CN-122022359-A - Intelligent scheduling method for drainage pipe network based on AI

CN122022359ACN 122022359 ACN122022359 ACN 122022359ACN-122022359-A

Abstract

The invention discloses an intelligent scheduling method of a drainage pipe network based on AI, which comprises the steps of carrying out system deployment and environment configuration on cloud and edge nodes, carrying out multi-objective optimization model training and parameter configuration, integrating a prediction model and a digital twin model to provide pipe network real-time monitoring data and rainfall prediction data, and generating a global scheduling strategy through the multi-objective optimization model based on the pipe network real-time monitoring data and the rainfall prediction data to carry out end cloud collaborative scheduling operation. The method is suitable for the fine operation and maintenance management and control of the drainage pipe network in the scenes of urban flood control and drainage, pump station energy consumption optimization, ecological flow guarantee and the like.

Inventors

WANG HAO
ZHANG DEQUAN
WANG HAO

Assignees

上海中井汉鼎数字技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260209

Claims (10)

1. An AI-based intelligent scheduling method for a drainage pipe network is characterized by comprising the following steps: Performing system deployment and environment configuration on the cloud and edge nodes; Training a multi-objective optimization model and configuring parameters; integrating the prediction model and the digital twin model to provide pipe network real-time monitoring data and rainfall prediction data, and And generating a global scheduling strategy based on the pipe network real-time monitoring data and the rainfall prediction data through a multi-objective optimization model, and performing end cloud collaborative scheduling operation.
2. The method as recited in claim 1, further comprising: Based on the real-time monitoring index and a preset target value, adjusting model parameters of the multi-target optimization model, and optimizing a scheduling strategy; when network interruption is detected, the edge node automatically switches to a local autonomous scheduling mode, and a cached scheduling strategy and a local lightweight model are called to make independent decisions; After the network is restored, the edge node automatically synchronizes the scheduling log and the operation data in the off-line period to the cloud end, and the cloud end calibrates the global scheduling strategy based on the synchronization data.
3. The method of claim 1, wherein the system deployment and environment configuration at the cloud and edge nodes comprises: Respectively installing an operating system and a dependent tool of corresponding versions in a cloud dispatching center and an edge node, and configuring network communication and security authentication; deploying a database, and introducing basic data and equipment parameters of a network; Deploying a cloud multi-target optimization model and an edge lightweight model, and configuring an end cloud communication interface.
4. The method of claim 1, wherein the performing multi-objective optimization model training and parameter configuration comprises: Collecting historical dispatching data, pipe network monitoring data and meteorological data, constructing a training data set, and labeling; Initializing multi-objective optimization model parameters, setting a learning rate to be 0.001, setting an experience playback pool size to be 10000, setting a target network updating frequency to be 100 steps, and performing iterative training until a loss function value is less than or equal to 0.01; A multi-objective weight coefficient omega 1 、ω 2 、ω 3 is configured.
5. The method of claim 1, wherein the predictive models include an AI rainfall predictive model and a waterlogging risk predictive model, and the rainfall predictive data includes rainfall, rainfall duration, and waterlogging risk level; The pipe network real-time monitoring data comprise a water level threshold, a flow threshold and a waterlogging risk index threshold.
6. The method of claim 1, wherein the generating a global scheduling policy based on the pipe network real-time monitoring data and the rainfall prediction data through the multi-objective optimization model, and performing the end cloud collaborative scheduling operation comprises: the cloud scheduling center generates a global scheduling strategy to each edge node through a multi-objective optimization model based on the pipe network real-time monitoring data and the rainfall prediction data; The edge node receives the cloud scheduling instruction, combines the local monitoring data, carries out local correction through a lightweight model, generates a final control instruction, and sends the final control instruction to the execution terminal; the execution terminal responds to the control instruction, adjusts the running state and feeds back the execution result to the edge node.
7. The method of claim 1, wherein the multi-objective optimization model optimizes global scheduling policies by a multi-objective rewards function: ; Wherein the method comprises the steps of For the waterlogging risk index, the value range is 01, and the risk is higher as the value is closer to 1 waterlogging; is the rated power of the pump station, The actual power of the pump station; for the actual flow-down rate of the pipe network, Is an ecological flow threshold; ω 1 、ω 2 、ω 3 is a target weight coefficient and ω 1 +ω 2 +ω 3 =1; r is a comprehensive rewarding value, the value range is 0 to 1, and the closer to 1, the better the scheduling effect is.
8. The method of claim 6, further comprising calculating and constraining the scheduling instruction transmission delay by the formula: ; Wherein, the The time delay for coding the instruction is less than 1ms; the transmission delay is required to be less than 3ms for instructions; for instruction decoding delay, less than 0.5ms is required; for the terminal response delay, less than 0.5ms is needed, ≤5ms。
9. The method of claim 2, further comprising calculating outage scene scheduling reliability by the formula: ; Wherein the method comprises the steps of To schedule system uptime during a outage, Is the total time length of the network disconnection.
10. An AI-based drainage network intelligent scheduling system according to claims 1 to 9, comprising: the multi-objective optimization scheduling module is configured to construct a multi-objective optimization model, and comprehensive optimization is realized through a multi-objective rewarding function and a dynamic weight coefficient; the real-time linkage control module is configured to realize the cooperative linkage of the valve and the gate group through a scheduling instruction; the closed loop feedback module is configured to be connected with the prediction model and the digital twin model, monitor the scheduling effect in real time, dynamically feed back to the multi-objective optimal scheduling module and realize iterative optimization of the scheduling strategy; The end cloud cooperative scheduling module comprises: The cloud scheduling center is responsible for global multi-objective optimization scheduling; An edge node responsible for local autonomous scheduling; and the data interaction module is configured to realize data interaction and synchronization of the monitoring data, the prediction data and the scheduling instructions.

Description

Intelligent scheduling method for drainage pipe network based on AI Technical Field The invention relates to the technical field of urban drainage pipe network operation and maintenance, artificial intelligence, digital twin and edge cloud cooperation, in particular to an AI-based intelligent scheduling method for a drainage pipe network. Background With the continuous acceleration of the urban process and the frequent occurrence of extreme rainfall events, the dispatching efficiency and the running reliability of the urban drainage pipe network become core factors influencing urban flood control safety, ecological environment sustainability and energy conservation, and the intelligent dispatching level is directly related to the operation and maintenance quality of urban infrastructure. However, the current urban drainage pipe network scheduling technology still has a plurality of outstanding core technical bottlenecks and industry pain points, which severely restrict the intelligent operation and maintenance upgrade of a drainage system, and are specifically embodied in the following five aspects: First, the scheduling policy lacks global multi-objective optimization capabilities. In the prior art, a manual experience driving scheduling mode or a single threshold triggering scheduling mode is adopted, only a single core target for flood control and drainage is focused, and three core requirements of accurate control of the risk of waterlogging, energy saving of pump station operation and ecological flow guarantee of a river channel are not comprehensively considered. Due to the lack of scientific multi-objective optimization algorithm support, serious waste of non-effective energy consumption exists in the operation process of the pump station, actual measurement statistical data shows that the non-effective energy consumption of the pump station in the traditional scheduling mode accounts for more than 30%, and meanwhile, the ecological flow unbalance of a river channel is caused by excessive drainage in a part of areas, so that the derivative problems such as ecological damage of a water body and the like are caused. Secondly, the response of the dispatching command is lagged, and the linkage control capability is insufficient. In a traditional scheduling system, control instructions of execution terminals such as valves, gates and pump stations need to be forwarded through multiple levels, and obvious transmission delay exists, so that the response delay of waterlogging emergency scheduling is generally more than 10 minutes. In addition, a unified cooperative linkage control mechanism is lacking among all execution terminals, and the valve/gate group cannot be synchronously and dynamically regulated according to the real-time running state of the pipe network, so that the distribution of drainage capacity is unbalanced, and finally the contradiction and coexistence phenomenon of 'unsmooth drainage' of a part of areas and 'excessive drainage' of another part of areas occur. Third, there is a lack of closed-loop management mechanisms for "predict-schedule-feedback". The existing scheduling system, an AI prediction model (such as a rainfall prediction model and a waterlogging risk prediction model) and a digital twin simulation model are in independent running states of mutual fracture, so that prediction data cannot be fully utilized to pre-judge and formulate a targeted scheduling plan in advance, and scheduling effects cannot be fed back in real time and scheduling strategies cannot be optimized dynamically and iteratively through the digital twin model. The scheduling process presents an open-loop operation mode of 'one-time instruction issue', so that a scheduling strategy is seriously disjointed from the actual operation state of the pipe network, and the adaptability to the complex dynamic working condition of the pipe network is seriously insufficient. Fourth, the architecture is not suitable enough, and the extreme scene is not reliable enough. The existing scheduling technology is not effectively adapted to a Hongshan Meng Bianyuan end cloud collaborative architecture, a scheduling decision excessively depends on a cloud computing power support and network communication link, and an edge node is only used as a simple instruction execution unit and lacks of local autonomous decision scheduling capability. When the extreme abnormal conditions such as network interruption and cloud failure are encountered, the scheduling system is easy to run out, the basic running scheduling guarantee requirement of the drainage pipe network cannot be guaranteed, and the emergency disposal capacity of the drainage system is seriously affected. Fifth, prior art fragmentation lacks a unified integration scheme. The related art focuses on single function optimization of a dispatching link, a unified integrated dispatching scheme covering a whole process of prediction, decision-making, execution and feedback is not fo