Remote control time delay supplement method for large equipment

Abstract

This invention belongs to the field of remote control technology, specifically a method for compensating for delays in remote control of large equipment. The method includes the following steps: Step 1: Establishing a control prototype dataset, generated by sampling data within a unit circle; specifically, the dataset formation involves collecting data during online or manual control of the large equipment. Step 2: Piecewise continuous function approximation, where the input features and output have a clear nonlinear relationship. This is used to linearly segment and classify the control data of the large equipment, allowing each data point to infinitely approximate a linear piecewise continuous function. This method for compensating for delays in remote control of large equipment can memorize, extract, and automatically design actions based on the motion data features in the three-dimensional coordinate system of the large equipment. For example, the pressure and flow values ​​of the hydraulic cylinder piston rod extension at key control nodes, combined with the current motion trajectory, can yield the control dataset for the current hydraulic cylinder position.

Description

Technical Field

[0001] This invention relates to the field of remote control technology, and in particular to a method for compensating for delays in remote control of large equipment. Background Technology

[0002] In existing technologies, with the development of wireless remote control technology, based on 5G technology, more and more large-scale equipment is adopting wireless control technology. This serves two purposes: firstly, it enables personnel protection in complex and dangerous environments; secondly, it allows for resource exploration and extraction in environments unsuitable for human access. Because large equipment generally operates slowly, the latency issues associated with remote control can be effectively avoided. However, due to the large size of the equipment, even slow-moving remote control requires consistent execution; otherwise, it can easily lead to danger. Current solutions mostly rely on strengthening network infrastructure and maintaining continuous network supply to address this issue. However, these solutions are still affected by network base stations, satellites, weather, and the location of the large equipment. Network latency or outages can easily damage large equipment, resulting in incalculable losses. Summary of the Invention

[0003] To address the technical problem that existing methods for remote control of large equipment rely solely on a single network support path, this invention proposes a delay compensation method for remote control of large equipment.

[0004] The present invention proposes a method for compensating for delays in remote control of large equipment, comprising the following steps:

[0005] Step 1: Establish the prototype dataset by sampling data within a unit circle; the formation of the specific dataset includes the collection of data during online or manual operation of large equipment.

[0006] Step 2: Piecewise continuous function approximation, where the input features and output have a clear nonlinear relationship, thereby linearly segmenting and classifying the control data of large equipment, so that each data point infinitely approximates the linear piecewise continuous function.

[0007] Step 3: The formal definition of the control indication function is used to assign a unique constant value to a point in the control set S without delay, so that the corresponding logical approximation closely matches the original indication function.

[0008] Step 4: Indicator function approximation, used to approximate the zero-delay control.

[0009] Step 5: Obtain the weights, discretize the function, and apply it to real-world remote control of large-scale equipment with no latency.

[0010] Step 6: Set up a neural network accelerator; this will enable the neural network to accelerate the computation of data packets.

[0011] Preferably, the control data without delay includes the acquisition or control of data information on the hydraulic cylinders, hydraulic pumps, and solenoid valves necessary for the hydraulic system on the large equipment. The position information of the current piston rod of the hydraulic cylinder is obtained by controlling the pressure valves and flow valves on the hydraulic cylinder pipeline, and the start and stop actions are controlled by controlling the solenoid valves to realize the acquisition or control of data information on the three-dimensional coordinate system of the large equipment.

[0012] Preferably, the motion data features in the three-dimensional coordinate system of large equipment are memorized, extracted, and automatically designed. The motion data in the three-dimensional coordinate system includes the pressure value and flow value of the hydraulic cylinder piston rod extension at the key control node. Combined with the current motion trajectory, the control dataset of the current hydraulic cylinder position can be obtained.

[0013] Preferably, in step one, the prototype dataset is established by generating an indicator function through sampling data within a unit circle. Where x is the input feature vector, i.e., the device status data;

[0014] To obtain zero-delay operation data samples, the zero-delay operation setting label value of large equipment is taken from {-1,1}, and the corresponding indicator function y(x) is accessed through a neural network and represented using continuous feature basis.

[0015] Preferably, in step two, the data generates a function based on the data within a unit interval. The control involves delayed sampling.

[0016] In step two, a piecewise continuous function approximation is performed. The piecewise continuous function is composed of several continuous function segments, with intervals or jumps between each segment.

[0017] Preferably, in step two, the piecewise continuous function approximation is improved by increasing the number of basis elements M, thereby comprehensively enhancing the approximation in the following formula:

[0018] ;

[0019] M is the index of the basis functions, used to construct the basis function number of the approximation function. Increasing M can improve the approximation accuracy.

[0020] Preferably, in step three, a unique constant value is assigned to each point in the no-delay control set S, and a different constant value is assigned to other points not in set S. The indicator function is defined on set S by using +1 to indicate that a point is in set S and -1 to indicate that a point is not in set S:

[0021] .

[0022] Since the indicator function is a piecewise constant, that is, a special subclass of piecewise continuous functions, it can be effectively approximated using any basis and a sufficiently large M, thereby enabling remote, delay-free control of large equipment.

[0023] Preferably, since the indicator function in step four takes values ​​of y(x) within ±1, when approximating the indicator function, a weighted sum is used and the sign is substituted. In this context, the approximation formula is simplified as follows:

[0024] ; The weights are the weights of each basis function or neuron connection. Then, through logistic regression, the logistic function tanh(αt) is used to approximate sign(t). The logistic approximation of the indicator function y(x) is:

[0025] .

[0026] Preferably, obtaining the weights in step five includes taking values ​​for each x and y(x) on {±1}, and deriving the softmax cost function for logistic regression as follows:

[0027] .

[0028] Step five involves obtaining weights, including adjusting the weights. and the internal parameters, minimize the logical approximation in the formula over all x, that is:

[0029] .

[0030] Preferably, the neural network accelerator in step six includes an architecture consisting of input vectors, weights and activated memory, and multiple data pathways for performing neuron computations.

[0031] The beneficial effects of this invention are as follows:

[0032] It can memorize, extract, and automatically design motion data features in the three-dimensional coordinate system of large equipment. For example, the extension amount of the hydraulic cylinder piston rod at key control nodes, pressure value, and flow value, combined with the current motion trajectory, can yield the control dataset of the current hydraulic cylinder position. When a delay occurs in remote network control, the features of the original data are back-matched, and the automatic design of motion using the features replaces the continuity of the large equipment control motion until the network is normal, and then it returns to normal. Attached Figure Description

[0033] Figure 1This is a schematic diagram of the original space top-down binary classification real dataset and the original space dataset viewed from a side view, which is the basis for the delay compensation method for remote control of large equipment proposed in this invention.

[0034] Figure 2 This is a schematic diagram of a piecewise continuous function for a delay compensation method for remote control of large equipment proposed in this invention.

[0035] Figure 3 This is a schematic diagram illustrating four sets S of a delay compensation method for remote control of large equipment proposed in this invention.

[0036] Figure 4 This is a schematic diagram of the accelerator architecture for a method to compensate for delays in remote control of large equipment proposed in this invention. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0038] Reference Figure 1-4 A method for compensating for delays in remote control of large equipment includes the following steps:

[0039] Step 1: Establish the prototype dataset by sampling data within a unit circle; the formation of the specific dataset includes the collection of data during online or manual operation of large equipment.

[0040] Real-time control data can be used to control hydraulic cylinders, hydraulic pumps, and necessary solenoid valves in the hydraulic system of large equipment. This data is obtained by monitoring the pressure and flow valves on the hydraulic cylinder pipelines to acquire the current position information of the piston rod. The start and stop actions are then controlled by controlling the solenoid valves. Similarly, if the large equipment also rotates, angle sensors are used to acquire the rotation angle information. After the above operations, data information in the three-dimensional coordinate system of the large equipment can be collected or controlled to achieve data-driven control of the large equipment. Furthermore, the automatic feature design module in this category is used to collect features from the above-mentioned data, and the point sets formed by the data are used to automatically generate features.

[0041] Specifically, in the operation of large equipment, the motion data features in the three-dimensional coordinate system of the large equipment are memorized, extracted, and automatically designed. For example, the pressure and flow values ​​of the hydraulic cylinder piston rod extension at key control nodes, combined with the current motion trajectory, can yield the control dataset for the current hydraulic cylinder position. When these control datasets are placed into the automatic feature design module, the process is as follows:

[0042] Furthermore, in step one, the prototype dataset is established, such as... Figure 1 As shown, an indicator function is generated by sampling data within a unit circle. Where x is the input feature vector, i.e., the device status data;

[0043] Obtaining zero-delay control data samples mainly involves using available large-scale equipment control data to approximate the data generation indicator function as closely as possible.

[0044] Furthermore, in this application, the delay-free operation setting label value of large equipment is taken from {-1,1}, and the corresponding indicator function y(x) is accessed through a neural network and represented using a continuous feature basis.

[0045] like Figure 2 As shown, step two involves piecewise continuous function approximation, where the input features and output have a clear nonlinear relationship. This is used to linearly segment and classify the control data of large equipment, so that each data point infinitely approximates the linear piecewise continuous function.

[0046] Furthermore, in step two, the data generates a function within a unit interval. The control involves delayed sampling.

[0047] Furthermore, in step two, the piecewise continuous function approximation is performed. The piecewise continuous function is composed of several continuous function segments, with intervals or jumps between each segment.

[0048] Furthermore, in step two, the piecewise continuous function approximation is improved by increasing the number of basis elements M, thus comprehensively enhancing the approximation in the following formula:

[0049]

[0050] M is the index of the basis functions, used to construct the basis function number of the approximation function. Increasing M can improve the approximation accuracy.

[0051] like Figure 3 As shown, step three involves the formal definition of the control instruction function, which assigns a unique constant value to each point in the delay-free control set S. This ensures that the corresponding logical approximation closely matches the original instruction function, thereby achieving data classification for large equipment. After establishing a linear relationship, when a delay occurs in remote network control, the characteristics of the original data are matched in reverse. The automatic design of actions based on these characteristics replaces the continuity of the large equipment control actions until the network returns to normal, at which point the system returns to normal.

[0052] Furthermore, in step three, a unique constant value is assigned to each point in the no-delay control set S, and a different constant value is assigned to other points not in set S. By using +1 to indicate that a point is in set S and -1 to indicate that a point is not in set S, the indicator function is defined on set S as follows:

[0053] .

[0054] Furthermore, since the indicator function is a piecewise constant, that is, a special subclass of piecewise continuous functions, it can be effectively approximated using any basis and a sufficiently large M, thereby enabling remote, delay-free control of large equipment.

[0055] Step 4: Indicator function approximation, used to approximate the zero-delay control.

[0056] Furthermore, since the indicator function in step four takes values ​​of y(x) within ±1, when approximating the indicator function, a weighted sum is used to substitute the sign. In this context, the approximation formula is simplified as follows:

[0057] Here are the weight parameters, i.e., the weights of each basis function or neuron connection. Then, through logistic regression, the logistic function tanh(αt) is used to approximate sign(t), and the logistic approximation of the indicator function y(x) is:

[0058] .

[0059] Step 5: Obtain the weights, discretize the function, and apply it to real-world remote control of large-scale equipment with no latency.

[0060] Furthermore, obtaining the weights in step five involves taking values ​​for y(x) on {±1} for each x, and deriving the softmax cost function for logistic regression as follows:

[0061] .

[0062] Furthermore, obtaining the weights in step five includes adjusting the weights. and the internal parameters, minimize the logical approximation in the formula over all x, that is:

[0063] .

[0064] To facilitate rapid computation within data packets, step six involves setting up a neural network accelerator. This accelerator uses a neural network to accelerate computation on the data packets. During remote control, the receiving end of the network transmission needs a mechanism to filter out network latency jitter to obtain valid data packets. Existing technologies employ minimum latency techniques to filter out network latency jitter, thereby obtaining valid time-based data packets.

[0065] like Figure 4 As shown, the specific implementation process is as follows: within a certain time window, such as 1 second, the time data packet with the smallest transmission delay is selected, and the minimum delay within this time window is calculated using this time data packet. However, this calculation process requires a large amount of computation. Using a neural network to calculate the data within the data packet not only greatly improves the calculation speed but also enables large-scale equipment to have intelligent effects. However, when using a neural network to calculate the data within the data packet, the required calculation speed needs to be guaranteed. Therefore, to improve the calculation speed of the neural network, a neural network accelerator needs to be installed in the large-scale equipment.

[0066] The neural network accelerator includes an architecture designed to achieve low-power hardware acceleration for executing highly accurate neural networks within data packets. The accelerator consists of input vectors, weights, and activation memory, as well as multiple data pathways for performing neuron computations. This allows each accelerator to execute the same neural network, ultimately accelerating the computation of data packets within the neural network.

[0067] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A large equipment remote control delay supplement method, characterized by: Includes the following steps: Step 1: Establish the prototype dataset by sampling data within a unit circle; the specific dataset formation includes the collection of data during online or manual operation of large equipment. Step 2: Piecewise continuous function approximation, where the input features and output have a clear nonlinear relationship, thereby linearly segmenting and classifying the control data of large equipment, so that each data point infinitely approximates the linear piecewise continuous function; The step two data in the unit interval for data generation function The operation has a delay sampling; In step two, the piecewise continuous function approximation is performed. The piecewise continuous function is composed of several continuous function segments, with intervals or jumps between each segment. In step two, the piecewise continuous function approximation is improved by increasing the number of basis elements M, thus significantly enhancing the approximation in the following formula: ； is an index of the basis function, and is the number of basis functions used to construct the approximation function. Increasing M can improve the approximation accuracy. Step 3: The formal definition of the control indication function is used to assign a unique constant value to a point in the control set S without delay, so that the corresponding logic approximation closely matches the original indication function. Step 4: Indicator function approximation, used to approximate the zero-delay control; Step 5: Obtain the weights, discretize the function, and apply it to real-world remote control of large-scale equipment with no latency. Step 6: Set up a neural network accelerator; this will enable the neural network to accelerate the computation of data packets.

2. The method of claim 1, wherein the method further comprises: The real-time control data includes the hydraulic cylinders, hydraulic pumps, and solenoid valves necessary for the hydraulic system on large equipment. It obtains the current position information of the piston rod of the hydraulic cylinder by controlling the pressure valves and flow valves on the hydraulic cylinder pipeline, and then controls the start and stop actions by controlling the solenoid valves to realize the acquisition or control of data information in the three-dimensional coordinate system of the large equipment.

3. The method of claim 2, wherein the method further comprises: The motion data features of large equipment in the three-dimensional coordinate system are memorized, extracted, and automatically designed. The motion data in the three-dimensional coordinate system includes the pressure value and flow value of the hydraulic cylinder piston rod extension at the key control node. Combined with the current motion trajectory, the control dataset of the current hydraulic cylinder position can be obtained.

4. The method of claim 1, wherein the method further comprises: In step one, the prototype dataset is established by generating an indicator function through data sampling within a unit circle. Where x is the input feature vector, i.e., the device status data; To obtain zero-delay operation data samples, the zero-delay operation setting label value of large equipment is taken from {-1,1}, and the corresponding indicator function y(x) is accessed through a neural network and represented using continuous feature basis.

5. The method for compensating for delays in remote control of large equipment according to claim 1, characterized in that: In step three, a unique constant value is assigned to each point in the no-delay control set S, and a different constant value is assigned to other points not in set S. The indicator function is defined on set S by using +1 to indicate that a point is in set S and -1 to indicate that a point is not in set S: ； Since the indicator function is a piecewise constant, that is, a special subclass of piecewise continuous functions, it can be effectively approximated using any basis and a sufficiently large M, thereby enabling remote, delay-free control of large equipment.

6. The method of claim 5, wherein the method further comprises: Since the indicator function in step four takes values ​​of y(x) within ±1, when approximating the indicator function, a weighted sum is used to substitute the sign. In this context, the approximation formula is abbreviated as: ; The weights are the weights of each basis function or neuron connection. Then, through logistic regression, the logistic function tanh(αt) is used to approximate sign(t). The logistic approximation of the indicator function y(x) is: 。 7. The method of claim 6, wherein the method further comprises: Step five involves obtaining weights by taking values ​​for each x and y(x) on the interval {±1}, and deriving the softmax cost function for logistic regression as follows: ； Step five involves obtaining weights, including adjusting the weights. and the internal parameters, minimize the logical approximation in the formula over all x, that is: 。 8. The method for compensating for delays in remote control of large equipment according to claim 1, characterized in that: The neural network accelerator in step six includes an architecture consisting of input vectors, weights and activated memory, and multiple data pathways for performing neuron computations.

Images