Afe direct-current bus power pooling coordination control method and system for drag-suction operation

Abstract

This invention relates to the field of ship power control technology, specifically disclosing a method and system for coordinated power pooling control of the AFE DC bus in trailing suction hopper (AFE) operations. The method includes collecting multi-source heterogeneous data during the AFE operation, including process parameters, power parameters, and environmental parameters, and performing spatiotemporal alignment processing on the data. The spatiotemporally aligned multi-source heterogeneous data is then input into a preset process stage identification model to determine the current operation process stage, which includes a preparation stage, a dredging stage, a sludge removal stage, and a transfer stage. This invention, through multi-source heterogeneous data acquisition and process stage identification, combined with a four-dimensional power demand prediction model, achieves multi-timescale hierarchical control of the AFE DC bus. Specifically, it constructs a multi-timescale hierarchical control mechanism for the AFE DC bus based on the collaboration of power-type energy storage units and energy-type energy storage units.

Description

Technical Field

[0001] This invention relates to the field of ship power control technology, and in particular to an AFE DC bus power pooling coordinated control method and system for trailing suction hopper operations. Background Technology

[0002] As a core technology in the field of ship dredging, trailing suction hopper dredging requires its power control system to ensure efficient coordination across multiple stages, including dredging, sludge removal, and transfer. Traditional control methods often employ static power allocation strategies, which cannot adapt to dynamic changes in operating conditions in real time. This results in frequent fluctuations in DC bus voltage, low system energy efficiency, and a lack of predictive control over process risks, thus limiting operational efficiency and equipment reliability.

[0003] Specifically, existing technologies struggle to achieve spatiotemporal alignment and deep fusion of multi-source heterogeneous data, resulting in insufficient identification accuracy during process stages. Furthermore, during process stage switching, energy distribution imbalances between the main power pool and redundant power pools can easily lead to bus voltage collapse or equipment overload, highlighting the urgent need for multi-timescale power coordination and adaptive risk warning. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose an AFE DC bus power pooling coordinated control method and system for trailing suction hopper (ATH) operations, in order to improve the stability of existing ship power systems.

[0005] To achieve the above objectives, a first aspect of the present invention proposes an AFE DC bus power pooling coordinated control method for trailing suction hopper operations, applied to a ship's power control system, comprising the following steps:

[0006] Collect multi-source heterogeneous data during the sweeping suction hopper operation, including process parameters, power parameters, and environmental parameters, and perform spatiotemporal alignment processing on the multi-source heterogeneous data;

[0007] Multi-source heterogeneous data after spatiotemporal alignment is input into a preset process stage identification model to determine the current operation process stage, which includes the preparation stage, dredging stage, dredging stage, and transfer stage.

[0008] Based on the current operational process stage and the multi-source heterogeneous data, a four-dimensional power demand prediction model is used to generate power demand curves and process risk warnings for future periods.

[0009] Based on the power demand curve and the process risk warning, multi-time-scale hierarchical control is implemented on the AFE DC bus.

[0010] The aforementioned multi-time-scale hierarchical control of the AFE DC bus includes:

[0011] In response to millisecond-level high-frequency power fluctuations, the power storage unit is controlled to perform power compensation in order to smooth out power pulsations caused by process interference.

[0012] In response to the second-level process stage switching requirements, it coordinates the energy allocation between the main power pool and the redundant power pool, and uses energy-type energy storage units to absorb or release energy, thereby achieving mid-frequency power adaptation and energy shifting.

[0013] In response to minute-level steady-state operation requirements, a multi-objective model predictive control strategy is used to optimize low-frequency power allocation. The multi-objective model predictive control strategy aims to stabilize bus voltage, optimize system energy efficiency, and ensure compliance with operation schedules. The optimization process also considers the state of charge constraints of the energy storage unit.

[0014] To achieve the above objectives, a second aspect of the present invention provides an AFE DC bus power pooling coordinated control system for trailing suction duct operations, comprising:

[0015] The data processing module is used to collect multi-source heterogeneous data during the shovel suction operation, and to perform hierarchical filtering, spatiotemporal alignment and tagging storage on the multi-source heterogeneous data.

[0016] The stage identification module is used to output the current operation stage and confidence level based on the multi-source heterogeneous data using the built-in process stage identification model.

[0017] The four-dimensional prediction module is used to output power demand curves and process risk warnings using the built-in four-dimensional power demand prediction model.

[0018] A multi-timescale control module for executing hierarchical control strategies includes: a power-type energy storage unit control unit for performing millisecond-level control; a power pool coordination unit for performing second-level control, wherein the power pool coordination unit works in conjunction with the energy-type energy storage unit; and an MPC optimization unit for performing minute-level control, wherein the MPC optimization unit uses the state of charge of the energy-type energy storage unit as an optimization constraint.

[0019] The AFE collaborative control module is connected to the multi-timescale control module and is used to adjust the working mode of the AFE hybrid topology or activate hardware redundancy according to the control instructions.

[0020] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described AFE DC bus power pooling coordinated control method for scraper suction operations.

[0021] The power pooling coordinated control method and system for AFE DC bus in the present invention, which is oriented towards trailing suction hopper operation, realizes multi-time-scale hierarchical regulation of AFE DC bus through multi-source heterogeneous data acquisition and process stage identification, combined with a four-dimensional power demand prediction model; specifically, it constructs a multi-time-scale hierarchical regulation mechanism for AFE DC bus based on the collaboration of power-type energy storage unit and energy-type energy storage unit.

[0022] Among them, the power-type energy storage unit is dedicated to the rapid smoothing of millisecond-level high-frequency power pulsations, ensuring the instantaneous stability of the bus voltage; the energy-type energy storage unit is responsible for energy scheduling and shifting at the second to minute level, optimizing power flow and steady-state energy efficiency during the switching of different process stages; the two complement each other and work together to achieve full-frequency domain power support from instantaneous to steady state, thereby significantly improving the system's adaptability to complex operating conditions, operating efficiency, and overall safety and reliability. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the AFE DC bus power pooling coordinated control method for scraper suction operation provided by the present invention.

[0024] Figure 2 This is a schematic diagram comparing the rate of change of process parameters with the stage identification results in the AFE DC bus power pooling coordinated control method for scraper suction operation provided by this invention.

[0025] Figure 3 This is a simulation diagram of DC bus voltage stability under multi-timescale regulation in the AFE DC bus power pooling coordinated control method for scraper suction operation provided by the present invention.

[0026] Figure 4 This is a schematic diagram comparing the predicted and actual values ​​of the power prediction model in the AFE DC bus power pooling coordinated control method for scraper suction operation provided by this invention.

[0027] Figure 5 This is a simulation diagram of the system dynamic performance under different AFE operating modes in the AFE DC bus power pooling coordinated control method for scraper suction operation provided by the present invention;

[0028] Figure 6 This is a schematic diagram of the main / redundant power pool output power curves during the stage switching process in the AFE DC bus power pooling coordinated control method for scraper suction operation provided by this invention.

[0029] Figure 7 This is a schematic diagram of the power allocation and operation progress change curves before and after multi-objective optimization in the AFE DC bus power pooling coordinated control method for trailing suction duct operations provided by this invention.

[0030] Figure 8 This is a schematic diagram illustrating the implementation of the AFE DC bus power pooling coordinated control system for scraper suction operations provided by the present invention.

[0031] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0032] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0033] The following describes, with reference to the accompanying drawings, an AFE DC bus power pooling coordinated control method, system, and electronic equipment for trailing suction duct operations according to embodiments of the present invention.

[0034] Example 1:

[0035] Figure 1 This is a flowchart illustrating an embodiment of the AFE DC bus power pooling coordinated control method for scraper suction operation according to an embodiment of the present invention, specifically including the following:

[0036] S1. This method first collects multi-source heterogeneous data during the drag suction hopper operation. This multi-source heterogeneous data includes process parameters, power parameters, and environmental parameters, among which:

[0037] The specific process parameters include the depth of the rake head lowering, the mud tank level, the ship's forward speed, the mud concentration, the ship's roll angle, the ship's transfer speed, the rake head lifting height, and the mud discharge speed.

[0038] Power parameters include high-frequency power data, DC bus voltage and current data;

[0039] Environmental parameters include wave data.

[0040] The data acquisition process is achieved through a network of sensors installed in various parts of the ship, such as depth sensors, level gauges, speedometers, concentration sensors, attitude sensors, and power monitoring units. These sensors operate at different sampling frequencies; for example, process parameters may be sampled at the second level, power parameters at the millisecond level, and environmental parameters at the minute level, resulting in heterogeneity of data in both time and space.

[0041] S2. The acquired multi-source heterogeneous data needs to undergo spatiotemporal alignment processing. Spatiotemporal alignment processing uses the control system's clock as a reference and aligns process parameters, power parameters, and environmental parameters with different sampling frequencies to a unified time slice through timestamp correction.

[0042] Specifically, timestamp correction employs interpolation or resampling algorithms. For example, linear interpolation is used to upscale low-frequency parameters to a high-frequency time series, while moving averages are used to downscale high-frequency parameters to a uniform time resolution. The uniform time slices can be set at millisecond or second intervals, depending on the real-time requirements of the control system. Spatiotemporal alignment ensures that all parameters have consistent values ​​at the same time point, thus providing standardized data for subsequent model inputs.

[0043] S3. The spatiotemporally aligned multi-source heterogeneous data is input into a pre-defined process stage identification model to determine the current operational process stage. The operational process stages include preparation, dredging, sludge removal, and transfer. The process stage identification model is built based on a random forest algorithm, and its input features are the current values ​​and rates of change of process parameters. The rate of change is obtained by calculating the derivative or difference of the parameters within a short time window; for example, the rate of change in the depth of the dredging head can reflect the dynamics of the dredging operation.

[0044] For example, the input feature dimensions of the process stage identification model are further explained here. For instance, the input features include the current value of the rake head lowering depth and its one-second rate of change, the current value of the mud tank level and its five-second rate of change, etc., totaling ten features. The random forest model contains one hundred trees, each with a maximum depth of ten. The confidence threshold is set to 0.7; a value below this triggers a review mechanism.

[0045] For example, the process stage identification model outputs the current operation process stage and its corresponding confidence level. The confidence level is a value between 0 and 1, representing the reliability of the model's stage identification results. For instance, a confidence level higher than 0.9 indicates that the model is highly confident that the current stage is dredging; a confidence level lower than 0.6 may indicate that the working condition is complex and requires further verification.

[0046] Based on the identified operational stages, the system extracts specific power fluctuation and process risk characteristics for each stage. Power fluctuation characteristics include the mid-frequency fluctuation amplitude of the trailing suction pump power during the dredging stage and the low-frequency trend of the discharge pump power during the sludge removal stage. The mid-frequency fluctuation amplitude can be obtained by calculating the variance of the power signal within a specific frequency band, while the low-frequency trend can be extracted using moving averages or filtering. Process risk characteristics may involve excessive mud concentration or excessive ship roll angle; these characteristics are used for subsequent risk warnings.

[0047] Specifically, during the dredging stage, the mid-frequency fluctuation amplitude of the trailing suction pump power is extracted from the 0.5Hz to 5Hz components using a bandpass filter, and then its root mean square value is calculated. During the sludge discharge stage, the low-frequency trend of the discharge pump power is extracted from the components below 0.1Hz using a low-pass filter, and a linear trend is fitted. Process risk characteristics include mud concentration gradient, ship roll rate, etc. These characteristics are compared with preset thresholds to generate a risk level.

[0048] like Figure 2 This demonstrates the correlation between the rate of change in the rake head's descent depth and the process stages identified by the system throughout the entire rake suction operation. Specifically:

[0049] The blue curve represents the change in the depth of the dredging head per unit time. Its fluctuations reflect the dynamic nature of the dredging head's movements during actual operations. Between 0 and 20 seconds, the rate of change remains close to zero, indicating that the vessel is not performing dredging actions during the preparation phase. Between 20 and 40 seconds, the rate of change rises rapidly and remains at a high value, indicating that the dredging head is being lowered, which the system identifies as the dredging stage. Between 40 and 60 seconds, the rate of change drops to zero, indicating that the dredging head has stabilized at a certain depth and has not continued its movements, corresponding to the mud removal stage. After 60 seconds until termination, the rate of change fluctuates slightly but remains generally low, which the system identifies as the transfer stage.

[0050] The red stepped curve represents the process stage label output by the random forest model. It changes abruptly at different time points, indicating the system's ability to identify process stage switching. Figure 2 As can be seen, the mutations in each stage closely follow the trend of parameter change rate, indicating that the process stage identification model can accurately identify different operating states based on the dynamic behavior of parameters.

[0051] S4. After determining the current operational process stage, the system uses a four-dimensional power demand forecasting model to generate power demand curves for future periods and process risk warnings based on the current operational process stage and multi-source heterogeneous data. The four-dimensional power demand forecasting model can be trained using machine learning methods and is an advanced forecasting tool that integrates time series analysis, load characteristics, and environmental factors. Its inputs include aligned multi-source heterogeneous data and stage identification results, and its output is a power demand curve for the next few seconds to minutes, expressed as a time function, accompanied by process risk warnings, such as voltage drop risk or equipment overload risk.

[0052] For example, the basic implementation of a four-dimensional power demand forecasting model can be based on time series forecasting techniques, such as an autoregressive integral moving average model. The model inputs are historical power data, process parameters, and environmental parameters, and the output is the power value at a future time point. Process risk warnings are implemented through logistic regression or a rule engine, for example, triggering an alert when the predicted power exceeds the equipment's rated value. The forecast results are visualized as curves and used in the control module.

[0053] For example, the power demand curve is generated considering the dynamic changes in each stage of the process. For instance, during the dredging stage, power demand may fluctuate frequently; during the sludge removal stage, power demand may show a steady upward trend. Process risk warnings are calculated based on historical data and real-time parameters; for example, a blockage risk warning is triggered when the mud concentration exceeds a set threshold.

[0054] The power demand curve is stored as an array, with each element corresponding to the power value at a future time point, with a time interval of 1 second. Process risk warnings include risk type, level, and recommended measures; for example, a high-risk warning suggests reducing the operating speed. The curves and warnings are sent to the control module via a communication protocol.

[0055] S5. Based on the power demand curve and process risk warning, the system implements multi-time-scale hierarchical control of the AFE DC bus. Multi-time-scale hierarchical control includes three levels:

[0056] 1. In response to millisecond-level high-frequency power fluctuations, control the power storage unit to perform power compensation in order to smooth out power pulsations caused by process interference;

[0057] 2. Responding to the process stage switching requirements at the second level, it coordinates the energy allocation between the main power pool and the redundant power pool, and uses energy-type energy storage units to absorb or release energy, thereby realizing medium-frequency power adaptation and energy shifting.

[0058] 3. To meet minute-level steady-state operating requirements, optimize low-frequency power allocation based on a multi-objective model predictive control strategy.

[0059] Specifically, millisecond-level control is executed every 0.1 milliseconds to monitor voltage fluctuations; second-level control is executed every second to switch processing stages; and minute-level control is executed every sixty seconds to optimize power distribution. Control commands are sent to the execution unit via fieldbus.

[0060] Ultimately, the multi-objective model predictive control strategy aims to stabilize bus voltage, optimize system energy efficiency, and ensure compliance with work schedules, while also considering the state of charge constraints of energy storage units during the optimization process.

[0061] For example, in millisecond-level regulation, power storage units such as supercapacitors or flywheel energy storage systems are used for rapid response. When high-frequency power fluctuations are detected, the power storage units inject or absorb power to maintain DC bus voltage stability.

[0062] In second-level control, the energy distribution between the main power pool and the redundant power pool is coordinated by the controller. For example, when switching from the dredging stage to the sludge discharge stage, the redundant power pool releases power in advance, the main power pool adjusts slowly, and energy storage units such as battery systems help absorb excess energy.

[0063] In minute-level control, the multi-objective model predictive control strategy solves the optimization problem. Its objective function is the weighted sum of the bus voltage deviation term, the system energy efficiency loss term, and the operation progress deviation term. The constraints include the state of charge range of the energy storage unit to ensure that the system operates efficiently under steady state.

[0064] like Figure 3 The dynamic response of the DC bus voltage under different control strategies when encountering disturbances during the snorkeling operation was demonstrated, which was used to verify the technical advantages of the multi-timescale control mechanism proposed in step S5.

[0065] The red dashed line represents the no-control strategy, which reflects that the voltage fluctuates drastically after being disturbed at 20 seconds, with a fluctuation range of more than 10 volts. This indicates that the system cannot suppress the sudden power and has serious instability.

[0066] The blue dotted line represents the use of a single time scale for regulation, with compensation provided only by energy storage units at the second level. The fluctuation amplitude is reduced but still reaches 6 volts, and the recovery time is relatively long, making it impossible to quickly smooth out interference.

[0067] The green solid line indicates the multi-timescale hierarchical control mechanism adopted in this embodiment, including millisecond-level power storage unit response, second-level energy coordination control, and minute-level optimal power allocation strategy. After a 20-second disturbance, only a slight voltage deviation of about 2 volts is generated, and it quickly recovers to a stable value, which is significantly better than other schemes.

[0068] The spatiotemporal alignment process described above is further described in detail below: Using the control system's clock as a reference, timestamp correction employs GPS time or an internal high-precision clock source. Parameters with different sampling frequencies are aligned through buffer and queue mechanisms; for example, high-frequency power data is stored in a circular buffer, while low-frequency process data is synchronized via event triggering. A uniform time slice is set to a ten-millisecond interval to ensure data real-time performance. The aligned data is timestamped and stored in a database for subsequent module access.

[0069] The training and usage process of the aforementioned process stage identification model is described in further detail below: Model training uses historical rake-suction operation data, including multiple sets of process parameters and their corresponding stage labels. The random forest algorithm generates multiple decision trees through Bootstrap aggregation. Each tree votes on the input features, and the final stage is determined by the majority vote. The confidence score is calculated based on the voting ratio; for example, a high consensus rate results in a high confidence score. During runtime, the model performs identification once per second, outputting the current stage and confidence score, and updating the feature library in real time.

[0070] It should also be noted that in this embodiment, the power-type energy storage unit is connected to the DC bus via a bidirectional converter, responding to millisecond-level commands; the energy-type energy storage unit monitors the state of charge through a battery management system and participates in energy buffering in second-level and minute-level regulation; the main power pool and redundant power pool are controlled by an AFE converter group to achieve power distribution. The multi-objective model predictive control strategy runs on an embedded processor, solving the optimization problem once per minute.

[0071] Optionally, extended applications of this method include: for example, adapting it to other ship operations such as crane or propulsion control, requiring only adjustments to process parameters and model training data. Economic or environmental indicators can also be optimized by modifying the multi-objective function.

[0072] Example 2:

[0073] This embodiment describes in detail an AFE DC bus power pooling coordinated control method for trailing suction duct operations. Based on Embodiment 1, this method focuses on enhancing the four-dimensional power demand prediction model, AFE hardware coordinated control, and oscillation suppression function.

[0074] Specifically, this embodiment achieves more accurate power prediction, hardware adaptation, and system stability assurance by constructing a fused input vector, a prediction model based on a long short-term memory network and attention mechanism, prediction mode selection, AFE hybrid topology working mode adjustment, and emergency lockout control mode.

[0075] In this embodiment, the specific architecture of the four-dimensional power demand prediction model can be further extended. The Long Short-Term Memory (LSTM) network includes multiple hidden layers, and the number of neurons in each layer can be adjusted according to the input dimension, for example, one hundred neurons. The attention mechanism employs additive attention or dot-product attention, calculating attention weights and then summing them weighted over the hidden states. Prediction confidence is generated through the softmax function of the output layer or a separate uncertainty module. During model training, the Adam optimizer and early stopping strategy are used to prevent overfitting. During online prediction, the input data is preprocessed in real time, including missing value imputation and outlier detection.

[0076] The core of the four-dimensional power demand forecasting model lies in constructing a fused input vector that incorporates process features, load features, environmental features, and risk features. Process features are derived from process parameters in multi-source heterogeneous data. These parameters, after spatiotemporal alignment, have their current values, historical trends, and rates of change extracted as features. Load features include statistics from high-frequency power data, DC bus voltage, and current data, such as mean, variance, and spectral components. Environmental features encompass the amplitude and frequency characteristics of wave data. Risk features are based on the confidence level of the model output and historical risk events, such as warning indicators triggered by excessive mud concentration or excessive ship roll angle, based on process stage identification. The fused input vector, through feature concatenation and normalization, forms a high-dimensional vector, which serves as the input to the forecasting model.

[0077] For example, the fused input vector is fed into a prediction model built on a Long Short-Term Memory (LSTM) network and an attention mechanism to calculate the power demand curve and prediction confidence for future time periods. The LSM network is a recurrent neural network capable of capturing long-term dependencies in time series data. Its unit structure includes an input gate, a forget gate, and an output gate, used to process the time-series data with the fused input vector. The attention mechanism gives the model the ability to focus on key features, dynamically adjusting the importance of inputs at different time steps by calculating feature weights.

[0078] The forecasting model outputs a power demand curve, which represents the power values ​​for the next few seconds to minutes in time series form. It also outputs a prediction confidence score, a scalar between 0 and 1 that reflects the model's reliability in predicting the results. The prediction confidence score is quantified based on the model's output probability or uncertainty, for example, by estimating the prediction variance using the Monte Carlo dropout method.

[0079] Optionally, the prediction mode is selected based on the prediction confidence level. The system presets two thresholds, namely a first threshold and a second threshold, wherein the first threshold is higher than the second threshold.

[0080] 1. When the prediction confidence level is higher than the first threshold, a long-step prediction is executed, covering a longer future time range, such as five to ten minutes. This is suitable for stable operating conditions, such as smooth operations during the sludge removal phase. When the prediction confidence level is between the first and second thresholds, a medium-step prediction is executed, covering a medium future time range, such as one to five minutes. This is suitable for operating conditions with slight fluctuations, such as routine operations during the dredging phase.

[0081] 2. When the prediction confidence level is lower than the second threshold, short-step prediction is executed, triggering the power storage unit to enter overclocking response mode. Short-step prediction covers a relatively short time range in the future, such as ten seconds to one minute, and is suitable for scenarios with drastic changes in operating conditions, such as rapid switching during the transfer phase. Overclocking response mode refers to the power storage unit operating at a frequency higher than the rated frequency, such as increasing the switching frequency to twice the normal value, to provide faster power compensation.

[0082] like Figure 4 The comparison of the prediction performance of the four-dimensional power demand prediction model within a 60-second time window is shown to verify the prediction accuracy of the model under complex working conditions.

[0083] Figure 4 The solid black line represents the actual load power value collected by the system in real time. It can be seen that the power shows a continuous upward trend in the first 20 seconds, simulating the typical state of the sludge pump gradually increasing its load; in the middle 20 to 40 seconds, the power rapidly climbs to a high level, simulating the intensified load during the sludge discharge stage; after 40 seconds, it slightly decreases, reflecting the power release process as the operation nears its end. Simultaneously, the curve contains periodic disturbances with an amplitude of 5 kilowatts, simulating mid-frequency fluctuations such as sludge concentration fluctuations or motor response vibrations.

[0084] Figure 4 The purple dashed line in the middle represents the predicted output of the four-dimensional power demand forecasting model. It is based on process characteristics, load characteristics, environmental characteristics, and risk characteristics to form a fused input vector, which is then processed through a long short-term memory network and an attention mechanism.

[0085] It can be observed that Figure 4 The overall trend of the predicted curve is highly consistent with the actual curve, with only a slight deviation of no more than 3 kilowatts at individual inflection points, indicating that the model can accurately capture the power trend under future operating conditions.

[0086] For example, the training process of a four-dimensional power demand prediction model further includes:

[0087] Training data is derived from historical data collection and extraction (DME) records, including multiple sets of fused input vectors and their corresponding actual power values. The Long Short-Term Memory (LSTM) network is optimized using backpropagation and gradient descent algorithms, with the loss function employing mean squared error or mean absolute error. The attention mechanism involves query, key, and value vectors derived from the hidden states, with weights generated via a softmax function. Prediction confidence is calculated based on the statistical distribution of prediction errors, for example, by using Gaussian process regression or a Bayesian neural network to output the uncertainty interval. After model deployment, an online update mechanism fine-tunes parameters based on real-time data to adapt to environmental changes.

[0088] Optionally, this embodiment also includes an AFE hardware cooperative control step. AFE refers to an active front-end converter, whose hybrid topology includes uncontrolled rectifiers and fully controlled switching devices, such as IGBTs or MOSFETs. The operating mode of the AFE hybrid topology is adjusted according to the current process stage. For example:

[0089] During the preparation phase, the control is set to low-power mode, reducing the output ratio of the uncontrolled rectifier and lowering the modulation frequency. The output ratio of the uncontrolled rectifier refers to the proportion of the output power of the uncontrolled rectifier in the total power, for example, reducing it to below 30%; the modulation frequency refers to the switching frequency of the switching devices, for example, reducing it to below 5000Hz to reduce switching losses and electromagnetic interference.

[0090] During the dredging phase, the control mode is set to high dynamic response, reducing the proportion of uncontrolled rectifier output and increasing the modulation frequency. For example, the proportion of uncontrolled rectifier output is reduced to below 20%, and the modulation frequency is increased to above 10,000 Hz to enhance the response capability to high-frequency power fluctuations.

[0091] During the sludge removal stage, the system is controlled in a high-efficiency mode, increasing the output ratio of the uncontrolled rectifier and setting a moderate modulation frequency. For example, the output ratio of the uncontrolled rectifier is increased to over 50%, and the modulation frequency is set to around 8000Hz to optimize energy efficiency.

[0092] During the transfer phase, the control mode is a power compensation mode, which focuses on the coordination of power-type energy storage units, such as adjusting the output of the uncontrolled rectifier to 40 percent and dynamically adjusting the modulation frequency to support power compensation.

[0093] For example, AFE hardware-coordinated control is implemented using a digital signal processor or FPGA. Switching between operating modes is based on the output of a process stage identification model, taking into account power demand forecasts. For instance, when the dredging stage is predicted to begin, the system switches to a high dynamic response mode in advance to ensure rapid response. Adjustments to the modulation frequency and the proportion of uncontrolled rectifier output are achieved through PWM signals and relay control, with specific parameters stored in a lookup table and dynamically retrieved based on real-time load. Low-power modes are suitable for ship standby, high dynamic response modes for fluctuating loads, high-efficiency modes for steady-state loads, and power-compensated modes for transient states. This adaptive adjustment significantly improves system energy efficiency and hardware lifespan.

[0094] Here's an explanation of the parameter adjustments for the AFE hardware-coordinated control: In low-power mode, the reduced output of the uncontrolled rectifier is achieved through a bypass switch, and the modulation frequency is reduced by adjusting the PWM counter value. In high dynamic response mode, increasing the modulation frequency may increase switching losses, thus requiring coordinated thermal management. In high-efficiency mode, the increased output of the uncontrolled rectifier is achieved by controlling relay switching, and a moderate modulation frequency balances efficiency and dynamics. In power compensation mode, the AFE converter collaborates with the power storage unit, achieving power injection or absorption through current loop control.

[0095] like Figure 5 The dynamic recovery process of DC bus voltage under four AFE operating modes after encountering equal-amplitude load disturbances is demonstrated.

[0096] Figure 5 The black dashed line in the middle represents the low-power mode response curve. This mode is suitable for the preparation phase and emphasizes energy saving, but its recovery speed is the slowest, and the voltage has not fully stabilized after 2 seconds.

[0097] The red solid line represents the high dynamic response mode, which is activated during the dredging stage. It can quickly detect disturbances and pull the voltage back from the lowest point to a stable value in a short time, almost completing the recovery within 0.8 seconds, demonstrating a powerful dynamic adjustment capability.

[0098] The blue dotted line corresponds to the high-efficiency mode, which is usually used in the sludge discharge stage, balancing response and energy efficiency. The voltage tends to stabilize after about 1.5 seconds, and the fluctuation is also small.

[0099] The green solid line represents the power-compensated mode response, which is applied in the transfer phase. Its response speed is between high dynamics and high efficiency, combining adaptability and robustness, and can complete more than 90% voltage recovery within 1 second.

[0100] Figure 5 This embodies the key idea of ​​adjusting the operating mode of the AFE hybrid topology according to the current operating stage. It also clearly demonstrates the specific strategies of lowering the modulation frequency for low power consumption, increasing the frequency for high dynamic response, increasing the proportion of uncontrolled rectifiers for high efficiency, and flexibly switching power compensation modes. By comparing the system dynamic performance under different modes, it is confirmed that the mode switching design not only improves the system's stage adaptability but also significantly improves bus stability and hardware lifespan.

[0101] This embodiment further includes an oscillation suppression step, which aims to prevent control oscillations caused by unstable output of the prediction model when the operating conditions change drastically.

[0102] Optionally, before selecting a prediction mode based on the prediction confidence level, the variance of the prediction confidence level fluctuation within a preset sliding time window is calculated. The preset sliding time window can be set to tens of seconds to several minutes, such as a thirty-second window, sliding once per second. The variance of the fluctuation is obtained by calculating the sample variance of the confidence values ​​built into the window, and the formula can be expressed as:

[0103] ;

[0104] in, For the i-th confidence value, Here, N represents the mean of the confidence level within the window, and N is the window size. The variance reflects the stability of the prediction confidence level; a larger value indicates more volatile confidence levels.

[0105] The variance of the fluctuation is then compared with a preset oscillation threshold. This oscillation threshold is an empirical value, such as 0.1, set based on the critical point of system stability in historical data. When the variance of the fluctuation exceeds the oscillation threshold, it is determined that the system is in a state of drastic change in operating conditions. The system forcibly interrupts the selection logic of the prediction model based on prediction confidence and activates the emergency lockout control mode. The emergency lockout control mode includes three sub-steps: discarding the prediction model output, locking the hardware state, and executing the hysteresis exit logic. Specifically:

[0106] 1. Instead of using the power demand curve generated by the four-dimensional power demand forecasting model, the predicted model output is replaced by a recursive least squares method to linearly extrapolate the actual load power over a preset historical period, generating a temporary power baseline command. The recursive least squares method is an online estimation algorithm that fits a linear model by minimizing the sum of squared errors. The formula can be expressed as:

[0107] ;

[0108] in, Let t be the power value at time t, and a and b be coefficients obtained through recursive updates. The preset time period can be set as the data window of the most recent ten seconds.

[0109] 2. Locking the hardware status means forcibly locking the AFE converter into voltage source control mode and prohibiting power-type energy storage units from performing charging operations. Power-type energy storage units are only allowed to perform unidirectional discharge support when the DC bus voltage drops beyond the safety limit. Voltage source control mode prioritizes maintaining bus voltage stability for the AFE converter, ignoring power distribution optimization; prohibiting charging operations prevents overcharging of energy storage units under unstable operating conditions; the safety limit can be set to 90% of the rated voltage, for example, triggering discharge when the 700V bus voltage drops to 630V. Unidirectional discharge support means that the power-type energy storage unit only provides power output and does not absorb power, in order to quickly support the voltage.

[0110] 3. The delayed exit logic involves continuously monitoring the fluctuation variance. Only when the fluctuation variance remains below a preset percentage of the oscillation threshold for a predetermined recovery judgment period will the emergency lockout control mode be released, and the selection logic for the prediction mode based on prediction confidence be restored. The recovery judgment period can be set from several seconds to several minutes, for example, five consecutive seconds; the preset percentage can be set to 80%, i.e., the fluctuation variance is below 0.08. The delayed exit logic prevents frequent mode switching and ensures stable system recovery. In emergency lockout control mode, the system prioritizes basic functions, such as bus voltage stability, while sacrificing some optimization objectives, such as energy efficiency.

[0111] Optionally, the recursive least squares method implementation in the oscillation suppression step can be further explained: the recursive least squares method updates the parameters through a recursive formula. Each time a new data point is received, the coefficients a and b are updated. The formula can be expressed as:

[0112] ;

[0113] in, Here is the gain matrix. Let covariance matrix be the variance matrix. For the input vector, This is the forgetting factor, typically set to 0.98. After the temporary power baseline command is generated by linear extrapolation, it is used to replace the predicted curve until the emergency mode is lifted.

[0114] Specifically, the hardware interaction of the emergency lockout control mode is explained here: Locking the AFE converter to voltage source control mode is achieved by modifying the control loop parameters, such as setting the current reference value to 0 and prioritizing voltage adjustment. Disabling charging of the power storage unit is achieved by sending commands via communication protocols, such as sending a disable signal via the CAN bus. Unidirectional discharge support is achieved through hardware protection circuitry, such as a comparator monitoring the voltage and triggering a discharge switch. For example, during the dredging phase, if sudden changes in wave data cause fluctuations in prediction confidence, the system may activate the emergency lockout control mode, switch to recursive least squares extrapolation of power, and lock the hardware state to prevent voltage collapse. During the sludge removal phase, if the prediction confidence is high, the system performs long-step prediction and adjusts the AFE to a high-efficiency mode to optimize energy efficiency. This integrated design improves the system's robustness under complex operating conditions.

[0115] In this embodiment, the risk warning function of the four-dimensional power demand prediction model can be further expanded. Process risk warnings are based on predicted output and feature analysis; for example, when the predicted power exceeds equipment capacity, an overload risk is triggered; when environmental characteristics indicate severe sea conditions, an operation interruption risk is triggered. Warning information is displayed through a human-machine interface or automatically triggers adjustments to control strategies.

[0116] In this embodiment, the operating mode adjustment of the AFE hybrid topology considers a trade-off between energy efficiency and dynamics. Low-power modes minimize losses, high-dynamic-response modes prioritize response speed, high-efficiency modes balance both, and power-compensated modes focus on stability. The mode switching logic is implemented based on a finite state machine, and the state transition conditions include process stage changes and prediction results.

[0117] Optionally, the delayed exit logic of the above-mentioned oscillation suppression steps can be further optimized. For example, the recovery determination time and preset percentage can be dynamically adjusted based on historical performance, such as by learning the optimal value through machine learning algorithms. Before exiting emergency mode, the system may execute a test cycle to gradually restore the use of the predictive model.

[0118] Optionally, to meet practical applications, the following parameter setting examples can be provided: First threshold set to 0.8, second threshold set to 0.6, oscillation threshold set to 0.1, recovery determination time set to five seconds, and preset percentage set to 80%. AFE modulation frequency range from 5000Hz to 20000Hz, and uncontrolled rectifier output percentage from 10% to 60%. These parameters can be customized according to ship specifications.

[0119] Example 3:

[0120] This embodiment describes in detail an AFE DC bus power pooling coordinated control method for trailing suction duct operations. This method focuses on enhancing the power compensation of power-type energy storage units, the energy coordination between the main power pool and the redundant power pool, and the multi-objective model predictive control strategy.

[0121] Specifically, by real-time monitoring of DC bus voltage pulsation and load power abrupt changes, calculating compensation power values, coordinating power pool energy allocation, and achieving low-frequency power allocation based on multi-objective optimization, this embodiment achieves more refined power management and improved system stability.

[0122] This method, in the multi-timescale hierarchical control of Example 1, further illustrates the handling of millisecond-level high-frequency power fluctuations, specifically including power compensation by controlling power-type energy storage units. Power compensation aims to smooth power pulsations caused by process interference and ensure the instantaneous stability of the DC bus voltage. Power-type energy storage units are typically composed of supercapacitors or high-speed flywheel systems, with response times within milliseconds, enabling rapid power injection or absorption. The system monitors the voltage pulsation amplitude of the DC bus and the sudden increase in load power in real time. The voltage pulsation amplitude is acquired by a voltage sensor, and the difference between its peak and trough values ​​within a sampling period is calculated; the sudden increase in load power is acquired by a power sensor, and its instantaneous rate of change is calculated. When the voltage pulsation amplitude exceeds a preset voltage threshold, and the sudden increase in load power exceeds the power mutation threshold set for the current operating process stage, the system triggers the power compensation mechanism.

[0123] The power compensation mechanism includes calculating the compensation power value, which is the product of the load power surge and the process correction factor. The load power surge is an instantaneous power change measured in kilowatts, obtained by the difference between the current power value and the power value at the previous sampling point. The process correction factor is a dimensionless coefficient, with different preset values ​​depending on whether the current stage is dredging or transfer. For example, during dredging, the process correction factor might be set to 0.8 to reflect the sensitivity to power fluctuations at this stage; during transfer, it might be set to 0.5 to reflect the smoothness of power demand at this stage. The formula for calculating the compensation power value is:

[0124] ;

[0125] This value is sent to the control unit of the power storage unit, which instructs it to output or absorb the corresponding power to offset the effects of sudden load changes.

[0126] Optionally, the specific implementation of power compensation can be further explained. The preset voltage threshold is set based on the rated voltage of the DC bus. For example, if the rated voltage is 700 volts, the threshold might be set to 10 volts, representing the allowable voltage fluctuation range. The power surge threshold is dynamically adjusted according to the operational phase. For example, during the dredging phase, the threshold might be set to 50 kilowatts because the load changes frequently during this phase; during the transfer phase, the threshold might be set to 20 kilowatts. The monitoring process is executed by a digital signal processor with a sampling frequency of 10 kHz, ensuring millisecond-level response. The power storage unit is connected to the DC bus via a bidirectional DC-DC converter. When a compensation command is issued, the converter adjusts its duty cycle to control the power flow direction. For example, when the load power suddenly increases, the power storage unit discharges; when the load power suddenly decreases, the power storage unit charges. This rapid compensation effectively suppresses voltage pulsations and prevents overvoltage or undervoltage of the equipment.

[0127] Optionally, the setting of the aforementioned process correction coefficients can be optimized based on historical data or machine learning models. The system maintains a lookup table to store the coefficient values ​​corresponding to different process stages, which are determined through experiments or simulations. For example, during the dredging stage, due to the start-up and shutdown of the skid suction pump and changes in slurry concentration, the coefficient value is higher to emphasize the compensation intensity; during the transfer stage, due to the relatively stable load, the coefficient value is lower to conserve energy storage resources. The coefficient update mechanism can be adjusted according to real-time performance evaluation; for example, if the voltage is still unstable after compensation, the system automatically increases the coefficient. The calculation of the compensation power value also includes limiting processing to ensure that it does not exceed the rated capacity of the power-type energy storage unit, such as the maximum discharge power of a supercapacitor being one hundred kilowatts.

[0128] Optionally, this embodiment further illustrates the handling of second-level process stage switching requirements, specifically including coordinating the energy allocation between the main power pool and the redundant power pool. The main power pool typically consists of a main AFE converter and a generator, responsible for basic power supply; the redundant power pool consists of an auxiliary converter and a backup power supply, providing additional power support. Energy allocation coordination aims to achieve mid-frequency power adaptation and energy shifting, ensuring a smooth transition during stage switching. The system utilizes energy-type energy storage units for energy absorption or release. These energy-type energy storage units are typically battery systems with response times on the order of seconds, suitable for handling power changes lasting several seconds.

[0129] Specifically, energy allocation coordination is applied in scenarios such as process phase transitions. When the process phase identification model predicts that the operation is about to transition from the dredging phase to the sludge discharge phase, the redundant power pool releases a preset proportion of its rated power at a predetermined time in advance, while limiting the power change rate of the main power pool to less than a preset first change rate threshold. The preset time can be set to several seconds, such as five seconds; the preset proportion can be set to 30%, meaning the redundant power pool outputs 30% of its rated power; and the first change rate threshold can be set to 50 kilowatts per second, indicating that the power increase rate of the main power pool must not exceed this value. This coordination avoids power surges during phase transitions. For example, when the sludge discharge phase starts, the sludge discharge pump requires significant power, and the early release of the redundant power pool alleviates the pressure on the main power pool.

[0130] Another energy allocation coordination scenario involves the transition from the sludge removal phase to the transfer phase. When it is predicted that the operation will transition from the sludge removal phase to the transfer phase, the main power pool is controlled to reduce the power of the sludge removal pump, and the rate of decrease is limited to less than a preset second rate of change threshold. Simultaneously, the redundant power pool is controlled to absorb excess power and store it in an energy storage unit. The second rate of change threshold can be set to 30 kilowatts per second, indicating that the power reduction rate of the main power pool must not exceed this value; absorbing excess power refers to the redundant power pool transferring excess energy to an energy storage unit, such as battery charging. This coordination prevents voltage spikes caused by sudden power drops and achieves energy recovery.

[0131] like Figure 6 This demonstrates the coordinated control strategy for energy output between the main power pool and the redundant power pool during the transition from the dredging stage to the discharge stage in a trailing suction hopper operation.

[0132] Figure 6The solid blue line represents the output power of the main power pool, which remains stable for the first 40 seconds before gradually increasing linearly from 600 kW to 800 kW. This demonstrates a controlled ramp-up strategy, effectively limiting the rate of power change and preventing fluctuations in bus voltage due to sudden power increases. The dashed red line represents the behavior of the redundant power pool, showing that it releases 100 kW of power 35 seconds before the main power pool responds, and then rapidly drops to zero within 10 seconds after the phase switching critical point of 40 seconds. This dynamic adjustment strategy of releasing power before withdrawing indicates that the redundant power pool plays a role in slow start-up in the control logic, i.e., by actively outputting some power before phase switching, it alleviates the dynamic ramp-up burden on the main power pool.

[0133] Figure 6 It embodies the division of labor mechanism between controlling the proportion of rated power released early by the redundant power pool and limiting the power change rate of the main power pool to climb in a manner that is less than the preset change rate threshold, and demonstrates the time coupling relationship and power balancing process between the coordinated release and the limited climb strategy.

[0134] For example, energy allocation coordination is implemented through a central controller. The controller receives the predictive output of the process stage identification model and generates allocation instructions based on the power demand curve. The primary power pool and redundant power pool are connected via communication protocols such as Modbus or CAN bus, and the instructions include power setpoints and rate-of-change limits. The energy storage unit monitors the state of charge through a battery management system to ensure that absorption or release operations are within safe limits. For instance, during the dredging-to-discharge phase, the release instruction from the redundant power pool is sent five seconds in advance, and the power of the primary power pool is adjusted in a linear ramp-up manner, with the rate of change enforced by a PID controller. During the discharge-to-transfer phase, the power of the primary power pool is adjusted in a linear ramp-down manner, and the power absorption of the redundant power pool is achieved through current loop control.

[0135] Optionally, the optimization of energy allocation coordination is further explained here. The preset time, preset ratio, and rate of change threshold can be dynamically adjusted according to real-time load and environmental conditions. For example, if the state of charge of the energy storage unit is low, the system may increase the release ratio of the redundant power pool to compensate; if environmental parameters indicate strong winds and waves, the system may tighten the rate of change threshold to enhance stability. The coordination logic is based on a finite state machine, with states including normal mode, transition mode, and emergency mode to ensure robustness.

[0136] Further explanation of handling minute-level steady-state operation requirements includes optimizing low-frequency power allocation based on a multi-objective model predictive control (MADC) strategy. MADC is an advanced control algorithm that aims to achieve bus voltage stability, system energy efficiency optimization, and compliance with operational schedules, while considering the state-of-charge (SOC) constraints of energy storage units during the optimization process. This strategy generates an optimal control sequence by constructing an objective function and solving for its minimum value, thus guiding power allocation.

[0137] For example, the implementation of a multi-objective model predictive control strategy begins with constructing an objective function. This objective function is a weighted sum of three terms: bus voltage deviation, system energy efficiency loss, and work schedule deviation. The bus voltage deviation represents the squared difference between the actual bus voltage and the reference voltage, reflecting voltage stability; the system energy efficiency loss represents the total system losses, including converter losses and line losses, reflecting energy efficiency levels; and the work schedule deviation represents the squared difference between the actual work progress and the planned progress, reflecting compliance. The weighted sum balances the importance of each objective through weighting coefficients; for example, voltage deviation has a higher weight to ensure safety.

[0138] The mathematical expression for the objective function can be represented as:

[0139] ;

[0140] in, , and These are the weighting coefficients. This is the actual bus voltage. For reference voltage, For system energy efficiency loss, This refers to the actual work progress. To plan the work schedule.

[0141] The optimal control sequence is obtained by minimizing the objective function under the given constraints. These constraints include: during the dredging stage, the rake head depth must not be lower than a set lower threshold and the mud concentration must not be higher than a set upper threshold; during the mud discharge stage, the mud tank level must not be higher than a set upper threshold and the discharge speed must not be lower than a set lower threshold; the modulation frequency range of the AFE; and the state of charge range of the energy storage unit. The lower and upper thresholds are set based on process requirements, for example, a lower limit of five meters for the rake head depth, an upper limit of 30% for the mud concentration, an upper limit of 90% for the mud tank level, and a lower limit of two meters per second for the discharge speed. The AFE modulation frequency range is determined by hardware, for example, 5000Hz to 20000Hz; and the state of charge range of the energy storage unit ensures battery safety, for example, 20% to 80%.

[0142] For example, minimizing the objective function is achieved through optimization algorithms, such as quadratic programming or genetic algorithms. The optimization process is executed in each control cycle, for example, once per minute, with the current system state and the output of the predictive model as input. The optimal control sequence includes the power setpoints for each power unit over the next few minutes, such as the power output of the main power pool, the activation state of the redundant power pool, and the charging and discharging power of the energy storage unit. The first element of the sequence is applied to the system, and the remainder is used for rolling optimization. This predictive control combines real-time performance with foresight, improving steady-state performance.

[0143] Optionally, weighting coefficients , and Determined through expert experience or multi-objective optimization, for example Set it to 0.5. Set it to 0.3. Set to 0.2; reference voltage The system has a voltage of 700 volts and energy efficiency losses. The formula, calculated using the efficiency model, is expressed as follows:

[0144] ;

[0145] in, For input power, Output power. Work progress. and Expressed as a percentage, data is obtained from sensors and schedules. Constraint handling employs a penalty function method, transforming constraint violations into additional terms in the objective function to ensure feasibility.

[0146] For example, the state of charge (SOC) range is enforced as a hard constraint in the optimization process; for instance, discharging is prohibited if the SOC is below 20%, and charging is prohibited if it is above 80%. The SOC prediction is obtained by integrating the power value, expressed by the formula:

[0147] ;

[0148] in, The initial state of charge, For battery power, This is the maximum battery capacity. This management extends battery life and ensures system reliability.

[0149] In this embodiment, during the dredging stage, the power storage unit compensates for the high-frequency fluctuations of the scraper pump; when the sludge discharge stage is predicted, the redundant power pool releases power in advance, while the main power pool slowly ramps up; and during the steady-state sludge discharge stage, multi-objective model predictive control optimizes power allocation to ensure voltage stability and energy efficiency. This multi-level coordination significantly reduces equipment stress and energy waste.

[0150] like Figure 7 The study demonstrates the differences in power allocation and operation progress of the main power pool during a typical steady-state trailing suction hopper operation, with and without the implementation of a multi-objective optimization control strategy.

[0151] Figure 7 The red dashed line represents the power output without optimized control. It can be seen that there are large fluctuations and random disturbances. The maximum power exceeds 800 kilowatts and drops to 600 kilowatts at the lowest, indicating that the system has failed to operate stably, has low energy efficiency, and has an adverse effect on voltage stability.

[0152] The green solid line represents the output result of the main power pool after applying the multi-objective model predictive control strategy proposed in this embodiment. Its overall distribution is smoother and more stable, with the average power controlled at around 750 kW and the fluctuation controlled within 10 kW, effectively achieving the synchronous optimization of system energy efficiency and voltage stability targets.

[0153] at the same time, Figure 7 The purple and blue curves represent the work progress curves under the two strategies, respectively. As can be seen from the figure, the optimized work progress curve is closer to linear growth and finally reaches 100% of the target in 60 minutes. However, under the unoptimized condition, there are multiple growth lags, indicating that the work deviates from the planned progress.

[0154] In this embodiment, the real-time performance of the multi-objective model predictive control strategy can be further optimized. For example, the control cycle can be adjusted according to the system load, such as shortening it to thirty seconds under high-fluctuation conditions. The optimization algorithm employs an efficient solver, such as QPOASES, to meet real-time requirements. Model prediction is based on the output of a four-dimensional power demand prediction model, enhancing accuracy.

[0155] For example, the parameter settings for this method can be specifically exemplified as follows: the preset voltage threshold is 10 volts, the power surge threshold is 50 kW during the dredging stage and 20 kW during the transfer stage. The process correction coefficient is 0.8 during the dredging stage and 0.5 during the transfer stage. The preset time is 5 seconds, the preset ratio is 30%, the first rate of change threshold is 50 kW per second, and the second rate of change threshold is 30 kW per second. The aforementioned weighting coefficients can all be optimized through on-site debugging.

[0156] In this embodiment, after the power-type energy storage unit processes millisecond-level events, the energy-type energy storage unit takes over second-level events, forming a seamless transition. For example, if power compensation is insufficient to smooth out fluctuations, energy allocation coordination may be activated in advance to provide additional support. The coordination logic is based on a priority mechanism, with millisecond-level events taking precedence over second-level events. In terms of hardware, the power-type energy storage unit is implemented using supercapacitor banks and bidirectional converters; the main power pool and redundant power pool are implemented using AFE converter banks and generators; and the energy-type energy storage unit is implemented using a lithium battery system and a battery management system. The controller uses an industrial PLC or embedded computer running a real-time operating system.

[0157] Example 4:

[0158] like Figure 8 As shown, corresponding to the above method embodiments, the present invention also proposes an AFE DC bus power pooling coordinated control system for scraper suction hopper operations, comprising:

[0159] The data processing module is used to collect multi-source heterogeneous data during the shovel suction operation, and to perform hierarchical filtering, spatiotemporal alignment and tagging storage on the multi-source heterogeneous data.

[0160] The stage identification module is used to output the current operation stage and confidence level based on the multi-source heterogeneous data using the built-in process stage identification model.

[0161] The four-dimensional prediction module is used to output power demand curves and process risk warnings using the built-in four-dimensional power demand prediction model.

[0162] A multi-timescale control module for executing hierarchical control strategies includes: a power-type energy storage unit control unit for performing millisecond-level control; a power pool coordination unit for performing second-level control, wherein the power pool coordination unit works in conjunction with the energy-type energy storage unit; and an MPC optimization unit for performing minute-level control, wherein the MPC optimization unit uses the state of charge of the energy-type energy storage unit as an optimization constraint.

[0163] The AFE collaborative control module is connected to the multi-timescale control module and is used to adjust the working mode of the AFE hybrid topology or activate hardware redundancy according to the control instructions.

[0164] The system's workflow begins with the data processing module, which collects multi-source heterogeneous data in real time during the scraper suction operation, including process parameters, power parameters, and environmental parameters. It then performs hierarchical filtering, spatiotemporal alignment, and tagged storage to provide standardized data for subsequent modules.

[0165] Subsequently, the stage identification module uses a built-in process stage identification model, such as the random forest algorithm, to process this data and output the current operation stage and confidence level.

[0166] The four-dimensional prediction module generates power demand curves for future periods and process risk warnings based on current stage and multi-source data through a four-dimensional power demand prediction model.

[0167] The multi-timescale control module executes a hierarchical control strategy based on the prediction results: millisecond-level control is handled by the power-type energy storage unit control unit to process high-frequency power fluctuations and achieve rapid power compensation; second-level control is handled by the power pool coordination unit and the energy-type energy storage unit in collaboration to process energy allocation and shift during process stage switching; minute-level control is handled by the MPC optimization unit based on multi-objective model predictive control to optimize low-frequency power allocation and consider the state of charge constraints of the energy-type energy storage unit.

[0168] Finally, the AFE collaborative control module adjusts the operating mode of the AFE hybrid topology according to the control instructions, such as low power consumption, high dynamic response, high efficiency, or power compensation mode, or activates hardware redundancy, to complete the pooled coordinated control of DC bus power and ensure system stability and energy efficiency optimization.

[0169] Example 5:

[0170] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0171] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0172] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0173] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0174] The memory 103 stores a computer program corresponding to the AFE DC bus power pooling coordinated control method for a scraper suction duct operation according to the above embodiments of the present invention. This computer program is executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0175] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0176] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for coordinated power pooling control of AFE DC bus for scraper suction operation, characterized in that, Applied to ship electrical control systems, the following steps are included: Collect multi-source heterogeneous data during the sweeping suction hopper operation, including process parameters, power parameters, and environmental parameters, and perform spatiotemporal alignment processing on the multi-source heterogeneous data; The spatiotemporally aligned multi-source heterogeneous data is input into a preset process stage identification model to determine the current operation process stage. Based on the current operational process stage and the multi-source heterogeneous data, a four-dimensional power demand prediction model is used to generate power demand curves and process risk warnings for future periods. Based on the power demand curve and the process risk warning, multi-time-scale hierarchical control is implemented on the AFE DC bus.

2. The method according to claim 1, characterized in that, The operational process includes a preparation stage, a dredging stage, a dredging stage, a dredging stage, and a transfer stage. The spatiotemporal alignment processing of the multi-source heterogeneous data specifically includes: Using the clock of the control system as a reference, the process parameters, power parameters, and environmental parameters with different sampling frequencies are aligned to a unified time slice through timestamp correction. The process parameters include the depth of the rake head lowering, the mud tank level, the ship's forward speed, the mud concentration, the ship's roll angle, the ship's transfer speed, the rake head lifting height, and the mud discharge speed. The power parameters include high-frequency power data, DC bus voltage and current data; The environmental parameters include wave data; The multi-time-scale hierarchical control of the AFE DC bus includes: In response to millisecond-level high-frequency power fluctuations, the power storage unit is controlled to perform power compensation in order to smooth out power pulsations caused by process interference. In response to the second-level process stage switching requirements, it coordinates the energy allocation between the main power pool and the redundant power pool, and uses energy-type energy storage units to absorb or release energy, thereby achieving mid-frequency power adaptation and energy shifting. In response to minute-level steady-state operation requirements, a multi-objective model predictive control strategy is used to optimize low-frequency power allocation. The multi-objective model predictive control strategy aims to stabilize bus voltage, optimize system energy efficiency, and ensure compliance with operation schedules. The optimization process also considers the state of charge constraints of the energy storage unit.

3. The method according to claim 2, characterized in that, The determination of the current operational process stage specifically includes: The current values ​​and rates of change of process parameters are extracted as input features. These process parameters include the depth of the rake head, the mud tank level, the ship's forward speed, the mud concentration, the ship's roll angle, and the ship's transfer speed. Using the process stage identification model built based on the random forest algorithm, the current process stage and its corresponding confidence level are output; Based on the identified operational process stages, extract the power fluctuation characteristics and process risk characteristics specific to each stage; The power fluctuation characteristics include the mid-frequency fluctuation amplitude of the rake suction pump power during the dredging stage or the low-frequency trend of the discharge pump power during the discharge stage.

4. The method according to claim 2, characterized in that, The method of generating power demand curves and process risk warnings for future periods using a four-dimensional power demand forecasting model specifically includes: Construct a fused input vector that includes process characteristics, load characteristics, environmental characteristics, and risk characteristics; The fused input vector is input into a prediction model constructed based on a long short-term memory network and an attention mechanism to calculate the power demand curve and prediction confidence for future time periods; a prediction mode is selected based on the prediction confidence. When the prediction confidence level is higher than the first threshold, long-step prediction is performed. When the prediction confidence level is between the first threshold and the second threshold, medium-step prediction is performed; When the prediction confidence level is lower than the second threshold, short-step prediction is performed and the power-type energy storage unit is triggered to enter the overclocking response mode.

5. The method according to claim 2, characterized in that, The power compensation of the controlled power type energy storage unit specifically includes: Real-time monitoring of DC bus voltage ripple amplitude and load power surges; When the voltage ripple amplitude exceeds the preset voltage threshold and the load power surge exceeds the power surge threshold set for the current operation process stage, a compensation power value is calculated. The compensation power value is the product of the sudden increase in load power and the process correction coefficient, wherein the process correction coefficient takes different preset values ​​depending on whether the current stage is dredging or transfer.

6. The method according to claim 2, characterized in that, The coordination of energy allocation between the main power pool and the redundant power pool specifically includes: When the process stage identification model predicts that the operation status is about to change from the dredging stage to the sludge discharge stage, it controls the redundant power pool to release a preset proportion of the rated power in advance by a preset time, and limits the power change rate of the main power pool to less than a preset first change rate threshold. When it is predicted that the operation status is about to change from the sludge discharge stage to the transfer stage, the main power pool is controlled to reduce the power of the sludge discharge pump and the rate of decrease is limited to less than a preset second rate of change threshold. At the same time, the redundant power pool is controlled to absorb excess power and store it in the energy storage unit.

7. The method according to claim 2, characterized in that, The optimization of low-frequency power allocation based on the multi-objective model predictive control strategy specifically includes: Construct an objective function, which is a weighted sum of the bus voltage deviation term, the system energy efficiency loss term, and the work progress deviation term; Under the premise of satisfying the constraints, the minimum value of the objective function is solved to obtain the optimal control sequence; The constraints include: The dredging stage has a rake head depth that is not lower than its set lower threshold and a mud concentration that is not higher than its set upper threshold; the mud tank level during the mud discharge stage is not higher than its set upper threshold and the mud discharge speed is not lower than its set lower threshold; the modulation frequency range of the AFE; and the state of charge range of the energy storage unit.

8. The method according to claim 2, characterized in that, The method also includes an AFE hardware cooperative control step: Adjust the working mode of the AFE hybrid topology according to the current operation process stage; During the preparation phase, the control mode is low-power mode, which reduces the output ratio of the uncontrolled rectifier and the modulation frequency. During the dredging stage, the control mode is set to a high dynamic response mode, reducing the output ratio of the uncontrolled rectifier and increasing the modulation frequency. During the sludge discharge stage, control the mode to high efficiency, increase the output ratio of the uncontrolled rectifier, and set a medium modulation frequency. During the transfer phase, the control mode is switched to power compensation mode.

9. The method according to claim 4, characterized in that, The method also includes an oscillation suppression step: Before selecting a prediction mode based on the prediction confidence, the fluctuation variance of the prediction confidence within a preset sliding time window is calculated, and the fluctuation variance is compared with a preset oscillation threshold. When the fluctuation variance exceeds the oscillation threshold, it is determined that the current operating condition is in a state of drastic change. The system forcibly interrupts the selection logic of the prediction mode based on the prediction confidence and activates the emergency lockout control mode. The emergency lockout control mode specifically includes: The power demand curve generated by the four-dimensional power demand prediction model is shielded, and the recursive least squares method is used to linearly extrapolate the actual load power in the past preset period to generate a temporary power reference command. The AFE converter is forcibly locked into voltage source control mode, and the power energy storage unit is prohibited from performing charging operations. The power energy storage unit is only allowed to perform unidirectional discharge support when the DC bus voltage drop exceeds the safety limit. The fluctuation variance is continuously monitored. The emergency lockout control mode is released and the selection logic of the prediction mode based on the prediction confidence is restored only when the fluctuation variance is continuously lower than a preset percentage of the oscillation threshold within a preset recovery judgment period.

10. The method according to claim 6, characterized in that, The method also includes dynamic decoupling and process concession steps for redundant power pool capacity constraints, specifically including: Before controlling the release of power from the redundant power pool, the capacity pressure index of the redundant power pool is calculated in real time. The capacity pressure index is a normalized value generated by weighting the current base load rate of the redundant power pool, the temperature gradient of the energy storage unit, and the remaining energy state. The system compares the capacity pressure index with a preset decoupling threshold. When the capacity pressure index is higher than the decoupling threshold, the system determines that the redundant power pool does not have the support capability and triggers the process concession protection strategy. Specifically, the process concession protection strategy includes: Ignore the instruction that controls the redundant power pool to release a preset proportion of the rated power at a preset time in advance, force the redundant power pool to enter a self-regulating mode, and prioritize the power supply stability of the auxiliary load connected to the redundant power pool. The control constraints of the main power pool are reconstructed in real time. The preset first rate of change threshold is multiplied by an attenuation coefficient less than 1 to generate a corrected ramp limit value. The attenuation coefficient is negatively correlated with the capacity pressure index. The main power pool is controlled to utilize only its own energy reserves and perform a gentler power ramp-up according to the modified ramp-up limit value until the target power of the sludge discharge stage is reached.

11. A coordinated control system for AFE DC bus power pooling for scraper suction operations, characterized in that, The system for implementing the control method according to any one of claims 1 to 10, the system comprising: The data processing module is used to collect multi-source heterogeneous data during the shovel suction operation, and to perform hierarchical filtering, spatiotemporal alignment and tagging storage on the multi-source heterogeneous data. The stage identification module is used to output the current operation stage and confidence level based on the multi-source heterogeneous data using the built-in process stage identification model. The four-dimensional prediction module is used to output power demand curves and process risk warnings using the built-in four-dimensional power demand prediction model. A multi-timescale control module for executing hierarchical control strategies includes: a power-type energy storage unit control unit for performing millisecond-level control; a power pool coordination unit for performing second-level control, wherein the power pool coordination unit works in conjunction with the energy-type energy storage unit; and an MPC optimization unit for performing minute-level control, wherein the MPC optimization unit uses the state of charge of the energy-type energy storage unit as an optimization constraint. The AFE collaborative control module is connected to the multi-timescale control module and is used to adjust the working mode of the AFE hybrid topology or activate hardware redundancy according to the control instructions.

Images