A method for omnidirectional current generation control in a circular marine engineering basin

Abstract

The present application relates to the field of physical model test of ocean engineering, in particular to a kind of omnidirectional flow generation control method for circular ocean engineering pool, comprising the following steps: S1, flow field digital modeling and space-time mapping construction;S2, dynamic adaptive partitioning and control unit mapping;S3, flow field prediction and control coupled calculation;S4, multi-layer annular array collaborative flow generation execution;S5, real-time error feedback construction based on multidimensional flow field perception;S6, predictive compensation type advance control correction;S7, full closed loop dynamic iterative control;The present application constructs space-time three-dimensional flow field model coupling pool structure and water fluid characteristics, realizes pool area dynamic adaptive partitioning, uses time sequence memory network to advance forecast flow field, obtains optimal control vector of flow generation unit by constraint optimization solution, drives multi-layer annular array collaborative flow generation, constructs spatial error field closed loop feedback, and combined with water attenuation and boundary reflection implements advance compensation control, so that the regulation precision is high, adaptive ability is strong, and flow field stability is good.

Description

Technical Field

[0001] This invention relates to the field of physical model testing in marine engineering, specifically to a method for omnidirectional flow control in a circular marine engineering pool. Background Technology

[0002] Circular marine engineering pools are the core infrastructure for conducting hydrodynamic model tests on equipment such as ships, offshore wind power foundations, marine platforms, and underwater vehicles. The simulation accuracy, hierarchical control capability, and dynamic response speed of the artificial current generation system directly determine the realism of the marine environment simulation and the reliability of the test data.

[0003] Existing traditional flow control methods for circular water tanks mostly employ fixed grid static zoning and a single passive feedback regulation mode, which have many technical drawbacks:

[0004] First, the flow field zoning mechanism is rigid and cannot adaptively reconstruct the zoning boundary based on the flow field gradient, turbulence intensity and error distribution, resulting in insufficient control precision in local flow regime change areas.

[0005] Secondly, the modeling did not incorporate the inherent structural parameters of the circular pool and the hydrodynamic characteristics of the water body, resulting in a large deviation from the actual flow mechanism of the pool.

[0006] Third, relying solely on real-time error lag adjustment results in significant control response lag.

[0007] Fourth, the lack of a closed-loop feedback mechanism for the spatial error field makes it impossible to introduce the actual flow field deviation into the prediction and control process in real time, and the failure to consider the influence of disturbances makes it impossible to meet the requirements of high-precision model tests.

[0008] Based on the above, a method for omnidirectional flow control in a circular marine engineering pool is provided. Summary of the Invention

[0009] The purpose of this invention is to provide an omnidirectional flow control method for a circular marine engineering pool, so as to solve the problems mentioned in the background art.

[0010] This invention provides a method for omnidirectional current generation control in a circular marine engineering pool, comprising the following steps:

[0011] S1. Flow field digital modeling and spatiotemporal mapping construction

[0012] Based on multi-source marine environmental data of the target sea area, the velocity vector, flow direction angle and turbulence characteristic parameters corresponding to different water depths are obtained. Combined with the structural parameters of the circular marine engineering pool and the hydrodynamic characteristics of the water body, a spatiotemporally coupled three-dimensional target flow field mapping model is constructed to generate a continuous time series of target flow field vector fields.

[0013] S2, Dynamic Adaptive Partitioning and Control Unit Mapping

[0014] Based on the three-dimensional target flow field mapping model, the test area of ​​the water tank is dynamically partitioned according to the flow field gradient change rate and local turbulence intensity, so as to form an adaptive set of control sub-regions. The spatial mapping relationship between each control sub-region and the circumferentially distributed layered flow generation units is established, wherein the partition boundary changes in real time according to the flow field error gradient.

[0015] S3, Coupled Calculation of Flow Field Prediction and Control

[0016] Historical flow field time series data, real-time flow field monitoring data and operating parameters are input into a deep learning prediction model based on a time series memory network. The model outputs the three-dimensional flow field prediction result for the next moment. Based on the error between the prediction result and the target flow field, a control optimization function is constructed, and the optimal control vector for each layered flow generation unit is calculated.

[0017] S4, Multi-layer Ring Array Cooperative Flow Generation Execution

[0018] The optimal control vector is converted into a frequency conversion drive signal and an outlet angle control signal, which are applied to the multi-layer distributed flow-generating units arranged in a ring array along the circular pool, so that the flow-generating units at different heights can work together to generate the target basic flow field, wherein each flow-generating unit forms a radially symmetrical drive structure in space.

[0019] S5. Construction of Real-Time Error Feedback Based on Multi-Dimensional Flow Field Sensing

[0020] By deploying multi-dimensional flow field monitoring modules at different spatial coordinates and water depths in the pool, three-dimensional flow velocity, flow direction and turbulence intensity data are collected in real time, and the current actual flow field state matrix is ​​constructed. The actual flow field state matrix is ​​spatially aligned with the target flow field to obtain the spatial error field, and the spatial error field is used as the feedback input variable of the deep learning prediction model to participate in the flow field prediction correction at the next moment.

[0021] S6, Predictive Compensation-Type Leading Control Correction

[0022] Based on the future flow field evolution trend output by the deep learning prediction model and the introduced spatial error field feedback, the control vector of each layered flow-generating unit is time-shifted and compensated. At the same time, the water energy attenuation model and the boundary reflection disturbance model are introduced to compensate and correct the control signal.

[0023] S7, Full Closed-Loop Dynamic Iterative Control

[0024] The closed-loop iterative process of "flow field acquisition - prediction modeling - error calculation - advance compensation - control update" is continuously executed to form a stable omnidirectional controllable stratified flow field environment in the water tank test area.

[0025] In a preferred technical solution, the structural parameters of the pool in step S1 include the radial geometric dimensions of the pool, the effective water depth, the curvature of the pool wall boundary, and the spatial layout of the annular stratified flow-generating units; the hydrodynamic properties of the water body include the kinematic viscosity of the water body, the viscous drag coefficient, the water flow reflection attenuation characteristics of the pool wall, and the flow field coupling disturbance characteristics between different control sub-regions; by coupling the spatial constraints of the pool structure, the constitutive properties of the water body fluid, and the time-domain ocean current operating parameters, a four-dimensional three-dimensional flow field characterization model containing spatial coordinates (x, y, z) and time t is established.

[0026] In a preferred technical solution, the process of forming the set of control sub-regions in the dynamic adaptive partitioning process in step S2 includes: spatial discretization of the water tank test area based on the three-dimensional target flow field mapping model, calculating the flow field gradient change rate and local turbulence intensity of each spatial sampling point, and dividing the spatial area that meets the preset threshold condition into the initial control sub-region;

[0027] Based on this, using the flow field error gradient as the driving variable, boundary fusion or boundary subdivision operations are performed on adjacent initial control sub-regions, so that the set of control sub-regions changes over time to form an adaptive reconstructed structure.

[0028] Specifically, when the local error gradient increases, a subdivision operation is performed on the corresponding control sub-region; when the local error gradient decreases, a merging operation is performed on adjacent control sub-regions to achieve dynamic adjustment of the control sub-region set.

[0029] In a preferred technical solution, the method for constructing the control optimization function and solving the optimal control vector in step S3 includes: comparing the spatial difference between the three-dimensional flow field prediction result of the deep learning prediction model output at the next moment and the target flow field to form a flow field error distribution;

[0030] Based on the flow field error distribution, a comprehensive control evaluation index is constructed. This evaluation index includes both the flow field deviation minimization target and the flow generation unit energy consumption constraint target. The flow field deviation is used to characterize the spatial consistency difference between the target flow field and the predicted flow field, and the energy consumption constraint is used to characterize the operational intensity limit of each layered flow generation unit.

[0031] The control variables of each layered flow generation unit are taken as the optimization decision objects. Through iterative optimization of physical boundary constraints, the control evaluation index is gradually converged and solved so that the system obtains the optimal control solution under the conditions of power limitation, flow field stability constraint and boundary reflection suppression.

[0032] The optimal control solution corresponds to the optimal control vector of each stratified flow-generating unit, which is used to generate variable frequency drive signals and water outlet angle control signals, thereby realizing dynamic approximation and real-time correction control of the target flow field.

[0033] In a preferred technical solution, the control variables of each stratified flow-generating unit are used as the optimization decision objects. Specifically, the rotational speed, outlet angle, and running time of each stratified flow-generating unit are used as control variables, and the control variables are used as the decision objects for optimization calculation. Based on this, control evaluation indicators are constructed according to the deviation between the target flow field and the predicted flow field, and the control evaluation indicators are gradually adjusted and converged through iterative optimization.

[0034] In the optimization solution process, physical boundary constraints are set, including upper limit constraints on the power output of the flow generation unit, constraints on the suppression of reflection interference at the boundary of the pool, and constraints on maintaining the stability of the flow field.

[0035] By continuously iteratively correcting the control variables under the premise of satisfying the above constraints, the system gradually obtains the optimal control solution that minimizes the flow field error, and the optimal control solution is used to generate the actual drive control commands for each layered flow generation unit.

[0036] In a preferred embodiment, the spatial error field in step S5 participates in predictive control in the following ways:

[0037] The spatial error field constructed in step S5 is used as the feedback input information of the deep learning prediction model and fused with historical flow field time series data and real-time monitoring data so that the prediction model can simultaneously consider the actual flow field deviation when generating the three-dimensional flow field prediction result for the next moment.

[0038] Furthermore, the spatial weights of the prediction results are adjusted based on the spatial error field, so that regions with larger errors receive higher correction priority during the prediction process, thereby improving the fit between the flow field prediction results and the actual flow state of the pool.

[0039] In a preferred embodiment, the spatial error field in step S5 is constructed by calculating the spatial difference between the target flow field and the actual flow field, and its expression is:

[0040]

[0041] in: For spatial error field; The target flow field vector; This is the actual measured flow field vector;

[0042] Spatial error weighting coefficients are generated based on the absolute value distribution of the error field, which are used to adjust the influence weights of different control sub-regions in subsequent prediction and control calculations.

[0043] In a preferred technical solution, step S6, predictive compensation-type advanced control correction, includes: adjusting the control vectors of each layered flow-generating unit in advance based on the future flow field evolution trend output by the deep learning prediction model, so that the control commands are applied to the flow field evolution process in advance.

[0044] Simultaneously, the predicted control results are corrected by combining the spatial error field, and the influence of water energy attenuation law and pool wall boundary reflection disturbance is introduced to physically constrain and compensate the control signal, so as to reduce the impact of prediction deviation on flow field stability.

[0045] In a preferred embodiment, during the predictive compensation-type advance control correction process in step S6, the future flow field state is based on the output of the deep learning prediction model. ), construct control correction quantity Its expression is:

[0046]

[0047] in: This is the pseudo-inverse of the flow field response matrix; To predict the compensation weight coefficient; For the target flow field; To measure the flow field in real time, )for Time and Space Point The predicted flow field vector.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] 1. This invention couples the geometric parameters of the water tank structure, the hydrodynamic characteristics of the water body, and the time-domain ocean working condition data to construct a four-dimensional spatiotemporal three-dimensional flow field mapping model, which fits the real flow mechanism of a circular water tank and avoids the simulation deviation caused by the general mathematical fitting model.

[0050] 2. A dynamic adaptive partitioning mechanism driven by the combined flow field gradient, turbulence intensity and error gradient can automatically complete the subdivision and merging of control sub-regions, realize real-time reconstruction of partition boundaries, and significantly improve the fineness of flow field control and adaptive capability.

[0051] 3. Introducing temporal memory network deep learning for advanced flow field prediction, combined with spatial error field closed-loop feedback, the actual flow field deviation is incorporated into the prediction process in real time, effectively overcoming the control delay problem of traditional lag regulation;

[0052] 4. Construct a comprehensive optimization index that balances the accuracy of flow field control with the energy consumption of the flow generation unit, superimpose multiple physical constraints such as power, boundary reflection, and flow field stability, and iteratively solve for the optimal control vector to balance control performance and engineering operation safety.

[0053] 5. By adopting a multi-composite compensation mechanism of predictive time advance, spatial error correction, water energy attenuation and pool wall boundary reflection, various disturbances are effectively suppressed, and the uniformity of the flow field, the controllability of the stratification and the long-term operational stability are improved.

[0054] 6. The overall system forms a closed-loop control architecture of perception, modeling, partitioning, prediction, optimization, execution, feedback, and proactive compensation, with a high degree of automation and intelligence, fully meeting the requirements of high-precision marine environment simulation tests in circular marine engineering pools. Attached Figure Description

[0055] Figure 1 This is a flowchart of the omnidirectional flow control method for water tanks according to the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 This invention provides a method for omnidirectional current generation control in a circular marine engineering pool, comprising the following steps:

[0058] S1. Flow field digital modeling and spatiotemporal mapping construction

[0059] Based on multi-source marine environmental data of the target sea area, the velocity vector, flow direction angle and turbulence characteristic parameters corresponding to different water depths are obtained. Combined with the structural parameters of the circular marine engineering pool and the hydrodynamic characteristics of the water body, a spatiotemporally coupled three-dimensional target flow field mapping model is constructed to generate a continuous time series of target flow field vector fields.

[0060] The structural parameters of the water tank include its radial geometric dimensions, effective water depth, wall boundary curvature, and the spatial layout of the annular stratified flow-generating units. The hydrodynamic properties of the water body include its kinematic viscosity, viscous drag coefficient, water flow reflection attenuation characteristics at the pool wall, and the flow field coupling disturbance characteristics between different control sub-regions. By coupling the spatial constraints of the water tank structure, the constitutive properties of the water body fluid, and the time-domain ocean current parameters, a four-dimensional three-dimensional flow field characterization model containing spatial coordinates (x, y, z) and time t is established.

[0061] S2, Dynamic Adaptive Partitioning and Control Unit Mapping

[0062] Based on the three-dimensional target flow field mapping model, the test area of ​​the water tank is dynamically partitioned according to the flow field gradient change rate and local turbulence intensity, so as to form an adaptive set of control sub-regions. The spatial mapping relationship between each control sub-region and the circumferentially distributed layered flow generation units is established, wherein the partition boundary changes in real time according to the flow field error gradient.

[0063] In the dynamic adaptive partitioning process, the formation process of the control sub-region set includes: based on the three-dimensional target flow field mapping model, the water tank test area is spatially discretized, the flow field gradient change rate and local turbulence intensity of each spatial sampling point are calculated, and the spatial area that meets the preset threshold condition is divided into the initial control sub-region;

[0064] Based on this, using the flow field error gradient as the driving variable, boundary fusion or boundary subdivision operations are performed on adjacent initial control sub-regions, so that the set of control sub-regions changes over time to form an adaptive reconstructed structure.

[0065] Specifically, when the local error gradient increases, a subdivision operation is performed on the corresponding control sub-region; when the local error gradient decreases, a merging operation is performed on adjacent control sub-regions to achieve dynamic adjustment of the control sub-region set.

[0066] S3, Coupled Calculation of Flow Field Prediction and Control

[0067] Historical flow field time series data, real-time flow field monitoring data and operating parameters are input into a deep learning prediction model based on a time series memory network. The model outputs the three-dimensional flow field prediction result for the next moment. Based on the error between the prediction result and the target flow field, a control optimization function is constructed, and the optimal control vector for each layered flow generation unit is calculated.

[0068] The method for constructing the control optimization function and solving the optimal control vector includes: comparing the spatial difference between the three-dimensional flow field prediction result of the deep learning prediction model output at the next moment and the target flow field to form a flow field error distribution;

[0069] Based on the flow field error distribution, a comprehensive control evaluation index is constructed. This evaluation index includes both the flow field deviation minimization target and the flow generation unit energy consumption constraint target. The flow field deviation is used to characterize the spatial consistency difference between the target flow field and the predicted flow field, and the energy consumption constraint is used to characterize the operational intensity limit of each layered flow generation unit.

[0070] The control variables of each layered flow generation unit are taken as the optimization decision objects. Through iterative optimization of physical boundary constraints, the control evaluation index is gradually converged and solved so that the system obtains the optimal control solution under the conditions of power limitation, flow field stability constraint and boundary reflection suppression.

[0071] The optimal control solution corresponds to the optimal control vector of each stratified flow-generating unit, which is used to generate variable frequency drive signals and water outlet angle control signals, thereby realizing dynamic approximation and real-time correction control of the target flow field.

[0072] Specifically, the control variables of each stratified flow-generating unit are used as the optimization decision objects. Specifically, the rotation speed, outlet angle and running time of each stratified flow-generating unit are used as control variables, and the control variables are used as the decision objects for optimization calculation. On this basis, according to the deviation between the target flow field and the predicted flow field, control evaluation indicators are constructed, and the control evaluation indicators are gradually adjusted and converged through iterative optimization.

[0073] In the optimization solution process, physical boundary constraints are set, including upper limit constraints on the power output of the flow generation unit, constraints on the suppression of reflection interference at the boundary of the pool, and constraints on maintaining the stability of the flow field.

[0074] By continuously iteratively correcting the control variables under the premise of satisfying the above constraints, the system gradually obtains the optimal control solution that minimizes the flow field error, and the optimal control solution is used to generate the actual drive control commands for each layered flow generation unit.

[0075] S4, Multi-layer Ring Array Cooperative Flow Generation Execution

[0076] The optimal control vector is converted into a frequency conversion drive signal and an outlet angle control signal, which are applied to the multi-layer distributed flow-generating units arranged in a ring array along the circular pool. This enables the flow-generating units at different heights to collaboratively generate the target basic flow field. Each flow-generating unit forms a radially symmetrical drive structure in space, ensuring uniform distribution of the flow field across the entire area and controllable stratified flow.

[0077] S5. Construction of Real-Time Error Feedback Based on Multi-Dimensional Flow Field Sensing

[0078] By deploying multi-dimensional flow field monitoring modules at different spatial coordinates and water depths in the pool, three-dimensional flow velocity, flow direction and turbulence intensity data are collected in real time, and the current actual flow field state matrix is ​​constructed. The actual flow field state matrix is ​​spatially aligned with the target flow field to obtain the spatial error field, and the spatial error field is used as the feedback input variable of the deep learning prediction model to participate in the flow field prediction correction at the next moment.

[0079] The spatial error field in step S5 participates in predictive control in the following ways:

[0080] The spatial error field constructed in step S5 is used as the feedback input information of the deep learning prediction model and fused with historical flow field time series data and real-time monitoring data so that the prediction model can simultaneously consider the actual flow field deviation when generating the three-dimensional flow field prediction result for the next moment.

[0081] Furthermore, the spatial weights of the prediction results are adjusted based on the spatial error field, so that regions with larger errors receive higher correction priority during the prediction process, thereby improving the fit between the flow field prediction results and the actual flow state of the pool.

[0082] Specifically, the spatial error field is constructed by calculating the spatial difference between the target flow field and the actual flow field, and its expression is:

[0083]

[0084] in: For spatial error field; The target flow field vector; This is the actual measured flow field vector;

[0085] Spatial error weighting coefficients are generated based on the absolute value distribution of the error field, which are used to adjust the influence weights of different control sub-regions in subsequent prediction and control calculations.

[0086] S6, Predictive Compensation-Type Leading Control Correction

[0087] Based on the future flow field evolution trend output by the deep learning prediction model and the introduced spatial error field feedback, the control vector of each layered flow-generating unit is time-shifted and compensated. At the same time, the water energy attenuation model and the boundary reflection disturbance model are introduced to compensate and correct the control signal.

[0088] Among them, the predictive compensation type advanced control correction includes: based on the future flow field evolution trend output by the deep learning prediction model, adjusting the control vector of each layer flow-generating unit in advance so that the control command acts on the flow field evolution process in advance;

[0089] Simultaneously, the predicted control results are corrected by combining the spatial error field, and the influence of water energy attenuation law and pool wall boundary reflection disturbance is introduced to physically constrain and compensate the control signal, so as to reduce the impact of prediction deviation on flow field stability.

[0090] Specifically, in the process of predictive compensation-type advanced control correction, the future flow field state is based on the output of a deep learning prediction model. ), construct control correction quantity Its expression is:

[0091]

[0092] in: This is the pseudo-inverse of the flow field response matrix; To predict the compensation weight coefficient; For the target flow field; To measure the flow field in real time, )for Time and Space Point The predicted flow field vector.

[0093] S7, Full Closed-Loop Dynamic Iterative Control

[0094] The closed-loop iterative process of "flow field acquisition - prediction modeling - error calculation - advance compensation - control update" is continuously executed to form a stable omnidirectional controllable stratified flow field environment in the water tank test area.

[0095] The fully closed-loop dynamic iterative control process includes a cyclical process of continuously executing flow field acquisition, flow field prediction, spatial error calculation, advance compensation correction, and control command update.

[0096] By introducing spatial error field feedback in each iteration cycle, the predictive control results can be dynamically corrected and work together with the physical compensation model to achieve stable convergence of the flow field in the pool test area and maintain the omnidirectional controllable stratified flow state.

[0097] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for omnidirectional current generation control in a circular marine engineering pool, characterized in that, Includes the following steps: S1. Flow field digital modeling and spatiotemporal mapping construction Based on multi-source marine environmental data of the target sea area, the velocity vector, flow direction angle and turbulence characteristic parameters corresponding to different water depths are obtained. Combined with the structural parameters of the circular marine engineering pool and the hydrodynamic characteristics of the water body, a spatiotemporally coupled three-dimensional target flow field mapping model is constructed to generate a continuous time series of target flow field vector fields. S2, Dynamic Adaptive Partitioning and Control Unit Mapping Based on the three-dimensional target flow field mapping model, the test area of ​​the water tank is dynamically partitioned according to the flow field gradient change rate and local turbulence intensity, so as to form an adaptive set of control sub-regions. The spatial mapping relationship between each control sub-region and the circumferentially distributed layered flow generation units is established, wherein the partition boundary changes in real time according to the flow field error gradient. S3, Coupled Calculation of Flow Field Prediction and Control Historical flow field time series data, real-time flow field monitoring data and operating parameters are input into a deep learning prediction model based on a time series memory network. The model outputs the three-dimensional flow field prediction result for the next moment. Based on the error between the prediction result and the target flow field, a control optimization function is constructed, and the optimal control vector for each layered flow generation unit is calculated. S4, Multi-layer Ring Array Cooperative Flow Generation Execution The optimal control vector is converted into a frequency conversion drive signal and an outlet angle control signal, which are applied to the multi-layer distributed flow-generating units arranged in a ring array along the circular pool, so that the flow-generating units at different heights can work together to generate the target basic flow field, wherein each flow-generating unit forms a radially symmetrical drive structure in space. S5. Construction of Real-Time Error Feedback Based on Multi-Dimensional Flow Field Sensing By deploying multi-dimensional flow field monitoring modules at different spatial coordinates and water depths in the pool, three-dimensional flow velocity, flow direction and turbulence intensity data are collected in real time, and the current actual flow field state matrix is ​​constructed. The actual flow field state matrix is ​​spatially aligned with the target flow field to obtain the spatial error field, and the spatial error field is used as the feedback input variable of the deep learning prediction model to participate in the flow field prediction correction at the next moment. S6, Predictive Compensation-Type Leading Control Correction Based on the future flow field evolution trend output by the deep learning prediction model and the introduced spatial error field feedback, the control vector of each layered flow-generating unit is time-shifted and compensated. At the same time, the water energy attenuation model and the boundary reflection disturbance model are introduced to compensate and correct the control signal. S7, Full Closed-Loop Dynamic Iterative Control The closed-loop iterative process of "flow field acquisition - prediction modeling - error calculation - advance compensation - control update" is continuously executed to form a stable omnidirectional controllable stratified flow field environment in the water tank test area.

2. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: The structural parameters of the pool in step S1 include the radial geometric dimensions of the pool, the effective water depth, the curvature of the pool wall boundary, and the spatial layout of the annular stratified flow-generating units; the hydrodynamic properties of the water body include the kinematic viscosity of the water body, the viscous drag coefficient, the water flow reflection attenuation characteristics of the pool wall, and the flow field coupling disturbance characteristics between different control sub-regions; by coupling the spatial constraints of the pool structure, the constitutive properties of the water body fluid, and the time-domain ocean current operating parameters, a four-dimensional three-dimensional flow field characterization model containing spatial coordinates (x, y, z) and time t is established.

3. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: In the dynamic adaptive partitioning process in step S2, the formation process of the control sub-region set includes: based on the three-dimensional target flow field mapping model, the water tank test area is spatially discretized, the flow field gradient change rate and local turbulence intensity of each spatial sampling point are calculated, and the spatial area that meets the preset threshold condition is divided into the initial control sub-region. Based on this, using the flow field error gradient as the driving variable, boundary fusion or boundary subdivision operations are performed on adjacent initial control sub-regions, so that the set of control sub-regions changes over time to form an adaptive reconstructed structure. Specifically, when the local error gradient increases, a subdivision operation is performed on the corresponding control sub-region; when the local error gradient decreases, a merging operation is performed on adjacent control sub-regions to achieve dynamic adjustment of the control sub-region set.

4. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: The method for constructing the control optimization function and solving the optimal control vector in step S3 includes: comparing the spatial difference between the three-dimensional flow field prediction result of the deep learning prediction model output at the next moment and the target flow field to form a flow field error distribution; Based on the flow field error distribution, a comprehensive control evaluation index is constructed. This evaluation index includes both the flow field deviation minimization target and the flow generation unit energy consumption constraint target. The flow field deviation is used to characterize the spatial consistency difference between the target flow field and the predicted flow field, and the energy consumption constraint is used to characterize the operational intensity limit of each layered flow generation unit. The control variables of each layered flow generation unit are taken as the optimization decision objects. Through iterative optimization of physical boundary constraints, the control evaluation index is gradually converged and solved so that the system obtains the optimal control solution under the conditions of power limitation, flow field stability constraint and boundary reflection suppression. The optimal control solution corresponds to the optimal control vector of each stratified flow-generating unit, which is used to generate variable frequency drive signals and water outlet angle control signals, thereby realizing dynamic approximation and real-time correction control of the target flow field.

5. The omnidirectional current generation control method for a circular marine engineering pool according to claim 4, characterized in that: Specifically, the control variables of each stratified flow-generating unit are used as the optimization decision objects. The rotational speed, outlet angle, and running time of each stratified flow-generating unit are used as control variables, and these control variables are used as the decision objects for optimization calculation. Based on this, control evaluation indicators are constructed according to the deviation between the target flow field and the predicted flow field, and the control evaluation indicators are gradually adjusted and converged through iterative optimization. In the optimization solution process, physical boundary constraints are set, including upper limit constraints on the power output of the flow generation unit, constraints on the suppression of reflection interference at the boundary of the pool, and constraints on maintaining the stability of the flow field. By continuously iteratively correcting the control variables under the premise of satisfying the above constraints, the system gradually obtains the optimal control solution that minimizes the flow field error, and the optimal control solution is used to generate the actual drive control commands for each layered flow generation unit.

6. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: The spatial error field in step S5 participates in predictive control in the following ways: The spatial error field constructed in step S5 is used as the feedback input information of the deep learning prediction model and fused with historical flow field time series data and real-time monitoring data so that the prediction model can simultaneously consider the actual flow field deviation when generating the three-dimensional flow field prediction result for the next moment. Furthermore, the spatial weights of the prediction results are adjusted based on the spatial error field, so that regions with larger errors receive higher correction priority during the prediction process, thereby improving the fit between the flow field prediction results and the actual flow state of the pool.

7. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: The spatial error field in step S5 is constructed by calculating the spatial difference between the target flow field and the actual flow field, and its expression is: in: For spatial error field; The target flow field vector; This is the actual measured flow field vector; Spatial error weighting coefficients are generated based on the absolute value distribution of the error field, which are used to adjust the influence weights of different control sub-regions in subsequent prediction and control calculations.

8. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: The step S6, predictive compensation-type advanced control correction, includes: adjusting the control vectors of each layered flow-generating unit in advance based on the future flow field evolution trend output by the deep learning prediction model, so that the control commands are applied to the flow field evolution process in advance; Simultaneously, the predicted control results are corrected by combining the spatial error field, and the influence of water energy attenuation law and pool wall boundary reflection disturbance is introduced to physically constrain and compensate the control signal, so as to reduce the impact of prediction deviation on flow field stability.

9. The omnidirectional current generation control method for a circular marine engineering pool according to claim 1, characterized in that: In step S6, during the predictive compensation-type advance control correction process, the future flow field state is based on the output of the deep learning prediction model. ), construct control correction quantity Its expression is: in: This is the pseudo-inverse of the flow field response matrix; To predict the compensation weight coefficient; For the target flow field; To measure the flow field in real time, )for Time and Space Point The predicted flow field vector.

Images