Closed-loop control system and method based on deep-sea ore raising pipe ore pulp concentration monitoring

Abstract

The invention discloses a closed-loop control system and method based on deep-sea ore raising pipe ore pulp concentration monitoring, and the system comprises a water surface support platform control center which is used for outputting a control strategy; the hydraulic ore injector is used for receiving the control strategy and carrying out self-adaptive adjustment according to the control strategy; wherein the water surface support platform control center comprises an intelligent control unit and a controller, and the intelligent control unit is used for collecting operation parameter data of the hydraulic ore injector, processing the operation parameter data and predicting in real time to obtain a solid phase volume concentration predicted value; the controller is used for receiving the solid phase volume concentration predicted value and determining a control strategy according to the solid phase volume concentration predicted value; the intelligent control unit comprises a data acquisition module, a data processing module and a prediction module; the concentration of ore pulp in the pipeline can be more accurately controlled through the system, and the system can better adapt to complex working conditions such as submarine topography and ore supply quantity fluctuation.

Description

Technical Field

[0001] This invention relates to the field of slurry concentration control technology, and in particular to a closed-loop control system and method for slurry concentration monitoring based on deep-sea ore lifting pipes. Background Technology

[0002] Deep-sea ore lifters are the core conveying devices of deep-sea mining systems. Their function is to transport seabed-mined ore (which exists in the form of slurry, a mixture of solid mineral particles and liquid) through vertical pipes to mining vessels or platforms on the sea surface. Deep-sea ore lifter slurry concentration monitoring refers to the technical process of real-time or periodic detection of the solid content of ore slurry in the conveying pipelines of deep-sea mining systems.

[0003] Currently, two main types of ore hoisting systems are used in deep-sea mining: one is the traditional hydraulic hoisting system, which uses submersible pumps (multi-stage centrifugal pumps) to generate upward-flowing slurry in risers, lifting the ore from the seabed to the surface support vessel. This system relies on submersible pumps as the core power source, but in practical applications, the rapid wear of the pump impeller (approximately 3mm within 1.5 hours) and high risk of clogging lead to frequent maintenance and low production efficiency. Another type is the closed-loop hoisting system: seawater is driven by a surface-mounted pump to circulate in a vertical closed pipeline system connecting the seabed mining site and the surface work vessel. Ore is injected into the high-pressure closed pipeline system through a special hydraulic ore injector. This injector usually consists of a flow control device, multiple ore bins, an ore feeder, and a mixer. Through the sequential control of valves, the ore is quantitatively supplied to the mixer by the ore feeder. Under the carrying force of the high-pressure water flow, it is fully mixed with the circulating seawater to form a stable slurry, which is then injected into the riser pipe. A closed loop is then formed through a closed pipeline (the water supply pipe is connected to the riser pipe). This allows the pump to only drive the seawater circulation without directly contacting the slurry containing coarse particles, thus avoiding the risks of pump wear and particle blockage in traditional hydraulic hoisting systems.

[0004] However, during the process of using a deep-sea closed-loop circulation system for ore hoisting, the instantaneous switching of ore bins can easily cause fluctuations in the slurry concentration in the pipeline. The existing control logic is unable to cope with the instantaneous flow rate fluctuations caused by switching between multiple bins, resulting in unstable slurry concentration in the hoisting pipe. If it is too high, it will easily clog the pipe, and if it is too low, it will reduce the efficiency of commercial mining. It is also difficult to adapt to complex working conditions such as seabed topography and fluctuations in ore supply. Summary of the Invention

[0005] In view of the above-mentioned prior art, the present invention provides a closed-loop control system and method for monitoring slurry concentration in deep-sea ore lifting pipes, which mainly solves the technical problems existing in the background art.

[0006] To achieve the above objectives, the technical solution of this invention is implemented as follows: In a first aspect, the present invention provides a closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring, comprising: The surface support platform control center is used to output control strategies; A hydraulic ore injector is used to receive a control strategy and perform adaptive adjustments according to the control strategy. The water surface support platform control center includes an intelligent control unit and a controller. The intelligent control unit is used to collect the operating parameter data of the hydraulic ore injector, process the operating parameter data, and predict the solid volume concentration in real time. The controller is used to receive the solid volume concentration prediction value and determine the control strategy based on the solid volume concentration prediction value. The intelligent control unit includes: The data acquisition module is used to collect the operating parameter data of the hydraulic ore injector; The data processing module is used to process the operating parameter data, filter the processed operating parameter data, and obtain key features; The prediction module is used to train the XGB-PSO concentration prediction model by inputting the key features into it, to obtain an optimized XGB-PSO concentration prediction model, to input real-time operating parameter data into the optimized XGB-PSO concentration prediction model, and to output the predicted value of the solid volume concentration of the slurry in the ore lifting pipe. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is:

[0007]

[0008] Among them, V newThe updated velocity is represented by w; the inertia weight is w; the current velocity is v; c1 is the individual learning factor; r1 is a random number 1; P best Let X be the individual's historical best position, and C be the current position; c2 be the group learning factor; r2 be the random number 2; G be the random number 2. best X is the globally optimal position for the group; new The updated position; V new For the updated speed; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

[0009] Optionally, the hydraulic ore injector includes: an ore feed bin, a water pump, a flow control device, an ore tank, a mixer, a lifting pipe, a separator, and a storage device; The output end of the ore feed hopper is connected to the input end of the flow direction control device via a feed pipe. A feed valve is installed on the feed pipe. The output end of the water supply pump is connected to the input end of the flow direction control device via a water supply pipe. A water supply valve and a flow meter are installed on the water supply pipe. The output end of the flow direction control device is connected to several ore tanks via conveying pipes. A control valve is installed on the conveying pipes. The output end of each ore tank is connected to the input end of the mixer. A torque sensor is installed inside the mixer. The mixer is connected to the separator via a lifting pipe. Pressure sensors are installed at both the inlet and outlet ends of the lifting pipe. The separator has a first output port and a second output port. The first output port is connected to the input end of the water supply pump via a recovery pipe. The second output port is connected to the storage device via a slurry conveying pipe.

[0010] Optionally, the operating parameter data includes feeder speed data, water supply pipe flow data, injector valve timing status data, ore lifting pipe differential pressure data, and mixer power data.

[0011] Optionally, the step of processing the operating parameter data and filtering the processed operating parameter data to obtain key features includes: Step S1: Use leverage value analysis to remove outliers from the collected data; Step S2: Use the LASSO regression algorithm to screen out key feature variables from the data after outlier removal.

[0012] Optionally, step S1 specifically includes: Step S11: Normalize the operating parameters; Step S12: Construct a feature matrix based on the normalized data, and calculate the hat matrix based on the feature matrix; Step S13: Calculate the lever value for each normalized running parameter data point using the diagonal elements of the hat matrix; Step S14: Preset an anomaly detection threshold, compare the anomaly detection threshold with the leverage value, and remove leverage values ​​that exceed the anomaly detection threshold to obtain data filtered for anomalies.

[0013] Optionally, step S2 specifically includes: Step S21: Perform data standardization on the data after outlier removal; Step S22: Construct a penalty objective function that includes an L1 regularization term. The expression of the penalty objective function is:

[0014] in, The target variable is pulp concentration. As input features, The penalty coefficient, is the regression coefficient.

[0015] Optionally, determining the control strategy based on the predicted solid volume concentration includes: A preset concentration safety threshold is set, the predicted solid phase volume concentration is compared with the concentration safety threshold, and a control strategy is determined based on the comparison result. If the predicted solid volume concentration is greater than the concentration safety threshold, the rotation speed of the ore feed hopper is reduced or the water supply valve is opened to dilute the slurry in the mixer. If the predicted solid volume concentration is less than the concentration safety threshold, it is determined whether the predicted solid volume concentration is within the optimal concentration range, and the hydraulic ore injector is controlled to make adaptive adjustments based on the determination result.

[0016] Optionally, determining whether the predicted solid volume concentration is within the optimal concentration range, and controlling the hydraulic ore injector to adaptively adjust based on the determination result, includes: If the predicted solid volume concentration is within the optimal concentration range, steady-state control is performed, and the rotation speed of the ore feed hopper is finely adjusted to maintain the slurry concentration within the optimal concentration range. If the predicted solid volume concentration is not in the optimal concentration range, transient compensation is performed to increase the rotation speed of the ore feed hopper and compensate for the concentration fluctuations caused by the switching gap of the ore tank.

[0017] Secondly, the present invention also provides a closed-loop control method based on deep-sea ore lifting pipe slurry concentration monitoring, comprising the following steps: The real-time operating parameter data of the hydraulic ore injector is collected by the intelligent control unit. The real-time operating parameter data is input into the optimized XGB-PSO concentration prediction model to predict the slurry concentration in the ore lifting pipe and output the predicted value of the solid phase volume concentration at the current moment. The current solid volume concentration prediction value is sent to the controller, which compares the current solid volume concentration prediction value with a preset concentration safety threshold and determines the control strategy based on the comparison result. The controller sends the control strategy to the hydraulic ore injector, which responds to adjust its operating parameters, thereby adjusting the slurry concentration entering the ore lifting pipe at the next moment, so that the slurry concentration entering the storage device meets the requirements. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is:

[0018]

[0019] Among them, V new The updated velocity is represented by w; the inertia weight is w; the current velocity is v; c1 is the individual learning factor; r1 is a random number 1; P best Let X be the individual's historical best position, and C be the current position; c2 be the group learning factor; r2 be the random number 2; G be the random number 2. best X is the globally optimal position for the group; new The updated position; V new For the updated speed; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

[0020] The beneficial effects of this invention are as follows: The closed-loop control system based on slurry concentration monitoring in deep-sea ore lifting pipes provided by this invention collects various operating parameter data through a data acquisition module, then filters the operating parameter data to obtain key features, and then uses an optimized XGB-PSO concentration prediction model to predict the slurry concentration in the ore lifting pipe, obtaining the predicted value of the solid volume concentration of the slurry in the pipe. Then, based on the predicted value of the solid volume concentration, the corresponding control strategy is determined and targeted adjustments are made, thereby more accurately controlling the concentration of slurry in the ore lifting pipe and better adapting to complex working conditions such as seabed topography and fluctuations in ore supply.

[0021] The XGB-PSO concentration prediction model optimized in the intelligent control unit is used to predict the solid volume concentration in the ore lifting pipe at the current moment. Based on the predicted solid volume concentration at the current moment, the feeding / water supply operation parameters of the hydraulic ore injector are adjusted for the next moment, thereby adjusting the slurry concentration entering the ore lifting pipe at the next moment. Finally, slurry with the required concentration is obtained. According to this method, the pulsation caused by the ore tank switching gap can be avoided from affecting the subsequent slurry concentration.

[0022] The closed-loop control system based on slurry concentration monitoring in deep-sea ore lifting pipes provided by this invention can also avoid the problem of densitometers being directly installed in the ore lifting pipes and used to detect the slurry concentration in the pipes, which would cause the densitometers to wear due to direct contact with the slurry. At the same time, using this system is easier to maintain and can also reduce the maintenance costs of deep-sea equipment. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the adaptive optimization principle of the optimized XGB-PSO model provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the seawater circulation process in the closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the slurry formation process in a closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring provided in an embodiment of the present invention. Figure 5This is a schematic diagram of the workflow of the water surface support platform control center provided in this embodiment of the invention; Figure 6 This is a schematic flowchart of a closed-loop control method based on deep-sea ore lifting pipe slurry concentration monitoring provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Optional embodiments of the present invention are described in detail below; however, in addition to these detailed descriptions, the present invention may have other embodiments.

[0025] Example 1 Please refer to the attached document. Figure 1 and attached Figure 2 This invention provides a closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring, comprising: The surface support platform control center is used to output control strategies; A hydraulic ore injector is used to receive a control strategy and perform adaptive adjustments according to the control strategy. The water surface support platform control center includes an intelligent control unit and a controller. The intelligent control unit is used to collect the operating parameter data of the hydraulic ore injector, process the operating parameter data, and predict the solid volume concentration in real time. The controller is used to receive the solid volume concentration prediction value and determine the control strategy based on the solid volume concentration prediction value. The intelligent control unit includes: The data acquisition module is used to collect the operating parameter data of the hydraulic ore injector; The data processing module is used to process the operating parameter data, filter the processed operating parameter data, and obtain key features; The prediction module is used to train the XGB-PSO concentration prediction model by inputting the key features into it, to obtain an optimized XGB-PSO concentration prediction model, to input real-time operating parameter data into the optimized XGB-PSO concentration prediction model, and to output the predicted value of the solid volume concentration of the slurry in the ore lifting pipe. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is:

[0026]

[0027] Among them, V new The updated velocity is represented by w; the inertia weight is w; the current velocity is v; c1 is the individual learning factor; r1 is a random number 1; P best Let X be the individual's historical best position, and C be the current position; c2 be the group learning factor; r2 be the random number 2; G be the random number 2. best X is the globally optimal position for the group; new The updated position; V new For the updated speed; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

[0028] Specifically, in existing technologies, physical flow meters and densitometers are used to monitor the concentration inside the ore lifting pipe. However, physical flow meters are contact sensors that need to be built into the ore lifting pipe and must directly contact the slurry inside to obtain flow data. Because they directly contact the slurry, they are prone to problems such as local blockage and wear. While densitometers are non-contact sensors that do not need to directly contact the slurry, they calculate the solid volume concentration indirectly. In this way, when the ore bin is switched, the densitometer may not be able to capture the fluctuation in real time due to the sudden increase in slurry concentration, which can easily lead to a lag in the dynamic concentration pulsation response. In this technical solution, the current slurry concentration is predicted by the intelligent control unit of the water surface support platform control center. Specifically, the data acquisition module collects the operating parameter data of the hydraulic ore injector. This data reflects the operating status of the lifting pipe, such as resistance and water volume, thus indicating blockage and transport efficiency, providing a basis for subsequent prediction of the solid volume concentration within the lifting pipe. Next, the data processing module processes the operating parameter data, identifying key features that significantly influence the prediction of the solid volume concentration within the lifting pipe. The optimized XGB-PSO concentration prediction model in the prediction module then performs the prediction to obtain the predicted solid volume concentration of the slurry within the pipe. The optimized XGB-PSO concentration prediction model is based on the XGBoost algorithm, which is used to construct an XGBoost model, and then the PSO algorithm is applied to... The XGBoost model performs hyperparameter optimization, which better captures complex patterns in the operating parameter data, enhances the model's robustness and generalization ability, and thus improves the accuracy of solid volume concentration prediction, obtaining the predicted solid volume concentration value at the current moment. This predicted solid volume concentration value is then sent to the controller, which determines the control strategy based on the predicted solid volume concentration value to adjust the operating parameters of the hydraulic ore injector at the next moment, thereby adjusting the slurry concentration at the next moment to obtain a slurry with the required concentration. In other words, this technical solution can adjust the slurry concentration at the next moment based on the slurry concentration at the previous moment to achieve dynamic closed-loop control. This avoids the problem of pulsation in the slurry concentration in the pipeline caused by the instantaneous switching of the ore bin, which leads to instability in the slurry concentration in the lifting pipe, ensuring its stability in complex deep-sea conditions.

[0029] Furthermore, this control system can overcome the problem of easy failure of deep-sea sensors (coarse-particle slurry can cause wear on pumps and sensors). This invention uses the power of easily maintained surface pumps and pipeline pressure difference data to predict the subsequent solid volume concentration, avoiding the problem of high wear of the density meter due to direct contact with the slurry. It also significantly reduces the maintenance cost of deep-sea equipment.

[0030] As an optional implementation method, please refer to the appendix. Figure 3 and 4 The hydraulic ore injector includes: an ore feed bin, a water pump, a flow control device, an ore tank, a mixer, a lifting pipe, a separator, and a storage device; The output end of the ore feed hopper is connected to the input end of the flow direction control device via a feed pipe. A feed valve is installed on the feed pipe. The output end of the water supply pump is connected to the input end of the flow direction control device via a water supply pipe. A water supply valve and a flow meter are installed on the water supply pipe. The output end of the flow direction control device is connected to several ore tanks via conveying pipes. A control valve is installed on the conveying pipes. The output end of each ore tank is connected to the input end of the mixer. A torque sensor is installed inside the mixer. The mixer is connected to the separator via a lifting pipe. Pressure sensors are installed at both the inlet and outlet ends of the lifting pipe. The separator has a first output port and a second output port. The first output port is connected to the input end of the water supply pump via a recovery pipe. The second output port is connected to the storage device via a slurry conveying pipe.

[0031] Specifically, when deep-sea ore hoisting is required, control commands are sent from the surface support platform control center to the hydraulic ore injector. The hydraulic ore injector's function is to "inject" low-pressure ore into the high-pressure water circuit. The control commands open the feed valve of the ore feed hopper, and the flow control device determines the corresponding ore tank, opening the corresponding control valve. The ore in the ore feed hopper flows through the flow control device and control valve into the corresponding ore tank. If the ore tank is detected to be full, the valve of another ore tank is controlled to open. When the flow control device switches to another ore tank, the ore from the ore feed hopper enters the other ore tank. Simultaneously, the water supply valve of the water pump opens, allowing water from the pump to enter the water supply pipe. Seawater in the water supply pipe is then fed into the corresponding ore tank via the flow control device. If the flow control device switches to another ore tank, the seawater similarly enters that tank and then flows into the mixer. In the mixer, the ore and seawater are mixed to obtain a slurry, which is then transported to the pumping station. In the mining pipe; during the above process, the data acquisition module of the intelligent control unit collects the current water supply pipe flow rate, feeder speed, ore lifting pipe pressure difference, injector valve timing status, ore lifting pipe pressure difference data and mixer power data, and sends them to the data processing module and prediction module. The module predicts the solid volume concentration in the ore lifting pipe and outputs the predicted solid volume concentration value. The predicted solid volume concentration value is sent to the controller, which judges and determines the control strategy. The control strategy is sent to the hydraulic ore injector to control the operating parameters of the hydraulic ore injector at the next moment. Based on the concentration prediction at the previous moment, the concentration at the next moment is adjusted so that the solid volume concentration of the slurry that finally enters the storage device through the ore lifting pipe meets the requirements. Then, under the action of the hydraulic pressure difference at both ends of the ore lifting pipe, the slurry in the ore lifting pipe is transported upward until it enters the separator. In the separator, the ore and seawater are separated. The separated ore is stored in the storage device, and the separated seawater is treated and then enters the water supply pump to realize a closed loop. Meanwhile, since the ore feed silo does not fill all the ore tanks simultaneously before the subsequent mixing of ore and seawater, but instead has multiple ore tanks, in this embodiment there are two ore tanks. In actual applications, the number of ore tanks can be set according to the actual situation, and there is no limit to the number of ore tanks. The ore feed silo is filled with ore one by one. That is, when one ore tank is full of ore, the flow control device will switch to the next ore tank, and the ore from the ore feed silo will be injected into the next ore tank. During the process of the flow control device switching from one ore tank to the next ore tank, there will be a switching gap. This switching gap makes the switching process take time, during which the ore supply is briefly interrupted, causing the slurry concentration in the pipeline to pulsate. In this technical solution, the solid volume concentration value in the lifting pipe is predicted by the optimized XGB-PSO concentration prediction model in the intelligent control unit. The adjustment strategy is then determined based on the solid volume concentration prediction value, thereby avoiding the impact of the pulsation caused by the ore tank switching gap on the subsequent slurry concentration.

[0032] It should be noted that the specific steps for detecting that the ore tank is full and switching to another ore tank are as follows: by volume integration based on the feeder speed (soft measurement): given the feeder speed and the feeder's single-cycle displacement, when the calculated volume reaches the ore tank's capacity, the water surface support platform control center determines it to be "full" and triggers the switch (i.e., switching from one ore tank to another ore tank).

[0033] The solid volume concentration predicted by the intelligent control unit is the slurry concentration at the mixer outlet, which is about to enter the bottom of the lifting pipe. The optimized XGB-PSO concentration prediction model predicts the slurry concentration at time T in real time based on the feed-forward parameters on the feed side (feeder speed data, water flow data, injector valve timing status data) and the feedback parameters on the conveying side (lifting pipe pressure difference data, mixer power data). Based on this, the feed and water supply parameters at time T+1 are adjusted to achieve dynamic closed-loop control. For example, if the predicted concentration at time T is high, the controller immediately reduces the feeder speed or increases the water injection, so that the concentration of the new slurry entering the pipe at time T+1 will decrease.

[0034] As an optional implementation, the operating parameter data includes feeder speed data, water supply pipe flow data, injector valve timing status data, ore lifting pipe differential pressure data, and mixer power data.

[0035] Specifically, the feeder speed data refers to the real-time speed of the rotary feeder, a key parameter for controlling slurry concentration, detected by a speed sensor on the motor drive shaft of the rotary feeder installed at the bottom of the ore feed hopper; the water supply pipe flow rate data refers to the real-time flow velocity of the water supply pipe, which affects the slurry concentration and dilution efficiency inside the ore lifting pipe, detected by a flow meter on the water supply pipe; the injector valve timing status data is obtained by directly collecting the switching timing signals of the valve limit switches, installed on the actuators of various key valves in the hydraulic ore injector. The parameters are: the drive shaft of the flow control device, and the valve stems of the feed valve and water supply valve; the pressure difference of the ore lifting pipe is the pressure difference between the two ends of the ore lifting pipe, that is, the difference between the inlet pressure and the outlet pressure of the ore lifting pipe. It can reflect the overall density of the solid-liquid two-phase flow and the flow resistance of the slurry in the ore lifting pipe. It is a key indicator for judging pipe blockage, wear or deposition; the power data of the mixer is the real-time load data in the mixer, which is detected by the torque sensor in the mixer; by coordinating the acquisition of the above parameters, a basis can be provided for the prediction of the subsequent solid phase volume concentration.

[0036] In another embodiment, in addition to acquiring the operating parameter data of the hydraulic ore injector, the data acquisition module can also acquire other physical parameter data, including the sound frequency of particles impacting the inner wall of the lifting pipe obtained by an acoustic sensor, the vibration frequency inside the lifting pipe obtained by a vibration sensor, and the concentration of solid particles in the slurry inside the lifting pipe obtained by an optical turbidimeter. The operating parameter data and physical parameter data are processed and then input into the optimized XGB-PSO concentration prediction model to predict the solid phase volume concentration, so as to enhance the characteristic capture of solid phase content and obtain more accurate prediction results.

[0037] As an optional implementation, the process of processing the operating parameter data and filtering the processed operating parameter data to obtain key features includes: Step S1: Use leverage value analysis to remove outliers from the collected data; Step S2: Use the LASSO regression algorithm to screen out key feature variables from the data after outlier removal.

[0038] As an optional implementation, step S1 specifically includes: Step S11: Normalize the operating parameters; Step S12: Construct a feature matrix based on the normalized data, and calculate the hat matrix based on the feature matrix; Step S13: Calculate the lever value for each normalized running parameter data point using the diagonal elements of the hat matrix; Step S14: Preset an anomaly detection threshold, compare the anomaly detection threshold with the lever value, and remove the sampling points corresponding to the lever values ​​that exceed the anomaly detection threshold to obtain the dataset filtered by anomalies.

[0039] Specifically, the leverage value analysis method is used to remove outliers from the collected data. The aim is to address the impact of the unique "particle impact noise" in deep-sea solid-liquid two-phase flow monitoring on the accuracy of subsequent concentration prediction models. The specific steps are as follows: The collected raw operating parameter data—including the ore lifting pipe pressure difference data P, water supply pipe flow rate data Q, mixer power data W, feeder speed data R, and discrete variable injector valve timing status S—are used to construct the raw dataset. Continuous variables (P, Q, W, R) are extracted from the raw dataset to construct an n×p dimensional feature matrix X, where n is the number of sampling points and p is the feature dimension. The formula is then used to... The hat matrix H is calculated, reflecting the geometric relationship between observed and predicted values. The leverage value of each normalized operating parameter data point is calculated using the diagonal elements of the hat matrix H (the main diagonal elements of the hat matrix H are extracted as the leverage value of the i-th sampled operating parameter data point; the leverage value quantifies the deviation of the data point from the data center in the feature space). Then, an anomaly detection threshold is set. When the leverage value calculated from the sensor data combination exceeds this threshold at a certain moment, the data is automatically identified as an outlier and removed from the dataset to prevent it from participating in subsequent calculations. This process needs to be iteratively executed, by refitting the model, updating the leverage value calculation, and repeating the threshold detection and removal operations until no new outliers appear in continuous iterations, ultimately obtaining the dataset filtered for outliers. High leverage values ​​(leverage values ​​exceeding the anomaly detection threshold) typically correspond to non-logical spikes generated by the instantaneous impact of large deep-sea ore particles on the pressure sensor probe; that is, although the value is high, it has no correlation with fluid pressure. Removing these spikes prevents the model from misjudging "impact noise" as "precursor to pipe blockage," improving the accuracy of the model's concentration prediction.

[0040] As an optional implementation, step S2 specifically includes: Step S21: Perform data standardization on the data after outlier removal; Step S22: Construct a penalty objective function that includes an L1 regularization term. The expression of the penalty objective function is:

[0041] in, The target variable is pulp concentration. As input features, The penalty coefficient is... These are the regression coefficients; Specifically, the dataset (containing P, Q, W, R, and S) after outlier removal is standardized to eliminate differences in physical dimensions and prepare for LASSO regression; then, a penalty objective function is used to select coefficients. The corresponding physical quantities are used as key characteristic variables. The key characteristic variables selected specifically include: the pressure difference in the ore lifting pipe. p: Physical characterization of the gravitational density of the solid-liquid mixture; Mixer power W: Physical characterization of the viscous resistance of the fluid at high concentrations; Water supply pipe flow rate Q: Physical characterization of the dilution effect and the speed of the conveying carrier; Feeder speed RPM: Physical characterization of the volumetric conveying rate of the ore source (feedforward signal); Injector valve timing state: Physical characterization of the flow field pulsation state caused by the switching of the hydraulic ore injector (transient characteristics).

[0042] In this technical solution, the lever value analysis method is introduced to automatically remove sensor outliers, and the LASSO regression algorithm is used to screen and solve the problem of "multicollinearity" (parameter interference), so as to obtain the key feature variables with the greatest impact on concentration and eliminate multicollinearity interference. Combined with the particle swarm optimization algorithm to globally optimize the hyperparameters of the XGBoost model, the XGB-PSO concentration prediction model can still maintain high-precision prediction capability when facing changes in seabed topography or fluctuations in ore supply.

[0043] Specifically, the PSO algorithm is used to globally optimize the hyperparameters of the XGBoost model. This is achieved by mapping the four core hyperparameters of the XGBoost model to the position coordinates of particles in the PSO algorithm, where each particle's position is a four-dimensional vector. During particle swarm initialization, multiple particles are randomly generated within the hyperparameter value range and randomly assigned initial velocities. The learning rate controls the step size of the response to concentration changes; this parameter determines the magnitude of the model's correction to the residuals. In this invention, The optimization goal is to enable the model to quickly track the second-level concentration fluctuations caused by the switching of ore tanks by the flow direction control device, and to prevent control instability caused by response lag. The maximum tree depth is used to control the fitting complexity of nonlinear rheological properties. This parameter determines The model's nonlinear expressive power was enhanced; due to the significant differences in slurry rheological properties between deep-sea and shallow-water environments, and the severe influence of flow velocity coupling on the relationship between pressure differential and concentration in the lifting pipe, the optimal PSO method was used to find the best PSO solution. The aim is to establish a model structure that can accurately map such high-dimensional nonlinear fluid dynamics relationships and avoid underfitting; The splitting threshold is used to control the filtering ability of noise. This parameter determines the conservatism of the model's splitting nodes. For the instantaneous spike noise generated by the pressure sensor caused by the random collision of coarse ore particles in the ore lifting pipe, optimizing this parameter can force the model to ignore such pseudo-signals that are not related to concentration changes, thereby improving the system's anti-interference robustness. Next, the fitness of each particle is evaluated. The hyperparameters corresponding to its position are substituted into the XGBoost model for pre-training, and the root mean square error (RMSE) is calculated as the fitness value. RMSE can quantify the model's prediction error and directly reflect the quality of the hyperparameter combination. The smaller the RMSE, the higher the fitness, indicating that the deviation between the model's predicted value and the true value is smaller, and the hyperparameter combination is better. The historical best position of each particle (individual best) and the historical best position of the entire swarm (global best) are recorded. The individual historical best P best This represents the model parameter configuration found by the particle in previous iterations that best suits the current seabed topography and slurry concentration prediction; the swarm global optimal G best This represents the globally optimal parameter combination found across the entire domain that minimizes the prediction residual of solid volume concentration within the ore-lifting pipe; then, position updates are performed according to the PSO update rule; the optimization is iterated continuously until the termination condition is met, finding the globally optimal hyperparameter combination that drives all particles toward G. best The optimized model output is able to overcome the shortcomings of traditional empirical formulas in terms of accuracy degradation under varying deep-sea conditions (such as water depth changes and backwater back pressure fluctuations). This hyperparameter combination represents the optimal balance between the model's ability to quickly capture early signs of pipe blockage and the random noise of the filtered fluid, enabling it to adapt to the dynamic changes in the deep-sea operating environment. Then, the hyperparameters corresponding to this globally optimal hyperparameter combination are assigned to the XGBoost model to obtain the optimized XGB-PSO concentration prediction model.

[0044] The real-time key feature variables selected using the LASSO regression algorithm ( The parameters (p, W, Q, RPM, Svalve) are input into the XGBoost model loaded with the optimal parameters. That is, the real-time key feature variables are input into the optimized XGB-PSO concentration prediction model, and the model outputs the current solid volume concentration prediction value (SC%).

[0045] It should be noted that this technical solution is not a simple application of existing general algorithms, but rather an improvement on the physical characteristics of deep-sea closed-loop ore hoisting systems. In this invention, particles represent the hyperparameters of the prediction model. This invention creatively establishes a mapping relationship between these abstract mathematical parameters and specific physical phenomena of deep-sea ore hoisting (valve pulsation, rheological nonlinearity, particle impact noise), providing a basis for subsequent concentration prediction based on this mapping relationship. Targeted noise handling: By introducing physical correlation constraints through leverage value analysis, the problem of false signal interference in the transport of large particles in the deep sea is specifically solved, thereby distinguishing between "true concentration increase" and "ore impact noise" and avoiding the impact of noise on the prediction of solid phase volume concentration.

[0046] Unique feature engineering (variable / feature selection): General prediction models usually only focus on continuous variables. This technical solution introduces the discrete feature of 'hydraulic injector valve timing state' into the XGB-PSO model for the first time. This enables the model to sense 'operating condition switching' and accurately predict transient concentration fluctuations caused by tank switching that traditional models cannot capture.

[0047] Adaptive parameter optimization: To address the issue of back pressure variation caused by changes in water depth during deep-sea operations, PSO is used to dynamically find the optimal tree depth and learning rate, thus solving the technical challenge of decreased accuracy of fixed parameter models under varying operating conditions.

[0048] As an optional implementation, determining the control strategy based on the predicted solid volume concentration includes: A preset concentration safety threshold is set, the predicted solid phase volume concentration is compared with the concentration safety threshold, and a control strategy is determined based on the comparison result. If the predicted solid volume concentration is greater than the concentration safety threshold, the rotation speed of the ore feed hopper is reduced or the water supply valve is opened to dilute the slurry in the mixer. If the predicted solid volume concentration is less than the concentration safety threshold, it is determined whether the predicted solid volume concentration is within the optimal concentration range, and the hydraulic ore injector is controlled to make adaptive adjustments based on the determination result.

[0049] The step of determining whether the predicted solid volume concentration is within the optimal concentration range and adaptively adjusting based on the determination result includes: If the predicted solid volume concentration is within the optimal concentration range, steady-state control is performed, and the rotation speed of the ore feed hopper is finely adjusted to maintain the slurry concentration within the optimal concentration range. If the predicted solid volume concentration is not in the optimal concentration range, transient compensation is performed to increase the rotation speed of the ore feed hopper and compensate for the concentration fluctuations caused by the switching gap of the ore tank.

[0050] Specifically, the controller sends a control strategy to the hydraulic ore injector for response. The control strategy includes control instructions on how specific devices in the hydraulic ore injector should adjust their parameters, so that the specific devices in the hydraulic ore injector can respond and execute according to these control instructions.

[0051] The predicted solid volume concentration is obtained by the optimized XGB-PSO concentration prediction model, and the concentration safety threshold is set according to the actual situation. If the predicted solid volume concentration is greater than the concentration safety threshold, it means that the slurry concentration in the ore feed pipe is high. The controller sends a control command to the ore feed hopper to reduce the rotation speed of the ore feed hopper to reduce the ore conveying, or opens the water injection valve to dilute the slurry in the mixer by adding water to reduce the concentration of the slurry. If the predicted solid volume concentration is less than the safe concentration threshold, it is necessary to further determine whether the predicted solid volume concentration is within the optimal concentration range. If it is within the optimal concentration range, the slurry concentration needs to be maintained within the optimal concentration range. Therefore, the rotation speed of the ore feed hopper may be slightly adjusted or not adjusted at all. If it is not within the optimal concentration range, and a switching signal is detected from the flow direction control device, a concentration drop may occur. The controller sends a control command to the ore feed hopper to increase the rotation speed in advance to compensate for the concentration fluctuation caused by the switching gap. It should be noted that when the flow control device issues a switching signal, there is an inertial delay in the liquid flow during the switching process, which may lead to temporary flow instability. Alternatively, uneven mixing of the residual slurry in the lifting pipe before switching with the newly flowing material may cause local concentration fluctuations, resulting in a drop in concentration. In this technical solution, the data acquisition module collects operating parameter data in real time, and makes real-time predictions based on the collected operating parameter data to obtain the real-time solid volume concentration prediction value. Then, the real-time solid volume concentration prediction value is compared with the concentration safety threshold and the optimal concentration range, and the slurry concentration is adjusted according to the comparison results. This method enables early perception and feedforward compensation control of concentration fluctuations at the switching moment, thereby achieving closed-loop control of slurry concentration. This adaptive adjustment method not only effectively prevents the risk of pipeline blockage, but also ensures that the lifting system always operates in the optimal efficiency range, reducing unnecessary drainage and energy consumption, while also adapting to complex working conditions such as seabed topography and fluctuations in ore supply.

[0052] Example 2 Please refer to the attached document. Figure 5 and 6 The present invention also provides a closed-loop control method based on deep-sea ore lifting pipe slurry concentration monitoring, comprising the following steps: The real-time operating parameter data of the hydraulic ore injector is collected by the intelligent control unit. The real-time operating parameter data is input into the optimized XGB-PSO concentration prediction model to predict the slurry concentration in the ore lifting pipe and output the predicted value of the solid phase volume concentration at the current moment. The current solid volume concentration prediction value is sent to the controller, which compares the current solid volume concentration prediction value with a preset concentration safety threshold and determines the control strategy based on the comparison result. The controller sends the control strategy to the hydraulic ore injector, which responds to adjust its operating parameters, thereby adjusting the slurry concentration entering the ore lifting pipe at the next moment, so that the slurry concentration entering the storage device meets the requirements. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is:

[0053]

[0054] Among them, V new The updated velocity is represented by w; the inertia weight is w; the current velocity is v; c1 is the individual learning factor; r1 is a random number 1; P best Let X be the individual's historical best position, and C be the current position; c2 be the group learning factor; r2 be the random number 2; G be the random number 2. best X is the globally optimal position for the group; new The updated position; V new For the updated speed; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

[0055] The above are merely specific embodiments 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. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A closed-loop control system based on deep-sea ore lift pipe slurry concentration monitoring, characterized in that, include: The surface support platform control center is used to output control strategies; A hydraulic ore injector is used to receive a control strategy and perform adaptive adjustments according to the control strategy. The water surface support platform control center includes an intelligent control unit and a controller. The intelligent control unit is used to collect the operating parameter data of the hydraulic ore injector, process the operating parameter data, and predict the solid volume concentration in real time. The controller is used to receive the solid volume concentration prediction value and determine the control strategy based on the solid volume concentration prediction value. The intelligent control unit includes: The data acquisition module is used to collect the operating parameter data of the hydraulic ore injector; The data processing module is used to process the operating parameter data, filter the processed operating parameter data, and obtain key features; The prediction module is used to train the XGB-PSO concentration prediction model by inputting the key features into it, to obtain an optimized XGB-PSO concentration prediction model, to input real-time operating parameter data into the optimized XGB-PSO concentration prediction model, and to output the predicted value of the solid volume concentration of the slurry in the ore lifting pipe. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is: where V new is the updated velocity; w is the inertia weight; V is the current velocity; c1 is the individual learning factor; r1 is random number 1; P best is the individual historical best position, X is the current position; c2 is the group learning factor; r2 is random number 2; G best is the group global best position; X new is the updated position; V new is the updated velocity; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

2. The closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring according to claim 1, characterized in that, The hydraulic ore injector includes: an ore feed silo, a water pump, a flow control device, an ore tank, a mixer, a lifting pipe, a separator, and a storage device; The output end of the ore feed hopper is connected to the input end of the flow direction control device via a feed pipe. A feed valve is installed on the feed pipe. The output end of the water supply pump is connected to the input end of the flow direction control device via a water supply pipe. A water supply valve and a flow meter are installed on the water supply pipe. The output end of the flow direction control device is connected to several ore tanks via conveying pipes. A control valve is installed on the conveying pipes. The output end of each ore tank is connected to the input end of the mixer. A torque sensor is installed inside the mixer. The mixer is connected to the separator via a lifting pipe. Pressure sensors are installed at both the inlet and outlet ends of the lifting pipe. The separator has a first output port and a second output port. The first output port is connected to the input end of the water supply pump via a recovery pipe. The second output port is connected to the storage device via a slurry conveying pipe.

3. The closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring according to claim 2, characterized in that, The operating parameter data includes feeder speed data, water supply pipe flow data, injector valve timing status data, ore lifting pipe pressure difference data, and mixer power data.

4. The closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring according to claim 1, characterized in that, The process of processing the operating parameter data and filtering the processed operating parameter data to obtain key features includes: Step S1: Use leverage value analysis to remove outliers from the collected data; Step S2: Use the LASSO regression algorithm to screen out key feature variables from the data after outlier removal.

5. The closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring according to claim 4, characterized in that, Step S1 specifically includes: Step S11: Normalize the operating parameters; Step S12: Construct a feature matrix based on the normalized data, and calculate the hat matrix based on the feature matrix; Step S13: Calculate the lever value for each normalized running parameter data point using the diagonal elements of the hat matrix; Step S14: Preset an anomaly detection threshold, compare the anomaly detection threshold with the leverage value, and remove leverage values ​​that exceed the anomaly detection threshold to obtain data filtered for anomalies.

6. The closed-loop control system based on deep-sea ore lifter slurry concentration monitoring according to claim 4, characterized in that, Step S2 specifically includes: Step S21: Perform data standardization on the data after outlier removal; Step S22: Construct a penalty objective function that includes an L1 regularization term. The expression of the penalty objective function is: in, The target variable is pulp concentration. As input features, The penalty coefficient, is the regression coefficient.

7. The closed-loop control system based on deep-sea ore lifting pipe slurry concentration monitoring according to claim 2, characterized in that, The step of determining the control strategy based on the predicted solid volume concentration includes: A preset concentration safety threshold is set, the predicted solid phase volume concentration is compared with the concentration safety threshold, and a control strategy is determined based on the comparison result. If the predicted solid volume concentration is greater than the concentration safety threshold, the rotation speed of the ore feed hopper is reduced or the water supply valve is opened to dilute the slurry in the mixer. If the predicted solid volume concentration is less than the concentration safety threshold, it is determined whether the predicted solid volume concentration is within the optimal concentration range, and the hydraulic ore injector is controlled to make adaptive adjustments based on the determination result.

8. The closed-loop control system based on deep-sea ore lifter slurry concentration monitoring according to claim 7, characterized in that, The step of determining whether the predicted solid volume concentration is within the optimal concentration range, and controlling the hydraulic ore injector to adaptively adjust based on the determination result, includes: If the predicted solid volume concentration is within the optimal concentration range, steady-state control is performed, and the rotation speed of the ore feed hopper is finely adjusted to maintain the slurry concentration within the optimal concentration range. If the predicted solid volume concentration is not in the optimal concentration range, transient compensation is performed to increase the rotation speed of the ore feed hopper and compensate for the concentration fluctuations caused by the switching gap of the ore tank.

9. A closed-loop control method based on deep-sea ore lift pipe slurry concentration monitoring, characterized in that, Includes the following steps: The real-time operating parameter data of the hydraulic ore injector is collected by the intelligent control unit. The real-time operating parameter data is input into the optimized XGB-PSO concentration prediction model to predict the slurry concentration in the ore lifting pipe and output the predicted value of the solid phase volume concentration at the current moment. The current solid volume concentration prediction value is sent to the controller, which compares the current solid volume concentration prediction value with a preset concentration safety threshold and determines the control strategy based on the comparison result. The controller sends the control strategy to the hydraulic ore injector, which responds to adjust its operating parameters, thereby adjusting the slurry concentration entering the ore lifting pipe at the next moment, so that the slurry concentration entering the storage device meets the requirements. The process of constructing the optimized XGB-PSO concentration prediction model includes: Construct the XGBoost model; The hyperparameter combination of the XGBoost model is defined as a particle P in the PSO algorithm. i Each particle contains a four-dimensional vector: ; in, The learning rate; Maximum tree depth; This is the splitting threshold; This represents the number of iterations. N particles are randomly generated to form an initial population, and their positions and velocities are assigned randomly. The hyperparameters represented by each particle are then assigned to the XGBoost model. The velocity and position of particles are updated to move them closer to their individual historical optimum and the global optimum. The formula for updating the velocity and position of particles is: where V new is the updated velocity; w is the inertia weight; V is the current velocity; c1 is the individual learning factor; r1 is random number 1; P best is the individual historical best position, X is the current position; c2 is the group learning factor; r2 is random number 2; G best is the group global best position; X new is the updated position; V new is the updated velocity; Calculate the root mean square error (RMSE) value and use it as the fitness value of the particle. When the preset number of iterations is reached or the RMS error value no longer decreases, output the parameter combination corresponding to the globally optimal particle to obtain the optimized XGB-PSO concentration prediction model.

Images