Intelligent prediction and self-adaptive control method for state of closed hydraulic system of roof drill

Abstract

An intelligent prediction and adaptive control method for the state of the closed hydraulic system of a roof drill, which uses an intelligent model based on reinforcement learning to intelligently control and adaptively adjust the pressure, flow rate, and temperature of the hydraulic system. First, the future state of the pressure, flow rate, and temperature of the hydraulic system is predicted by a GRU model to provide early warning of possible overpressure and pressure failure. Then, the current working state of the system is judged in real time by an SVM model, and based on the GRU prediction results and SVM state judgment, adjustment instructions are generated and sent to an embedded microprocessor. The embedded microprocessor adjusts the pressure relief valve, servo valve, and buffer in real time according to the instructions to ensure stable operation of the system. At the same time, the data collection module continuously collects sensor data and feedback information data, and combines with the adaptive learning mechanism to optimize and update the intelligent control model in real time. This method can effectively improve the response speed and safety of the hydraulic system, and avoid overpressure and pressure failure.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence control, and particularly relates to an intelligent prediction and self-adaptive regulation method for the state of a closed hydraulic system of a raise borer. BACKGROUND

[0002] The raise borer, as an important equipment for mine and underground engineering construction, has a vital role in the stability and safety of production, and the performance of the hydraulic system directly affects the stability and construction efficiency of the raise borer. In actual application, the pressure, flow and temperature parameters of the hydraulic system are usually affected by external load, borer working conditions and other factors, and thus overpressure, pressure retention, pressure fluctuation and other fault conditions are prone to occur, which may even cause system damage or operation accidents. The traditional hydraulic system monitoring method mainly relies on preset thresholds to determine whether the pressure, flow and temperature parameters are normal, and when a parameter exceeds the set value, the control system directly triggers the alarm device to alarm or directly takes adjustment measures for adjustment. However, the traditional method cannot predict and judge the dynamic changes of the system in real time, resulting in slow response speed and thus difficulty in effectively dealing with sudden problems under complex working conditions. In addition, the control of the hydraulic system usually relies on single sensor data and manually set control strategies, lacking real-time prediction and self-adaptive adjustment capability of the system state.

[0003] In order to effectively deal with sudden problems under complex working conditions, and to ensure the safe and stable operation of the raise borer, it is urgent to provide a hydraulic self-adaptive regulation scheme capable of predicting and judging the dynamic changes of the system in real time. SUMMARY

[0004] In view of the problems existing in the prior art, the application provides an intelligent prediction and self-adaptive regulation method for the state of a closed hydraulic system of a raise borer.

[0005] In order to achieve the above purpose, the application provides an intelligent prediction and self-adaptive regulation method for the state of a closed hydraulic system of a raise borer, comprising the following steps:

[0006] Step 1: In the historical operation process of the closed hydraulic system of the raise borer, a sensor group installed in the closed hydraulic system of the raise borer is used to collect historical operation data and send it to a feedback unit for recording as historical operation data;

[0007] Step 2: The historical operation data are preprocessed to obtain initial parameter data;

[0008] Step 3: The initial parameter data are extracted to obtain pressure key features, flow key features and temperature key feature data;

[0009] Step 4: A GRU prediction model is constructed and trained;

[0010] S41: For the key feature data, a sliding window technique is used to divide the key feature data into time series segments of fixed length, and the length of each time series segment is t; a data set is constructed based on the obtained time series segment data, and the data set is divided into a training set and a test set according to a set proportion;

[0011] S42: A GRU prediction model is constructed based on a gated recurrent unit;

[0012] S43: A loss function L of the GRU prediction model is constructed according to formula (1);

[0013]

[0014] In the formula, N is the data amount, P' t+k , Q' t+k and T' t+k are prediction values of pressure, flow and temperature at future k time, P t+k , Q t+k and T t+k are target values of pressure, flow and temperature at future k time;

[0015] S44: The training set is used as input data to train the GRU prediction model, a back propagation algorithm is used to calculate the gradient of the loss function L with respect to the model parameters, at the same time, an ADAM optimization algorithm is used to update the model parameters, and the loss function L is minimized, and finally a trained GRU prediction model is obtained;

[0016] Step five: Real-time prediction is performed by using the GRU prediction model;

[0017] S51: In the real-time operation process of the closed hydraulic system of the raise borer, a sensor group installed in the closed hydraulic system of the raise borer is used to collect real-time operation data;

[0018] S52: The real-time operation data is first divided into time series segments of fixed length by using a sliding window technique, and then is input into the GRU prediction model as input data for prediction, and pressure, flow and temperature prediction data are output;

[0019] Step six: Control is implemented based on the GRU prediction result;

[0020] The prediction data is sent to the parameter regulation unit for analysis and decision making, and the decision information is sent to the processing unit. The processing unit obtains corresponding control instructions according to the decision information, and controls the action parameters of the relief valve, buffer and high-speed servo valve of the closed hydraulic system according to the corresponding control instructions until the closed hydraulic system returns to normal. In the regulation process, the feedback unit collects and records the pressure data, flow data, temperature data, drilling rig working condition information data before regulation, the pressure data, flow data, temperature data, drilling rig working condition information data after regulation, the adjustment value data of the relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, and the prediction value data of the GRU prediction model, as regulation feedback data; at the same time, the regulation feedback data is sent to the parameter regulation unit, and then the update data of a complete regulation period is obtained;

[0021] Step seven: continuously optimize the GRU prediction model using feedback data to obtain an optimized GRU prediction model;

[0022] S71: repeatedly execute steps five and six multiple times until the set regulation period is reached, and obtain update data of several regulation periods;

[0023] S72: adopt the incremental learning method, input the update data into the GRU prediction model for continuous update and optimization, and finally obtain the optimized GRU prediction model;

[0024] Step eight: build and train the SVM prediction model;

[0025] S81: for the update data obtained in step seven for several regulation periods, first use the sliding window technology to divide the real-time running data into fixed-length time sequence segments, and based on the obtained time sequence segment data, build a data set, and then divide the data set into a training set and a test set according to a set proportion;

[0026] S82: build the SVM prediction model based on the support vector mechanism;

[0027] S83: use the training set as input data to train the SVM prediction model, and obtain the trained SVM prediction model;

[0028] Step nine: use the SVM prediction model for real-time prediction;

[0029] S91: in the real-time running process of the closed hydraulic system of the raise boring machine, use the sensor group installed in the closed hydraulic system of the raise boring machine to collect real-time running data;

[0030] S92: use the sliding window technology to divide the prediction data into fixed-length time sequence segments, and then input them into the SVM prediction model as input data for prediction, and output the state judgment result;

[0031] Step ten: Implementing regulation based on SVM prediction results;

[0032] The state judgment data is sent to the parameter regulation unit, which obtains adjustment decisions based on the state judgment results using the mapping relationship between pressure, flow rate, temperature, and control strategy, and sends the adjustment decisions to the processing unit. The processing unit obtains corresponding control instructions according to the adjustment decisions, and controls the action parameters of the pressure relief valve, buffer, and high-speed servo valve of the closed hydraulic system according to the corresponding control instructions until the closed hydraulic system returns to normal. In the regulation process, the feedback unit collects and records the pressure data, flow rate data, temperature data, rig working condition information data before regulation, the pressure data, flow rate data, temperature data, rig working condition information data after regulation, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve, and the adjustment value data of the buffer, the prediction value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model as regulation feedback data. At the same time, the regulation feedback data is sent to the parameter regulation unit, and then the update data of a complete regulation period is obtained;

[0033] Step eleven: continuously optimizing the SVM prediction model using feedback data to obtain an optimized SVM prediction model;

[0034] S111: repeatedly performing steps nine and ten multiple times until a set regulation period is reached, and obtaining update data for several regulation periods;

[0035] S112: using incremental learning to input the update data into the SVM prediction model for continuous updating and optimization, and finally obtaining an optimized SVM prediction model;

[0036] Step twelve: using the optimized GRU prediction model and the optimized SVM prediction model to regulate the closed hydraulic system of the raise boring rig in real time;

[0037] S121: during the real-time operation of the closed hydraulic system of the raise boring rig, using the sensor group installed in the closed hydraulic system of the raise boring rig to collect real-time operation data;

[0038] S122: first using the sliding window technology to divide the real-time operation data into fixed-length time sequence segments, and the obtained time sequence segments include pressure time sequence data P t , flow rate time sequence data Q t , temperature time sequence data T t , working condition information time sequence data C t , pressure relief valve adjustment value time sequence data L zt , and servo valve adjustment value time sequence data Lvt Time series data of buffer adjustment value L bt ;

[0039] S123: Input the obtained time series segments as input data into the optimized GRU prediction model for prediction, and output the predicted pressure value P′ at time t+k. t+k Flow forecast value Q′ t+k and temperature prediction value T′ t+k ;

[0040] S124: The predicted pressure value P′ at time t+k. t+k Flow forecast value Q′ t+k Temperature prediction value T′ t+k Time series data of operating condition information C t Time series data of pressure relief valve adjustment value L zt Time series data of servo valve integer values ​​L vt Time series data of buffer adjustment value L bt The input data is fed into the optimized SVM prediction model for prediction, and the output is the system's state label S. t The system's status label S t The numbers 0, 1, and 2 represent the normal state, the overpressure state, and the suffocation state, respectively.

[0041] S125: Based on the set maximum pressure threshold P max Pressure fluctuation threshold P fltut Minimum flow threshold Q min and maximum temperature threshold T max Adjustments are made based on forecast data and state prediction results;

[0042] If P′ t+k >P max And S t =0 or S t If the value is 1, the processing unit controls the alarm device to issue a warning of potential overpressure. Simultaneously, it executes overpressure prevention measures and triggers overpressure regulation, increasing the opening degree L of the pressure relief valve. zt To release excess system pressure;

[0043] If P′ t+k <P max And S t If the value is 1, the following instruction is generated: Increase the opening degree L of the pressure relief valve. zt To release excess system pressure, adjust the opening L of the servo valve. vt This is to optimize flow control and avoid insufficient flow caused by overvoltage;

[0044] If Q′t+k Q min , and S t =0 or S t =2, the processing unit controls the alarm device to act to give a warning of possible pressure build-up, at the same time, executes pressure build-up prevention measures, and triggers flow supplement regulation action to increase the opening L of the high-speed servo valve vt to increase the flow rate;

[0045] If Q' t+k >Q min , and S t =2, the following instruction is generated: increase the opening L of the servo valve to increase the flow rate, keep the low opening L of the pressure relief valve to prevent further pressure accumulation; vt zt

[0046] If T' t+k >T max , and S t =0, the processing unit controls the alarm device to act to give a warning of possible pressure build-up, at the same time, executes pressure build-up prevention measures, and triggers flow supplement regulation action to increase the opening L of the high-speed servo valve vt to increase the flow rate;

[0047] If the predicted P' t+k will fluctuate greatly in a short time, and is close to P fluct , at the same time, S t =0 or S t =1, the damping coefficient L of the buffer is adjusted, at the same time, the opening L of the servo valve is fine-tuned according to the predicted flow rate change to ensure stable flow rate. bt vt

[0048] As a preferred, in step two, the preprocessing process includes data cleaning by using interpolation method, data filtering by using Gaussian filter, and detection of outliers by using Isolation Forest algorithm, and normalization processing of the data after cleaning, filtering and outlier detection.

[0049] As a preferred, in step one, the historical operation data includes pressure data, flow rate data and temperature data; the pressure data includes pressure data at the oil outlet of the hydraulic pump, pressure data at the oil inlet of the hydraulic motor, pressure data at the inlet side of the pressure relief valve, and pressure data at the oil outlet of the oil tank; the flow rate data includes flow rate data at the oil outlet of the hydraulic pump, flow rate data at the oil inlet of the hydraulic motor, flow rate data at the oil outlet of the hydraulic motor, and flow rate data of the main oil return pipeline; the temperature data includes temperature data at the oil outlet of the hydraulic pump, temperature data of the oil tank, and temperature data of the hydraulic motor.

[0050] ​​​​As a preferred, in step three, the pressure key features include instantaneous pressure value, pressure fluctuation amplitude and pressure change rate; the flow key features include instantaneous flow value, flow change rate and cumulative flow; the temperature key features include instantaneous temperature value, temperature change rate and steady-state temperature interval.

[0051] As a preferred, in step four, the process of updating model parameters is as follows:

[0052] The weight matrix W in the model is updated according to formula (2) by gradient descent method z , W r , W h , W z and bias term b z , b r , b h , b o ;

[0053]

[0054] In the formula, θ represents weight or bias, η is the learning rate, is the gradient of the loss function to the parameter.

[0055] As a preferred, in step six, the pressure data includes hydraulic pump discharge port pressure data, hydraulic motor inlet pressure data, pressure relief valve inlet side pressure data, oil tank outlet pressure data; the flow data includes hydraulic pump discharge port flow data, hydraulic motor inlet flow data, hydraulic motor outlet flow data, main oil return pipeline flow data; the temperature data includes hydraulic pump discharge port temperature data, oil tank temperature data, hydraulic motor temperature data; the rig working condition information data includes the working mode of the top drive bit, including drilling mode, lifting mode and pushing mode; the pressure relief valve adjustment value data includes the opening adjustment value of the pressure relief valve; the servo valve adjustment value data includes valve port opening parameter, flow parameter and flow rate parameter; the buffer adjustment value data includes the damping coefficient adjustment value of the buffer; the prediction value data of the GRU prediction model includes future pressure trend data, flow trend data and temperature trend data.

[0056] As a preferred, in step eleven, the system state label data predicted by the SVM prediction model includes the state result of the system operating condition.

[0057] Further, to ensure the accuracy and intelligence of the regulation, in S125 of step twelve, during the regulation process, the feedback unit collects and records the pressure data, flow data, temperature data, rig working condition information data before regulation, the pressure data, flow data, temperature data, rig working condition information data after regulation, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, the adjustment value data of the buffer, the predicted value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model as regulation feedback data, and based on the regulation feedback data, the model parameters are continuously optimized and updated through adaptive learning.

[0058] The application provides an intelligent closed-loop hydraulic adaptive regulation method based on a combination of a gated recurrent unit (GRU) model and a support vector machine (SVM) model, which aims to realize intelligent control and adaptive adjustment of the system by monitoring the pressure, flow and temperature parameters of the hydraulic system in real time. The method first predicts the future state of the pressure, flow and temperature of the hydraulic system through the GRU prediction model, thereby warning of possible overpressure and pressure failure in advance, then judges the current working state of the system (such as normal, overpressure, and pressure) in real time through the SVM prediction model, and generates decision information based on the prediction results of the GRU and the state judgment of the SVM. The decision information is sent to the processing unit, which generates corresponding control instructions according to the decision information, and adjusts the actions of the pressure relief valve, servo valve and buffer in the hydraulic system in real time according to the instructions to ensure the stable operation of the system. At the same time, the method can optimize and update the machine learning model in real time through continuous collection of new sensor data and feedback information data combined with an adaptive learning mechanism, which can effectively improve the response speed and safety of the hydraulic system; the method can realize accurate control of pressure, flow and temperature and timely response to abnormal states during the operation of the hydraulic system, which can effectively avoid and solve the problems of overpressure, pressure and pressure fluctuation in complex working conditions, thereby ensuring the stable, safe and efficient operation of the system in complex working conditions.

[0059] Specifically, compared with the traditional technical solution, the application has the following advantages:

[0060] I. Intelligent prediction and adaptive adjustment: based on the time series prediction of the GRU prediction model, the potential faults of the hydraulic system can be identified in advance, and the current working condition of the system can be judged in real time combined with the SVM model, thereby realizing intelligent prediction and adaptive adjustment, and improving the response speed and decision accuracy of the system.

[0061] II. Multi-model cooperation: The present application combines GRU prediction model and SVM prediction model, which can not only predict future pressure, flow and temperature trends using GRU prediction model to avoid sudden failures, but also judge the state of the system in real time through SVM prediction model, greatly improving the intelligent level of the system, and thus achieving more accurate control decisions.

[0062] III. Continuous optimization and adaptive learning: Through real-time data and working condition characteristics feedback, adaptive learning is performed to continuously optimize control strategies and update model parameters, thereby gradually improving the regulation performance and stability of the system, enabling the system to adapt to changing working conditions and avoiding the static control limitations of traditional methods.

[0063] IV. Precise control and efficient response: The present application uses intelligent adjustment of pressure relief valves, servo valves and buffers in combination and joint regulation, which can break through the single on-off control or lagging response of traditional technology, and effectively ensure the stable operation of the system under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 is a flowchart of the present application;

[0065] Figure 2 is a control principle block diagram in the present application;

[0066] Figure 3 is a circuit control block diagram of the intelligent prediction and adaptive regulation system of the closed hydraulic system state of the raiseboring machine in the present application. DETAILED DESCRIPTION

[0067] The present application will be further described below in conjunction with the accompanying drawings.

[0068] As shown in Figure 1 and Figure 2 , the present application provides an intelligent prediction and adaptive regulation method for the closed hydraulic system state of a raiseboring machine, which adopts an intelligent prediction and adaptive regulation system for the closed hydraulic system state of a raiseboring machine, as shown in Figure 3 , an intelligent prediction and adaptive regulation system for the closed hydraulic system state of a raiseboring machine includes a pressure sensor group, a flow sensor group, a temperature sensor group, a power module, an intelligent control module, a hydraulic power part, a hydraulic system adjustment component and an auxiliary component.

[0069] The pressure sensor group includes a plurality of pressure sensors, which are respectively installed at the oil outlet of the hydraulic pump, the oil inlet of the hydraulic motor, the oil inlet side of the pressure relief valve and the oil outlet of the oil tank, for real-time acquisition of pressure change signals and sending to the intelligent control module.

[0070] The flow sensor group includes multiple flow sensors, which are respectively installed at the oil outlet of the hydraulic pump, the oil inlet of the hydraulic motor, the oil outlet of the hydraulic motor, and the main oil return pipeline, for collecting flow change signals in real time and sending them to the intelligent control module to ensure the balance of the closed-loop system flow;

[0071] The temperature sensor group includes multiple temperature sensors, which are respectively installed at the oil outlet of the hydraulic pump, in the oil tank, and at the hydraulic motor, for collecting temperature signals in real time and sending them to the intelligent control module to prevent overheating from causing system failure;

[0072] The intelligent control module includes a feedback unit, a parameter adjustment unit, and a processing unit. The feedback unit is used to receive data of the hydraulic system pressure, flow, and temperature before and after regulation, the regulation values of the pressure relief valve, the buffer, and the high-speed servo valve, the predicted value data of the machine learning model, and the system state label data predicted by the machine learning model, so as to form a closed-loop control logic;

[0073] The parameter adjustment unit is used to preprocess data, extract features, construct a GRU prediction model, and construct an SVM prediction model. Meanwhile, it is used to predict the pressure, flow, and temperature change trend based on the GRU prediction model and to judge the state based on the SVM prediction model, including the normal operation state of the closed-loop hydraulic system, the overpressure state of the closed-loop hydraulic system, and the pressure holding state of the closed-loop hydraulic system. Meanwhile, the parameter adjustment unit is used to train and optimize the GRU prediction model and the SVM prediction model based on the real-time feedback data of the feedback unit to realize real-time optimization and update of the machine learning model and achieve self-adaptive adjustment. The parameter adjustment unit is used to decide whether to adjust the pressure relief valve, the servo valve, and the buffer in the closed-loop hydraulic system, for example, when it is predicted that the closed-loop hydraulic pressure will exceed the safety threshold, a decision will be made to decide whether to send decision information to the processing unit, and the processing unit generates control instructions to adjust the pressure relief valve according to the decision information. When it is predicted that the closed-loop hydraulic system will have a large fluctuation, a decision will be made to decide whether to send decision information to the processing unit, and the processing unit generates control instructions to adjust the buffer according to the decision information. When it is predicted that the closed-loop system is unstable, a decision will be made to decide whether to send decision information to the processing unit, and the processing unit generates control instructions to realize accurate adjustment of the flow according to the decision information;

[0074] The processing unit is used to generate corresponding control instructions according to the received decision information and send them to the corresponding execution components, including the pressure relief valve, the buffer, and the servo valve, to dynamically adjust the parameters of the work of each control element, such as the opening degree of the valve of the pressure relief valve and the high-speed servo valve, and the damping of the buffer;

[0075] The hydraulic power component includes a hydraulic pump, a hydraulic motor and a supplementary oil valve group, the hydraulic pump is used to provide a supply of pressure oil to provide a power source for the closed hydraulic system, the hydraulic motor is used to receive the pressure oil provided by the hydraulic pump and convert it into mechanical rotary motion to drive the drill bit or other execution components, and the supplementary oil valve group is used to supplement hydraulic oil on the low pressure side to maintain stable pressure in the closed circuit.

[0076] The hydraulic system regulating component includes a pressure relief valve, a buffer and a servo valve; the pressure relief valve is used to perform pressure relief action according to the received instructions to actively regulate the pressure in the hydraulic system to avoid excessive pressure; the buffer is used to perform energy absorption action according to the received instructions to absorb and release sudden pressure fluctuations in the hydraulic system to prevent hydraulic shock conditions and ensure smooth operation of the system; the servo valve is a high-speed servo valve used to act according to the received instructions to adjust the flow of hydraulic oil according to the real-time needs of the hydraulic system to ensure that the hydraulic pump and motor work in the best state.

[0077] The auxiliary component includes a power module, a hydraulic oil tank and a display screen; the power module is used to supply power to various electrical components; the hydraulic oil tank is used to provide a supplementary oil function and collect hydraulic oil discharged by the pressure relief valve; the display screen is used to display real-time data for the operator to observe and monitor the data in real time, and the display screen has a human-computer interaction function to facilitate the operator to set the rig conditions and special needs;

[0078] The intelligent control module is connected with the pressure sensor group, the flow sensor group, the temperature sensor group, the power module, the hydraulic pump, the hydraulic motor, the supplementary oil valve group, the pressure relief valve, the buffer, the servo valve and the display screen.

[0079] The method specifically includes the following steps:

[0080] Step 1: During the historical operation of the closed hydraulic system of the raise borer, the sensor group installed in the closed hydraulic system of the raise borer is used to collect historical operation data and send it to the feedback unit for recording as historical operation data;

[0081] Step 2: Preprocess the historical operation data to obtain initial parameter data;

[0082] Step 3: Extract features from the initial parameter data to obtain pressure key features, flow key features and temperature key features data;

[0083] Step 4: Build and train a GRU prediction model;

[0084] S41: For the key feature data, use the sliding window technique to divide the key feature data into time series segments of fixed length, and make the length of each time series segment t; based on the obtained time series segment data, construct a data set, and then divide the data set into a training set and a test set according to a set proportion;

[0085] S42: Construct a GRU prediction model based on a gated recurrent unit (GRU);

[0086] The GRU prediction model includes an input layer, a GRU layer, and an output layer.

[0087] In the input layer, receive the input parameters X t of each time step t X t = [P t , Q t , T t , C zt , L vt , L bt ];

[0088] In the GRU layer, the hidden state of each time step depends on the hidden state of the previous time step and the current input data, including the update gate z t = σ (W z *[h t-1 , X t ] + b z ), where z t is the update gate, which determines that the hidden state at the current time is a weighted combination of the memory h t-1 at the previous time and the current input X t , σ is an activation function, W z is a weight matrix, h t-1 is the hidden state at the previous time, and b z is a bias term; the reset gate r t = σ (W r *[h t-1 , X t ] + b r ) controls the influence of the hidden state at the previous time on the candidate hidden state calculation; the candidate hidden state h' t = tanh (W h *[r t *h t-1 , X t ] + b h ) is calculated from the current input and the adjusted hidden state at the previous time, representing the "new memory" at the current time, and tanh is the hyperbolic tangent activation function; the final hidden state h t = (1-z t )*h't +z t *h t-1 The final hidden state is a weighted combination of the updated gate z t the previous time hidden state h t-1 and the current candidate hidden state h' t ;

[0089] In the output layer, the future pressure, flow and temperature [P' t+k , Q' t+k , T' t+k ] are predicted according to the final hidden state h t , and output, W o *h t +b o , W o is the weight matrix of the output layer, b o is the output bias term, and [P' t+k , Q' t+k , T' t+k ] is the predicted value at the future time.

[0090] S43: Construct the loss function L of the GRU prediction model according to formula (1);

[0091]

[0092] where N is the data amount, P' t+k , Q' t+k and T' t+k are the predicted values of pressure, flow and temperature at the future k time, P t+k , Q t+k and T t+k are the target values of pressure, flow and temperature at the future k time;

[0093] S44: Use the training set as input data to train the GRU prediction model, use the back propagation algorithm to calculate the gradient of the loss function L with respect to the model parameters, at the same time, use the ADAM optimization algorithm to update the model parameters, and minimize the loss function L, finally get the trained GRU prediction model;

[0094] Step five: Real-time prediction using the GRU prediction model;

[0095] S51: During the real-time operation of the open-end drill closed hydraulic system, the sensor group installed in the open-end drill closed hydraulic system is used to collect real-time operation data;

[0096] S52: first, using the sliding window technique, the real-time running data is divided into fixed-length time series segments, which are then input into the GRU prediction model for prediction, outputting pressure, flow and temperature prediction data;

[0097] Step six: based on the GRU prediction results, implement regulation and control;

[0098] The prediction data is sent to the parameter regulation unit for analysis and decision-making, and the decision information is sent to the processing unit, which obtains the corresponding control instructions according to the decision information, and then controls the action parameters of the relief valve, buffer and high-speed servo valve of the closed hydraulic system according to the corresponding control instructions until the closed hydraulic system returns to normal. In the regulation and control process, the feedback unit collects and records the pressure data, flow data, temperature data, rig working condition information data before regulation and control, the pressure data, flow data, temperature data, rig working condition information data after regulation and control, the adjustment value data of the relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, and the prediction value data of the GRU prediction model, as regulation and control feedback data; at the same time, the regulation and control feedback data is sent to the parameter regulation unit, and then the update data of a complete regulation and control cycle is obtained;

[0099] Step seven: use the feedback data to continuously optimize the GRU prediction model, and obtain the optimized GRU prediction model;

[0100] S71: repeatedly execute steps five and six for multiple times until the set regulation and control cycle is reached, and obtain update data for several regulation and control cycles; as an optimization, the set regulation and control cycle is 500 times;

[0101] S72: using the incremental learning method, input the update data into the GRU prediction model for continuous update and optimization, and finally obtain the optimized GRU prediction model;

[0102] Step eight: build and train the SVM prediction model;

[0103] S81: for the update data obtained in step seven for several regulation and control cycles, first, using the sliding window technique, the real-time running data is divided into fixed-length time series segments, and based on the obtained time series segment data, a data set is constructed, and then the data set is divided into a training set and a test set according to a set proportion;

[0104] S82: build an SVM prediction model based on support vector machine (SVM);

[0105] S83: use the training set as input data to train the SVM prediction model, and obtain the trained SVM prediction model;

[0106] In the training process, the input parameters X tand the state Y of the closed-circuit hydraulic system t Data as state label as input data, get the state label S of the system t

[0107] Step nine: real-time prediction using SVM prediction model

[0108] S91: During the real-time operation of the closed-circuit hydraulic system of the raise borer, the sensor group installed in the closed-circuit hydraulic system of the raise borer is used to collect real-time operation data

[0109] S92: Using sliding window technology, the prediction data is divided into fixed-length time sequence segments, which are then input into the SVM prediction model as input data for prediction, and the state judgment result is output

[0110] Step ten: based on the SVM prediction result to implement regulation and control

[0111] The state judgment data is sent to the parameter regulation unit, which uses the mapping relationship between pressure, flow, temperature and control strategy to obtain adjustment decisions based on the state judgment result, and sends the adjustment decisions to the processing unit, which obtains the corresponding control instructions according to the adjustment decisions, and then controls the action parameters of the pressure relief valve, buffer and high-speed servo valve of the closed-circuit hydraulic system according to the corresponding control instructions until the closed-circuit hydraulic system returns to normal. During the regulation and control process, the feedback unit collects and records the pressure data, flow data, temperature data, and borer working condition information data before regulation and control, the pressure data, flow data, temperature data, and borer working condition information data after regulation and control, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve, and the adjustment value data of the buffer, the adjustment value data of the buffer, the prediction value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model, as regulation and control feedback data. At the same time, the regulation and control feedback data is sent to the parameter regulation unit, and then the update data of a complete regulation and control cycle is obtained

[0112] Step eleven: continuously optimize the SVM prediction model using feedback data to obtain an optimized SVM prediction model

[0113] S111: repeatedly execute steps nine and ten for multiple times until the set regulation and control cycle is reached, and obtain update data for several regulation and control cycles

[0114] S112: using incremental learning, input the update data into the SVM prediction model for continuous update and optimization, and finally obtain the optimized SVM prediction model

[0115] Step twelve: use the optimized GRU prediction model and the optimized SVM prediction model to regulate and control the closed-circuit hydraulic system of the raise borer in real time ​

[0116] S121: During the real-time operation of the closed-loop hydraulic system of the raise boring machine, a sensor group installed in the closed-loop hydraulic system of the raise boring machine is used to collect real-time operation data;

[0117] S122: The real-time operation data is divided into time series segments of fixed length using the sliding window technique, and the obtained time series segments include pressure time series data P t , flow time series data Q t , temperature time series data T t , working condition information time series data C t , relief valve adjustment value time series data L zt , servo valve integral value time series data L vt , and buffer adjustment value time series data L bt .

[0118] S123: The obtained time series segments are input as input data into the optimized GRU prediction model for prediction, and output pressure prediction value P′ t+k , flow prediction value Q′ t+k , and temperature prediction value T′ t+k .

[0119] S124: The pressure prediction value P′ t+k , flow prediction value Q′ t+k , temperature prediction value T′ t+k , working condition information time series data C t , relief valve adjustment value time series data L zt , servo valve integral value time series data L vt , and buffer adjustment value time series data L bt at time t+k in the future are input as input data into the optimized SVM prediction model for prediction, and output system state label S t , the system state label S t includes 0, 1 and 2, respectively representing normal state, overpressure state and pressure holding state; specifically, when S t =0 indicates that the system works in the normal range without overpressure or pressure holding, when S t =1 indicates that the system pressure exceeds the safety threshold, and S t =2 indicates that the hydraulic oil flow is blocked and the system is in the pressure holding state;

[0120] S125: Based on the set maximum pressure threshold P max , pressure fluctuation threshold P fluct , minimum flow threshold Q min , and maximum temperature threshold Tmax Adjustments are made based on forecast data and state prediction results;

[0121] If P′ t+k >P max And S t =0 or S t If the value is 1, the processing unit controls the alarm device to issue a warning of potential overpressure. Simultaneously, it executes overpressure prevention measures and triggers overpressure regulation, increasing the opening degree L of the pressure relief valve. zt To release excess system pressure;

[0122] If P′ t+k <P max And S t If the value is 1, the following instruction is generated: Increase the opening degree L of the pressure relief valve. zt To release excess system pressure, adjust the opening L of the servo valve. vt This is to optimize flow control and avoid insufficient flow caused by overvoltage;

[0123] If Q′ t+k min And S t =0 or S t If the value is 2, the processing unit controls the alarm device to issue a warning of potential pressure buildup. Simultaneously, it executes pressure buildup prevention measures and triggers a flow replenishment adjustment, increasing the opening L of the high-speed servo valve. vt To increase traffic;

[0124] If Q′ t+k Q min And S t If the value is 2, the following instruction is generated: Increase the opening degree L of the servo valve. vt To increase flow rate and maintain a low opening degree L of the pressure relief valve. zt To prevent further accumulation of stress;

[0125] If T′ t+k >T max , and s t If the value is 0, the processing unit controls the alarm device to issue a warning of potential pressure buildup. Simultaneously, it executes pressure buildup prevention measures and triggers flow replenishment regulation, increasing the opening L of the high-speed servo valve. vt To increase traffic;

[0126] If predict P′ t+k It will fluctuate significantly in a short period of time, and be close to P. fluct Meanwhile, S t =0 or S t =1, then adjust the damping coefficient L of the buffer. bt Meanwhile, the opening degree L of the servo valve is finely adjusted based on the predicted flow rate changes.​vt , to ensure the flow stability.

[0127] As a preferred, in step two, the preprocessing process includes data cleaning by using interpolation method, data filtering by using Gaussian filter, and detecting outliers of data by using Isolation Forest algorithm, and normalizing the data after cleaning, filtering and outlier detection.

[0128] As a preferred, in step one, the historical operation data includes pressure data, flow data and temperature data; the pressure data includes pressure data at the hydraulic pump oil outlet, pressure data at the hydraulic motor oil inlet, pressure data at the inlet side of the pressure relief valve, pressure data at the oil tank oil outlet; the flow data includes flow data at the hydraulic pump oil outlet, flow data at the hydraulic motor oil inlet, flow data at the hydraulic motor oil outlet, flow data of the main oil return pipeline; the temperature data includes temperature data at the hydraulic pump oil outlet, temperature data of the oil tank, temperature data of the hydraulic motor.

[0129] As a preferred, in step three, the pressure key features include instantaneous pressure value, pressure fluctuation amplitude and pressure change rate; the flow key features include instantaneous flow value, flow change rate and cumulative flow; the temperature key features include instantaneous temperature value, temperature change rate and steady state temperature interval.

[0130] As a preferred, in S44 of step four, the process of updating model parameters is as follows:

[0131] By gradient descent method, the weight matrix W in the model is updated according to formula (2) z , W r , W h , W z and the bias term b z , b r , b h , b o ;

[0132]

[0133] In the formula, θ represents weight or bias, η is learning rate, is the gradient of the loss function to the parameters.

[0134] As a preferred, in step six, the pressure data includes pressure data at the hydraulic pump discharge port, pressure data at the hydraulic motor inlet, pressure data at the inlet side of the pressure relief valve, and pressure data at the oil tank outlet; the flow data includes flow data at the hydraulic pump discharge port, flow data at the hydraulic motor inlet, flow data at the hydraulic motor outlet, and flow data of the main oil return pipeline; the temperature data includes temperature data at the hydraulic pump discharge port, temperature data of the oil tank, and temperature data of the hydraulic motor; the rig working condition information data includes the working mode of the top drive rig cutter disc, including drilling mode, lifting mode and pushing mode; the pressure relief valve adjustment value data includes the opening adjustment value of the pressure relief valve; the servo valve adjustment value data includes valve port opening parameter, flow parameter and flow rate parameter; the buffer adjustment value data includes the damping coefficient adjustment value of the buffer; the prediction value data of the GRU prediction model includes future pressure trend data, flow trend data and temperature trend data.

[0135] As a preferred, in step eleven, the system state label data predicted by the SVM prediction model includes the state result of the system operating condition.

[0136] In order to ensure the accuracy and intelligence of the regulation, in S125 of step twelve, during the regulation process, the feedback unit collects and records the pressure data, flow data, temperature data, rig working condition information data before regulation, the pressure data, flow data, temperature data, rig working condition information data after regulation, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, the adjustment value data of the buffer, the prediction value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model, as regulation feedback data, and based on the regulation feedback data, the model parameters are continuously optimized and updated through adaptive learning.

[0137] The application provides an intelligent closed-circuit hydraulic adaptive regulation method based on a combination of a gated recurrent unit (GRU) model and a support vector machine (SVM) model, aiming to realize intelligent control and adaptive adjustment of the system by monitoring the pressure, flow and temperature parameters of the hydraulic system in real time. The method first predicts the future state of the pressure, flow and temperature of the hydraulic system through the GRU prediction model, thereby early warning possible overpressure and pressure failure, then judges the current working state of the system (such as normal, overpressure and pressure) in real time through the SVM prediction model, and generates decision information based on the prediction results of the GRU and the state judgment of the SVM. The decision information is sent to the processing unit, which generates corresponding control instructions according to the decision information, and adjusts the action of the pressure relief valve, servo valve and buffer in the hydraulic system in real time according to the instructions to ensure stable operation of the system. At the same time, the method can optimize and update the machine learning model in real time by continuously collecting new sensor data and feedback information data and combining the adaptive learning mechanism, which can effectively improve the response speed and safety of the hydraulic system; the method can realize accurate control of pressure, flow and temperature and timely response to abnormal state during the operation of the hydraulic system, which can effectively avoid and solve the problems of overpressure, pressure and pressure fluctuation in complex working conditions, so as to ensure stable, safe and efficient operation of the system under complex working conditions.

[0138] The method can make the raise boring machine respond to pressure regulation in advance when it is under high dynamic load during drilling (such as encountering hard rock or impact load) by using the collected data and the adaptive algorithm built in the intelligent control unit, for example, by dynamically adjusting the opening of the pressure relief valve and the damping parameters of the buffer, so that the system can realize the optimal state of pressure regulation under different construction environments (such as hard rock, soft rock, mudstone, etc.).

Claims

1. An intelligent prediction and adaptive control method for the state of a closed hydraulic system of a roof drill, characterized in that, Comprising the following steps: Step one: in the historical operation process of the closed hydraulic system of the raise drill rig, the sensor group installed in the closed hydraulic system of the raise drill rig is used to collect historical operation data and send it to the feedback unit for recording as historical operation data; Step two: pre-processing the historical operation data to obtain initial parameter data; Step three: extracting features from the initial parameter data to obtain pressure key features, flow key features and temperature key features data; Step four: constructing and training the GRU prediction model; S41: for the key feature data, use the sliding window technique to divide the key feature data into fixed-length time sequence segments, and make the length of each time sequence segment t; based on the obtained time sequence segment data, a data set is constructed, and then the data set is divided into a training set and a test set according to a set proportion; S42: constructing a GRU prediction model based on a gated recurrent unit; S43: constructing a loss function L of the GRU prediction model according to formula (1); where N is the data amount, P t ′ +k , Q t ′ +k , and T t ′ +k are predicted values of pressure, flow, and temperature at future k time, P t+k , Q t+k , and T t+k are target values of pressure, flow, and temperature at future k time S44: using the training set as input data to train the GRU prediction model, using a back propagation algorithm to calculate the gradient of the loss function L with respect to the model parameters, at the same time, using an ADAM optimization algorithm to update the model parameters, and minimizing the loss function L, finally obtaining the trained GRU prediction model; Step five: using the GRU prediction model for real-time prediction; S51: in the real-time operation process of the closed hydraulic system of the raise drill rig, the sensor group installed in the closed hydraulic system of the raise drill rig is used to collect real-time operation data; S52: first, use the sliding window technique to divide the real-time operation data into fixed-length time sequence segments, and then input them into the GRU prediction model as input data for prediction, outputting pressure, flow and temperature prediction data; Step six: implementing regulation and control based on the GRU prediction results; Send the prediction data to the parameter control unit for analysis and decision making, and send the decision information to the processing unit, which obtains the corresponding control instructions according to the decision information, and then controls the action parameters of the pressure relief valve, buffer and high-speed servo valve of the closed hydraulic system according to the corresponding control instructions until the closed hydraulic system returns to normal. In the regulation and control process, the feedback unit collects and records the pressure data, flow data, temperature data, drill rig working condition information data before regulation and control, the pressure data, flow data, temperature data, drill rig working condition information data after regulation and control, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, and the prediction value data of the GRU prediction model, as regulation and control feedback data; at the same time, send the regulation and control feedback data to the parameter control unit, and then obtain the update data of a complete regulation and control cycle; Step seven: continuously optimizing the GRU prediction model using the feedback data to obtain the optimized GRU prediction model; S71: repeatedly execute steps five and six for several times until a set regulation and control cycle is reached, and obtain update data of several regulation and control cycles; S72: In an incremental learning manner, the update data is input into the GRU prediction model for continuous updating and optimization, and finally an optimized GRU prediction model is obtained; Step eight: build and train the SVM prediction model; S81: For the update data obtained in step seven for several regulation periods, first use the sliding window technology to divide the real-time running data into fixed-length time sequence segments, and based on the obtained time sequence segment data, build a data set, and then divide the data set into a training set and a test set according to a set proportion; S82: Build an SVM prediction model based on support vector mechanism; S83: Use the training set as input data to train the SVM prediction model, and obtain a trained SVM prediction model; Step nine: use the SVM prediction model for real-time prediction; S91: In the real-time running process of the closed hydraulic system of the raise drill rig, the sensor group installed in the closed hydraulic system of the raise drill rig is used to collect real-time running data; S92: Use the sliding window technology to divide the prediction data into fixed-length time sequence segments, and then input them into the SVM prediction model as input data for prediction, and output the state judgment result; Step ten: implement regulation based on the SVM prediction result; The state judgment data is sent to the parameter regulation unit, which obtains adjustment decisions based on the state judgment result using the mapping relationship between pressure, flow rate, temperature and control strategy, and sends the adjustment decisions to the processing unit, which obtains corresponding control instructions according to the adjustment decisions, and then controls the action parameters of the pressure relief valve, buffer and high-speed servo valve of the closed hydraulic system according to the corresponding control instructions until the closed hydraulic system returns to normal. In the regulation process, the feedback unit collects and records the pressure data, flow rate data, temperature data, rig working condition information data before regulation, the pressure data, flow rate data, temperature data, rig working condition information data after regulation, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, the adjustment value data of the buffer, the prediction value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model as regulation feedback data. At the same time, the regulation feedback data is sent to the parameter regulation unit, and a complete regulation period update data is obtained; Step eleven: continuously optimize the SVM prediction model using the feedback data to obtain an optimized SVM prediction model; S111: Repeat steps nine and ten multiple times until a set regulation period is reached, and obtain update data for several regulation periods; S112: In an incremental learning manner, the update data is input into the SVM prediction model for continuous updating and optimization, and finally an optimized SVM prediction model is obtained; Step twelve: use the optimized GRU prediction model and the optimized SVM prediction model to regulate the closed hydraulic system of the raise drill rig in real time; S121: In the real-time running process of the closed hydraulic system of the raise drill rig, the sensor group installed in the closed hydraulic system of the raise drill rig is used to collect real-time running data; S122: first divide the real-time running data into fixed-length time series segments using the sliding window technique, and the obtained time series segments include pressure time series data P t , flow time series data Q t , temperature time series data T t , working condition information time series data C t , pressure relief valve adjustment value time series data L zt , servo valve integral value time series data L vt , buffer adjustment value time series data L bt ; S123: input the obtained time series segment as input data into the optimized GRU prediction model for prediction, output the pressure prediction value P at future time t+k t +k , the flow prediction value Q t +k , and the temperature prediction value T t +k ;​​​ S124: the pressure prediction value P at future time t+k is calculated t +k t +k t +k t zt vt bt t t The system state label S includes 0, 1 and 2, respectively representing normal state, overpressure state and pressure holding state.​​​​​​​​​​​ S125: regulating based on the set maximum pressure threshold P max , a pressure fluctuation threshold P fluct , a minimum flow threshold Q min and a maximum temperature threshold T max , according to the prediction data and the state prediction result. If P t ′ +k > P max , and S t = 0 or S t = 1, the processing unit controls the alarm device to give a warning of possible overpressure, at the same time, executes overpressure prevention measures, and triggers overpressure regulation action to increase the opening degree L zt of the pressure relief valve to release excess system pressure; If P t +k <P nax , and S t = 1, the following instructions are generated: increase the opening degree L zt of the pressure relief valve to release excess system pressure, and adjust the opening degree L vt of the servo valve to optimize flow control and avoid the problem of insufficient flow caused by overpressure;​ If Q t +k <Q min , and S t = 0 or S t = 2, the processing unit controls the alarm device to act to give a warning that pressure build-up is likely to occur, at the same time, executes pressure build-up prevention measures, and triggers flow supplement adjustment action to increase the opening degree L vt of the high-speed servo valve to increase the flow rate.​ If Q t ′ +k Q min And S t If the value is 2, the following instruction is generated: Increase the opening degree L of the servo valve. vt To increase flow rate and maintain a low opening degree L of the pressure relief valve. zt To prevent further accumulation of stress; If T t ′ +k >T max And S t If the value is 0, the processing unit controls the alarm device to issue a warning of potential pressure buildup. Simultaneously, it executes pressure buildup prevention measures and triggers flow replenishment regulation, increasing the opening L of the high-speed servo valve. vt To increase traffic; If the predicted P t +k fluct will fluctuate greatly in a short time, and is close to P t = 0 or S t = 1, the damping coefficient L bt of the buffer is adjusted, and simultaneously, the opening L vt of the servo valve is fine-tuned according to the predicted flow change to ensure the flow stability.​​ 2. The method of claim 1, wherein the method further comprises: In step two, the preprocessing process includes data cleaning using interpolation method, data filtering using Gaussian filter, anomaly value detection using Isolation Forest algorithm, and normalization of the cleaned, filtered and anomaly value detected data.

3. The method of claim 1 or 2, wherein the method further comprises: In step one, the historical operation data includes pressure data, flow data and temperature data; the pressure data includes pressure data at the hydraulic pump discharge port, pressure data at the hydraulic motor inlet, pressure data at the inlet side of the pressure relief valve, and pressure data at the oil tank outlet; the flow data includes flow data at the hydraulic pump discharge port, flow data at the hydraulic motor inlet, flow data at the hydraulic motor outlet, and flow data of the main oil return pipeline; the temperature data includes temperature data at the hydraulic pump discharge port, temperature data of the oil tank, and temperature data of the hydraulic motor.

4. The method of claim 3, wherein the method further comprises: In step three, the pressure key features include instantaneous pressure value, pressure fluctuation amplitude and pressure change rate; the flow key features include instantaneous flow value, flow change rate and cumulative flow; and the temperature key features include instantaneous temperature value, temperature change rate and steady state temperature interval.

5. The method of claim 1, wherein the method further comprises: In S44 of step four, the process of updating the model parameters is as follows: The weight matrix W in the model is updated according to formula (2) by gradient descent method z , W r , W h , W z and the bias term b z , b r , b h , b o ; In the formula, θ represents a weight or bias, η is a learning rate, is the gradient of the loss function with respect to the parameters.

6. The method of claim 1, wherein the method further comprises: In step six, the pressure data includes pressure data at the hydraulic pump discharge port, pressure data at the hydraulic motor inlet, pressure data at the inlet side of the pressure relief valve, and pressure data at the oil tank outlet; the flow data includes flow data at the hydraulic pump discharge port, flow data at the hydraulic motor inlet, flow data at the hydraulic motor outlet, and flow data of the main oil return pipeline; the temperature data includes temperature data at the hydraulic pump discharge port, temperature data of the oil tank, and temperature data of the hydraulic motor; the rig working condition information data includes the working mode of the top drive rig cutter, including drilling mode, lifting mode and pushing mode; the adjustment value data of the pressure relief valve includes the opening adjustment value of the pressure relief valve; the adjustment value data of the servo valve includes the valve opening parameter, flow parameter and flow rate parameter; the adjustment value data of the buffer includes the damping coefficient adjustment value of the buffer; and the prediction value data of the GRU prediction model includes future pressure trend data, flow trend data and temperature trend data.

7. The method of claim 1, wherein the method further comprises: In step eleven, the system state label data predicted by the SVM prediction model includes the state result of the system operation condition.

8. The method of claim 1, wherein the method further comprises: In S125 of step twelve, in the regulation process, the feedback unit collects and records the pressure data, flow data, temperature data, rig working condition information data before regulation, the pressure data, flow data, temperature data, rig working condition information data after regulation, the adjustment value data of the pressure relief valve, the adjustment value data of the servo valve and the adjustment value data of the buffer, the adjustment value data of the buffer, the prediction value data of the GRU prediction model, and the system state label data predicted by the SVM prediction model as regulation feedback data, and performs adaptive learning based on the regulation feedback data, and continuously optimizes and updates the model parameters.

Images