Methods for controlling and monitoring the concentration of overflow from the loading tank

Abstract

This invention discloses a method for controlling and monitoring the concentration of overflow in mud silt, relating to the field of waterway dredging technology. The apparatus includes a concentration monitoring system, an overflow control system, and control devices. The method collects vertical detection data to construct a continuous vertical concentration-gradation profile model characterizing the sediment content and particle size distribution inside the mud silt. Based on this, a one-dimensional transport-diffusion-settlement evolution equation is introduced to establish a dynamic prediction model, predicting the evolution trend of sediment morphology over time. A discrete layer optimization decision-making strategy is adopted to construct a multi-objective evaluation function including concentration deviation, gradation deviation, and fine particle loss. The optimal overflow port height and opening sequence are solved through a layer mass balance mechanism. This invention integrates a risk-constrained stopping strategy and a phased adaptive control law, solving problems such as large monitoring blind spots, control lag, and severe fine particle loss in traditional overflow control, thereby improving mud silt loading efficiency and automation levels.

Description

Technical Field

[0001] This invention belongs to the field of waterway dredging technology, and in particular to the method for controlling and monitoring the concentration of overflow from loading tanks. Background Technology

[0002] Trailing suction hopper dredgers are core equipment for waterway dredging and reclamation projects, and their operational efficiency directly affects the project cycle and economic benefits. During construction, the trailing suction hopper dredger uses centrifugal pumps to suck seabed mud into the mud chamber. As loading progresses, the upper layer of low-concentration mud is discharged through the overflow pipe to replace the high-concentration sediment that follows. This is a key technological step to increase the effective loading capacity per voyage and optimize the bulk density structure of the mud chamber.

[0003] Existing overflow control technologies for mud tanks mainly rely on manual experience or simple setpoint feedback mechanisms. Traditional devices typically use overflow outlets at fixed heights or control the overflow based on a single monitoring point's turbidity threshold; that is, the overflow is activated when the concentration at the monitoring point is below a set value and deactivated when it is above the set value. Monitoring methods are mostly limited to single-point sampling or simple liquid level monitoring, lacking a comprehensive understanding of the complex flow field and sediment settling state inside the mud tank. Control actions are often triggered based on the instantaneous state at the current moment.

[0004] The core flaw of existing solutions lies in the lack of dynamic prediction capabilities for the vertical stratification evolution of mud silt chambers and the absence of multi-objective refined control mechanisms. Specifically, isolated monitoring data fails to reconstruct the entire field profile, and lagging control logic leads to excessive loss of fine particles or backmixing of high-concentration layers. Single-point threshold-based control ignores the nonlinear distribution of the vertical concentration gradient in the mud silt chamber, failing to accurately distinguish the boundary between clear and turbid water layers, resulting in coarse overflow height adjustment and difficulty in retaining effective soil while discharging clear water. Existing technologies only provide feedback based on the current state, lacking prediction of the temporal evolution of physical processes such as sediment deposition and diffusion, making it difficult to overcome control deviations caused by sensor delays and mechanical lags, easily leading to concentration overshoot or payload loss. The lack of a multi-objective collaborative optimization framework for average concentration, gradation maintenance, and discharge efficiency makes it difficult to automatically seek the optimal overflow strategy under complex operating conditions, resulting in a trade-off between siltation efficiency and construction quality. Summary of the Invention

[0005] The purpose of this invention is to provide a method for controlling and monitoring the concentration of overflow in a cargo hold, in order to solve the aforementioned problems in the prior art.

[0006] Technical solution: A method for controlling and monitoring the concentration of overflow during cargo loading, comprising:

[0007] Vertical detection data were collected from multiple discrete depth points within the water depth range of the mud chamber. The vertical detection data were used to construct a vertical concentration-gradation profile model of the mud chamber, which characterizes the continuous function of sand content and particle size distribution inside the mud chamber.

[0008] Read the preset engineering target parameters, map the engineering target parameters to the mud tank vertical concentration-gradation profile model for multi-objective coupling calculation, and generate overflow control commands that match the current working conditions. The overflow control commands include overflow port height commands and bottom opening degree commands.

[0009] In response to the overflow control command, the overflow control device is driven to adjust the overflow inlet to the vertical position corresponding to the overflow port height command, and adjust the bottom opening to the effective flow area corresponding to the bottom opening opening degree command.

[0010] Beneficial effects: This invention integrates a risk-constrained stopping strategy with a phased adaptive control law, solving problems such as large monitoring blind spots, control lag, and severe loss of fine particles in traditional overflow control, thereby improving mud tank loading efficiency and automation level. Attached Figure Description

[0011] Figure 1 This is a flowchart of the method for controlling and monitoring the concentration of overflow in the tank in the embodiments of this application.

[0012] Figure 2 This is a flowchart illustrating the steps of a one-dimensional partial differential equation describing the physical movement of sediment in an embodiment of this application.

[0013] Figure 3 This is a flowchart illustrating the steps involved in constructing a discrete layer state model of the mud chamber in an embodiment of this application.

[0014] Figure 4 This is a flowchart illustrating the steps involved in making a risk-constrained overflow stop decision in an embodiment of this application. Detailed Implementation

[0015] Example 1 describes the physical carrier and hardware architecture used to achieve control and monitoring of the overflow concentration in the cargo hold.

[0016] Step 101, a mud tank overflow concentration control and monitoring device, comprising: a concentration monitoring system, arranged inside the mud tank or near the overflow port, for collecting vertical detection data of multiple discrete depth points within the water depth range of the mud tank.

[0017] The concentration monitoring system is a component for sensing the water and sediment conditions inside the mud tank. In this embodiment, the concentration monitoring system can adopt a distributed architecture, using multiple independent monitoring units deployed at different locations within the mud tank, or employing a mobile single monitoring unit to achieve coverage of the entire water space within the mud tank. Vertical detection data specifically refers to a comprehensive data packet containing depth information, sediment concentration information, turbidity information, or particle size distribution information. This type of data is transmitted in the form of digital signals and serves as direct input to subsequent algorithm models. For example, vertical detection data can be represented as a two-dimensional array of depth and corresponding concentration values. Furthermore, the concentration monitoring system can be configured to withstand harsh environments such as high pressure, high humidity, and sediment abrasion, and its outer shell material is preferably made of corrosion-resistant stainless steel or high-strength engineering plastics. In some embodiments, the location of the concentration monitoring system is not limited to near the overflow outlet; it can also be placed in the central area of ​​the mud tank or in a relatively static area away from the mud inlet to obtain more representative background concentration data.

[0018] Step 102, overflow control system, including a position-adjustable overflow inlet and an adjustable bottom opening for discharging water-sand mixture from the mud chamber.

[0019] An overflow control system is a mechanical actuator that performs sediment discharge. An adjustable overflow inlet refers to an overflow pipe whose top inlet height can be continuously or progressively adjusted vertically. This adjustment mechanism allows the system to selectively discharge water layers of different depths, prioritizing the discharge of clear water from the upper layers with lower sediment content. An adjustable bottom opening refers to the outlet located at the bottom or below the side wall of the overflow pipe, whose effective flow area can be changed by mechanical devices. For example, the opening area can be continuously adjusted using a sliding gate, rotary valve, or telescopic sleeve. Changing the opening degree of the bottom opening effectively alters the hydraulic resistance characteristics of the overflow system, thereby controlling the overflow flow rate. By jointly adjusting the height of the overflow inlet and the opening degree of the bottom opening, dual control over the source layer and discharge rate of the discharged water can be achieved.

[0020] In some alternative implementations, the overflow control system can also be equipped with auxiliary adjustment devices. For example, a variable-size screen can be installed at the overflow inlet. The screen size is the mesh size; by changing different screen sizes or adjusting the overlap angle of double-layer screens, the porosity of the flow cross-section can be changed, physically intercepting coarse particles larger than a certain size and preventing their loss. Although this auxiliary adjustment method is not the primary flow control method, it can serve as a last line of defense to protect high-value coarse particles.

[0021] Step 103, control device, which is connected to the concentration monitoring system and the overflow control system, includes a memory and a processor.

[0022] The control unit is the brain of the entire system, responsible for data aggregation, calculation, and command issuance. It establishes a bidirectional data connection with field sensors and actuators via wired cables or wireless communication modules. The memory stores control programs, historical data, model parameters, and engineering target configurations. The processor runs the program code in memory, performing complex numerical calculations and logical judgments. In this embodiment, the control unit can be implemented using an industrial-grade programmable logic controller (PLC), an embedded industrial computer, or a high-performance server. Considering the complexity of the field environment, the control unit is typically installed in the dredger's cab or a dedicated electrical control cabinet, possessing good shock resistance and electromagnetic shielding performance.

[0023] Step 104: The memory stores a computer program. When the computer program is executed by the processor, it implements the steps of the overflow concentration control method and generates overflow control instructions to drive the overflow control system to operate.

[0024] This embodiment emphasizes the integration of hardware and software. The computer program is not merely a simple logic switch control, but encapsulates complete algorithms for profile modeling, evolution prediction, optimization decision-making, and risk control. When the processor executes this program, it can convert the acquired physical signals into mathematical models, perform calculations, and output specific control commands. Overflow control commands typically include, but are not limited to: the target overflow port height (e.g., 5.5 meters), the target bottom opening area (e.g., 0.8 square meters), and the rate limit for action execution. Driving the overflow control system involves sending control signals to drive components such as motors and hydraulic pumps via electrical interfaces, causing them to move into position according to the commands.

[0025] Step 105: The overflow control system includes a fixing device, a telescopic device, and an automatic opening and closing device; one end of the fixing device is fixedly connected to the bottom of the mud tank, and the other end is connected to the bottom of the telescopic device.

[0026] The fixing device forms the base of the overflow system, ensuring the stability of the entire system under ship rolling and water flow impact. The fixing device is typically securely installed on the bottom plate of the mud tank using welding or flange connections. It not only provides support but may also include a flow guiding structure to direct water flow. The telescopic device is a key component for adjusting the height of the overflow inlet. In this embodiment, the telescopic device can employ a multi-stage hydraulic cylinder structure, achieving extension and retraction through the injection and discharge of hydraulic oil; alternatively, it can use a screw-nut transmission mechanism, where a motor drives the screw to rotate and raise or lower the sleeve. This telescopic structure allows the overflow pipe to change length over a wide range, adapting to water level changes at different loading stages.

[0027] Step 106: The telescopic device is configured to extend and retract vertically to adjust the overflow inlet to the height specified by the overflow control command.

[0028] This embodiment describes the specific functions of the telescopic device. Vertical telescopic movement means that the direction of the telescopic motion is mainly parallel to the direction of gravity. Adjusting the overflow inlet indicates that the overflow inlet is fixed to the movable end of the telescopic device (usually the top). To ensure adjustment accuracy, the telescopic device is typically equipped with a position feedback sensor (such as a rope displacement sensor or a magnetostrictive displacement sensor), which can provide real-time feedback of the current actual height to the control device, forming a closed-loop position control. When a new height command is received, the control device calculates the height difference and drives the telescopic device to move until the deviation between the actual height and the commanded height is within the allowable range.

[0029] Step 107: An automatic opening and closing device is installed at the bottom opening of the overflow control system and configured to adjust the effective flow area of ​​the bottom opening in response to the overflow control command.

[0030] The automatic opening and closing device is the final actuator for controlling flow. It is located at the bottom opening and uses the relative movement of mechanical components to block or allow water flow. Specifically, it can be implemented using an electric butterfly valve, adjusting the area by changing the valve disc angle; or a gate valve, changing the height of the flow cross-section by raising or lowering the gate. Responding to overflow control commands indicates that the device has automatic execution capability, requiring no manual intervention. Its adjustment process can also be closed-loop, i.e., it has an opening feedback signal. In some high-precision control scenarios, the automatic opening and closing device can also have a flow characteristic curve correction function, that is, based on the valve's nonlinear characteristics, it converts the target area command into a corresponding stroke or angle command, achieving more linear flow control.

[0031] Step 108: The concentration monitoring system includes a first signal sensor, a second signal sensor, a vertical sampler, and a turbidity meter; the first signal sensor and the second signal sensor are respectively fixedly installed on both sides of the overflow control system near the overflow inlet.

[0032] This embodiment describes the component composition and layout of the monitoring system. Installing the first and second signal sensors on either side of the overflow inlet allows for a symmetrical arrangement, reducing the impact of unilateral flow field distortion on the measurements. The signal sensors effectively function as signal relays, position references, and cable guides. They are fixed to the top structure of the overflow pipe and rise and fall with the overflow outlet, ensuring that the monitoring equipment always uses the overflow outlet as a relative reference, facilitating direct acquisition of depth data relative to the overflow outlet.

[0033] Step 109: The vertical sampler is connected to the first signal sensor through the first rigid tube and the first fixed pulley, and is configured to move in the vertical direction and collect sediment sample signals in layers.

[0034] Vertical samplers are used to acquire solid water samples for more precise particle size distribution analysis. A rigid tube connection prevents the flexible cable from drifting and tangling under water flow, ensuring the sampler accurately reaches the designated depth. The first fixed pulley changes the direction of transmission and supports the rigid tube. Vertical movement of the sampler can be achieved by raising and lowering the rigid tube using a winch or servo motor. Layered sampling means the sampler can remain at different depths to collect a certain volume of mud, or a flow-through sampling structure can be used to continuously acquire sample characteristics at different depths. The sample signal can be a direct physical sample or an electrical signal generated by in-situ analysis of the sample by internal sensors within the sampler.

[0035] Step 110: The turbidity meter is connected to the second signal sensor via the second rigid tube and the second fixed pulley, and is configured to move in the vertical direction and detect the stratified sand content signal in real time.

[0036] Turbidimeters provide high-frequency, real-time concentration data. Similar to vertical samplers, they achieve vertical movement via a rigid tube and pulley system. Turbidimeters typically operate based on optical scattering or transmission principles, enabling rapid response to changes in suspended particle concentration in water. By controlling the lifting and lowering speed of the rigid tube, rapid profile scanning across the water depth range of the mud chamber can be achieved. Real-time monitoring emphasizes the timeliness of the data and its importance for dynamic control. Stratified sediment concentration signals correspond to turbidity values ​​at different depths, which are converted into mass concentration values ​​after calibration. The dual monitoring mechanism (sampler + turbidity meter) ensures both the real-time nature of the data (via the turbidity meter) and the accuracy and multidimensionality of the data (via sampler calibration), providing a solid data foundation for subsequent high-precision modeling.

[0037] Example 2, as follows Figure 1 The diagram illustrates the method and flow for controlling the overflow concentration in the silt tank. This embodiment upgrades traditional single-point threshold-based switching control to multi-objective optimization control based on full-field information by constructing a continuous digital profile model, thus solving the technical problem of simultaneously controlling the discharge of clean water and the retention of fine particles during sediment discharge.

[0038] Step 201: Collect vertical detection data at multiple discrete depth points within the water depth range of the mud chamber, and use the vertical detection data to construct a vertical concentration-gradation profile model of the mud chamber that characterizes the continuous function of sand content and particle size distribution inside the mud chamber.

[0039] Data acquisition is fundamental to model building. In this embodiment, the control system drives the vertical sampler and turbidimeter to perform vertical scanning within the mud chamber. The system sets a series of discrete depth sampling points, for example, every 0.5 meters as a sampling layer. At each sampling point, the current depth value and the corresponding sensor reading are recorded. To eliminate random fluctuations, multiple measurements can be taken at each point for several seconds, and the average value is calculated. The vertical detection data can be represented in set form: S in ={(z1,C1,P1),(z2,C2,P2),...,(z n C n ,P n )}, where z i For depth, C i For sediment content, P i This provides the particle size distribution data at this depth.

[0040] The process of building the model involves transforming these discrete points into a continuous mathematical function. For the sediment concentration C(z), this embodiment preferably uses the cubic spline interpolation method. Cubic spline interpolation ensures the continuity of the interpolation function and its first and second derivatives, resulting in a smooth and natural curve that closely matches the physical law of gradual sediment concentration change under gravity settling. For the particle size distribution P(z,d), bilinear interpolation or a fitting method based on physical characteristics can be used. That is, first, it is assumed that the particle size distribution conforms to a specific probability distribution function (such as a log-normal distribution or a Rosin-Rammler distribution), and then the distribution parameter function that varies with depth is fitted according to the characteristic parameters of each measuring point. The final vertical concentration-gradation profile model of the sediment chamber contains two functions: the sediment concentration per unit volume function C(z) and the particle distribution function P(z,d).

[0041] Step 202: Read the preset engineering target parameters, map the engineering target parameters to the mud tank vertical concentration-gradation profile model for multi-objective coupling calculation, and generate overflow control commands that match the current working conditions. The overflow control commands include overflow port height commands and bottom opening degree commands.

[0042] Engineering target parameters are guiding indicators for the control system, typically set by operators based on specific construction tasks. For example, P... target =[C target ,θ target ,γ max ,T max[ ] represents the desired average sediment concentration target value, the desired fine particle proportion target value, the maximum allowable fine particle loss rate, and the maximum overflow duration, respectively. Based on the constructed C(z) and P(z,d) models, the system calculates the achievable state indicators if overflow is stopped at the current moment or if overflow occurs in a specific manner through numerical integration. The calculated state indicators are then compared with P... target Compare and assess the current deviation.

[0043] The process of generating instructions is essentially an optimization decision-making process. The system needs to find an optimal combination of parameters (z) within the feasible control space (i.e., the adjustable range of the overflow height and the adjustable range of the opening). ovf A opt This ensures that the future system state is as close as possible to the engineering goals. For example, if the upper layer of clear water is found to be thick and has an extremely low sediment content, the system will generate a lower overflow outlet height command and a larger opening command to quickly discharge this portion of water. Conversely, if the upper layer concentration is already close to the target value, the system may increase the overflow outlet height or decrease the opening.

[0044] Step 203: In response to the overflow control command, drive the overflow control device to adjust the overflow inlet to the vertical position corresponding to the overflow port height command, and adjust the bottom opening to the effective flow area corresponding to the bottom opening opening degree command.

[0045] This embodiment represents the final execution stage of the control closed loop. The control device converts the calculated digital commands into physical drive signals. For overflow port height commands, the system calculates the difference between the target height and the current height, drives the motor of the telescopic device to rotate forward or backward, and simultaneously monitors the feedback from the displacement sensor in real time. The system stops operating when the error is less than a preset dead zone (e.g., ±10mm). For opening commands, the system drives the actuator of the automatic opening and closing device to adjust the valve or gate to the specified position. To prevent frequent actions from causing mechanical wear, a control cycle is typically set (e.g., adjustment every 30 seconds or 1 minute), or an adjustment threshold is set, triggering mechanical action only when the calculated target change exceeds a certain range.

[0046] Step 204, constructing a vertical concentration-gradation profile model of the mud chamber using vertical detection data to characterize the continuous function of sediment content and particle size distribution inside the mud chamber, includes: analyzing the vertical detection data and extracting the instantaneous sediment content measurement values ​​and discrete particle size distribution functions at multiple discrete depth coordinates within the water depth range of the mud chamber.

[0047] Data analysis is the process of converting raw sensor signals into physical quantities. For turbidimeters, this requires converting voltage or current signals into sand content per unit volume (e.g., kg / m³) based on a pre-calibrated turbidity-concentration curve. For sampler data, this may involve analyzing image analysis results or laser particle size analyzer data to extract mass fractions for different particle size ranges. The final extracted data is standardized, meaning that for each discrete depth z... i Each of these corresponds to a scalar concentration value C. i and a vector or functional form P describing the particle size distribution i (d)

[0048] Step 205: Perform spatial interpolation or numerical fitting on multiple instantaneous sediment concentration measurements to construct a volumetric sediment concentration function that describes the continuous change of sediment concentration with depth.

[0049] This embodiment emphasizes the transformation from discrete to continuous. Besides cubic spline interpolation, in cases with few data points or high noise, least squares methods can be used for polynomial fitting or exponential function fitting. For example, assuming that concentration increases exponentially with depth, the fitted function can be C(z) = a*exp(b*z) + c, and the coefficients a, b, and c can be determined through regression analysis. The constructed C(z) function allows the system to query concentration estimates at any depth, not only limited to the sampling point location, but also crucial for subsequent integral calculations.

[0050] Step 206: Perform spatial interpolation on multiple discrete particle size distribution functions to construct a particle distribution function that describes the continuous variation of particle size distribution with depth and particle size.

[0051] Spatial variations in particle size distribution are often more complex than variations in concentration. Typically, coarser particles tend to settle in the lower layers, while finer particles float in the upper layers. Spatial interpolation needs to be performed in two dimensions: interpolation in the depth dimension and description in the particle size dimension. The constructed particle distribution function P(z,d) actually defines a scalar field, that is, the probability density of particles with size d at depth z. This function enables the system to accurately calculate the total mass of particles within a specific particle size range (e.g., fine particles d < 0.063 mm) at any water depth.

[0052] Step 207: Combine the volumetric unit sand content function and particle distribution function into a mud chamber vertical concentration-gradation profile model to provide a basis for querying concentration and gradation at any depth.

[0053] The combined model is a complete information entity. It not only contains C(z) and P(z,d), but may also include relevant metadata, such as data acquisition time and effective water depth range [z]. min ,z maxThis model serves as a unified data interface for subsequent optimization algorithm modules. Whether subsequent algorithms need to calculate average concentration, total fine particulate matter, or predict sedimentation trends, they all obtain basic data from this model, ensuring data source consistency.

[0054] Example 3 describes the process of generating optimal control commands through mathematical calculations using a pre-constructed profile model and engineering target parameters.

[0055] Step 301: The engineering target parameters include the expected average sand content target value, the expected total proportion of fine particles target value, and the maximum allowable loss ratio of fine particles.

[0056] These parameters are used to calculate the quality requirements for construction. The expected average sand content target value C. target This is a core indicator of mud cargo loading, directly related to the transportation efficiency of a single voyage. The expected overall proportion of fine particles is θ. target This reflects the requirements for soil quality. For example, some hydraulic reclamation projects require the soil to have a certain degree of cohesion to prevent excessive loss of fine particles. The maximum allowable loss ratio of fine particles, L... max It is a constraint boundary that limits the extent to which fine particles are sacrificed in order to increase concentration.

[0057] Step 302: Map the engineering target parameters to the vertical concentration-gradation profile model of the mud hopper for multi-objective coupled calculation, and generate overflow control commands that match the current working conditions. These commands include: calculating the real-time average sediment concentration C'(t) and the real-time total proportion of fine particles θ in the mud hopper at the current moment through integral calculation based on the volume unit sediment concentration function C(z) and particle distribution function P(z,d). fine The formula for calculating C'(t) is: C'(t) = (1 / H) * ∫[z min ->z max ]C(z)dz;θ fine (t)=(1 / (H*C'(t)))*∫[z min ->z max ](C(z)*∫[0->d fine ]P(z,d)dd)dz; where H represents the effective water depth of the mud tank, z min and z max d represents the upper and lower boundaries of the water body, respectively. fine This represents the upper limit threshold of fine particle size; it calculates the concentration deviation between the real-time average sediment concentration and the target value of the expected average sediment concentration, as well as the gradation deviation between the real-time total proportion of fine particles and the target value of the expected total proportion of fine particles.

[0058] This embodiment transforms the model function into a global statistical index. The integral operation provides a precise assessment of the overall state of the mud tank. The first formula calculates the average concentration throughout the water column, obtained by integrating the concentration distribution along the depth direction and then dividing by the total water depth. The second formula calculates the mass percentage of fine particles in all solids. Inner layer integral ∫[0->d] fine P(z,d)dd calculates the local proportion of fine particles at a certain depth z. Multiplying it by the local concentration C(z) at that depth and integrating along the depth yields the total mass of fine particles. Finally, dividing by the total solid mass (i.e., H*C'(t)) gives the overall proportion of fine particles. The integral-based calculation method is more accurate than a simple arithmetic mean because it considers the weighted contribution of different depth layers to the overall composition.

[0059] Step 303: Obtain the cumulative fine particle loss ratio from the start of overflow to the current time, and construct a multi-objective comprehensive evaluation function that includes concentration, gradation and loss terms by combining concentration deviation and gradation deviation.

[0060] A multi-objective comprehensive evaluation function J is constructed by combining the engineering objective parameters. The definition of the multi-objective comprehensive evaluation function J is as follows: J=w1*|C'(t)-C target |+w2*|θ fine (t)-θ fine , target |+w3*L fine (t); where C target θ is the target value for the desired average sediment concentration. fine , target To determine the target value for the overall proportion of fine particles, L fine (t) represents the cumulative fine-particle loss ratio from the start of the overflow to the current time, where w1, w2, and w3 are preset non-negative weighting coefficients. The cumulative fine-particle loss ratio L... fine (t) is obtained by accumulating the mass of fine particles discharged from the overflow over time. Within each control time step Δt, the increase in the mass of fine particles discharged from the overflow is ΔM. fine,ovf (t)=Q ovf (t)·Δt·C(z ovf (t),t)·φ fine (z ovf (t),t), where Q ovf (t) represents the overflow flow rate, C(z) ovf (t),t) represents the local concentration at the current overflow height (provided by the profile model), φ fine (z ovf (t), t) represents the proportion of fine particles at that height (obtained by integrating the particle distribution function over the fine particle size range). The cumulative fine particle loss proportion L... fine (t)=Στ ΔM fine,ovf (τ) / M fine,total (0), where M fine,total (0) represents the total mass of fine particles in the mud tank at the start of the overflow, which can be obtained by double integration of the vertical concentration-gradation profile model of the mud tank over the entire water depth range and the fine particle size range.

[0061] The evaluation function J serves as a compass for optimization. It integrates three interdependent objectives (concentration proximity, gradation proximity, and loss minimization) into a scalar index. The first term measures the difference between the current concentration and the target concentration; a smaller difference is better. The second term measures the difference between the gradation and the target concentration. The third term is a penalty term, which increases with the increase in fine particle loss. The weighting coefficients w1, w2, and w3 reflect the preferences of engineering decisions. For example, if the project prioritizes yield and is less concerned about fine particle loss, a larger w1 and a smaller w3 can be set. By adjusting the weights, this method can adapt to different types of dredging conditions.

[0062] In determining the overflow height, this embodiment employs an optimization method based on a weighted minimization of local concentration and fine particle proportion. Specifically, the optimal overflow height is calculated using the following formula:

[0063] In another alternative implementation, z ovf (t)=arg min z∈[zmin,zmax] λ1C(z,t)+λ2φ fine (z,t); where z ovf (t) represents the optimal overflow height selected at time t, in meters; arg min represents the value of the independent variable that minimizes the objective function; z represents the candidate overflow height variable, in meters; [z min ,z max ] represents the adjustable range of the overflow outlet height, determined by the physical structure of the equipment; λ1 represents the weighting coefficient of the concentration term, a non-negative real number, the reciprocal of cubic meters per kilogram, used to normalize the concentration value; λ2 represents the weighting coefficient of the fine particle proportion term, a non-negative real number, dimensionless; C(z,t) represents the volumetric sand content at time t and depth z, in kilograms per cubic meter, provided by the mud tank vertical concentration profile model; φ fine (z,t) represents the mass ratio of fine particles at time t and depth z, which is a dimensionless real number with a value range from zero to one.

[0064] Fine particle mass ratio φ fine (z,t) is obtained by integrating the particle distribution function over the fine particle size range:

[0065] φ fine (z,t)=∫0 dfineP(z,d,t)dd; where P(z,d,t) represents the mass fraction density function of a particle with diameter d at time t, depth z; d fine This indicates the upper limit threshold for fine particle size, expressed in millimeters or micrometers.

[0066] The physical meaning of the above optimization formula is that the overflow outlet should be preferentially located in the depth region with lower concentration and less fine particle content. When λ1 is larger, the system tends to select the low-concentration layer for overflow; when λ2 is larger, the system pays more attention to protecting fine particles. In typical engineering applications, the ratio of λ1 to λ2 can be adjusted according to the project's emphasis on yield and soil quality.

[0067] Step 304: With the goal of minimizing the multi-objective comprehensive evaluation function J, search for the optimal overflow port height and bottom opening degree within the preset feasible region, and convert the search results into overflow control commands.

[0068] Within a preset feasible region (i.e., the allowable operating range of the equipment), the system searches for a set of control variables (height and opening) that minimizes the predicted value of J at the next moment. Since the J function may exhibit nonlinear characteristics, this embodiment preferably employs a sequential quadratic programming (SQP) algorithm for solving the problem. The SQP algorithm has advantages in terms of fast convergence speed and high computational efficiency when handling such nonlinear constraint optimization problems. If computational resources permit, global optimization algorithms such as genetic algorithms (GA) can also be used to avoid getting trapped in local optima. The optimal parameters obtained from the solution are then converted into specific control commands and sent to the actuator. For example, if the optimization results indicate that the overflow port should be raised to reduce the loss of fine particles, the system generates a corresponding raising command.

[0069] Example 4, such as Figure 2 As shown, this embodiment introduces a one-dimensional partial differential equation that can describe the physical movement of sediment, solving the time lag problem caused by relying solely on static profiles for control, and enabling early prediction of the backmixing and settling trends of high-concentration sediment layers.

[0070] Step 401, the method further includes constructing a dynamic model capable of predicting the evolution of sediment morphology within the mud bin over time. Specific steps include: establishing a one-dimensional transport-diffusion-sedimentation evolution equation describing the movement of suspended sediment within the mud bin. The equation defines the partial derivative of the volumetric sediment concentration per unit volume with respect to time as the sum of the negative gradient of the vertical transport flux, the diffusion term caused by the turbulent diffusion coefficient, and the sedimentation term caused by the particle settling velocity. The equation is expressed as: ΨC(z,t) / Ψt=-Ψ[v(z,t)C(z,t)] / Ψz+DΨ²C(z,t) / Ψz²-w sΨC(z,t) / Ψz; where Ψ is the partial derivative, C(z,t) represents the volumetric sediment concentration per unit volume at depth z and time t, v(z,t) represents the effective vertical velocity, D represents the effective turbulent diffusion coefficient, and w s This indicates the particle settling velocity.

[0071] In this embodiment, the one-dimensional transport-diffusion-sedimentation evolution equation comprehensively considers the three main mechanisms affecting the vertical distribution of sediment. The first term is the convection term, which describes the sediment migration driven by the overall flow of the water body. The effective vertical velocity v(z,t) is usually the superposition of the upward velocity caused by overflow suction and the background velocity caused by the internal circulation of the sediment chamber. The second term is the diffusion term, which describes the concentration gradient smoothing effect caused by fluid turbulence fluctuations. The effective turbulent diffusion coefficient D is positively correlated with the Reynolds number and stirring intensity within the sediment chamber. The third term is the sedimentation term, which describes the relative downward motion of solid particles under the action of gravity.

[0072] Specifically, the particle settling velocity w s It is a key physical parameter. In some implementations, w s The settling velocity can be estimated using Stokes' theorem, which states that the settling velocity is directly proportional to the square of the particle size and inversely proportional to the fluid viscosity. Considering the presence of particles of various sizes within the sludge chamber, a weighted average particle size can be used to calculate the average settling velocity, or separate equations can be established and solved for different particle size components. For example, in a condition where fine particles dominate, w... s The value of is usually small, and the influence of the sedimentation term is relatively weak; however, for operating conditions with more coarse particles, the sedimentation term will dominate the evolution of the concentration profile, leading to a rapid increase in the bottom concentration. By introducing this equation, the control system no longer makes decisions based on past measurements, but rather on the simulation results of future physical processes.

[0073] Particle settling velocity w s It is a key physical parameter determining the vertical distribution and evolution of sediment. According to Stokes' law of settling, the settling velocity of a spherical particle in still water is proportional to the square of its particle size:

[0074] w s =(ρ s -ρ w )gd 2 / 18μ; where, w s ρ represents the particle settling velocity, measured in meters per second. s This represents the density of sediment particles, expressed in kilograms per cubic meter (kg / m³). For quartz sand, this value is typically taken as approximately 2650 kg / m³. wThe density of water is expressed in kilograms per cubic meter (kg / m³), typically around 1000 kg / m³ for freshwater and around 1025 kg / m³ for seawater; g represents the acceleration due to gravity, expressed as 9.81 m / s²; d represents the particle diameter, expressed in meters; and μ represents the dynamic viscosity of water, expressed in pascals per second (Pascals per second), approximately 0.001 Pascals per second for freshwater at room temperature.

[0075] Because the mud chamber contains a mixture of particles of various sizes, a weighted average particle size d is typically used in engineering applications. 50 The representative settling velocity can be calculated, or the evolution equations for different particle size components can be established separately and then superimposed for solution. For fine particles (e.g., silt and clay with a particle size of less than 0.063 mm), the settling velocity is small, the effect of the settling term is relatively weak, and the diffusion effect is dominant; for coarse particles (e.g., medium-coarse sand with a particle size of more than 0.2 mm), the settling velocity is large, and the settling term will dominate the evolution of the concentration profile, resulting in a rapid increase in the bottom concentration.

[0076] In the one-dimensional transport-diffusion-sedimentation evolution equation, the vertical flux q(z,t) represents the mass of sediment transported downwards through a unit horizontal area per unit time, and its complete expression is:

[0077] q(z,t) = v(z,t)·C(z,t); where q(z,t) represents the vertical sediment mass flux at depth z and time t, in kilograms per square meter per second; v(z,t) represents the effective vertical velocity at depth z and time t, in meters per second, with positive values ​​indicating downward motion and negative values ​​indicating upward motion; C(z,t) represents the volumetric sediment content at depth z and time t, in kilograms per cubic meter.

[0078] The effective vertical velocity v(z,t) is a combined effect of multiple physical mechanisms and can be decomposed into the following components:

[0079] v(z,t)=v overflow (z,t)+v circulation (z,t); where v overflow (z,t) represents the upward velocity component caused by the overflow suction. When the overflow port is opened, the water near the overflow port is pumped out, forming an upward compensating flow inside the chamber. This component is usually negative (upward); v circulation (z,t) represents the background vertical velocity component caused by the circulation inside the mud chamber, which may be driven by the inlet jet, temperature gradient, or density difference.

[0080] Under typical operating conditions, v overflow The magnitude is determined by the overflow flow Q. ovf With mud chamber cross-sectional area A tank The ratio determines that, i.e., |v overflow |≈Qovf / A tank .

[0081] Step 402: Input the volumetric unit sediment concentration function in the vertical concentration-gradation profile model of the mud tank as the spatial distribution state C(z,t0) at the initial time t0 into the one-dimensional transport-diffusion-settlement evolution equation.

[0082] This embodiment achieves the connection between the static and dynamic models. The mud tank vertical concentration-gradation profile model provides a snapshot at time t0. The system discretizes the continuous function C(z) obtained through interpolation or fitting and assigns it to the grid nodes of the evolution equation as the starting point for time step calculation. To improve the accuracy of the initial conditions, before inputting the equation, the Kalman filter algorithm can be used to fuse measurement data from multiple recent times to smooth the initial state and reduce the impact of measurement noise on prediction accuracy.

[0083] Step 403: Based on the one-dimensional transport-diffusion-sedimentation evolution equation, time step calculation is performed to predict the future concentration distribution value C(z,t+Δt) at one or more future time steps, providing a feedforward control basis for adjusting the overflow height.

[0084] Time-step calculation refers to the process of solving equations step by step along a time axis. The system can set the prediction time domain, such as 5 to 10 minutes into the future. Through iterative calculations, the system can generate a series of future concentration profile sequences. These predictions can reveal the rising rate of the high-concentration sediment interface. For example, if the prediction shows that the high-concentration turbid water layer at the bottom will rise to the current overflow height in 3 minutes, the control system can issue a command in advance to raise the overflow position and prevent turbid water from being drawn in. The feedforward control mechanism effectively compensates for the mechanical delay of the actuator's action and the transmission delay of the sensor measurement.

[0085] Step 404 involves time-stepping calculations based on the one-dimensional transport-diffusion-sedimentation evolution equation, including solving using boundary conditions and finite difference schemes. Specific steps include: at the upper boundary depth z of the mud tank water... min and lower bound depth z max Upper and lower boundary conditions are set at each point. The upper boundary condition stipulates that the convective flux and diffusion flux at this point cancel each other out, making the total upward solid flux zero. The lower boundary condition stipulates that the total solid flux at this point equals the deposition flux determined by the particle settling velocity. Their mathematical expressions are as follows:

[0086] Upper boundary: (v(z,t)C(z,t)-DΨC(z,t) / Ψz)|z=z min =0;

[0087] Lower boundary: (v(z,t)C(z,t)-DΨC(z,t) / Ψz+w s C(z,t))|z=z max =0;

[0088] Where Ψ is the partial derivative, the upper boundary condition indicates that the total solid flux at the surface of the mud tank is zero, and the lower boundary condition indicates that the total solid flux at the bottom of the mud tank is equal to the particle settling velocity w. s The determined deposition flux.

[0089] Boundary conditions are necessary constraints for the boundary value solutions of partial differential equations. The upper boundary condition is set at the free surface, which physically means that sediment cannot cross the water surface to enter the air; therefore, the total flux, including convection and diffusion, is zero. The lower boundary condition is set at the bottom of the mud tank or at the mud-water interface. Here, not only convection and diffusion must be considered, but also the continuous deposition of sediment at the bottom of the tank due to gravity, forming a sediment layer. Therefore, the total flux at the bottom is set to be equal to the deposition flux term. Correctly setting the boundary conditions ensures mass conservation during the numerical simulation, meaning that the change in the total sediment volume in the mud tank is only caused by overflow discharge and bottom deposition, and there will be no spurious increases or decreases in the calculation.

[0090] Step 405: Discretize the vertical space of the mud hopper into grid points j, and the time axis into time steps n. Use an explicit finite difference scheme to numerically discretize the one-dimensional transport-diffusion-settlement evolution equation. Combine the upper and lower boundary conditions to obtain the iterative formula for the discrete node prediction sequence: C j (n+1) =C j n +Δt[-v(z j (C) (j+1) n -C (j-1) n ) / (2Δz)+D(C (j+1) n -2C j n +C (j-1) n ) / (Δz) 2 -w s (C (j+1) n -C (j-1) n ) / (2Δz)];where, C j (n+1) C represents the predicted concentration at the j-th spatial grid point at the (n+1)-th time step. j n Let Δt be the known concentration at the nth time step, Δz be the time step size, and Δz be the spatial step size.

[0091] This embodiment describes the specific algorithm for numerical solution. The explicit finite difference scheme used has the advantages of fast computation speed and ease of programming implementation, making it suitable for running on embedded industrial control computers with limited computing power. The terms in the formula correspond to the time derivative, convection, diffusion, and sedimentation terms in the original differential equation, respectively. The spatial derivative is approximated using a central difference scheme to obtain second-order truncation error accuracy. The time derivative is approximated using a forward difference scheme.

[0092] In practical implementation, to ensure the stability of numerical calculations, the selection of the time step Δt and spatial step Δz must satisfy the Courant-Friedrich-Lévy (CFL) conditions. For example, for convection-dominated processes, v*Δt / Δz must be less than or equal to 1. Under typical engineering parameters, if the spatial step Δz is set to 0.1 meters and the effective vertical velocity v is approximately 0.05 meters per second, then the time step Δt should be controlled within 1 second. Through high-frequency iterative calculations, the system can update future concentration distribution predictions in real time, providing accurate input data for optimization decisions.

[0093] Example 5 describes how to transform the continuous mud tank control problem into a discrete-layer mass balance optimization problem. This example simplifies the complex fluid dynamics problem into algebraic operations by establishing a virtual water-sand layer model, enabling the application of modern optimization algorithms to find the globally optimal control sequence.

[0094] Step 501: The multi-objective coupled calculation adopts a discrete layer optimization decision strategy. This strategy constructs a discrete layer state model of the mud chamber, divides the vertical depth range of the mud chamber into multiple continuous and non-overlapping virtual water and sediment layers, and initializes the layer state variables of each virtual water and sediment layer based on the vertical concentration-gradation profile model of the mud chamber. The layer state variables include layer volume, total solid mass of the layer, and average concentration of the layer.

[0095] like Figure 3 As shown, the specific steps include: dividing the vertical depth range of the mud chamber into N virtual water-sand layers, and initializing the layer volume V of each virtual water-sand layer k. k (t) and total mass of the solid layer M k (t).

[0096] The system logically divides a continuous body of water into several horizontal slices, for example, N equals 20 layers. Each layer is treated as a homogeneous control volume. The layer volume V... k (t) equals the thickness of the layer multiplied by the cross-sectional area of ​​the mud chamber. The total mass M of the solid layer. k (t) equals the layer volume multiplied by the average concentration of that layer. The initialization process uses the C(z) model to integrate the concentration over the depth range of each layer, calculating the initial mass and volume of each layer. The discretization modeling method reduces the model dimensionality, allowing subsequent optimization calculations to be completed within milliseconds.

[0097] Step 502: Establish a layer extraction mapping relationship describing the overflow process, mapping the overflow height at any time to a mass extraction operation on one or more specific virtual water and sediment layers. That is, determine that the height of the overflow at time t corresponds to the m(t)th virtual water and sediment layer, and obtain the overflow flow rate Q at that time. ovf (t).

[0098] The layer extraction mapping defines the interaction mechanism between control variables and system states. The overflow height is a continuous variable, while the virtual layers are discrete. The system identifies the depth range of the current overflow layer through simple geometric judgment and marks that layer as the extracted layer m(t). Overflow flow rate Q ovf (t) can be obtained in real time by flow meter measurement, or it can be hydraulically estimated using Bernoulli's equation, utilizing the current overflow inundation depth and opening area. In some embodiments, if the overflow spans the boundary of two adjacent layers, the mapping relationship can be defined as extracting from both layers simultaneously according to the overlap ratio.

[0099] Step 503: Construct a layer mass balance update mechanism to calculate the volume reduction and solid mass reduction of the extracted layer at each discrete time step based on the mass extraction operation, and update the layer state variables of the remaining virtual water and sand layers based on the volume and mass reduction results.

[0100] For the m(t)th virtual water-sand layer extracted, the following volume and mass update calculations are performed within the time step Δt: V m (t)(t+Δt)=V m (t)(t)-Q ovf (t)·Δt;M m (t)(t+Δt)=M m (t)(t)-Q ovf (t)·Δt·C m (t)(t); where V m (t)(t+Δt) and M m (t)(t+Δt) represent the updated layer volume and the total solid mass of the layer, respectively, C m(t) (t) represents the average concentration of the layer at time t, and for other layers that are not extracted, k≠m(t), their volume and mass remain unchanged.

[0101] This embodiment is the engine for dynamic model extrapolation. Based on the law of conservation of mass, it simulates the changes in the internal state of the mud hopper caused by the overflow process. For the selected m(t)-th layer, its volume reduction equals the flow rate multiplied by the time step, and its mass reduction equals the volume reduction multiplied by the current concentration of that layer. This mechanism assumes that within a small time step, the concentration of the extracted water equals the instantaneous average concentration of that layer. For layers that are not extracted, it is assumed that their state remains static. By iteratively updating this process, the system can extrapolate the remaining volume and mass of each layer within the mud hopper at any future time under any given overflow height and flow rate sequence.

[0102] Based on the layer mass balance update mechanism, this embodiment further provides a precise calculation method for the fine particle loss. Within each time step Δt, the mass of fine particles overflowing from the extracted layer m(t) is calculated using the following formula: ΔM fine,ovf (t) = θ fine,m(t) (t)·△M ovf (t); where △M fine,ovf (t) represents the mass of fine particles discharged through the overflow within a time step t, in kilograms; θ fine,m(t) (t) represents the proportion of fine particles to the total mass of the solid in the extracted layer m(t) at time t. It is a dimensionless real number with a value ranging from zero to one; ΔM ovf (t) represents the total mass of solid discharged through overflow within a time step t, expressed in kilograms. Its calculation formula is ΔM. ovf (t)=Q ovf (t)·△t·C m(t) (t).

[0103] By summing the fine particle loss at each time step, the cumulative fine particle loss ratio from the start of overflow to any time t can be calculated: L fine (t)=Σ τ=0 t △M fine,ovf (τ) / M fine,total (0); where L fine (t) represents the cumulative proportion of fine particle loss from the start of overflow to time t, and is a dimensionless real number; Σ τ=0 t △M fine,ovf (τ) represents the summation of fine-particle overflow losses over all time steps from the initial time zero to the current time t; M fine,total (0) represents the total mass of fine particles in the mud chamber at the initial moment, in kilograms.

[0104] Step 504, the discrete layer optimization decision strategy further includes the step of solving the optimal control sequence: defining control variables including overflow port height and overflow flow rate, and setting multiple discrete time steps within a preset prediction time interval.

[0105] This embodiment describes the definition phase of the optimization problem. The control variable u(t) is defined as a vector sequence containing the suggested overflow height and opening degree for each future time step. The prediction time interval is the optimization window, for example, the next 10 minutes. Within this interval, the system may be divided into 60 discrete time steps. The optimization objective is to find the optimal control variable sequence corresponding to these 60 time steps, maximizing the overall benefit of the entire process.

[0106] Step 505: For each discrete time step: calculate the squared term of the concentration deviation between the predicted average sediment concentration and the target value of the expected average sediment concentration at that time; calculate the squared term of the gradation deviation between the predicted total proportion of fine particles and the target value of the total proportion of fine particles at that time; obtain the cumulative fine particle loss ratio at that time as the fine particle loss penalty term.

[0107] By using preset non-negative weighting coefficients, the concentration deviation square term, the gradation deviation square term, and the fine particle loss penalty term are linearly weighted and accumulated to generate a discrete total objective function for evaluating the control effect.

[0108] A discrete overall objective function is constructed, which is the result of weighted summation of the squared terms of concentration deviation, squared terms of gradation deviation, and fine particle loss penalty terms at all discrete time steps within the prediction time interval.

[0109] Step 506, constructing the discrete total objective function is achieved through a weighted summation method, specifically calculated as: J = ∑ n=0 N T [α1(C'(t n )-C target ) 2 +α2(θ fine (t n )-θ fine,target ) 2 +α3L fine (t n )]; where J is the discrete total objective function, t n N represents the nth discrete time step. T C'(t) represents the total number of time steps within the predicted time interval. n ), θ fine (t n ) and L fine (t nThe values ​​(C) represent the predicted average sediment concentration, total proportion of fine particles, and cumulative fine particle loss proportion for that time step. target and θ fine , target These represent the target values ​​for the expected average sediment concentration and the expected total proportion of fine particles, respectively; α1, α2, and α3 are non-negative weighting coefficients used to balance the importance of each target.

[0110] This embodiment provides specific mathematical standards for evaluating performance. The discrete overall objective function J accumulates the performance indicators over the entire prediction time domain, reflecting the idea of ​​global optimization and avoiding short-sighted behavior. For example, a certain control strategy may have mediocre concentration performance at the current moment, but it can retain more high-value fine particles later, thus winning in the overall score J. The setting of the weighting coefficients α1, α2, and α3 directly determines the characteristics of the control system. The weighting coefficients can be dynamically adjusted according to the engineering stage. For example, in the early stage of overflow, α1 is set to a larger value to pursue rapid concentration increase; in the later stage of overflow, α3 is set to a larger value to strictly limit leakage.

[0111] Step 507: Perform dynamic deduction based on the layer quality balance update mechanism. Under the premise of satisfying the preset constraints, search for a set of control variable sequences that minimizes the discrete total objective function, and convert the first time step data in the set of control variable sequences into overflow control commands.

[0112] This embodiment describes the solution and execution phase. The system utilizes a layer quality balance update mechanism as the system equation, aiming to minimize J through mathematical optimization. Considering that the objective function J contains squared terms and cumulative terms, and the constraints (such as maximum opening and maximum height) are linear or simple boundary constraints, this embodiment preferably uses the Sequential Quadratic Programming (SQP) algorithm for solving the problem. The SQP algorithm solves the nonlinear programming problem by approximating it as a quadratic programming problem at the current iteration point, resulting in fast convergence and high accuracy. In another optional implementation, if the system state equation is complex or exhibits non-convex characteristics, a Genetic Algorithm (GA) can be used. The Genetic Algorithm simulates the natural selection process, performing crossover, mutation, and selection on the control variable sequence population, and can find the global optimum with a relatively high probability.

[0113] After the search is completed, although the system obtains a series of optimal control sequences for the entire future time period, under the Rolling Time Control (RHC) framework, the system only adopts the data from the first time step of the sequence as the current overflow control command to be executed. In the next control cycle, the system will re-perform the entire optimization process based on the new measured data. The rolling optimization mechanism gives the system extremely strong anti-interference capability and robustness.

[0114] Example 6: This example, as a deep optimization scheme for overflow stopping logic, introduces the concepts of probability constraints and confidence intervals from statistics to solve the risk of overflow stopping too early or too late due to sensor noise or model bias.

[0115] Step 601, generating overflow control instructions that match the current operating condition, also includes executing overflow stop decisions based on risk constraints, such as... Figure 4 As shown, the specific steps include: predicting future time T based on the current state of the mud hopper. s To account for measurement noise and model error, the average concentration at this future time is modeled as a normally distributed random variable C(T). s )~N(μ(T s ),σ²(T s )), where μ(T s ) represents the predicted mean, σ(T) s ) represents the standard deviation of the forecast.

[0116] In practical engineering, predictions always contain errors. This embodiment no longer treats the prediction results as absolutely accurate numerical values, but rather as a probability distribution. The predicted mean μ(T) s The prediction standard deviation σ(T) can be directly taken from the model's predicted values. s Specifically, the system can maintain a sliding time window (e.g., the most recent 5 minutes) and record the residuals between the model's predicted values ​​and the actual sensor measurements at each moment in real time. By calculating the standard deviation of this set of residual samples, the prediction uncertainty of the current model can be dynamically evaluated. When the operating conditions are stable, the residuals are small, σ(T) s ) is relatively small; when the operating conditions fluctuate drastically, σ(T) s As the size increases, the system will automatically become more cautious.

[0117] Predicted standard deviation σ(T) s The residuals are obtained using a sliding time window-based statistical method. Specifically, the system maintains a fixed-length historical record window to record the residual sequence between the predicted and measured values ​​at recent times.

[0118] Let the current time be t, and the sliding window length be W time steps, then the residual sequence is defined as:

[0119] e i =C'(t i )-C measured (ti), i=1,2,…,W; where e i C'(t) represents the prediction residual at the i-th time step, in kilograms per cubic meter; i ) indicates at time t i The previously predicted average concentration for that time period; Cmeasured (ti) represents time t. i The actual measured average concentration value; W represents the number of time steps contained in the sliding window, which is a positive integer, typically ranging from 10 to 30.

[0120] Based on the above residual sequence, the predicted standard deviation is estimated using the following formula:

[0121] ; where Σ i=1 W This represents the summation of W residual samples within the window. W is the mean of the residuals, and W-1 is the degree of freedom correction factor.

[0122] The advantage of this dynamic estimation method is that when the operating conditions are stable and the model prediction is accurate, the residual is small, and σ(T) s The residual is relatively small, allowing the system to use a tighter safety margin; however, when operating conditions fluctuate drastically or sensor noise increases, the residual increases, and σ(T) s As the value increases accordingly, the system automatically becomes more cautious, requiring a higher forecast mean to trigger a stop decision.

[0123] Step 602: Read the preset confidence level parameter p0 and the concentration tolerance margin ε, and construct the probabilistic constraint: P(C(Ts)≥C target -ε)≥p0, the probability constraint requires that the probability that the random variable of concentration is greater than or equal to the expected average sediment concentration target value minus the allowable deviation margin of concentration is not less than the confidence level parameter.

[0124] This embodiment defines the risk tolerance in engineering. The confidence level parameter p0 is typically set to a high value, such as 0.95 or 0.99, indicating a requirement of 95% or 99% certainty. The allowable concentration margin ε is a small buffer value, allowing the final concentration to be slightly lower than the target value. The physical meaning of this probabilistic constraint is that the system must guarantee that, in the vast majority of cases (probability p0), the average concentration at the final stopping point will not be lower than the target value minus an acceptable error.

[0125] Step 603: Utilize the predicted mean and predicted variance, i.e., the standard normal distribution quantile z. p0 Transform the probabilistic constraints into a deterministic safety margin inequality: μ(Ts)-z p0 ·σ(Ts)≥C target –ε.

[0126] To facilitate computer calculation, the above probability form needs to be transformed into an algebraic inequality. According to the properties of the normal distribution, P(X≥x)≥p0 is equivalent to the mean minus z times the standard deviation being greater than or equal to x. Where z... p0It is the quantile of the standard normal distribution with a cumulative probability of p0. For example, when p0 is 0.95, z p0 Approximately 1.645. This inequality clearly shows that the predicted mean μ(Ts) must not only reach the target, but also exceed it by a safe distance (z). p0 •σ(Ts)). The safety distance is dynamic, increasing as the prediction error σ(Ts) increases. This means that if the current measurement noise is high, the system will require the predicted concentration to be much higher than the target value before stopping, effectively avoiding the risk of misjudgment caused by data fluctuations.

[0127] In addition to the lower bound constraint, this embodiment can further set an upper bound constraint on the concentration to avoid energy waste or equipment wear caused by excessive overflow time. Specifically, the preset upper bound allowable value of the concentration and the corresponding confidence level parameter are read to construct the upper bound probability constraint condition: P(C(T) s )≤C upper )≥q0; where P represents the probability operator; C(T)≥q0; s ) indicates the candidate stopping time T s The average concentration random variable at point C; upper This represents the preset upper limit of the allowable concentration, expressed in kilograms per cubic meter. Its value is typically slightly higher than the target concentration C. target , is used to limit the maximum allowable range of concentration overshoot; q0 represents the confidence level parameter of the upper bound constraint, which is a real number between zero and one, with a typical value of 0.90 to 0.99.

[0128] By utilizing the properties of the normal distribution, the above probability constraint can be transformed into a deterministic safety margin inequality:

[0129] μ(T s )+z q0 ·σ(T s )≤C upper ;wherein, μ(T s ) indicates the candidate stopping time T s The predicted mean of the average concentration at the location, in kilograms per cubic meter; z q0 z represents the quantile of the standard normal distribution with a cumulative probability of q0. It is a dimensionless real number; for example, when q0 is 0.95, z... q0 Approximately 1.645; σ(T) s ) indicates the candidate stopping time T s The predicted standard deviation of the average concentration at the location, in kilograms per cubic meter.

[0130] In practical decision-making, the system needs to simultaneously check whether the lower bound constraint and the upper bound constraint are satisfied. Only when a candidate stops at time T... sA target stopping time with two-way safety margin can only be determined when both of the following inequalities are satisfied:

[0131] μ(T s )-z p0 ·σ(T s )≥C target -ε;

[0132] μ(T s )+z q0 ·σ(T s )≤C upper ;

[0133] The first inequality ensures the concentration won't be too low, while the second inequality ensures it won't be too high. Through this two-way risk constraint mechanism, the system can achieve a balance between underload and overload risks, selecting the most robust stopping point.

[0134] Step 604: Detect whether the future time Ts satisfies the safety margin inequality. If it does, determine that the future time Ts is the target stopping time with a safety margin, and generate an overflow stop command to close the overflow control device when it approaches that time.

[0135] This embodiment represents the final decision-making stage. The system iterates through a series of candidate stopping times Ts. Once a time is found to satisfy the aforementioned safety margin inequality, that time is locked as the planned stopping point. As a preferred supplement to this embodiment, the system performs a conservative buffer operation before generating the final closing command. Specifically, within a preset buffer period (e.g., 30 seconds) before reaching the target stopping time Ts, the system first forcibly adjusts the bottom opening degree command to the minimum allowable opening degree A. min Maintain a small overflow. This buffer phase serves both as a final confirmation period to prevent a sudden rebound in concentration and as a mitigation of hydrodynamic shocks to protect pipeline equipment. Once everything is confirmed to be in order, close the overflow port and bottom opening.

[0136] Example 7 describes how, during actual overflow, the entire process is divided into different control stages based on the real-time concentration state within the sludge chamber, and differentiated control laws are used to balance emission efficiency and control accuracy. This example not only implements piecewise function logic but also supplements a simplified linear prediction model for the fine-tuning stage, serving as a lightweight alternative to the high-computing-power PDE model.

[0137] Step 701: Generate overflow control commands matching the current operating conditions. A phased adaptive control strategy is adopted, reading the preset proportional coefficient, the maximum allowable opening area, and the minimum allowable opening area, and comparing the real-time average sediment concentration with the product of the target value of the expected average sediment concentration and the proportional coefficient.

[0138] If the real-time average sediment concentration is less than the product, it is determined that the current stage is coarse adjustment. An instruction is generated to set the bottom opening degree to the maximum opening area to maximize the discharge efficiency of low-concentration water.

[0139] If the real-time average sediment concentration is greater than or equal to the product and less than the target value of the expected average sediment concentration, it is determined that the current stage is fine-tuning. Based on the degree of closeness between the real-time average sediment concentration and the target value of the expected average sediment concentration, the opening value that decreases linearly with the increase of concentration is calculated, and the bottom opening command is set to this opening value to suppress concentration overshoot.

[0140] Specifically, the effective opening area A(t) of the bottom opening is calculated based on the real-time average sediment concentration C'(t), and the calculation formula is a piecewise function: A(t) = {A max C'(t)<αC target A max -k(C'(t)-αC target ),αC target ≤C'(t) <C target A min ,C'(t)≥C target}, where A(t) is the bottom opening degree instruction at time t; A max and A min These represent the maximum and minimum allowable opening areas, respectively; α is a proportionality coefficient between 0 and 1, used to distinguish between the coarse adjustment stage and the fine adjustment stage; k is the opening attenuation coefficient, used to control the rate at which the opening decreases with increasing concentration during the fine adjustment stage.

[0141] In the simplified linear prediction model, the extrapolation adjustment coefficient β is used to correct for systematic biases that may arise from simple linear extrapolation. Its role in the calculation formula is as follows:

[0142] C'(t+△t)=C'(t)+β·dC' / dt·△t; where C'(t+△t) represents the predicted average concentration at the next moment, in kilograms per cubic meter; C'(t) represents the measured average concentration at the current moment, in kilograms per cubic meter; β represents the extrapolation adjustment coefficient, a dimensionless real number; dC' / dt represents the rate of change of concentration at the current moment, in kilograms per cubic meter per second, which is approximated by the difference C'(t)-C'(t-△t) / △t; △t represents the time step, in seconds.

[0143] The engineering significance and typical values ​​of the extrapolation adjustment factor β are as follows:

[0144] When β=1, it indicates complete trust in linear extrapolation, meaning it assumes that the future rate of change is the same as the current rate of change. This is suitable for operating conditions where the concentration change is relatively linearly stable.

[0145] When β < 1, it indicates a conservative correction to the linear extrapolation, and the predicted change is smaller than the result of the linear extrapolation. This is suitable for operating conditions where the concentration change rate may slow down, such as near saturation. Typical values ​​range from 0.7 to 0.9.

[0146] When β > 1, it indicates an aggressive correction to the linear extrapolation, with the predicted change being larger than the result of the linear extrapolation. This is suitable for special operating conditions where the rate of change may accelerate. Typical values ​​range from 1.0 to 1.2.

[0147] In practical engineering, β can be calibrated through regression analysis of historical data, or dynamically adjusted based on the current second derivative estimate. As a preferred embodiment, the initial value of β is set to 0.85, and it is corrected online during operation based on the cumulative deviation of the prediction error.

[0148] This embodiment discretizes the continuous control process into three logically clear stages: coarse adjustment stage, fine adjustment stage, and conservative stage.

[0149] During the coarse adjustment phase, when the real-time average sediment concentration C'(t) has not yet reached the target value C target When the water level reaches a certain percentage (i.e., less than α times the target value), the system determines that the current mud chamber is severely underloaded, and the primary task is to quickly replace the water. At this time, the system outputs the maximum opening command A. max The proportionality coefficient α is a boundary parameter for this stage, typically ranging from 0.7 to 0.8. This means that during the first 70% to 80% increase in concentration, the overflow system operates at full speed to minimize construction time.

[0150] During the fine-tuning phase, when the concentration enters the critical range of the target value (i.e., αC)... target To C target When the concentration is between (the two values), the system automatically switches the control law. At this point, the maximum opening is no longer maintained; instead, a negative feedback mechanism is introduced. As C'(t) increases, the opening A(t) decreases linearly. The opening attenuation coefficient k determines the sensitivity of the adjustment. If k is large, the opening will contract sharply with a small increase in concentration; this sudden braking strategy helps prevent concentration overshoot but may cause pipeline pressure fluctuations. If k is small, the adjustment process is smoother. This linear attenuation formula mathematically constitutes a proportional controller (P controller), whose function is to actively reduce the system gain when approaching the target, achieving a soft landing.

[0151] When the concentration eventually reaches or exceeds the target value C target At this point, the system enters a conservative phase (corresponding to the third case in the formula), locking the opening at the minimum sustaining opening A. min And prepare to trigger the stop judgment logic.

[0152] In some alternative implementations, to compensate for the lag in linear feedback control, especially when computational resources are limited and complex PDE evolution models cannot be run, this embodiment introduces a simplified linear prediction model to enhance the fine-tuning stage. Specifically, the system uses a difference formula to estimate the current concentration change rate dC / dt≈(C'(t)-C'(t-Δt)) / Δt. Based on this change rate, the concentration at the next time step is predicted. , where β is the empirical correction coefficient. When calculating the piecewise function, the system can use the predicted values. Instead of the measured value C'(t), the aperture is determined by looking up a table. For example, if the predicted value shows that the concentration will exceed αC at the next moment. target The system will enter the fine-tuning stage one time step in advance and begin to reduce the opening. Based on the predictive control of the simplified model, it can effectively suppress the concentration overshoot caused by the sensor response delay, and the calculation time is extremely low.

[0153] Example 8 describes how, when the device has multi-stage vertical overflow ports or planar movement capabilities, specific priority functions and planar indices can be used to further explore the control potential of the mud hopper in the vertical and horizontal directions and eliminate local dead zones.

[0154] Step 801: The overflow control device is equipped with multiple switchable overflow ports at different vertical heights; generating overflow control commands matching the current operating conditions includes executing a stratified optimal discharge strategy, specifically including: based on the mud tank vertical concentration-gradation profile model, extracting the height H of each multi-stage switchable overflow port k. k The local concentration value C(H) at the location k ,t) and the proportion of local fine particles φ fine (H k ,t).

[0155] This embodiment describes an improved hardware structure where multiple independent windows (e.g., upper, middle, and lower levels) are created at different heights on the sidewall of the overflow pipe, each equipped with an independent opening and closing valve. The system utilizes a full-field model to virtually sample the height of each window. Although physical sensors may only sample at certain points, the system can accurately estimate the height H of each window using models C(z) and P(z,d). k The local water quality characteristics at the location demonstrate the advantages of model-driven control: supporting dense control point decisions with a small number of monitoring points.

[0156] In a multi-stage switchable overflow structure, the effective opening area of ​​each overflow outlet is determined by both the open / closed state and the opening ratio, and its calculation formula is: A k (t)=b k (t)·u k (t)·A geom,k Among them, Ak (t) represents the effective opening area of ​​the k-th overflow outlet at time t, in square meters; b k (t) represents the opening / closing status indication of the k-th overflow outlet at time t, and is a binary variable. When b k When (t)=0, it indicates that the overflow outlet is in a closed state. When b k When (t)=1, it indicates that the overflow outlet is in the allowed-to-open state; u k (t) represents the opening degree of the k-th overflow outlet at time t, which is a dimensionless real number ranging from zero to one, where zero represents fully closed and one represents fully open; A geom,k This represents the maximum geometric opening area of ​​the k-th overflow outlet, in square meters, and is determined by the physical dimensions of the overflow outlet.

[0157] The instantaneous overflow flow rate of each overflow outlet can be estimated using the orifice outflow formula:

[0158] Q k (t)=C d,k ·A k (t)·sqrt2g[H w (t)-H k ] + ; where Q k (t) represents the instantaneous overflow flow rate of the k-th overflow outlet at time t, in cubic meters per second; C d,k The flow coefficient of the k-th overflow outlet is a dimensionless real number, typically ranging from 0.6 to 0.9, with the specific value determined by the geometry and edge conditions of the overflow outlet; A k (t) represents the effective opening area of ​​the k-th overflow outlet at time t, in square meters; g represents the gravitational acceleration, which is 9.81 m / s²; H w (t) represents the water level in the mud tank at time t, in meters; H k [x] represents the center height of the k-th overflow outlet, in meters; + This represents the positive value operator, which takes the value x itself when the value x inside the parentheses is greater than zero, and takes zero when x is less than or equal to zero. This operator is used to ensure that there is no outflow from the overflow outlet when the water level is lower than the overflow outlet height.

[0159] The total overflow flow rate of the mud tank is the sum of the flow rates of each overflow outlet:

[0160] Q ovf (t)=Σ k=1 K Q k (t); where Q is... ovf (t) represents the total overflow flow rate at time t, in cubic meters per second; K represents the total number of multi-stage switchable overflow outlets, which is a positive integer; Σ k=1K This represents the summation of the flow rates over all K overflow outlets.

[0161] Before implementing the stratified optimal discharge strategy, it is necessary to pre-set the reference concentration threshold C for determining the low-concentration water layer. ref and related weighting coefficients. Reference concentration threshold C ref This represents the critical value used to determine whether a water layer at a certain depth is a low-concentration, dischargeable aquifer, expressed in kilograms per cubic meter. Its value should be lower than the average concentration in the mud tank; a typical range is the target concentration C. target 30% to 50%. For example, when C target When set at 600 kg per cubic meter, C ref It can be set between 180 and 300 kg per cubic meter. When the local concentration C(H) at the height of a certain overflow outlet... k ,t) is lower than C ref When this occurs, it indicates that the water layer has discharge value.

[0162] Weighting coefficients η1 and η2 are used to balance the potentially conflicting objectives of discharging low-concentration water and retaining fine particles:

[0163] η1 represents the degree of importance attached to discharging low-concentration water bodies. It is a non-negative real number, typically ranging from 0.5 to 2.0. The larger the η1 value, the more the system tends to choose to open the overflow outlet with the lowest concentration.

[0164] η2 represents the degree of importance attached to retaining fine particulate matter. It is a non-negative real number, typically ranging from 0.1 to 1.0. The larger the η2 value, the more the system tends to avoid opening overflow outlets with a higher proportion of fine particles, even if the concentration at that point is low.

[0165] In a typical dredging project, if the project focuses on increasing output, η1=1.5 and η2=0.3 can be set; if the project focuses on retaining fine particles to improve soil quality, η1=0.8 and η2=0.8 can be set.

[0166] When the opening of a certain overflow point spans the boundaries of two or more adjacent virtual water and sediment layers, a weighted allocation mechanism is needed to determine the proportion extracted from each layer. Let the center height of the k-th overflow point be H. k The opening height range is [H] k -h k / 2,H k +h k / 2], where h k This refers to the opening height of the overflow port.

[0167] If the opening range is within the depth interval of the j-th virtual layer [z j bottom ,z jtop If there is overlap, then the overlap length is:

[0168] △h k,j =max (0,min (H k +h k / 2,z j top )-max(H k -h k / 2,z j bottom ));

[0169] Wherein, △h k,j The value represents the overlap length between the opening range of the k-th overflow outlet and the depth interval of the j-th virtual layer, in meters; max and min represent operations that take the larger and smaller values, respectively.

[0170] The traffic allocation weight extracted from the j-th virtual layer by the k-th overflow port is:

[0171] w k,j =△h k,j / h k Among them, w k,j This represents the weight allocation, a dimensionless real number ranging from zero to one, and the sum of all layers overlapping the overflow outlet equals one, i.e., Σ. j w k,j =1.

[0172] Therefore, the volumetric flow rate extracted from the j-th virtual layer through the k-th overflow port is:

[0173] Q k,j (t)=w k,j ·Q k (t); where Q is... k,j (t) represents the volumetric flow rate drawn from the j-th layer by the k-th overflow outlet at time t, in cubic meters per second; Q k (t) represents the total flow rate of the k-th overflow outlet. This weighting mechanism ensures that when the overflow outlet is located between two layers, the model can correctly calculate the impact on the quality of each layer, avoiding discontinuities at layer boundaries.

[0174] Step 802: Construct an overflow priority function. The overflow priority function is negatively correlated with the local concentration value and negatively correlated with the local fine particle ratio. It is used to evaluate the suitability of overflow from different heights.

[0175] Calculate the discharge priority index F for each multi-stage switchable overflow port. k (t), the calculation formula is: F k (t)=η1[C ref -C(Hk ,t)]+-η2φ fine (H k ,t), where C ref The reference concentration threshold for determining low-concentration water layers; [·]+ indicates positive value operation, that is, when the value in the parentheses is less than zero, it is zero; η1 and η2 are non-negative weighting coefficients, which respectively represent the degree of importance attached to the discharge of low-concentration water and the retention of fine particulate matter;

[0176] Excretion priority index F k (t) is the scoring mechanism used to solve the decision-making problem of which opening to make. The first term of the formula [C ref -C(H k ,t)] + Pay attention to concentration. If at a certain height H k The concentration is lower than the reference value C. ref (For example, the clear water boundary), the larger the difference, the higher the score, indicating that the water body at that location is more worthwhile to discharge; if the concentration is higher than C ref This term is zero, indicating no emission value. The second term in the formula is -η²φ. fine (H k η1, η2 is a penalty term. Even if the concentration is low at a certain point, if it contains a very high proportion of fine particles (e.g., thin colloidal mud), this term will reduce its total score and inhibit the extraction of that layer. By adjusting the weights η1 and η2, engineers can flexibly define which water layers are the worst and should be discharged.

[0177] Step 803: Calculate the discharge priority index for each multi-stage switchable overflow port using the discharge priority function, and select the discharge priority index F. k (t) The highest one or more overflow ports are taken as the target opening objects, and an overflow control command containing the target opening object identifier is generated to control the overflow control device to only open the target opening objects and close the remaining overflow ports.

[0178] This embodiment describes the decision-making and execution phase. The system processes the F values ​​of all K overflow ports. k (t) Sort the data and select the port with the highest score to open. For example, in the early stages of overflow, the uppermost window has the highest score, so the system opens the upper window; as the liquid level drops and the mud layer rises, the concentration near the upper window increases, causing the score to decrease, while the score of the middle window relatively increases, and the system automatically switches to the middle window. The dynamic switching mechanism realizes the stratification and stripping of the mud tank water, always maintaining the current optimal height for operation.

[0179] Step 804, the overflow control device has the ability to move the overflow inlet position within the mud tank horizontal plane; the method further includes performing a planar non-uniformity control step: dividing the mud tank horizontal plane into multiple planar regions, and estimating the regional average concentration of each planar region based on sensor data distributed at different locations.

[0180] This embodiment addresses a planar movable overflow device (e.g., an overflow pipe mounted on a gantry or rotating arm). The system establishes a two-dimensional planar grid model, dividing the mud tank water surface into P regions (e.g., port side, starboard side, bow, and stern). Through distributed sensors deployed in these regions, or by using mobile sensors for inspection, the system maintains a planar concentration heatmap in real time, recording the average concentration C of each region. p (t).

[0181] In the planar partitioning model, the average concentration of each planar region is obtained by integrating the vertical concentration profile within that region: C p (t)=1 / H p ∫ zmin,p zmax,p C p (z,t) dz; where C p (t) represents the average concentration of the p-th planar region at time t, in kilograms per cubic meter; H p H represents the effective water depth of the p-th planar region, in meters, and is calculated using the formula: H p =z max,p -z min,p ;z min,p z represents the upper boundary depth of the water body in the p-th planar region, in meters; max,p C represents the lower boundary depth of the water body within the p-th planar region, in meters; p (z,t) represents the volumetric sand content per unit volume at depth z within the p-th planar region at time t, expressed in kilograms per cubic meter; ∫ zmin,p zmax,p ·dz represents the definite integral operation along the depth direction from the upper bound to the lower bound.

[0182] When determining the planar area to which the current location of the overflow inlet belongs, the minimum distance criterion is used:

[0183] p*(t)=arg min p∈{1,2,… ,P} [(x ovf (t)-x p ) 2 +(y ovf (t)-y p ) 2Where p*(t) represents the planar region number to which the overflow inlet belongs at time t, and is a positive integer; arg min represents the value of the independent variable that minimizes the objective expression; p represents the candidate planar region number variable; {1,2,…,P} Let P represent the set of all planar region numbers, where P is the total number of planar regions; x ovf (t) represents the horizontal coordinate of the overflow inlet on the horizontal plane at time t, in meters; y ovf (t) represents the longitudinal coordinate of the overflow inlet on the horizontal plane at time t, in meters; x p The y-coordinate represents the horizontal coordinate of the center point of the p-th planar region, in meters; p This represents the longitudinal coordinate of the center point of the p-th planar region, in meters.

[0184] The physical meaning of this distance criterion is that the overflow inlet is assumed to draw water and sediment primarily from the planar region closest to its geometric location. This simplified assumption applies to working conditions where the horizontal flow within the mud tank is relatively slow.

[0185] Step 805: Calculate the regional concentration deviation between the regional average concentration of each planar area and the overall average concentration of the mud tank, and construct a planar non-uniformity index to characterize the differences in the horizontal distribution of the mud tank based on all regional concentration deviations; identify specific planar areas where the regional concentration deviation indicates a concentration lower than the overall average concentration, and set them as target operating areas.

[0186] Planar non-uniformity index J plane The complete definition of (t) is the sum of squares of the concentration deviations in each planar region:

[0187] J plane (t)=Σ p=1 P [△C p (t)] 2 ; among which, J plane (t) represents the plane concentration non-uniformity index at time t, with units of kilograms squared per cubic meter squared. The smaller the index value, the more uniform the concentration distribution in the plane direction within the mud tank; Σ p=1 P This represents the summation over all P planar regions; P represents the total number of planar regions, which is a positive integer; △C p (t) represents the regional concentration deviation of the p-th planar region at time t, in kilograms per cubic meter, and its calculation formula is:

[0188] △C p (t)=C p (t)-C total (t); where C p (t) represents the regional average concentration of the p-th planar region at time t; Ctotal (t) represents the overall average concentration of the mud tank at time t, and its calculation formula is:

[0189] C total (t)=Σ p=1 P V p (t)·C p (t) / Σ p=1 P V p (t); where V p (t) represents the water volume of the p-th planar region at time t, in cubic meters. When ΔC p When (t) is positive, it indicates that the concentration in the p-th region is higher than the overall average level; when ΔC p When (t) is negative, it indicates that the concentration in the p-th region is lower than the overall average. One of the goals of the control system is to adjust J by moving the overflow inlet position. plane (t) tends to be minimized.

[0190] Step 806: Generate a planar movement command to drive the overflow control device to move the overflow inlet position horizontally to the coordinate range of the target operation area, prioritize the discharge of low-concentration water in the area and reduce the planar non-uniformity index.

[0191] These two steps constitute the closed loop of planar control. Regional concentration deviation ΔC p (t)=C p (t)-C total (t) reflects local inhomogeneity. If the deviation in a certain area is negative and the absolute value is large, it indicates the presence of clear water pits or low-concentration dead zones. To calculate inhomogeneity, a planar inhomogeneity index J can be constructed. plane (t)=∑(ΔC p (t)) 2 .

[0192] In determining the target work area p target In this embodiment, the following comprehensive criterion formula is preferably adopted: p target =argmax p (-λ1ΔC p (t)+λ2f fine (p,t)). Among them, the first term is -λ1ΔC p (t) guides the overflow outlet to the area with the lowest concentration (ΔC) p When the value is large negative, the term is large positive); the second term λ2f fine (p,t) is used to correct for the effect of fine particle distribution. For example, if a region has a low concentration but a large number of floating fine particles, the system may avoid selecting that region. Determine p. targetThen, the control system calculates the distance from the current coordinates (x, y) to the center coordinates (x, y) of the target area. target ,y target The path drives the horizontal moving mechanism to perform displacement, actively eliminating uneven distribution on the plane and improving the consistency of the entire cabin.

[0193] To address the problem of isolated monitoring data and the inability to reconstruct the entire field profile, the embodiment constructs a continuous vertical concentration-gradation profile model of the mud tank by collecting vertical data from multiple points and using cubic spline interpolation and fitting techniques. This enables a holographic perception of the flow field and stratification state inside the mud tank, allowing control decisions to no longer be limited to local single points.

[0194] To address the issues of concentration overshoot and backmixing caused by control logic lag, this embodiment introduces a one-dimensional transport-diffusion-sedimentation evolution equation and a finite difference solution algorithm to establish a dynamic prediction model. This model enables feedforward control of future sediment deposition trends, allowing for early prediction of the rise of high-concentration layers and adjustment of the overflow outlet. Simultaneously, a risk-stopping strategy based on probability constraints is combined with a normal distribution model to handle prediction errors, effectively avoiding the risk of misjudgment caused by measurement noise and ensuring that the overflow stops within a safety margin.

[0195] To address the lack of multi-objective collaborative optimization, the implementation example adopts a discrete layer optimization decision-making strategy. By constructing a comprehensive evaluation function that includes concentration, gradation, and loss rate, the optimal overflow port height and opening are automatically calculated using a sequential quadratic programming algorithm. This achieves a leap from blind adjustment based on experience to mathematical optimization, minimizing the loss of fine particles while ensuring the average concentration.

[0196] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for controlling and monitoring the concentration of overflow from a cargo hold, characterized in that, include: Vertical detection data were collected from multiple discrete depth points within the water depth range of the mud chamber. The vertical detection data were used to construct a vertical concentration-gradation profile model of the mud chamber, which characterizes the continuous function of sand content and particle size distribution inside the mud chamber. Read the preset engineering target parameters, map the engineering target parameters to the mud tank vertical concentration-gradation profile model for multi-objective coupling calculation, and generate overflow control commands that match the current working conditions. The overflow control commands include overflow port height commands and bottom opening degree commands. In response to the overflow control command, the overflow control device is driven to adjust the overflow inlet to the vertical position corresponding to the overflow port height command, and adjust the bottom opening to the effective flow area corresponding to the bottom opening opening degree command; The vertical concentration-gradation profile model of the mud chamber, which uses vertical detection data to construct a continuous function characterizing the sand content and particle size distribution inside the mud chamber, includes: Analyze the vertical detection data to extract the instantaneous sediment concentration measurement values ​​and discrete particle size distribution functions at multiple discrete depth coordinates within the water depth range of the mud chamber; Spatial interpolation or numerical fitting is performed on multiple instantaneous sediment concentration measurements to construct a volumetric sediment concentration function that describes the continuous change of sediment concentration with depth. Spatial interpolation is performed on multiple discrete particle size distribution functions to construct a particle distribution function that describes the continuous variation of particle size distribution with depth and particle size. The volumetric unit sediment concentration function and particle distribution function are combined into a mud chamber vertical concentration-gradation profile model, providing a basis for querying concentration and gradation at any depth; The multi-objective coupled computation adopts a discrete layer optimization decision strategy, which constructs a discrete layer state model of the mud chamber; Discrete-layer optimization decision strategies further include the step of solving for the optimal control sequence: Define control variables including overflow outlet height and overflow flow rate, and set multiple discrete time steps within a preset prediction time interval; A discrete overall objective function is constructed, which is the result of a weighted sum of the squared terms of concentration deviation, squared terms of gradation deviation, and fine particle loss penalty term at all discrete time steps within the prediction time interval. Dynamic simulation is performed based on the layer quality balance update mechanism. Under the premise of satisfying the preset constraints, a set of control variable sequences that minimizes the discrete total objective function is searched, and the data of the first time step in the set of control variable sequences is converted into overflow control commands.

2. The method according to claim 1, characterized in that, The engineering target parameters include the expected average sediment concentration target value, the expected total proportion of fine particles target value, and the maximum allowable loss ratio of fine particles. These engineering target parameters are mapped to the vertical concentration-gradation profile model of the mud silt chamber for multi-objective coupled calculations, generating overflow control commands that match the current operating conditions, including: Based on the volumetric sediment concentration function C(z) and particle distribution function P(z,d), the real-time average sediment concentration C'(t) and the real-time total proportion of fine particles θ in the mud chamber at the current moment are calculated through integral operations. fine (t), the calculation formula is: C'(t)=(1 / H)*∫[z min ->z max ]C(z)dz; θ fine (t)=(1 / (H*C'(t)))*∫[z min ->z max ](C(z)*∫[0->d fine ]P(z,d)dd)dz; Where H represents the effective water depth of the mud chamber, z min and z max d represents the upper and lower boundaries of the water body, respectively. fine This indicates the upper limit threshold for fine particle size; A multi-objective comprehensive evaluation function J is constructed by combining the engineering objective parameters. The definition of the multi-objective comprehensive evaluation function J is as follows: J=w1*|C'(t)-C target |+w2*|θ fine (t)-θ fine , target |+w3*L fine (t); where C target θ is the target value for the desired average sediment concentration. fine , target To determine the target value for the overall proportion of fine particles, L fine (t) represents the cumulative fine particle loss ratio from the start of the overflow to the current time, and w1, w2, w3 are preset non-negative weighting coefficients; With the goal of minimizing the multi-objective comprehensive evaluation function J, the optimal overflow outlet height and bottom opening degree are searched within the preset feasible region, and the search results are converted into overflow control commands.

3. The method according to claim 1, characterized in that, The method further includes constructing a dynamic model capable of predicting the evolution of sediment morphology within the mud chamber over time. Specific steps include: A one-dimensional transport-diffusion-sedimentation evolution equation is established to describe the movement of suspended sediment within the mud chamber. The equation is expressed as follows: ΨC(z,t) / Ψt=-Ψ[v(z,t)C(z,t)] / Ψz+DΨ²C(z,t) / Ψz²-w s ΨC(z,t) / Ψz; where Ψ is the partial derivative, C(z,t) represents the volumetric sediment concentration per unit volume at depth z and time t, v(z,t) represents the effective vertical velocity, D represents the effective turbulent diffusion coefficient, and w s Indicates particle settling velocity; The volumetric sediment concentration function in the vertical concentration-gradation profile model of the mud chamber is used as the spatial distribution state C(z,t0) at the initial time t0 and input into the one-dimensional transport-diffusion-sedimentation evolution equation. Based on the one-dimensional transport-diffusion-sedimentation evolution equation, time-step calculations are performed to predict the future concentration distribution value C(z,t+Δt) at one or more future time steps, providing a feedforward control basis for adjusting the overflow outlet height.

4. The method according to claim 3, characterized in that, Time-step calculations based on the one-dimensional transport-diffusion-settlement evolution equations include solving the problem using boundary conditions and finite difference schemes. Specific steps include: At the upper boundary depth z of the mud chamber water min and lower bound depth z max The upper and lower boundary conditions are set at points respectively, and their mathematical expressions are as follows: Upper boundary: $(v(z,t)C(z,t)-D\frac{\Psi C(z,t)}{\Psi z})|_{z = z}$ min $= 0$; Lower boundary: (v(z,t)C(z,t) - DΨC(z,t) / Ψz + w s C(z,t))|z=z max = 0; Where Ψ is the partial derivative, the upper boundary condition indicates that the total solid flux at the surface of the mud tank is zero, and the lower boundary condition indicates that the total solid flux at the bottom of the mud tank is equal to the particle settling velocity w. s The determined sedimentation flux; The mud hopper is discretized vertically into grid points j, and the time axis is discretized into time steps n. An explicit finite difference scheme is used to numerically discretize the one-dimensional transport-diffusion-settlement evolution equation, yielding an iterative formula for the discrete node prediction sequence: C j (n+1) =C j n +Δt[-v(z j (C) (j+1) n -C (j-1) n ) / (2Δz)+D(C (j+1) n -2C j n +C (j-1) n ) / (Δz) 2 -w s (C (j+1) n -C (j-1) n ) / (2Δz)];where, C j (n+1) C represents the predicted concentration at the j-th spatial grid point at the (n+1)-th time step. j n Let Δt be the known concentration at the nth time step, Δz be the time step size, and Δz be the spatial step size.

5. The method according to claim 1, characterized in that, The multi-objective coupled computation employs a discrete-layer optimization decision-making strategy, which constructs a discrete-layer state model of the mud chamber. The specific steps include: The vertical depth range of the mud chamber is divided into N virtual water-sand layers, and the layer volume V of each virtual water-sand layer k is initialized. k (t) and total mass of the solid layer M k (t); Establish a layer extraction mapping relationship to determine that the height of the overflow outlet at time t corresponds to the m-th virtual water-sediment layer, and obtain the overflow flow rate Q at that time. ovf (t); A layer mass balance update mechanism is constructed. For the m(t)th virtual water-sand layer extracted, the following volume and mass update calculations are performed within a time step Δt: V m (t)(t+Δt)=V m (t)(t)-Q ovf (t)·Δt; M m (t)(t+Δt)=M m (t)(t)-Q ovf (t)·Δt·C m (t)(t); Among them, V m (t)(t+Δt) and M m (t)(t+Δt) represent the updated layer volume and the total solid mass of the layer, respectively, C m(t) (t) represents the average concentration of the layer at time t, and for other layers that are not extracted, k≠m(t), their volume and mass remain unchanged.

6. The method according to claim 1, characterized in that, The discrete overall objective function is constructed using a weighted summation method, and the specific calculation formula is as follows: J=∑ n=0 N T [α1(C'(t n )-C target ) 2 +α2(θ fine (t n )-θ fine,target ) 2 +α3L fine (t n )]; where J is the discrete total objective function, t n N represents the nth discrete time step. T C'(t) represents the total number of time steps within the predicted time interval. n ), θ fine (t n ) and L fine (t n The values ​​(C) represent the predicted average sediment concentration, total proportion of fine particles, and cumulative fine particle loss proportion for that time step. target and θ fine , target These represent the target values ​​for the expected average sediment concentration and the expected overall proportion of fine particles, respectively; α1, α2, and α3 are non-negative weighting coefficients used to balance the importance of each target.

7. The method according to claim 2, characterized in that, Generating overflow control instructions that match the current operating conditions also includes executing overflow stop decisions based on risk constraints, the specific steps of which include: Predicting future time T based on the current state of the mud hopper s The average concentration at this future time is modeled as a random variable C(T) that follows a normal distribution, taking into account measurement noise and model error. s )~N(μ(T s ),σ²(T s )), where μ(T s ) represents the predicted mean, σ(T) s () represents the standard deviation of the forecast; Read the preset confidence level parameter p0 and the allowable lower deviation margin ε of the concentration, and construct the probabilistic constraints: P(C(Ts)≥C target -ε)≥p0; Using the standard normal distribution quantile z p0 Transform probabilistic constraints into deterministic safety margin inequalities: μ(Ts)-z p0 ·σ(Ts)≥C target –e; Detecting future moments T s Does the safety margin inequality satisfy? If it does, then determine the future time T. s A target stopping time with a safety margin is defined, and an overflow stop command is generated to close the overflow control device as the time approaches that time.

8. The method according to claim 2, characterized in that, The overflow control command that matches the current operating conditions is generated using a staged adaptive control strategy. Specifically, the effective opening area A(t) of the bottom opening is calculated based on the real-time average sediment concentration C'(t). The calculation formula is a piecewise function: A(t) = {A max C'(t)<αC target A max -k(C'(t)-αC target ),αC target ≤C'(t) <C target A min ,C'(t)≥C target }, where A(t) is the bottom opening degree instruction at time t; A max and A min These represent the maximum and minimum allowable opening areas, respectively; α is a proportionality coefficient between 0 and 1, used to distinguish between the coarse adjustment stage and the fine adjustment stage; k is the opening attenuation coefficient, used to control the rate at which the opening decreases with increasing concentration during the fine adjustment stage.

Images