Industrial robot energy consumption prediction method based on time convolution network

Abstract

This invention discloses an energy consumption prediction method for industrial robots based on temporal convolutional networks, which utilizes a joint trajectory variable time-scale scaling module to scale the reference joint trajectory q r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k This allows for the discretization of the power and energy consumption equations, yielding the relationship between power and energy consumption and joint position q. s There is a nonlinear causal relationship between (t), joint velocity, and joint acceleration; thus, with q s (t), and h k As input, the causal relationship between power and energy consumption is extracted using a temporal convolutional layer, thereby obtaining predicted values ​​of power and energy consumption. That is, the industrial robot energy consumption prediction method based on temporal convolutional networks of this invention, by running multiple robot joint trajectories and collecting their energy consumption values ​​and joint motion variable values, completes the energy consumption prediction of industrial robot joint trajectories under different time scaling rules under a preset task without obtaining any robot electrical and dynamic parameters.

Description

Technical Field

[0001] This invention belongs to the field of industrial robot technology, specifically a method for predicting the energy consumption of industrial robots based on temporal convolutional networks. Background Technology

[0002] Calculating the energy consumption of industrial robots requires obtaining their inertial parameters, such as mass, center of gravity, and inertia tensor; friction coefficients of the transmission system, such as Coulomb friction coefficient and viscous friction coefficient; and electrical parameters of the drive system, such as armature inductance, back electromotive force constant, motor torque constant, rotor moment of inertia of permanent magnet synchronous motors, and conduction and switching loss coefficients of rectifiers and inverters. However, due to the commercial nature of industrial robots, users often cannot obtain these parameters. Therefore, in engineering applications, the electrical and dynamic parameters required for calculating the energy consumption of industrial robots must be obtained through complex parameter experiments, which makes the calculation of industrial robot energy consumption labor-intensive and prone to significant errors. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide an industrial robot energy consumption prediction method based on temporal convolutional networks. By running multiple robot joint trajectories and collecting their energy consumption values ​​and joint motion variable values, the method can predict the energy consumption of industrial robot joint trajectories under different time scaling rules under a preset task without obtaining any robot electrical and dynamic parameters.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An energy consumption prediction method for industrial robots based on a temporal convolutional network, wherein the temporal convolutional network includes a joint trajectory variable time-scale scaling module, a data preprocessing layer, a temporal convolutional layer, and a power output and integration module; the method includes the following steps:

[0006] Step 1: Use the joint trajectory variable time scale scaling module to scale the reference joint trajectory q r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k Based on variable time-scale scaling of trajectory q s (t) and the time-scaled time interval sequence h k The power and energy consumption equations for the industrial robot are constructed as follows:

[0007]

[0008]

[0009] Among them, Pac (t) represents the power of the industrial robot; E ac (t) represents the energy consumption of the industrial robot; q represents a nonlinear causal relationship function; s (t) represents the trajectory scaled by a variable time scale, i.e., the joint position; Represents the trajectory q scaled by the variable time scale s (t) is the first derivative with respect to time t, i.e., the joint velocity; Represents the trajectory q scaled by the variable time scale s (t) is the second derivative with respect to time t, i.e., the joint acceleration; N is the number of control segments on the time axis;

[0010] Step 2: Using h k q s (t), and As input, the data preprocessing layer is used to standardize and normalize the input data sequence, and the data sequence is divided into training set, validation set and test set according to energy consumption level using a hierarchical sampling method;

[0011] Step 3: Input the training set into the temporal convolutional layer to extract the variable time-scaled trajectory q. s (t), joint velocity and joint acceleration The characteristic information of the industrial robot, as well as its power, energy consumption, and joint trajectory q. s (t), velocity acceleration The causal relationship between them;

[0012] Step 4: The power output and integration module obtains the predicted power value based on the causal relationship extracted from the time convolutional layer, and then obtains the power prediction sequence and energy consumption prediction sequence of the industrial robot.

[0013] Step 5: Based on the power prediction sequence and energy consumption prediction sequence of the industrial robot, determine whether the error between the predicted value and the actual value of power and energy consumption is less than the set threshold: if yes, proceed to step 7; if no, proceed to step 6.

[0014] Step 6: Update the weights of the temporal convolutional network using the backpropagation algorithm, and then execute Step 3;

[0015] Step 7: Verify and test the prediction accuracy of power and energy consumption using the validation set and test set respectively;

[0016] Step 8: Obtain the joint trajectory q of any desired industrial robot. s (t) is used to predict the power and energy consumption of an industrial robot when executing the trajectory using a temporal convolutional network.

[0017] Furthermore, in step one, the reference joint trajectory q r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k The method steps are as follows:

[0018] 11) For the reference joint trajectory q r (t r On the time axis t r The numbers u0, u1, u2, ..., u are evenly distributed on the top. N-1 ,u N There are a total of N+1 control time points, and the time axis is divided into N control segments. Let Δ = u k+1 -u k If 0 ≤ k ≤ N, then Δ is a constant;

[0019] 12) For the control section [u k ,u k+1 Joint reference trajectory q on ] r (t r Dynamic scaling with varying time scales is performed, and the functional relationship before and after scaling is as follows:

[0020] t r =u(t),

[0021] q s (t)=q r (t r )=q r (u(t)),

[0022]

[0023]

[0024] Where u is a variable time-scale scaling function; and , respectively, are the first and second derivatives of u with respect to t; t is the scaled joint trajectory execution time variable;

[0025] 13) Order In the control section [u k ,u k+1 If the value inside is a constant, then u , t in control segment [u k ,u k+1 The state-space expression that satisfies time scaling is as follows:

[0026]

[0027] Among them, hk This represents a time interval sequence scaled by a variable time scale, and:

[0028] h k =t k+1 -t k

[0029] Among them, t k For u k The control time points obtained after scaling by a variable time scale.

[0030] Furthermore, the temporal convolutional layer consists of n residual blocks connected in series. Each residual block includes a causal dilated convolution, a weight normalization function, a ReLU activation function, a DropOut random deactivation mechanism, and a 1×1 convolution.

[0031] Furthermore, the principle of the causal dilated convolution is as follows:

[0032]

[0033] Where F = (f1, f2, ..., f m ) is the convolution kernel, f i Let X be the i-th convolutional kernel, i = 1, 2, ..., m; X = (x0, x1, ..., xm) k ) represents the input data, x k Indicates the t-th k Input data at time step 1; * represents the convolution operator; M is the kernel size, and d is the dilation factor.

[0034] Furthermore, the temporal convolutional layer employs parallel computation to extract the variable time-scale scaling trajectory q. s (t), joint velocity and joint acceleration The characteristics of the robot and the causal relationship between its power and energy consumption and its joint trajectory, speed, and acceleration.

[0035] Furthermore, in step five, the mean square error (MSE) is used to characterize the power prediction error:

[0036]

[0037] Among them, P ac (t k ) is the control time point t k The measured power value, The control time point t is predicted by the temporal convolutional network. k The predicted power value; L is the number of trajectory samples with varying time scales; N is the number of control segments on the time axis;

[0038] Energy consumption prediction error is characterized by mean absolute percentage error (MAPE):

[0039]

[0040] Among them, E ac,i Let i be the total energy consumption value of the i-th trajectory; This represents the total energy consumption of the i-th trajectory output by the temporal convolutional network.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention presents an industrial robot energy consumption prediction method based on temporal convolutional networks. It constructs a temporal convolutional network comprising a joint trajectory variable time-scale scaling module, a data preprocessing layer, a temporal convolutional layer, and power output and integration modules. The variable time-scale scaling module is used to scale the reference joint trajectory q... r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k This allows for the discretization of the power and energy consumption equations, yielding the relationship between power and energy consumption and joint position q. s (t), joint velocity and joint acceleration There is a non-linear causal relationship between them, thus, with joint position q s (t), joint velocity Joint acceleration and the time interval sequence h with variable time scale k As input, power and energy consumption are extracted from joint trajectory q using a temporal convolutional layer. s (t), velocity acceleration The causal relationship between them can be used to obtain the predicted values ​​of industrial robot power and energy consumption. That is, the industrial robot energy consumption prediction method based on temporal convolutional networks of this invention can predict the energy consumption of industrial robot joint trajectories under different time scaling rules under preset tasks by running multiple robot joint trajectories and collecting their energy consumption values ​​and joint motion variable values, without obtaining any robot electrical and dynamic parameters. Attached Figure Description

[0043] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0044] Figure 1 This is a structural diagram of a temporal convolutional network;

[0045] Figure 2 A diagram showing the non-linear mapping between the reference timeframe and the scaled timeframe;

[0046] Figure 3 This is a structural diagram of a temporal convolutional layer;

[0047] Figure 4 This is a schematic diagram illustrating the principle of causal dilated convolution. Detailed Implementation

[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0049] The energy consumption prediction method for industrial robots based on temporal convolutional networks in this embodiment uses a temporal convolutional network that includes a joint trajectory variable time-scale scaling module, a data preprocessing layer, a temporal convolutional layer, and a power output and integration module. Figure 1 As shown. Specifically, the energy consumption prediction method for industrial robots based on temporal convolutional networks in this embodiment includes the following steps:

[0050] Step 1: Use the joint trajectory variable time scale scaling module to scale the reference joint trajectory q r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k .

[0051] Specifically, the joint trajectory q will be referenced. r (t r The uniform time interval sequence Δ is converted into a variable time-scaled trajectory q. s (t) and the time-scaled time interval sequence h k The method steps are as follows:

[0052] 11) For the reference joint trajectory q r (t r On the time axis t r The numbers u0, u1, u2, ..., u are evenly distributed on the top. N-1 ,u N There are a total of N+1 control time points, and the time axis is divided into N control segments. Let Δ = u k+1 -u k If 0 ≤ k ≤ N, then Δ is a constant.

[0053] 12) For the control section [u k ,u k+1 Joint reference trajectory q on ] r (t r Dynamic scaling with varying time scales is performed, and the functional relationship before and after scaling is as follows:

[0054] tr =u(t),

[0055] q s (t)=q r (t r )=q r (u(t)),

[0056]

[0057]

[0058] Where u is a variable time-scale scaling function; and Let be the first and second derivatives of u with respect to t, respectively; t is the scaled joint trajectory execution time variable. The mapping relationship between the time variables before and after scaling and t is as follows: Figure 2 As shown:

[0059] 13) Order In the control section [u k ,u k+1 If the value inside is a constant, then u , t in control segment [u k ,u k+1 The state-space expression that satisfies time scaling is as follows:

[0060]

[0061] Among them, h k This represents a time interval sequence scaled by a variable time scale, and:

[0062] h k =t k+1 -t k

[0063] Among them, t k For u k The control time points obtained after scaling by a variable time scale.

[0064] Based on trajectory q with variable time scale s (t) and the time-scaled time interval sequence h k The discrete expression form of the power and energy consumption equations of the industrial robot is constructed as follows:

[0065]

[0066]

[0067] Among them, P ac (t) represents the power of the industrial robot; E ac (t) represents the energy consumption of the industrial robot; q represents a nonlinear causal relationship function; s (t) represents the trajectory scaled by a variable time scale, i.e., the joint position; Represents the trajectory q scaled by the variable time scale s (t) is the first derivative with respect to time t, i.e., the joint velocity; Represents the trajectory q scaled by the variable time scale s (t) is the second derivative with respect to time t, i.e., joint acceleration; N is the number of control segments on the time axis.

[0068] Step 2: Using h k q s (t), and As input, the data preprocessing layer is used to standardize and normalize the input data sequence, and the data sequence is divided into training set, validation set and test set according to energy consumption level using a hierarchical sampling method.

[0069] Specifically, by setting The sequence, i.e., u0, u1, u2, ..., u N-1 ,u N There are a total of N+1 control time points. This value allows for dynamic scaling of the reference joint trajectory on each control segment across different time scales. Based on the target robot's maximum allowable execution speed, acceleration, and other conditions, the following values ​​are set. The range of values ​​is randomly generated to form a sufficient number of groups. The value of the sequence, and for The sequence is filtered, and the corresponding variable time-scaled trajectory q is obtained. s (t). A variable time-scale scaling trajectory is run on the target robot, and the actual joint positions, velocities, accelerations, and power data are measured. The obtained data sequences are standardized and normalized using a data preprocessing layer to construct a dataset.

[0070] Step 3: Input the training set into the temporal convolutional layer to extract the variable time-scaled trajectory q. s (t), joint velocity and joint acceleration The characteristic information of the industrial robot, as well as its power, energy consumption, and joint trajectory q. s (t), velocity acceleration The causal relationship between them.

[0071] In this embodiment, the temporal convolutional layer consists of n residual blocks concatenated. Each residual block includes a causal dilated convolution, a weight normalization function, a ReLU activation function, a DropOut random deactivation mechanism, and a 1×1 convolution, such as... Figure 3 As shown.

[0072] The principle of causal dilated convolution is as follows:

[0073]

[0074] Where F = (f1, f2, ..., f m ) is the convolution kernel, f i Let X be the i-th convolutional kernel, i = 1, 2, ..., m; X = (x0, x1, ..., xm) k ) represents the input data, x k Indicates the t-th k Input data at time step 1; * represents the convolution operator; M is the kernel size, and d is the dilation factor.

[0075] In this embodiment, the hole coefficient of the causal dilated convolution is set to base 2 and increases exponentially with the number of residual blocks (d = 2). n To increase the receptive field; when K=2, d increases exponentially by 2, the structure of causal dilated convolution is as follows: Figure 4 As shown, this structure has excellent ability to extract, filter, and memorize causal relationships of joint trajectory feature information.

[0076] Furthermore, since convolution is used to extract information from the input data, the information extraction process can be performed in parallel, which greatly improves computational efficiency. Specifically, in this embodiment, the temporal convolutional layer uses parallel computation to extract the variable time-scale scaling trajectory q. s (t), joint velocity and joint acceleration The characteristics of the robot and the causal relationship between its power and energy consumption and its joint trajectory, speed, and acceleration.

[0077] Step 4: The power output and integration module obtains the predicted power value based on the causal relationship extracted from the time convolutional layer, and then obtains the power prediction sequence and energy consumption prediction sequence of the industrial robot.

[0078] Step 5: Based on the power prediction sequence and energy consumption prediction sequence of the industrial robot, determine whether the error between the predicted value and the actual value of power and energy consumption is less than the set threshold. If yes, proceed to step 7; otherwise, proceed to step 6.

[0079] In this embodiment, the mean square error (MSE) is used to characterize the power prediction error:

[0080]

[0081] Among them, P ac (t k ) is the control time point t k The measured power value, The control time point t is predicted by the temporal convolutional network. k The predicted power value; L is the number of trajectory samples with varying time scales; N is the number of control segments on the time axis;

[0082] Energy consumption prediction error is characterized by mean absolute percentage error (MAPE):

[0083]

[0084] Among them, E ac,i Let i be the total energy consumption value of the i-th trajectory; This represents the total energy consumption of the i-th trajectory output by the temporal convolutional network.

[0085] Step Six: Update the weights of the temporal convolutional network using the backpropagation algorithm, and then execute Step Three. Specifically, in this embodiment, the Glorot initializer is used to initialize the weights of the temporal convolutional network.

[0086] Step 7: Verify and test the prediction accuracy of power and energy consumption using the validation set and test set respectively.

[0087] Step 8: Obtain the joint trajectory q of any desired industrial robot. s (t) is used to predict the power and energy consumption of an industrial robot when executing the trajectory using a temporal convolutional network.

[0088] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A method for predicting the energy consumption of industrial robots based on temporal convolutional networks, characterized in that: The temporal convolutional network includes a joint trajectory time-scale scaling module, a data preprocessing layer, a temporal convolutional layer, and a power output and integration module. Each temporal convolutional layer consists of n residual blocks connected in series. Each residual block includes a causal dilated convolution, a weight normalization function, a ReLU activation function, a DropOut random deactivation mechanism, and a 1×1 convolution. The dilation factor d of the causal dilated convolution is base 2 and increases exponentially with the number of residual block levels n, i.e., d = 2^n. The temporal convolutional layer uses parallel computation to extract feature information and causal relationships. The method includes the following steps: Step 1: Use the joint trajectory time-scale scaling module to zoom the reference joint trajectory and uniform time interval sequence Convert to variable time-scaled trajectory and variable time scale scaling time interval sequence ; Reference joint trajectory and uniform time interval sequence Convert to variable time-scaled trajectory and variable time scale scaling time interval sequence The method steps are as follows: 11) Reference joint trajectory On the timeline Evenly distributed on top There are a total of N+1 control time points, and the time axis is divided into N control segments, let , ,but It is a constant; 12) Control section Joint reference trajectory For dynamic scaling with varying time scales, the functional relationship before and after scaling is as follows: in, For variable time scale scaling functions; and They are respectively right The first and second derivatives; The execution time variable is used for the scaled joint trajectory; 13) Order In the control section If the inner is a constant, then , In the control section The state-space expression that satisfies time scaling is as follows: in, This represents a time interval sequence scaled by a variable time scale, and: in, for The control time points obtained after scaling by a variable time scale; Trajectory based on variable time scale and variable time scale scaling time interval sequence The power and energy consumption equations for the industrial robot are constructed as follows: in, Indicates the power of the industrial robot; Indicates the energy consumption of industrial robots; Represents a nonlinear causal relationship function; This represents the trajectory scaled by a variable time scale, i.e., the joint position; Represents a trajectory scaled by a variable time scale Regarding time The first derivative, i.e., joint velocity; Represents a trajectory scaled by a variable time scale Regarding time The second derivative, i.e., joint acceleration; The number of control segments on the time axis; Step Two: , , and As input, the data preprocessing layer is used to standardize and normalize the input data sequence, and the data sequence is divided into training set, validation set and test set according to energy consumption level using a hierarchical sampling method; Step 3: Input the training set into the temporal convolutional layer to extract the variable time-scaled trajectory. Joint velocity and joint acceleration Feature information, power and energy consumption of industrial robots, and variable time-scale scaling trajectories. Joint velocity Joint acceleration The causal relationship between them; Step 4: The power output and integration module obtains the predicted power value based on the causal relationship extracted from the time convolutional layer, and then obtains the power prediction sequence and energy consumption prediction sequence of the industrial robot. Step 5: Based on the power prediction sequence and energy consumption prediction sequence of the industrial robot, determine whether the error between the predicted value and the actual value of power and energy consumption is less than the set threshold: if yes, proceed to step 7; if no, proceed to step 6. Step 6: Update the weights of the temporal convolutional network using the backpropagation algorithm, and then execute Step 3; Step 7: Verify and test the prediction accuracy of power and energy consumption using the validation set and test set respectively; Step 8: Obtain the variable time-scale scaling trajectory of any industrial robot to be executed. The power and energy consumption of an industrial robot executing this trajectory are predicted using a temporal convolutional network. The principle of the causal dilated convolution is as follows: in, For convolution kernel, For the first One convolutional kernel, ; For input data, Indicates the first Input data at any given time; This is the convolution operator; The kernel size is [size]. It is the expansion factor; The temporal convolutional layer uses parallel computing to extract variable time-scale scaling trajectories. Joint velocity and joint acceleration The characteristics and causal relationships between the power and energy consumption of industrial robots and the variable time-scale scaling trajectory, joint speed, and joint acceleration.

2. The industrial robot energy consumption prediction method based on temporal convolutional networks according to claim 1, characterized in that: In step five, the mean square deviation is used. MSE Characterizing power prediction error: in, To control the time point The measured power value, The control time points predicted by the temporal convolutional network The predicted power value; To scale the number of trajectory samples for varying time scales; The number of control segments on the time axis; Energy consumption prediction error is characterized by mean absolute percentage error (MAPE): in, For the first The total energy consumption of a track scaled by a variable time scale; The output of the temporal convolutional network is the first The total energy consumption of the track scaled by the variable time scale.

Images