An optimization control system and method for the process of cabin pumping and shore blowing of a trailing suction dredger
By constructing a neural network prediction model, selecting key parameters, and optimizing them using the PSO algorithm, the problem of accurately predicting the critical flow velocity during the dredging and shore-blowing process of trailing suction hopper dredgers was solved, thereby improving efficiency and reducing energy consumption.
Patent Information
- Application Number
- CN202210208945.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-03
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-03-03
AI Technical Summary
Existing trailing suction hopper dredgers cannot accurately predict critical flow velocities during the decanting and shore-blowing process, resulting in low efficiency and increased energy consumption, and failing to achieve optimal control.
A neural network prediction model was constructed. By collecting slurry operation parameters and relevant parameters of the trailing suction hopper dredger, slurry flow velocity, average particle size of slurry and sand, hopper door opening, and slurry pump speed were selected as input variables. The neural network was initialized using the PSO algorithm to predict the critical flow velocity. Wavelet threshold denoising and Leyte criterion were used to process the data to achieve accurate prediction.
It enables accurate prediction of critical flow velocity during the dredging and shore-blowing process of trailing suction hopper dredgers, improving efficiency, reducing energy consumption, and reducing reliance on operator experience.
Smart Images

Figure CN114661006B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to bow blowing dredging operation of a trailing suction hopper dredger, and in particular to an optimization control system and method for a cabin pumping and shore blowing process of a trailing suction hopper dredger. Background Art
[0002] Dredging is an underwater earthwork project that uses specific mechanical equipment to dig sediment, sand or rock to widen and deepen the excavation area. Trailing suction hopper dredgers occupy an important position in dredging construction with their advantages of self-propelled, self-dig, self-load and self-unload. People have higher and higher requirements for the performance, efficiency, automation level and environmental protection level of trailing suction hopper dredgers. Improving the performance and efficiency of dredgers has become the focus of dredging work. Bow blowing is one of the commonly used filling construction methods for trailing suction hopper dredgers. It is mostly used to take sand from offshore deep sea areas and transport it to nearshore areas for bow blowing to create land. A complete trailing suction hopper dredger dredging operation, from dredging and loading at the starting point to pumping out the cabin and blowing mud to the shore at the destination, the efficiency of each link will affect the efficiency of the entire dredger. The chamber pumping and shore blowing process is the process in which mud and sand enter the chamber pumping pipe from the mud chamber and are discharged together with the seawater in the chamber pumping pipe. The key parameter affecting the efficiency of the chamber pumping and shore blowing process is the critical flow velocity. When the critical flow velocity in the chamber pumping pipe is lower than the optimal critical flow velocity, pipe blockage will occur, causing the entire chamber pumping and shore blowing process to stagnate. If the mud flow velocity is higher than the optimal critical flow velocity, it will cause a huge burden on energy consumption and increase the burden on the equipment.
[0003] Due to the insufficiency of the existing prediction technology of critical flow velocity of the trailing suction hopper dredger, the optimal control with the critical flow velocity as the target during the process of cabin pumping and shore blowing of the trailing suction hopper dredger cannot be achieved, resulting in relatively low efficiency of cabin pumping and shore blowing.
[0004] Some optimization control systems for trailing suction hopper dredgers that have been published, such as the automatic dredging optimization control system for trailing suction hopper dredgers disclosed in the prior art patent application No. CN202010317070.5, include a control panel, an embedded logic controller (PAC), a sensor, a programmable logic controller (PLC) and a traditional trailing suction hopper dredger operating system, wherein the control panel and the embedded logic controller can realize the interaction of the human-machine interface, display the data collected by the sensor on the control panel in a digital form, and also can convert the control instructions input by the operator on the operation interface into machine language and pass them to the programmable logic controller; the programmable logic controller can control the corresponding dredging equipment according to the received control instructions. The automatic dredging optimization control system for trailing suction hopper dredger can only provide real-time data of the dredging process for auxiliary judgment, but the specific parameters of the key equipment in the dredging process still need to be adjusted by the operator, and the efficiency of the process of pumping cabins and blowing shore is based on the experience of the operator, and it still cannot realize the automatic optimization of the dredging parameters of intelligent ships, and cannot achieve true optimal control.
[0005] In the prior art, patent application No. CN200510123193.0 discloses an automatic optimization method for dredging construction of a trailing suction hopper dredger, which relates to a technology for automatically controlling the working state of a trailing suction hopper dredger during dredging construction, and specifically, an automatic optimization method for dredging construction of a trailing suction hopper dredger. The method is to set a plurality of subsystems on the trailing suction hopper dredger that can detect and control the working conditions of the dredger, including a high-pressure flushing pump system, a pump system, a rake pipe winch system, a main engine system, a draft loading system, a production meter system, a plane position positioning system, and a rake arm position indication system. In each subsystem, a corresponding sensor device is used to collect signals, and each valid data is stored in a database system. The entire process of dredging construction includes a trial excavation step, an automatic optimization step, and a formal dredging step. Through trial excavation and data collection and analysis at a new construction site where the situation is unfamiliar, guiding optimization construction parameters can be proposed, so that the dredger can achieve continuous high production effects under the premise of ensuring equipment safety. This method can only obtain a set of optimized construction parameters offline, but in the actual construction process, the construction parameters also need to change in real time to ensure work efficiency and safety.
[0006] The dredging methods mentioned in the above-mentioned prior art are all optimized based on the dredging process as a whole, which can shorten the time required for dredging to a certain extent. However, the optimal construction parameters of each process are different. From dredging and loading to pumping out the dredging and blowing the shore, it is necessary to change the set parameters many times. It is impossible to automatically switch the corresponding dredging optimization model according to the different dredging conditions, and it is impossible to keep the mud flow rate at the optimal flow rate, thereby failing to effectively reduce the workload of operators and improve the efficiency and safety of dredging.
[0007] At present, the existing SCADA control system of the trailing suction hopper dredger can only provide real-time data in the dredging process for auxiliary judgment. However, the critical flow velocity in the dredging process still needs to be judged by the operator based on experience, and the critical flow velocity cannot be accurately known.
[0008] Therefore, it is necessary to propose an optimization control system and method for the cabin pumping and shore blowing process of a trailing suction dredger. Summary of the invention
[0009] Purpose of the invention: The present invention provides an optimization control system for the chamber pumping and shore blowing process of a trailing suction dredger, which is used to predict the critical flow velocity of mud during the chamber pumping process.
[0010] The technical scheme is as follows: The present invention provides an optimization control system for the process of chamber pumping and shore blowing of a bucket-suction dredger, and the system includes: a collection module, which is used to collect mud operation parameters and relevant parameters of the bucket-suction dredger; a data processing module, which is used to screen out mud flow rate, average particle size of sediment, chamber door opening, and mud pump speed from the mud operation parameters and relevant parameters of the bucket-suction dredger as input variables, and is used to select the critical flow rate vc of the chamber pumping and shore blowing process of the bucket-suction dredger as an output parameter; and the data processing module is used to filter, process outliers, and normalize the input variables for preprocessing, and divide the preprocessed data into a training sample set and a verification sample set; a model construction module, which is used to establish a neural network prediction model, initialize the neural network using a PSO algorithm, and determine the minimum fitness function; a training verification module, which is used to train the neural network prediction model using data in a training sample set, and is used to select data in a verification sample set to verify the trained neural network prediction model.
[0011] Furthermore, the neural network prediction model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is a is an integer in the interval [0,10]; the initialization of the neural network is: the particle dimension M is: The fitness function of the neural network is: Where N is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij The expected output of the sample is preset; the fitness function value of the particle is evaluated by iteratively updating the speed and position of the particle until the fitness function value reaches a preset iteration accuracy and determines that the fitness function value is minimum.
[0012] Furthermore, the critical velocity v c for: Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the sinking speed in still water; d m is the average particle size; and, C n =4V m / (vπD 2 ), where D is the inner diameter of the pumping pipe, v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; the volume of sediment entering the pipeline per unit time V m It is related to the opening of the pumping door, and the mud flow velocity v is related to the mud pump speed.
[0013] Furthermore, the data processing module is used to filter the input variables using wavelet threshold denoising; and the data processing module is used to perform outlier processing on the filtered input variables using the Leyte criterion, and the input variables are measured N times and outlier processing is performed to obtain x1, x2, x3, ..., x k , …, x n ,, take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula: And, the data processing module is used to normalize the data after outlier processing to obtain X as: Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
[0014] Furthermore, the training and verification module is used to select the model with the smallest root mean square error RMSE and average percentage error MAPE of the data in the training sample set as the trained model:
[0015]
[0016]
[0017] Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
[0018] The optimization control system for the process of chamber pumping and shore blowing of a trailing suction dredger is used to build a neural network prediction model, to collect mud operation parameters and relevant parameters of the trailing suction dredger and to screen out mud flow rate, average particle size of sediment, chamber pumping door opening, and mud pump speed as model input variables, and to select a critical flow rate as a model output parameter; to preprocess the input variables and divide them into a training sample set and a verification sample set; to initialize the neural network using a PSO algorithm and determine a minimum fitness function; and to train the model, select a model with the smallest root mean square error and average percentage error as a trained model, and use the verification set data to verify the trained model to obtain a critical flow rate prediction model for the process of chamber pumping and shore blowing of a trailing suction dredger, so as to achieve long-term accurate prediction; and to adopt a method of predicting the critical flow rate to achieve the highest efficiency, thereby solving the problem of long-term reliance on operator experience in construction.
[0019] The present invention also provides a method for optimizing control of a tank pumping and shore blowing process of a trailing suction dredger, comprising the following steps:
[0020] (1) The sensors were used to collect the mud operation parameters and relevant parameters of the trailing suction hopper dredger during the process of vacuuming and blowing the bank. (2) The mud flow rate, average particle size of sediment, vacuum door opening, and mud pump speed were selected from the mud operation parameters and relevant parameters of the trailing suction hopper dredger as input variables, and the critical flow rate v of the vacuuming and blowing the bank process of the trailing suction hopper dredger was selected. c as output parameters; (3) filtering, outlier processing, and normalization of input variables for preprocessing, and dividing the preprocessed data into a training sample set and a verification sample set; (4) establishing a neural network prediction model, using the PSO algorithm to initialize the neural network, and determining the minimum fitness function; (5) using the data in the training sample set to train the neural network prediction model, and selecting the data in the verification sample set to verify the trained neural network prediction model.
[0021] Furthermore, in step (4), the neural network includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is Where a is an integer in the interval [0,10]. In the step of initializing the neural network using the PSO algorithm, the particle dimension M is obtained as: The fitness function is determined as: Where N is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij The expected output of the sample is preset; the fitness function value of the particle is evaluated by iteratively updating the speed and position of the particle until the fitness function value reaches a preset iteration accuracy and determines that the fitness function value is minimum.
[0022] Furthermore, in step (2), the critical velocity v c for: Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the sinking speed in still water; d m is the average particle size of the mud; and, C n =4V m / (vπD 2 );
[0023] Where v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; the volume of sediment entering the pipeline per unit time V m It is related to the opening of the pumping door, and the mud flow velocity v is related to the mud pump speed.
[0024] Furthermore, in step (3), the following steps are also included: filtering the input variables using wavelet threshold denoising; performing outlier processing on the filtered input variables using the Leyte criterion, measuring the input variables N times and filtering them to obtain x1, x2, x3, ..., x k , …, x n ,, take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula: The data after outlier processing is normalized to obtain X: Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
[0025] Furthermore, when the neural network prediction model is trained in step (5), the model with the smallest root mean square error RMSE and average percentage error MAPE of the data in the training sample set is selected as the trained model:
[0026]
[0027]
[0028] Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
[0029] The optimization control method for the suction hopper dredger's cabin pumping and shore blowing process constructs a neural network prediction model, collects mud operation parameters and relevant parameters of the suction hopper dredger, and screens out mud flow rate, average sediment particle size, suction door opening, and mud pump speed as model input variables, and selects critical flow rate as model output parameter; the input variables are preprocessed and divided into training sample set and verification sample set; the PSO algorithm is used to initialize the neural network and determine the minimum fitness function; the model is trained using the training sample set data, and the trained model is verified using the verification set data to predict the critical flow rate in the suction hopper dredger's cabin pumping and shore blowing process, so as to achieve long-term accurate prediction; the method of predicting the critical flow rate is used to achieve the highest efficiency, which solves the problem of long-term reliance on operator experience in construction. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 The present invention is a flow chart of an optimization control method for a chamber pumping and blowing process of a trailing suction dredger. DETAILED DESCRIPTION
[0031] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0032] like Figure 1 As shown, the present invention discloses an optimization control system for the chamber pumping and shore blowing process of a trailing suction dredger, wherein the chamber pumping and shore blowing is a process in which mud and sand enter the chamber pumping pipe from the mud chamber and are discharged along with the seawater in the chamber pumping pipe. The system comprises: an acquisition module, a data processing module, a model building module and a training verification module.
[0033] The collection module is used to collect mud operation parameters and related parameters of the trailing suction hopper dredger; the mud operation parameters include the concentration, flow rate and particle size of the mud mixture in the process of chamber pumping and shore blowing of the trailing suction hopper dredger; the related parameters of the trailing suction hopper dredger include the opening of the chamber pumping door, the maximum opening area of the chamber pumping door, the length and diameter of the chamber pumping pipe, the resistance coefficient of the pipe, the discharge pressure of the mud pump, etc., a total of 302 related parameters; some of these parameters remain unchanged, while others change during the process of chamber pumping and shore blowing, but not all parameters will affect the critical flow rate during the process of chamber pumping and shore blowing, so the parameters need to be screened before use.
[0034] According to the mathematical model and process related to chamber pumping and bank blowing, the data processing module is used to select the mud flow rate, average sediment particle size, chamber door opening, and mud pump speed from the mud operation parameters and the relevant parameters of the trailing suction dredger as input variables to select the critical flow rate v of the chamber pumping and bank blowing process of the trailing suction dredger. c As an output parameter;
[0035] According to the empirical formula, for the same trailing suction dredger, the mud volume concentration C in the suction pipe is n and the average particle size of sediment d m Determines the magnitude of the critical flow velocity, the critical flow velocity v output by the model building module c for:
[0036]
[0037] Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the sinking speed in still water; d m is the average particle size; and, due to the volume concentration of the mud C n Directly affects the critical flow velocity v c The size of the mud volume concentration C in the pumping tank pipeline n The analytical calculation is:
[0038] C n =4V m / (vπD 2 );
[0039] Where D is the inner diameter of the pumping pipe, v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; when the inner diameter D of the pumping pipeline remains unchanged and the sediment flow rate does not change much, the volume of sediment entering the pumping pipeline determines the mud concentration in the pipeline. The volume of sediment entering the pipeline is mainly controlled by the pumping door. The opening of the pumping door directly determines the volume of sediment entering the pumping pipeline. The speed of the mud pump directly affects the mud flow rate; that is, the volume of sediment entering the pipeline per unit time V m There is a positive correlation between the mud flow velocity v and the mud pump speed.
[0040] According to engineering experience and mechanism analysis, the volume concentration and particle size of the mud in the process of chamber pumping and shore blowing can greatly affect the critical flow velocity. Therefore, the model construction module is used to select mud flow velocity, chamber pumping door opening, average sediment particle size, and mud pump speed as factors affecting the critical flow velocity as input variables of the model.
[0041] The data processing module is used to filter, process outliers, and normalize the input variables for preprocessing. Specifically, the actual data of four input variables, namely, mud flow rate, chamber door opening, average sediment particle size, and mud pump speed, are extracted from the collected historical data, and the selected data are preprocessed. The data preprocessing includes: filtering the input variables using wavelet threshold denoising; and the data preprocessing also includes: processing outliers on the filtered input variables using the Leyte criterion, measuring the input variables N times and processing outliers to obtain x1, x2, x3, ..., x k , …, x n ,, take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula:
[0042]
[0043] And, data preprocessing also includes: normalizing the data after outlier processing to obtain X as:
[0044]
[0045] Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
[0046] The data processing module is used to divide the pre-processed data into a training sample set and a verification sample set, wherein the first 80% of the pre-processed data is used as the training sample set data, and the rest is used as the verification sample set data.
[0047] The model building module is used to establish a neural network prediction model, use the PSO algorithm to initialize the neural network, determine the minimum fitness function, and realize the prediction of the critical flow velocity during the cabin pumping and shore blowing process of the trailing suction dredger. Since the BP neural network is very sensitive to network parameters such as initial weights and learning rates, it is easy to fall into the local optimal solution; the PSO algorithm is based on swarm intelligence optimization, dynamically adjusts the speed and position of particles to find the global optimal solution, and has a strong global search capability. Therefore, the PSO algorithm is used to initialize the network connection weights and thresholds, and then optimize the BP neural network structure.
[0048] The pre-processed data of mud flow rate, average particle size of sediment, opening of the suction door, and mud pump speed are used as input variables of the neural network prediction model. The critical flow velocity v of the suction dredger during the suction and shore blowing process is c As the output parameter of the neural network prediction model.
[0049] The neural network prediction model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is a is an integer in the interval [0,10];
[0050] The PSO algorithm is used to initialize the weights and thresholds of the neural network. The initialization of the neural network is as follows: the number of particle swarms, learning factors, and inertia weights are set to empirical values. The particle dimension M is determined by the number of nodes in the input layer, hidden layer, and output layer. The particle dimension M is:
[0051]
[0052] The fitness function value is an important parameter to characterize the quality of particles. The purpose of minimizing the objective function can be achieved by minimizing the fitness function value. The mean square error of the BP neural network is used as the fitness function. The fitness function of the neural network is:
[0053]
[0054] Where N is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij is the preset sample expected output. From the above formula, it can be seen that the smaller the fitness function value, the smaller the network error and the better the particle adaptability. By dynamically adjusting the speed and position of the particle, the optimal particle can be determined by minimizing the fitness function. The fitness function value of the particle is evaluated by iteratively updating the speed and position of the particle until the fitness function value reaches a preset iteration accuracy and the fitness function value is determined to be minimum, and the historical optimal position of the particle and the global optimal position of the group are obtained.
[0055] The training verification module is used to train the neural network prediction model using the data in the training sample set, and to select the data in the verification sample set to verify the trained neural network prediction model, so as to obtain the critical flow velocity prediction model in the process of cabin pumping and shore blowing of the trailing suction dredger.
[0056] The training verification module is also used to use the root mean square error RMSE and the mean percentage error MAPE of the data in the training sample set as training evaluation indicators, and select the model with the smallest root mean square error RMSE and the mean percentage error MAPE as the trained model:
[0057]
[0058]
[0059] Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
[0060] When training the model, the model is fine-tuned and updated according to the above-mentioned root mean square error RMSE and average percentage error MAPE to obtain a relatively accurate prediction model. The model with the smallest root mean square error RMSE and average percentage error MAPE is selected as the trained model, and the trained model is verified using the validation set data to obtain a critical flow velocity prediction model for the process of vacuuming and blowing the shore of a trailing suction dredger.
[0061] The optimization control system of the suction hopper dredger during the cabin pumping and shore blowing process is used to construct a neural network prediction model, to collect mud operation parameters and relevant parameters of the suction hopper dredger, and to screen out mud flow rate, average particle size of sediment, cabin door opening, and mud pump speed as model input variables, and to select critical flow rate as model output parameter; to pre-process the input variables and divide them into training sample set and verification sample set; to initialize the neural network using the PSO algorithm and determine the minimum fitness function; and to train the model, select the model with the smallest root mean square error and average percentage error as the trained model, and use the verification set data to verify the trained model to obtain the critical flow rate prediction model during the cabin pumping and shore blowing process of the suction hopper dredger, so as to achieve long-term accurate prediction; when pumping and shore blowing, the current critical flow rate can be accurately understood, so that it can be adjusted according to the critical flow rate, and there will be no mud blocking due to too low flow rate, and there will be no excessive energy consumption due to too fast flow rate. The method of predicting the critical flow rate is adopted to achieve the highest efficiency, which solves the drawback of relying on the experience of operators for a long time.
[0062] The present invention provides a method for optimizing control of a tank pumping and shore blowing process of a trailing suction dredger, comprising the following steps:
[0063] (1) Use sensors to collect mud operation parameters and relevant parameters of the trailing suction hopper dredger during the process of vacuuming and blowing the shore;
[0064] In this step (1), the mud operation parameters include the concentration, flow rate and particle size of the mud mixture during the process of chamber pumping and shore blowing of the trailing suction dredger; the relevant parameters of the trailing suction dredger include the opening of the chamber pumping door, the maximum opening area of the chamber pumping door, the length and diameter of the chamber pumping pipeline, the resistance coefficient of the pipeline, the discharge pressure of the mud pump, etc., a total of 302 relevant parameters; some of these parameters remain unchanged, while others change during the process of chamber pumping and shore blowing, but not all parameters will affect the critical flow rate during the process of chamber pumping and shore blowing, so the parameters need to be screened before use.
[0065] (2) A neural network prediction model was established. The mud flow rate, average particle size of sediment, opening of the suction chamber door, and mud pump speed were selected from the mud operation parameters and related parameters of the trailing suction dredger as the input variables of the neural network prediction model. The critical flow velocity v of the suction chamber dredger during the suction chamber blowing process was selected. c As the output parameter of the neural network prediction model;
[0066] In step (2), according to the empirical formula, for the same trailing suction dredger, the volume concentration of mud in the suction pipe and the average particle size of the mud determine the critical flow velocity, and the output critical flow velocity v is:
[0067]
[0068] Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the sinking speed in still water; d m is the average particle size of the mud; and, since the volume concentration of the mud directly affects the critical velocity, the volume concentration of the mud in the pumping pipe is analyzed as follows:
[0069] C n =4V m / (vπD 2 );
[0070] Where v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; when the inner diameter D of the pumping pipeline remains unchanged and the sediment flow rate does not change much, the volume of sediment entering the pumping pipeline determines the mud concentration in the pipeline. The volume of sediment entering the pipeline is mainly controlled by the pumping door. The opening of the pumping door directly determines the volume of sediment entering the pumping pipeline. The speed of the mud pump directly affects the mud flow rate; that is, the volume of sediment entering the pipeline per unit time V m There is a positive correlation between the mud flow velocity v and the mud pump speed.
[0071] (3) Preprocessing the input variables by filtering, processing outliers, and normalizing them, and dividing the preprocessed data into a training sample set and a validation sample set;
[0072] In step (3), four construction parameters, namely, mud flow rate, chamber door opening, average sediment particle size, and mud pump speed, are extracted from the collected historical data as actual data of input variables, and the selected data are preprocessed. The specific steps of data preprocessing are as follows:
[0073] (31) Use wavelet threshold denoising to filter the input variables and remove noise and interference signals in the data;
[0074] (32) The Leyte criterion, i.e., the triple error method, is used to process the outliers of the filtered input variables. The input variables are measured N times and filtered to obtain x1, x2, x3, ..., x k , …, x n ,, take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula:
[0075]
[0076] (33) After outlier processing, the data is normalized to obtain X:
[0077] Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
[0078] (4) Using the neural network prediction model to predict the critical flow velocity during the process of vacuuming and blowing the shore by a trailing suction dredger;
[0079] In this step (4), since the BP neural network is very sensitive to network parameters such as initial weights and learning rate, it is easy to fall into the local optimal solution; the PSO algorithm is based on swarm intelligence optimization, dynamically adjusts the speed and position of particles to find the global optimal solution, and has a strong global search capability. Therefore, the PSO algorithm is used to initialize the network connection weights and thresholds, and then optimize the BP neural network structure.
[0080] Step (4) specifically includes the following steps:
[0081] (41) Construct a neural network prediction model. The neural network includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is Where a is an integer in the interval [0,10];
[0082] (42) The PSO algorithm is used to initialize the neural network and the particle dimension M is obtained as:
[0083]
[0084] (43) The fitness function is determined as:
[0085]
[0086] Where N is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij The expected output of the preset sample;
[0087] (44) Minimize the fitness function to determine the minimum objective function.
[0088] The particle's fitness function value is evaluated by iteratively updating the particle's speed and position until the fitness function value reaches a preset iteration accuracy and the fitness function value is determined to be minimum, thereby obtaining the particle's historical optimal position and the group's global optimal position.
[0089] (5) The neural network prediction model is trained using the data in the training sample set, and the root mean square error and average percentage error are used as training evaluation indicators. The data in the verification sample set are selected to verify the trained neural network prediction model.
[0090] In step (5), the first 80% of the preprocessed data is used as the training sample set, and the remaining data is used as the validation sample set.
[0091] When training the neural network prediction model in step (5), the model with the smallest root mean square error RMSE and average percentage error MAPE of the data in the training sample set is selected as the trained model:
[0092]
[0093]
[0094] Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
[0095] The optimization control method of the suction hopper dredger's cabin pumping and shore blowing process builds a neural network prediction model, collects mud operation parameters and relevant parameters of the suction hopper dredger, and selects mud flow rate, average particle size of sediment, cabin door opening, and mud pump speed as model input variables, and selects critical flow rate as model output parameter; the input variables are preprocessed and divided into training sample set and verification sample set; the PSO algorithm is used to initialize the neural network and determine the minimum fitness function; the model is trained using the training sample set data, and the trained model is verified using the verification set data to predict the critical flow rate in the suction hopper dredger's cabin pumping and shore blowing process, so as to achieve long-term accurate prediction; when pumping and blowing the cabin, the current critical flow rate can be accurately understood, so that it can be adjusted according to the critical flow rate, and there will be no mud blocking due to too low flow rate, and there will be no excessive energy consumption due to too fast flow rate. The method of predicting the critical flow rate is used to achieve the highest efficiency, which solves the problem of relying on the operator's experience for a long time.
[0096] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. An optimization control system for the process of vacuuming and blowing the bank of a trailing suction dredger, characterized in that: The system includes: The acquisition module is used to collect mud operation parameters and related parameters of the trailing suction dredger; The data processing module is used to select the mud flow rate, average sediment particle size, suction door opening, and mud pump speed from the mud operation parameters and related parameters of the trailing suction dredger as input variables to select the critical flow rate v of the suction dredger during the suction dredger blowing process. c as output parameters; and the data processing module is used to filter, process outliers, and normalize the input variables for preprocessing, and divide the preprocessed data into a training sample set and a verification sample set; The model building module is used to build a neural network prediction model, use the PSO algorithm to initialize the neural network, and determine the minimum fitness function; The training and verification module is used to train the neural network prediction model using the data in the training sample set, and to verify the trained neural network prediction model by selecting the data in the verification sample set.
2. The optimization control system for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 1 is characterized in that: The neural network prediction model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is a is an integer in the interval [0,10]; The initialization of the neural network is: particle dimension M: The fitness function of the neural network is: Where N is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij The expected output of the preset sample; The fitness function value of the particle is evaluated by iteratively updating the speed and position of the particle until the fitness function value reaches a preset iteration accuracy and determines that the fitness function value is minimum.
3. The optimization control system for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 1 is characterized in that: Critical flow velocity v c for: Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the still water sinking speed; d m is the average particle size; and C n =4V m / (vπD 2 ); Where D is the inner diameter of the pumping pipe, v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; the volume of sediment entering the pipeline per unit time V m It is related to the opening of the pumping door, and the mud flow velocity v is related to the mud pump speed.
4. The optimization control system for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 1 is characterized in that: The data processing module is used to filter the input variables by wavelet threshold denoising; and the data processing module is used to perform outlier processing on the filtered input variables by using the Leyte criterion, and the input variables are measured N times and outlier processing is performed to obtain x1, x2, x3, ..., x k , …, x n , take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula: And, the data processing module is used to normalize the data after outlier processing to obtain X as: Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
5. The optimization control system for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 4 is characterized in that: The training and verification module is used to select the model with the smallest root mean square error RMSE and average percentage error MAPE of the data in the training sample set as the trained model: Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
6. A method for optimizing control of the process of vacuuming and blowing the bank of a trailing suction dredger, characterized in that: The following steps are involved: (1) Use sensors to collect mud operation parameters and relevant parameters of the trailing suction hopper dredger during the process of vacuuming and blowing the shore; (2) From the mud operation parameters and the relevant parameters of the trailing suction dredger, the mud flow rate, the average particle size of the sediment, the opening of the suction door, and the mud pump speed are selected as input variables, and the critical flow rate v of the suction dredger during the suction process is selected. c As an output parameter; (3) Preprocessing the input variables by filtering, processing outliers, and normalizing them, and dividing the preprocessed data into a training sample set and a validation sample set; (4) Establish a neural network prediction model, use the PSO algorithm to initialize the neural network, and determine the minimum fitness function; (5) The neural network prediction model is trained using the data in the training sample set, and the trained neural network prediction model is verified using the data in the verification sample set.
7. The optimization control method for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 6 is characterized in that: In step (4), the neural network includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is n, the number of nodes in the output layer is l, and the number of neuron nodes in the hidden layer is Where a is an integer in the interval [0,10]; In the step of initializing the neural network using the PSO algorithm, the particle dimension M is obtained as: The fitness function is determined as: Where M is the total number of input samples; M is the particle dimension; Y ij is the actual output of the i-th sample of the model; T ij The expected output of the preset sample; The particle's speed and position are updated iteratively to evaluate the particle's fitness function value until the fitness function value reaches a preset iteration accuracy and determines that the fitness function value is minimum.
8. The optimization control method for the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 6 is characterized in that: In step (2), the critical velocity v c for: Among them, C n is the volume concentration of mud; g is the acceleration of gravity; D is the inner diameter of the pumping pipe; v t is the still water sinking speed; d m is the average particle size of the mud; and C n =4V m / (vπD 2 ); Where v is the mud flow rate, V m is the volume of sediment entering the pipeline per unit time; the volume of sediment entering the pipeline per unit time V m It is related to the opening of the pumping door, and the mud flow velocity v is related to the mud pump speed.
9. The method for optimizing control of the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 6, characterized in that: In step (3), the following steps are also included: Wavelet threshold denoising is used to filter the input variables; the Leyte criterion is used to process the outliers of the filtered input variables. The input variables are measured N times and filtered to obtain x1, x2, x3, ..., x k , …, x n , take x1,x2,x3,…,x k , …, x n The average of The data x k The condition for judging as an outlier is: data x k The corresponding residual R k Satisfies the following formula: The data after outlier processing is normalized to obtain X: Among them, X is the data after normalization. is the data after outlier processing, X min is the minimum value of the data after outlier processing, X max It is the maximum value in the data after outlier processing.
10. The method for optimizing control of the process of vacuuming and shore blowing of a trailing suction hopper dredger according to claim 6, characterized in that: When training the neural network prediction model in step (5), the root mean square error RMSE and average percentage error MAPE of the data in the training sample set are used as training evaluation indicators: Among them, y i represents the i-th true value in the training sample set, represents the i-th predicted value, and n represents the number of data in the training sample set.
