters of the target robot and the auxiliary parameters of each of the auxiliary robots according to the weight values ​​to obtain the characteristic information; wherein, obtaining the weight values ​​includes: obtaining constraints on the weight values, wherein the constraints include: the sum of the first weight value and all the second weight values ​​is 1, the first weight value is greater than any second weight value, and the second weight value of a multi-order auxiliary robot is the nth power of the second weight value of a first-order auxiliary robot, where n represents the order of the auxiliary robot; solving the constraints to obtain the weight values ​​corresponding to the target robot and each of the auxiliary robots.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the parameter prediction of the robot according to any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein the program executes the parameter prediction of the robot according to any one of claims 1 to 7 when running.