CN-121616118-B - Hydraulic engineering management method and system based on deep learning

CN121616118BCN 121616118 BCN121616118 BCN 121616118BCN-121616118-B

Abstract

The invention discloses a hydraulic engineering management method and a system based on deep learning, and relates to the technical field of data management.A cofferdam, a gate, a pump station, a construction surface and personnel states are uniformly managed by utilizing linkage state diagram data in a hydraulic junction scene where construction and local test operation coexist, controllable parameter data is obtained through perturbation identification, and water level envelope data and risk zone data are predicted and output by a nerve operator; and combining the evaluation function data to automatically generate the optimal action sequence data meeting the hard constraint, and simultaneously providing the anti-fact evidence chain data and the audit log data, so that the quick linkage treatment, the interpretable comparison and the traceable recovery are realized, the risk of overtaking and evacuation deficiency is reduced, and the blind shutdown is reduced.

Inventors

CHI HAO
QIU YANAN
JING LI
GAO TIANYU
BAO LI
CHANG LEI
Hou Junjiao
YAO XUEJIE

Assignees

湖南诚德建设有限公司

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (9)

1. The hydraulic engineering management method based on deep learning is characterized by comprising the following steps of: Step S1, acquiring hydraulic engineering object state data and constructing linkage state diagram data, wherein the hydraulic engineering object state data comprises cofferdam state data, gate state data, pump station state data, construction surface state data and personnel state data, and writing the hydraulic engineering object state data into a database; step S2, generating and executing perturbation control action data under the constraint of the linkage state diagram data, collecting control response data, inputting the control response data into a depth state space network, outputting controllable parameter data and writing the controllable parameter data into a database; S3, obtaining boundary condition data, inputting the boundary condition data and the controllable parameter data into a nerve operator prediction network, outputting water level envelope data and risk zone data, and writing the water level envelope data and the risk zone data into a database; s4, constructing evaluation function data based on the risk zone data, the construction progress data and the quality constraint data, and writing the evaluation function data into a database; S5, inputting the linkage state diagram data, the evaluation function data and the risk zone data into a constraint generation type arrangement network, outputting candidate action sequence data, evaluating the candidate action sequence data and outputting optimal action sequence data, and generating counterfactual evidence chain data; step S6, sending the optimal action sequence data to a control end and a command end for execution, and writing the optimal action sequence data, the anti-facts evidence chain data and the corresponding time stamp data into audit log data; Step S4 comprises the following sub-steps: Step S401, construction progress data and quality constraint data are obtained, wherein the construction progress data comprise critical path procedure duration data and resource occupation data, and the quality constraint data comprise concrete temperature control threshold data and maintenance continuity threshold data; In step S401, construction progress data and quality constraint data are obtained, key path procedure time length data and resource occupation data are exported from a construction planning system and written into a database, and concrete temperature control threshold value data and maintenance continuity threshold value data are set by a quality management standard or a laboratory and written into the database; Step S402, safety cost data are built based on the risk zone data, progress cost data are built based on construction progress data, and quality cost data are built based on quality constraint data; In step S402, safety cost data is constructed based on the risk zone data, and the safety cost data can be defined by using expected representation of constraint violation intensity: ; Wherein, the Corresponding to the upper limit constraint of the cofferdam water level, Outputting water level envelope data for a network Is used as a base material for the vehicle, Constraining the corresponding water level for the upper limit of the water level at the upstream of the cofferdam; The corresponding person is evacuated from the minimum time constraint, The time required for the evacuation of the personnel is calculated by the personnel position data and the personnel evacuation channel state data, Is the risk lead; Construction progress cost data based on construction progress data The critical path delay is defined as: ; Wherein, the Re-estimating the working procedure time length caused by construction working procedure switching action data; constructing quality cost data based on the quality constraint data, and defining the concrete temperature control overrun strength as: ; Wherein, the Is obtained by on-site temperature control collection, In order to maintain the length of time for the interruption, Maintenance continuity threshold data; step S403, fusing the safety cost data, the progress cost data and the quality cost data to generate evaluation function data, and writing the evaluation function data into a database; In step S403, the safety cost data, the progress cost data and the quality cost data are fused to generate evaluation function data, where the evaluation function data may be expressed as: ; Wherein, the 、 And Is weighted and written into the database.
2. The deep learning-based hydraulic engineering management method according to claim 1, wherein the step S1 includes the following sub-steps: S101, collecting cofferdam state data, wherein the cofferdam state data comprises water level data at the upstream of the cofferdam, water level data at the downstream of the cofferdam, cofferdam osmotic pressure data and cofferdam displacement data; step S102, gate state data are collected, wherein the gate state data comprise gate opening data, gate opening and closing instruction data, gate displacement feedback data and gate driving current data; step S103, collecting pump station state data, wherein the pump station state data comprises pump set start-stop state data, pump set rotating speed data, pump set outlet flow data and pump set motor temperature rise data Step S104, collecting construction surface state data and personnel state data, wherein the construction surface state data comprises construction process state data, construction equipment occupation state data and construction material presence data, and the personnel state data comprises personnel position data and personnel evacuation channel state data; Step S105, linkage state diagram data are built according to cofferdam state data, gate state data, pump station state data, construction surface state data and personnel state data, and the linkage state diagram data are written into a database.
3. The deep learning-based hydraulic engineering management method according to claim 2, wherein the step S2 includes the following sub-steps: Step S201, calculating perturbation safety boundary data based on linkage state diagram data, wherein the perturbation safety boundary data comprises cofferdam upstream water level safety margin data, gate opening and closing speed upper limit data and pump group current upper limit data; step S202, generating perturbation control action data under the constraint of perturbation safety boundary data, wherein the perturbation control action data comprises gate pulse opening action data and pump group step start-stop action data; step S203, performing perturbation control action data and collecting control response data, wherein the control response data comprises water level change rate data, outflow change data, gate displacement hysteresis data and pump group efficiency change characteristic data; step S204, the control response data is written into a database and is input into a deep state space network, and controllable parameter data is output.
4. The deep learning-based hydraulic engineering management method according to claim 3, wherein the controllable parameter data includes equivalent roughness parameter data, local loss parameter data, gate hysteresis parameter data and pump group efficiency attenuation parameter data, and the processing logic of the deep state space network is as follows: Performing time sequence coding on the control response data to obtain state embedded data; Outputting the equivalent roughness parameter data, the local loss parameter data, the gate hysteresis parameter data and the pump group efficiency attenuation parameter data based on the state embedded data; And the output controllable parameter data and the corresponding perturbation control action data are associated and written into a database for the input of a follow-up nerve operator prediction network.
5. The deep learning-based hydraulic engineering management method as claimed in claim 4, wherein the step S3 includes the sub-steps of: step S301, boundary condition data is obtained, wherein the boundary condition data comprises upstream water incoming data, rainfall radar data and hydraulic scheduling constraint data, and the boundary condition data is written into a database; Step S302, boundary condition data and controllable parameter data are input into a nerve operator prediction network, and water level envelope data and risk zone data of a future period are output; Step S303, physical constraint losses are applied to the nerve operator prediction network, wherein the physical constraint losses comprise flow continuity constraint losses and water level monotonicity constraint losses, so that water level envelope data and risk zone data meet preset engineering feasibility constraints.
6. The method for managing hydraulic engineering based on deep learning according to claim 5, wherein the step S5 comprises the following sub-steps: Step S501, determining action set data based on linkage state diagram data, wherein the action set data comprises gate opening adjustment action data, pump group start-stop action data, construction procedure switching action data and personnel evacuation action data; step S502, inputting action set data, evaluation function data and risk zone data into a constraint generation type arrangement network, and outputting candidate action sequence data with time labels; Step S503, screening out the hard constraints of the candidate action sequence data, wherein the hard constraints comprise upper limit constraints of water level at the upstream of the cofferdam, minimum time constraints of personnel evacuation and upper limit constraints of equipment working conditions; Step S504, performing evaluation function calculation on the screened candidate action sequence data, selecting candidate action sequence data with the optimal evaluation value, and outputting the candidate action sequence data as optimal action sequence data.
7. The deep learning-based hydraulic engineering management method as set forth in claim 6, wherein the generating logic of the feedback evidence chain data is: Generating at least one group of comparison action sequence data by taking the optimal action sequence data as target sequence data, wherein the comparison action sequence data is selected from candidate action sequence data or regenerated by a constraint generation type arrangement network under the same hard constraint; Respectively inputting the target sequence data and the comparison action sequence data into a nerve operator prediction network to obtain corresponding risk zone data; and calculating the risk increment data and the cost increment data of the target sequence data relative to the comparison action sequence data, and organizing the risk increment data and the cost increment data into anti-fact evidence chain data.
8. The deep learning-based hydraulic engineering management method according to claim 7, wherein the audit log data at least comprises trigger cause data, perturbation control action data, control response data, controllable parameter data, water level envelope data, risk zone data, candidate action sequence data, optimal action sequence data and anti-facts evidence chain data, and the data is written into a database after adding timestamp data and version number data to the data respectively.
9. The hydraulic engineering management system based on the deep learning is applied to the hydraulic engineering management method based on the deep learning as claimed in any one of claims 1 to 8, and is characterized by comprising a state diagram construction module, a perturbation identification module, a risk prediction module, an evaluation function module, a sequence arrangement module and an execution audit module; The state diagram construction module is used for collecting hydraulic engineering object state data and constructing linkage state diagram data, wherein the hydraulic engineering object state data comprises cofferdam state data, gate state data, pump station state data, construction surface state data and personnel state data, and the hydraulic engineering object state data is written into the database; The perturbation identification module is used for generating perturbation control action data under the constraint of the linkage state diagram data and executing the perturbation control action data, collecting control response data, inputting the control response data into the deep state space network, outputting controllable parameter data and writing the controllable parameter data into the database; the risk prediction module is used for acquiring boundary condition data, inputting the boundary condition data and the controllable parameter data into the nerve operator prediction network, outputting water level envelope data and risk zone data, and writing the water level envelope data and the risk zone data into the database; The evaluation function module is used for constructing evaluation function data based on the risk zone data, the construction progress data and the quality constraint data, and writing the evaluation function data into a database; The sequence arrangement module is used for inputting the linkage state diagram data, the evaluation function data and the risk zone data into the constraint generation type arrangement network, outputting candidate action sequence data, evaluating the candidate action sequence data and outputting optimal action sequence data, and generating the counterfactual evidence chain data; And the execution audit module is used for sending the optimal action sequence data to the control end and the command end for execution, and writing the optimal action sequence data, the anti-facts evidence chain data and the corresponding time stamp data into audit log data.

Description

Hydraulic engineering management method and system based on deep learning Technical Field The invention relates to the technical field of data management, in particular to a hydraulic engineering management method and system based on deep learning. Background In the stage of coexistence of construction and local test running, cofferdam water blocking, diversion tunnel overflow, gate debugging, pump station drainage and concrete pouring temperature control maintenance are often carried out simultaneously, and the short duration heavy rainfall in the flood season can enable upstream inflow to rapidly rise within 1-3 hours. At the moment, engineering management not only judges whether the water level at the upstream of the cofferdam approaches the upper limit, but also synthesizes the gate opening and closing speed, pump group availability, construction procedure switching cost and personnel evacuation channel state, and an executable linkage instruction is formed in a limited time. The existing scheme depends on threshold value alarm, manual consultation or off-line hydraulic simulation, is difficult to cope with controllable change caused by equipment and working condition drift in the construction period, and is also difficult to output auditable linkage action sequences under multi-scheme comparison, so that the problems of excessive shutdown conservation or concurrent risk of disposal lag are caused. At present, china patent application No. CN202510227875.3 discloses a BIM-based hydraulic engineering monitoring data management method, which is used for analyzing stability of hydraulic engineering data on a time sequence of a monitoring point, primarily screening stable reference points, further analyzing consistency of the stable reference points and local reference points at all times, screening available reference points, simultaneously considering aggregation conditions of the screened available reference points in regional distribution density to determine final reference points, and finally combining the unstable reference points and the reduced final reference points for compression storage. In the hydraulic junction scene of coexistence of construction and local test running and parameter drift, the technology is difficult to identify controllable parameters in real time and predict risk zones based on cofferdam state data, gate state data and pump station state data, and further linkage action sequence data and counterfactual evidence chain data meeting hard constraints are automatically generated. Disclosure of Invention The method solves the technical problems that in the prior art, under a water junction scene with concurrent construction and local test running and parameter drift, controllable parameters are identified in real time and risk zones are predicted based on cofferdam state data, gate state data and pump station state data, and then linkage action sequence data and counterfactual evidence chain data meeting hard constraint are automatically generated. In order to solve the technical problems, the invention provides the following technical scheme: the hydraulic engineering management method based on deep learning comprises the following steps: Step S1, acquiring hydraulic engineering object state data and constructing linkage state diagram data, wherein the hydraulic engineering object state data comprises cofferdam state data, gate state data, pump station state data, construction surface state data and personnel state data, and writing the hydraulic engineering object state data into a database; step S2, generating and executing perturbation control action data under the constraint of the linkage state diagram data, collecting control response data, inputting the control response data into a depth state space network, outputting controllable parameter data and writing the controllable parameter data into a database; S3, obtaining boundary condition data, inputting the boundary condition data and the controllable parameter data into a nerve operator prediction network, outputting water level envelope data and risk zone data, and writing the water level envelope data and the risk zone data into a database; s4, constructing evaluation function data based on the risk zone data, the construction progress data and the quality constraint data, and writing the evaluation function data into a database; S5, inputting the linkage state diagram data, the evaluation function data and the risk zone data into a constraint generation type arrangement network, outputting candidate action sequence data, evaluating the candidate action sequence data and outputting optimal action sequence data, and generating counterfactual evidence chain data; step S6, sending the optimal action sequence data to a control end and a command end for execution, and writing the optimal action sequence data, the anti-facts evidence chain data and the corresponding time stamp data into audit log data; Step S4 comprises the