Digital-twin-driven ultra-deep well hoisting system steel wire rope tension learning control method

Abstract

The present application relates to a digital twin driven ultra-deep well hoisting system steel wire rope tension learning control method. The method comprises: constructing a digital twin virtual model of the ultra-deep well hoisting system, wherein the digital twin virtual model comprises a physical layer twin model, a system layer twin model and an information layer twin model; constructing a tension control model of the steel wire rope of the ultra-deep well hoisting system; based on the digital twin virtual model and the tension control model, updating learning is carried out through an evaluation neural network to obtain a system control law u; the control voltage u of an electro-hydraulic servo valve is determined according to the system control law u L , and the hydraulic cylinder is driven to adjust the tension of the steel wire rope with the control voltage u L . In this way, by using the real-time interaction data of the digital twin model driven twin body and the physical entity, the dynamics behavior of the ultra-deep well hoisting system is learned in real time by using deep learning technology, an intelligent learning control method is constructed, and the tension of the steel wire rope of the ultra-deep well hoisting system is adjusted more accurately, so that the tension difference is smaller.

Description

Technical Field

[0001] This invention relates to the field of mining computer technology, specifically to a digital twin-driven method for learning and controlling the tension of wire ropes in ultra-deep well hoisting systems. Background Technology

[0002] Ultra-deep well hoisting systems are vertical shaft hoisting systems for mining depths greater than 1500m. Due to their depth, a dual-rope winding hoisting system is required. However, during the hoisting or lowering of the hoisting container, factors such as differences in drum diameter manufacturing, installation differences between the two wire ropes, and inconsistent elastic moduli can cause asynchronous movement at the ends of the two wire ropes, resulting in inconsistent tension. Prolonged operation under these conditions can easily lead to one wire rope exceeding its safe operating stress, causing a serious accident such as rope breakage. To avoid such accidents, it is necessary to actively adjust the tension of the two wire ropes to maintain as consistent a tension as possible.

[0003] The most commonly used control methods in the current technology are model-based control methods, but these control methods require complex and detailed modeling of the hoisting system. Since the hoisting system of an ultra-deep vertical shaft is a complex electromechanical-hydraulic system with multiple structures, especially since the wire rope is a typical continuous flexible body, it is difficult to obtain an accurate model of the wire rope. Summary of the Invention

[0004] The purpose of this invention is to provide a digital twin-driven method for learning and controlling the tension of wire ropes in ultra-deep well hoisting systems, thereby more accurately adjusting the tension of wire ropes in ultra-deep well hoisting systems and reducing tension differences.

[0005] To achieve the above objectives, this invention provides a digital twin-driven method for learning and controlling the tension of wire ropes in an ultra-deep well hoisting system, the method comprising:

[0006] A digital twin virtual model of an ultra-deep well hoisting system is constructed. The digital twin virtual model is used for real-time transmission of operating data in ultra-deep wells and for simulating the operating status of equipment in ultra-deep wells. The digital twin virtual model includes a physical layer twin model, a system layer twin model, and an information layer twin model.

[0007] A tension control model for the wire rope in an ultra-deep well hoisting system is constructed, wherein the tension control model is used to control the tension difference between two wire ropes in the ultra-deep well hoisting system in order to minimize the tension difference;

[0008] Based on the digital twin virtual model and the tension control model, the system control law u is obtained by updating and learning through an evaluation neural network.

[0009] The control voltage u of the electro-hydraulic servo valve is determined based on the system control law u. Land with the control voltage u L The hydraulic cylinder is driven to adjust the tension of the wire rope.

[0010] Optionally, the tension control model is constructed based on the following parameters: wire rope tension F z The angle γ between the wire rope and the plane of the upper platform, and the load pressure drop P of the hydraulic cylinder. L Stiffness k of tension / compression sensor f The effective working area A of the hydraulic cylinder piston rod P The total mass m of the hydraulic cylinder load, and the viscous damping coefficient B of the hydraulic cylinder. p The total leakage coefficient C of the hydraulic cylinder tl The total effective volume V of the hydraulic cylinder's inlet and outlet oil chambers t The bulk modulus β of hydraulic fluid e The load flow rate Q of the hydraulic cylinder L .

[0011] Optionally, the tension control model is shown in the following state space:

[0012]

[0013] Wherein, the system state variable is x e , F z P represents the tension of the wire rope. L k represents the load pressure drop of the hydraulic cylinder. f A represents the stiffness of the tension / compression sensor. P B represents the effective working area of ​​the hydraulic cylinder piston rod, m represents the total mass of the hydraulic cylinder load, and B represents the effective working area of ​​the hydraulic cylinder piston rod. p C represents the viscous damping coefficient of the hydraulic cylinder. tl V represents the total leakage coefficient of the hydraulic cylinder. t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e Let represent the bulk modulus of elasticity of the hydraulic fluid, d represent external disturbances, γ represent the angle between the wire rope and the plane containing the upper platform, and x1, x2, and x3 are defined system state variables. The speed of the floating headwheel. The speed at which the hydraulic piston rod moves.

[0014] Optionally, the step of determining the control voltage u of the electro-hydraulic servo valve according to the system control law u is... L ,include:

[0015] The system control law u is used as the load flow rate Q of the hydraulic cylinder. L The control voltage u is obtained through the following formula. L :

[0016]

[0017] Among them, Q r This indicates the electro-hydraulic servo valve at the rated pressure drop Δp r The rated flow rate, p s P represents the pressure of the hydraulic oil source. L Indicates the load pressure drop of the hydraulic cylinder, u max This is the maximum control voltage of the electro-hydraulic servo valve.

[0018] Optionally, the load flow rate Q of the hydraulic cylinder L This can be expressed by the following formula:

[0019]

[0020] Among them, the discharge coefficient C of the electro-hydraulic servo valve d The throttling window area gradient w of the servo valve and k are both estimated values. v p is the proportional coefficient of the electro-hydraulic servo valve. s ρ is the pressure of the hydraulic oil source, and ρ0 is the density of the hydraulic oil.

[0021] Optionally, the system control law u is obtained by the following formula:

[0022]

[0023] in, It is a positive definite matrix. Let W be the activation function, and W represent the weights. Let B be an estimate of W, and B = [0, 0, θ6]. T , V t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e This indicates the bulk modulus of elasticity of hydraulic fluid.

[0024] Optionally, the system control law u is obtained through the optimal control law u. * The obtained optimal control law u * It can be obtained through the following formula:

[0025]

[0026] in, B is a positive definite matrix, [0, 0, θ6] T , V t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e Let J(x) represent the bulk modulus of the hydraulic fluid, J*(x) be the penalty function, and J*(x) be the optimized value of J(x). It is the derivative of J*(x).

[0027] Optionally, the physical layer twin model is used for simulation modeling of ultra-deep downhole equipment. The physical layer twin model includes a system physical ontology, an environmental physical ontology, a sensor system, a wireless sensor network, and a real-time control system.

[0028] The physical components of the system include an asynchronous motor, a double drum, a hoisting container, a hoisting wire rope, a hoisting system mechanical mechanism, a floating sheave, a hydraulic cylinder, hydraulic pipelines, and a pump station. The physical components of the environment include the hoisting shaft, the environment of the hoisting shaft bottom platform, and the environment of the mine platform. The sensor system includes a pin force sensor, a displacement sensor, a tension sensor, a hydraulic pressure sensor, an acceleration sensor, a tilt sensor, and a rotary encoder sensor. The wireless sensor network includes a wireless network transmitting node, a wireless network receiving node, and a power supply system. The real-time control system includes a signal acquisition and conditioning system, an industrial control computer real-time system, a host computer real-time monitoring system, a power supply system, and power cables.

[0029] Optionally, the system-level twin model is used to simulate the operating status of the lifting container, the tension status of the wire rope, the status of the hydraulic cylinder, the status of the floating sheave, and the operating status of the wireless sensor network in a virtual environment, and to generate test data through the digital twin prototype.

[0030] Optionally, the information layer twin model is used to transmit real-time status information data collected by sensors during the operation of the ultra-deep well hoisting system, so that the system layer twin model drives a 3D visualization engine to render and construct the digital twin virtual model based on the real-time status information data. The real-time status information data includes the hoisting system's speed, position, acceleration, fluid pressure, tilt signal, and wire rope tension. The pin force sensor and displacement sensor are installed on the hydraulic cylinder, the tension sensor is installed and connected to the wire rope, the acceleration sensor and tilt sensor are installed on the hoisting container, and the wireless sensors are distributed in the pipeline shaft.

[0031] Through the above technical solution, based on the digital twin virtual model and the tension control model, an evaluation neural network is used for updating and learning, thereby minimizing the tension difference between the two steel wire ropes in the ultra-deep well hoisting system. The generated virtual model can be visualized, interacted with naturally, and monitored from multiple perspectives both online in real time and offline in non-real time. It can also visualize and predict information such as working conditions, improving the intelligence level of the dual-rope winding ultra-deep vertical shaft hoisting system. By utilizing the real-time interaction data between the digital twin model-driven twin and the physical entity, and using deep learning technology, the dynamic behavior of the ultra-deep well hoisting system is learned in real time, constructing an intelligent learning control method to more accurately adjust the tension of the steel wire ropes in the ultra-deep well hoisting system, making its tension difference smaller. Attached Figure Description

[0032] Figure 1 This is an exemplary flowchart of a digital twin-driven wire rope tension learning control method for an ultra-deep well hoisting system.

[0033] Figure 2 This is a schematic diagram of a double-rope winding hoist.

[0034] Figure 3 This is a schematic diagram of the active tension adjustment system of a mine hoist.

[0035] Figure 4 This is a schematic diagram for evaluating neural network structures.

[0036] Figure 5a and Figure 5b This is a schematic diagram of the tension difference in a wire rope obtained using the method of the present invention.

[0037] Figure 6a and Figure 6b This is a schematic diagram of the tension difference in a wire rope obtained using methods from related technologies. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] Figure 1 This is an exemplary flowchart of a digital twin-driven wire rope tension learning control method for ultra-deep well hoisting systems, as shown below. Figure 1 As shown, the method may include S1 to S4.

[0040] S1. Construct a digital twin virtual model of an ultra-deep well hoisting system, wherein the digital twin virtual model is used for real-time transmission of operating data in ultra-deep wells and for simulating the operating status of equipment in ultra-deep wells. The digital twin virtual model includes a physical layer twin model, a system layer twin model, and an information layer twin model.

[0041] S2. Construct a tension control model for the steel wire rope in the ultra-deep well hoisting system, wherein the tension control model is used to control the tension difference between the two steel wire ropes in the ultra-deep well hoisting system in order to minimize the tension difference;

[0042] S3. Based on the digital twin virtual model and the tension control model, the system control law u is obtained by updating and learning through an evaluation neural network;

[0043] S4. Determine the control voltage u of the electro-hydraulic servo valve according to the system control law u. Land with the control voltage u L The hydraulic cylinder is driven to adjust the tension of the wire rope.

[0044] Figure 2 This is a schematic diagram of a dual-rope winding hoist. In this invention, the physical layer twin model is used for simulation modeling of ultra-deep downhole equipment. The physical layer twin model includes a system physical entity, an environmental physical entity, a sensor system, a wireless sensor network, and a real-time control system.

[0045] The physical components of the system include an asynchronous motor, a double drum, a hoisting container, a hoisting wire rope, a hoisting system mechanical mechanism, a floating sheave, a hydraulic cylinder, hydraulic pipelines, and a pump station. The physical components of the environment include the hoisting shaft, the environment of the hoisting shaft bottom platform, and the environment of the mine platform. The sensor system includes a pin force sensor, a displacement sensor, a tension sensor, a hydraulic pressure sensor, an acceleration sensor, a tilt sensor, and a rotary encoder sensor. The wireless sensor network includes a wireless network transmitting node, a wireless network receiving node, and a power supply system. The real-time control system includes a signal acquisition and conditioning system, an industrial control computer real-time system, a host computer real-time monitoring system, a power supply system, and power cables.

[0046] Specifically, a 3D ontology model of the ultra-deep well hoisting system was constructed in ADAMS software according to the physical dimensions of the physical entities. The assembly motion relationships of each physical entity were defined. The 3D model of the ultra-deep well hoisting system was then imported into the Unity game engine. The relationships between the various components of the system were clarified, and the assembly relationships were re-bound in Unity to build a digital twin prototype of the physical layer. This enabled the virtual prototype to realize all the motion postures of the physical prototype. This achieved the physical layer twin of the ultra-deep well hoisting system.

[0047] In this invention, the system-level twin model is used to simulate the operating status of the lifting container, the tension status of the wire rope, the status of the hydraulic cylinder, the status of the floating sheave, and the operating status of the wireless sensor network in a virtual environment, and to generate test data through the digital twin prototype.

[0048] System-level twins are simulations of the software aspects of ultra-deep well hoisting systems. Through programming, abstract layers are replicated, and information systems such as the hoisting container status, wire rope tension status, hydraulic cylinder and floating sheave status, and wireless sensor network status are built in a virtual environment. This drives the digital twin prototype to interact with the wellbore environment, wireless sensor network data packet loss and latency, and the well tower environment, generating test data. The results are compared and the program algorithm is improved. Developing and testing code in a virtual environment using digital twin technology can identify algorithmic problems, including but not limited to "logic conflicts," "assignment errors," "memory leaks," and "data latency."

[0049] In this invention, the information layer twin model is used to transmit real-time status information data collected by sensors during the operation of the ultra-deep well hoisting system, so that the system layer twin model can drive a three-dimensional visualization engine to render and construct the digital twin virtual model based on the real-time status information data. The real-time status information data includes the hoisting system's speed, position, acceleration, fluid pressure, tilt signal, and wire rope tension. The pin force sensor and displacement sensor are installed on the hydraulic cylinder, the tension sensor is installed and connected to the wire rope, the acceleration sensor and tilt sensor are installed on the hoisting container, and the wireless sensors are distributed in the pipeline shaft.

[0050] The concept of digital twins is progressive. Simulation modeling at the physical level and virtual algorithm testing at the system level are replicas of real physical prototypes. Digital technology is used to build a virtual model that is highly similar to the real world, thereby enabling the understanding, analysis and optimization of physical objects.

[0051] Digital twins are no longer limited to the simple replication of ultra-deep well hoisting systems, but rather an expansion of information dimensions. Information-level digital twins can assist in real machine experiments, enabling state monitoring, data storage, and human-machine interaction during robot operation. Their core lies in data-driven approaches. This chapter mainly addresses the "data" and "connectivity" issues in the five-dimensional model of digital twins, analyzes the data transmission requirements in "data source change testing" and "robot motion control testing," and designs a reasonable data acquisition scheme.

[0052] Create expression templates for sensor data obtained by sensors based on Extensible Markup Language; establish the correlation between sensor data and sensors; and create a twin data information chain between sensor data and intelligent decision data based on the correspondence between sensor data and intelligent decision data.

[0053] In the dual-rope winding ultra-deep vertical shaft hoisting system, pin force sensors and displacement sensors are installed on the hydraulic cylinders, tension sensors are installed and connected to the wire ropes, acceleration sensors and tilt sensors are installed on the hoisting container, and wireless sensors are distributed throughout the pipe shaft. These sensors transmit the speed, position, acceleration, fluid pressure, and tilt signals of the dual-rope winding ultra-deep vertical shaft hoisting system to the control system. The control system collects data from the operation panel of the dual-rope winding ultra-deep vertical shaft hoisting system and the speed, position, acceleration, fluid pressure, and temperature of the hoist itself. It then transmits the collected data to the digital twin virtual model via signal communication, while simultaneously controlling the operation of the dual-rope winding ultra-deep vertical shaft hoist.

[0054] In a dual-rope winding ultra-deep vertical shaft hoisting system, pin force sensors and displacement sensors are installed on the hydraulic cylinders, tension sensors are connected to the wire ropes, acceleration sensors and tilt sensors are installed on the hoisting container, and wireless sensors are distributed throughout the shaft. These sensors collect real-time status data of the dual-rope winding ultra-deep vertical shaft hoisting system during operation. Based on this information, a 3D visualization engine is used to render and generate a virtual digital twin model of the dual-rope winding ultra-deep vertical shaft hoisting system, identical to the actual system. This achieves a faithful twin mapping between the dual-rope winding ultra-deep vertical shaft hoisting system and the virtual digital twin model. The generated virtual model can be visualized, interacted with, and monitored from multiple perspectives both online in real-time and offline in non-real-time. It also provides visualized display and prediction of operating conditions and other information, enhancing the intelligence level of the dual-rope winding ultra-deep vertical shaft hoisting system.

[0055] Figure 3 This is a schematic diagram of the active tension adjustment system for a mine hoist. According to... Figure 3 As shown, the active tension adjustment system is a typical servo valve-controlled hydraulic cylinder system. It uses an electro-hydraulic servo valve to control a hydraulic cylinder to drive a floating sheave to adjust the wire rope tension, and uses a corresponding tension sensor as feedback to form a closed-loop adjustment, thus achieving active adjustment of the wire rope tension. Figure 3 It can be deduced that the pin force of the hydraulic cylinder is the resultant force of the lateral tension of the string and the vertical lateral tension:

[0056] F x =F z +F z sinγ (1)

[0057] In the formula, γ is the angle between the wire rope and the plane containing the upper platform, and F x F is the force on the piston rod pin of the hydraulic cylinder. z This refers to the tension of the steel wire rope.

[0058] The pin force of the hydraulic cylinder in the tension active adjustment system is measured by the corresponding tension / compression sensor. If k... f To represent the stiffness of the tension / compression sensor, according to Hooke's Law, the pin force of the two hydraulic cylinders in the tension active adjustment system can be expressed as:

[0059] F x =k f (x p -x pf (2)

[0060] In the formula, F x x is the pin force of the hydraulic cylinder. p x is the effective displacement of the hydraulic cylinder piston rod. pf The effective displacement of the floating sheave driven by the piston rod of the hydraulic cylinder.

[0061] Ignoring pressure losses and oil dynamics within the hydraulic lines of the active tension control system, the load flow rate Q of the i-th hydraulic cylinder in the active tension control system... L It can be represented as:

[0062]

[0063] In the formula, A p C is the effective working area of ​​the hydraulic cylinder piston rod. tl V is the total leakage coefficient of the hydraulic cylinder. t P is the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. L =p1-p2 is the load pressure drop of the hydraulic cylinder, p1 is the pressure in the inlet chamber of the hydraulic cylinder, p2 is the pressure in the return chamber of the hydraulic cylinder, β e Q1 is the bulk modulus of the hydraulic fluid, Q2 is the flow rate of the hydraulic cylinder inlet chamber, and Q2 is the flow rate of the hydraulic cylinder return chamber.

[0064] The active tension adjustment system of a mine hoist is a typical electro-hydraulic servo valve-controlled hydraulic cylinder system; therefore, its load flow rate Q L It is determined by the valve core displacement x of the electro-hydraulic servo valve. v Controlled, therefore

[0065]

[0066] In the formula, C d Let ρ be the discharge coefficient of the electro-hydraulic servo valve, w be the throttling window area gradient of the electro-hydraulic servo valve, and ρ be the discharge coefficient of the electro-hydraulic servo valve. o p is the density of the hydraulic fluid. s The pressure of the hydraulic oil source.

[0067] In electro-hydraulic servo systems, the response speed of the electro-hydraulic servo valve is often much higher than that of the entire system. Therefore, when modeling the electro-hydraulic servo valve, its dynamic model can be ignored. Thus, the electro-hydraulic servo valve model can be represented as:

[0068] x v =k v u L (5)

[0069] In the formula, k v The proportional coefficient of the electro-hydraulic servo valve, u L This refers to the control voltage of the electro-hydraulic servo valve.

[0070] Furthermore, according to equations (4) and (5), the load flow of the hydraulic cylinder can be re-expressed as:

[0071]

[0072] The discharge coefficient C of the electro-hydraulic servo valve d Both the throttling window area gradient w of the servo valve and the servo valve are estimated values. As can be seen from the above formula, directly obtaining the control voltage of the electro-hydraulic servo valve is extremely difficult. However, an electro-hydraulic servo valve at its rated pressure drop Δp... r Rated flow rate Q r It is often definite. Therefore, we can conclude that:

[0073]

[0074] In the formula, u max Q is the maximum control voltage of the electro-hydraulic servo valve; r Rated flow rate of electro-hydraulic servo valve; ΔP r This is the rated pressure drop of the electro-hydraulic servo valve. From equations (6) and (7), we can obtain...

[0075]

[0076] The discharge coefficient C of the electro-hydraulic servo valve d Throttling window area gradient w and proportionality coefficient k v All are positive numbers greater than 0. Therefore, from equation (8), we can obtain:

[0077]

[0078] Finally, based on the above analysis, the control voltage of the electro-hydraulic servo valve can be expressed as:

[0079]

[0080] According to Newton's second law, the load force balance equation of the hydraulic cylinder in the tension active adjustment system can be expressed as:

[0081]

[0082] In the formula, m is the total mass of the hydraulic cylinder load, and B p denoted as the viscous damping coefficient of the hydraulic cylinder, and d represents external disturbance.

[0083] Substituting equation (1) into equation (11), we get

[0084]

[0085] In this invention, the tension control model is constructed based on the following parameters: wire rope tension F z The angle γ between the wire rope and the plane of the upper platform, and the load pressure drop P of the hydraulic cylinder. L Stiffness k of tension / compression sensor f The effective working area A of the hydraulic cylinder piston rod PThe total mass m of the hydraulic cylinder load, and the viscous damping coefficient B of the hydraulic cylinder. p The total leakage coefficient C of the hydraulic cylinder tl The total effective volume V of the hydraulic cylinder's inlet and outlet oil chambers t The bulk modulus β of hydraulic fluid e The load flow rate Q of the hydraulic cylinder L .

[0086] Define the system state variables as The model of the hydraulic cylinder and the tension of the wire rope it regulates can be represented in the following state-space form:

[0087]

[0088] In the formula, the correspondence between each simplified parameter and the parameters of the active tension adjustment system itself is as follows: F z This indicates the tension of the wire rope.

[0089] P L k represents the load pressure drop of the hydraulic cylinder. f A represents the stiffness of the tension / compression sensor. P B represents the effective working area of ​​the hydraulic cylinder piston rod, m represents the total mass of the hydraulic cylinder load, and B represents the effective working area of ​​the hydraulic cylinder piston rod. p C represents the viscous damping coefficient of the hydraulic cylinder. tl V represents the total leakage coefficient of the hydraulic cylinder. t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e Let represent the bulk modulus of elasticity of the hydraulic fluid, d represent external disturbances, γ represent the angle between the wire rope and the plane containing the upper platform, and x1, x2, and x3 are defined system state variables. The speed of the floating headwheel. The speed at which the hydraulic piston rod moves.

[0090] Assumption 1: The tension F of the two steel wire ropes in the tension active adjustment system z and its first derivative Second derivative and third derivative Both are bounded.

[0091] In the intelligent learning control method of this invention, firstly, based on the tension control model, a matrix form is constructed as follows:

[0092]

[0093] in, B = [0, 0, θ6] T

[0094] The present invention aims to design an optimal control quantity u* To ensure the system state reaches 0 in the optimal way, i.e., to minimize the following penalty function:

[0095]

[0096] In the formula, r x (x(s),u(s)) is the utility function.

[0097] Design of dynamic estimators for unknown systems:

[0098] In the system state equation, f(x) represents the unknown integrated disturbance. Therefore, to improve system control performance, it is necessary to estimate and compensate for the system disturbance. The filter variable x of the estimator... f and u f As shown below:

[0099]

[0100] In the formula, τ is the positive control gain. Therefore, the estimator can be designed as follows:

[0101]

[0102] In the formula, This is an estimate of the unknown integrated interference f. Estimation error.

[0103] Lemma 1. Considering the system equation (1) and the estimator, the estimation error e is bounded. In the formula, for The upper boundary; it can be seen that e converges to 0 exponentially, and thus, we can conclude Or F→0.

[0104] Proof. The first-order low-pass filter 1 / (τs+1) in Applying both sides of the equation, we get...

[0105]

[0106] In the formula, f f The filter value of f is obtained by... f f (0) = 0. Therefore, from equation (16), we can obtain... Combining equation (18), we can obtain

[0107] Furthermore, we can obtain The estimation error dynamics of the estimator can be obtained as follows:

[0108]

[0109] Define the Lyapunov function as Its derivative with respect to time is

[0110]

[0111] From the above formula, we can obtain that and Therefore, the estimation error e will approach 0 exponentially. or

[0112] According to Lemma 1, we obtain the estimated value. It will approach f exponentially and converge to a very small envelope of the true value of f. Therefore, system (1) can be rewritten as

[0113]

[0114] Optimal Controller Design Based on Evaluative Learning

[0115] This section will design an optimization control method based on evaluative learning to minimize the system tracking error z. The penalty function is constructed as follows.

[0116]

[0117] In the formula, and All are positive definite matrices. Let J* be the optimized value of J, i.e.,

[0118]

[0119] Furthermore, applying the Bellman optimality principle, we can obtain J * The derivative of (x) satisfy

[0120]

[0121] in,

[0122]

[0123] Combining optimal control theory, we obtain the optimal control law as follows:

[0124]

[0125] Next, we will use equation (25) to... The optimal solution is obtained. Based on the principle of adaptive dynamic programming, an evaluation neural network is used for the first time to construct the optimization penalty function. A comment learning algorithm, based on the weight error estimated by the evaluation neural network, is used to update the optimal control solution. Therefore, based on the evaluation neural network, the optimal solution J is obtained. *(x) is:

[0126]

[0127] In the formula, To evaluate the ideal weights of a neural network, Here, ε is the activation function, l represents the number of neurons in the hidden layer, and ε is the activation function. v (x) represents the construction error. Therefore, the optimal solution J... * (x) derivative along the x-direction

[0128]

[0129] In the formula,

[0130] Evaluating neural network architecture, such as Figure 4 As shown.

[0131] Assumption 2. Evaluate the weights θ and functions of the neural network. and its derivative Construction error ε v (·) and its derivative Both are bounded, that is, ||θ||≤θ M , Where, θ M σ N σ M and σ ε All are normal numbers.

[0132] Typically, the optimization weights θ of a neural network should be solvable, and their actual values ​​can be expressed by the following formula.

[0133]

[0134] Then, its derivative can be obtained as

[0135]

[0136] Based on equations (28) and (29), equation (26) can be rewritten as follows:

[0137]

[0138] And the actual optimal control law is

[0139]

[0140] For equation (32), we can obtain that the system state variable x→0 depends on the estimated weights of the evaluation neural network. Furthermore, substituting the estimated penalty function (30) into (25), we can obtain

[0141]

[0142] In the formula, Estimate the error for the Hamilton-Jacobi-Bellman equation. The known and unknown parts in equation (33) are defined as follows:

[0143]

[0144] Then, equation (34) can be rewritten as

[0145] Ω=-θ T X-ε HJB (35)

[0146] To obtain the real-time solution of equation (35), the following auxiliary matrix is ​​defined. for

[0147]

[0148] In the formula, c>0 is a positive constant. Therefore, the solution to equation (36) is...

[0149]

[0150] Furthermore, another auxiliary vector It can be defined as

[0151]

[0152] Based on equations (36) and (38), we can obtain that Τ=-Φθ+v, where For a bounded variable, that is, ||v||≤ε 1v In the formula ε 1v It is a positive number.

[0153] Based on this, we can rewrite for

[0154]

[0155] In the formula, To evaluate the estimation error of the neural network, the evaluation learning law can be derived based on equation (39) as follows:

[0156]

[0157] In the formula, Λ>0 represents the learning gain. The learning law in equation (40) includes the estimation error. This will make As it gradually approaches its true value, the system converges.

[0158] Define the Lyapunov function as Its derivative with respect to time is

[0159]

[0160] From equation (41), we can obtain the estimation error. It will gradually converge to Therefore, in an ideal situation, ε HJB =0, the control system is stable.

[0161] Design of Optimization Control Methods Based on Updated Bellman Equations

[0162] Applying the integral reinforcement learning function, the penalty function can be rewritten as follows:

[0163]

[0164] In the formula, T>0 represents the sampling frequency. Therefore, the penalty function based on the evaluation neural network can be obtained as follows:

[0165] J * (x)=W T ψ(x)+ζ(x) (43)

[0166] And J * derivative along x It can be obtained from the following formula

[0167]

[0168] In the formula, is the activation function, and N is the number of neurons in the evaluation neural network. To evaluate the weights of a neural network, and To evaluate the estimation error of the neural network.

[0169] To achieve continuous updating of the evaluation, the penalty function (43) can be substituted into the Bellman equation (42) to obtain...

[0170]

[0171] In the formula, k(x,u) is the integral reinforcement learning term, Δψ(t)=ψ(x(t))-ψ(x(tT)), and the error ζ x =ζ(x(t))-ζ(x(tT)) is bounded. Next, we define an adaptive law to approximate the weights W(30) online, introducing auxiliary variables. for

[0172]

[0173] In the formula, Let these be the design parameters. The solution to equation (46) is:

[0174]

[0175] For the penalty function (43), its estimated value can be obtained as follows:

[0176]

[0177] In the formula, and These are the estimated values ​​for J and W, respectively.

[0178] An adaptive law referencing sliding mode technology was designed as follows:

[0179]

[0180] In the formula, Γ>0 represents the learning parameter.

[0181] System stability proof:

[0182] The variable G is bounded, that is, λ max (G)≤α, where α is a constant greater than 0. Substituting k from equation (16) into equation (45), we get F=-GW+Θ, where, α1 is a constant greater than 0; therefore, we can obtain

[0183]

[0184] Since Δψ(t) satisfies the PEC (continuous excitation condition), then G is invertible; therefore, we can obtain G can be obtained -1 The derivative of M is

[0185]

[0186] In the formula, For bounded, ||Θ s ||≤α2, where α2 is a constant greater than 0. Since λ min (G)≤ξ、λ max (G)≤α,G -1 It is also bounded, that is, at λ min (G)>α、λ max (G)≤1 / ξ, the time-varying Lyapunov function Ψ3 is defined as follows:

[0187]

[0188] In the formula, K1>0 is a control parameter. The derivative of Ψ3 can be obtained as follows:

[0189]

[0190] In the formula, For a positive constant, 0 < K1 < ξ / (λ) max (Γ -1 )α2); therefore, in a finite time With Ψ3=0 and M=0, we can obtain

[0191] 1) For ζ(x) = 0, we can obtain ζ x =0, M=0 and Θ=Θ s Further, we can obtain and So, the weight update error It will approach 0 within a finite time t1.

[0192] 2) For ζ(x)≠0, we can obtain ζ x ≠0, M≠0, further we can obtain So show The system is bounded after a finite time t1. The entire control system is stable.

[0193] Ultimately, the true control law of the system can be obtained as follows:

[0194]

[0195] In the formula,

[0196] This yields the true control law u (which is also the load flow Q). L The voltage will be converted into the control voltage u of the electro-hydraulic servo valve via equation (10). L This drives the hydraulic cylinder to move and adjust the tension of the wire rope.

[0197] The following is an example illustration. Hydraulic system oil source pressure P s =15×10 6 Pa, the effective working area A of the double-rod hydraulic cylinder p =1.88×10 -3 m2, hydraulic system load mass m=200kg, hydraulic system viscous damping coefficient B p = 25000 N (m / s), the total volume V of the hydraulic cylinder's inlet and outlet chambers t =0.96×10 -3 m3, total leakage coefficient of hydraulic system C tl =9.2×10 -13 m 3 / (s / Pa), hydraulic oil bulk elastic modulus β e =6.9×10 8 Pa, stiffness k f=2×107. Control parameters: K1=500, Γ=20, N=10, Λ=15, c=5, τ=2, l=10.

[0198] Figure 5a and Figure 5b This is a schematic diagram of the wire rope tension difference obtained using the method of the present invention. Figure 6a and Figure 6b This is a schematic diagram of the tension difference in a wire rope obtained using methods in related technologies. As shown in the figure, the maximum tension difference in the wire rope obtained using the method of this invention does not exceed 40N, while the tension difference in the wire rope obtained using methods in related technologies can reach 200N. The tension difference obtained using the method of this invention is even smaller.

[0199] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A digital twin-driven method for learning and controlling the tension of wire ropes in ultra-deep well hoisting systems, characterized in that, The method includes: A digital twin virtual model of an ultra-deep well hoisting system is constructed. The digital twin virtual model is used for real-time transmission of operating data in ultra-deep wells and for simulating the operating status of equipment in ultra-deep wells. The digital twin virtual model includes a physical layer twin model, a system layer twin model, and an information layer twin model. A tension control model for the wire rope in an ultra-deep well hoisting system is constructed, wherein the tension control model is used to control the tension difference between two wire ropes in the ultra-deep well hoisting system in order to minimize the tension difference; Based on the digital twin virtual model and the tension control model, the system control law u is obtained by updating and learning through an evaluation neural network. The control voltage u of the electro-hydraulic servo valve is determined based on the system control law u. L and with the control voltage u L The hydraulic cylinder is driven to adjust the tension of the wire rope.

2. The method according to claim 1, characterized in that, The tension control model is constructed based on the following parameters: wire rope tension F z The angle γ between the wire rope and the plane of the upper platform, and the load pressure drop P of the hydraulic cylinder. L Stiffness k of tension / compression sensor f The effective working area A of the hydraulic cylinder piston rod P The total mass m of the hydraulic cylinder load, and the viscous damping coefficient B of the hydraulic cylinder. p The total leakage coefficient C of the hydraulic cylinder tl The total effective volume V of the hydraulic cylinder's inlet and outlet oil chambers t The bulk modulus β of hydraulic fluid e The load flow rate Q of the hydraulic cylinder L .

3. The method according to claim 1, characterized in that, The tension control model is shown in the following state space: Wherein, the system state variable is x e , F z P represents the tension of the wire rope. L k represents the load pressure drop of the hydraulic cylinder. f A represents the stiffness of the tension / compression sensor. P B represents the effective working area of ​​the hydraulic cylinder piston rod, m represents the total mass of the hydraulic cylinder load, and B represents the effective working area of ​​the hydraulic cylinder piston rod. p C represents the viscous damping coefficient of the hydraulic cylinder. tl V represents the total leakage coefficient of the hydraulic cylinder. t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e Let represent the bulk modulus of elasticity of the hydraulic fluid, d represent external disturbances, γ represent the angle between the wire rope and the plane containing the upper platform, and x1, x2, and x3 are defined system state variables. The speed of the floating headwheel. The speed at which the hydraulic piston rod moves.

4. The method according to claim 1, characterized in that, The control voltage u of the electro-hydraulic servo valve is determined according to the system control law u. L ,include: The system control law u is used as the load flow rate Q of the hydraulic cylinder. L The control voltage u is obtained through the following formula. L : Among them, Q r This indicates the electro-hydraulic servo valve at the rated pressure drop Δp r The rated flow rate, p s P represents the pressure of the hydraulic oil source. L Indicates the load pressure drop of the hydraulic cylinder, u max This is the maximum control voltage of the electro-hydraulic servo valve.

5. The method according to claim 1, characterized in that, The load flow rate Q of the hydraulic cylinder L This can be expressed by the following formula: Among them, the discharge coefficient C of the electro-hydraulic servo valve d The throttling window area gradient w of the servo valve and k are both estimated values. v p is the proportional coefficient of the electro-hydraulic servo valve. s ρ is the pressure of the hydraulic oil source, and ρ0 is the density of the hydraulic oil.

6. The method according to claim 1, characterized in that, The system control law u is obtained through the following formula: in, It is a positive definite matrix. Let W be the activation function, and W represent the weights. Let B be an estimate of W, and B = [0, 0, θ6]. T , V t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e This indicates the bulk modulus of elasticity of hydraulic fluid.

7. The method according to claim 1, characterized in that, The system control law u is obtained through the optimal control law u. * The obtained optimal control law u * It can be obtained through the following formula: in, B is a positive definite matrix, [0, 0, θ6] T , V t β represents the total effective volume of the hydraulic cylinder's inlet and outlet oil chambers. e Let J(x) represent the bulk modulus of the hydraulic fluid, J*(x) be the penalty function, and J*(x) be the optimized value of J(x). It is the derivative of J*(x).

8. The method according to claim 1, characterized in that, The physical layer twin model is used for simulation modeling of ultra-deep downhole equipment. The physical layer twin model includes the system physical ontology, the environment physical ontology, the sensor system, the wireless sensor network, and the real-time control system. The physical components of the system include an asynchronous motor, a double drum, a hoisting container, a hoisting wire rope, a hoisting system mechanical mechanism, a floating sheave, a hydraulic cylinder, hydraulic pipelines, and a pump station. The physical components of the environment include the hoisting shaft, the environment of the hoisting shaft bottom platform, and the environment of the mine platform. The sensor system includes a pin force sensor, a displacement sensor, a tension sensor, a hydraulic pressure sensor, an acceleration sensor, a tilt sensor, and a rotary encoder sensor. The wireless sensor network includes a wireless network transmitting node, a wireless network receiving node, and a power supply system. The real-time control system includes a signal acquisition and conditioning system, an industrial control computer real-time system, a host computer real-time monitoring system, a power supply system, and power cables.

9. The method according to claim 1, characterized in that, The system-level twin model is used to simulate the operating status of the lifting container, the tension of the wire rope, the hydraulic cylinder, the floating sheave, and the operating status of the wireless sensor network in a virtual environment, and generates test data through the digital twin prototype.

10. The method according to claim 1, characterized in that, The information layer twin model is used to transmit real-time status information data collected by sensors during the operation of the ultra-deep well hoisting system. This enables the system layer twin model to drive a 3D visualization engine to render and construct the digital twin virtual model based on the real-time status information data. The real-time status information data includes the hoisting system's speed, position, acceleration, fluid pressure, tilt signal, and wire rope tension. The pin force sensor and displacement sensor are installed on the hydraulic cylinder, the tension sensor is installed and connected to the wire rope, the acceleration sensor and tilt sensor are installed on the hoisting container, and the wireless sensors are distributed in the pipeline shaft.

Images