CN-121980960-A - Big data-based automatic optimization method and system for water pump room

CN121980960ACN 121980960 ACN121980960 ACN 121980960ACN-121980960-A

Abstract

The invention discloses a big data-based automatic optimization method and system for a water pump room. The invention relates to the technical field of industrial automation optimization, in particular to a water pump house automation optimization method and system based on big data, which comprises the steps of firstly extracting mixed characteristics by combining time sequence, frequency domain transformation and a water pump physical model, constructing a time sequence knowledge graph, realizing the conversion from unstructured big data to structured knowledge through characteristic screening and few fault data enhancement, and tamping a characteristic basis of subsequent prediction; and finally, constructing a multi-objective optimization framework, optimizing pump set configuration parameters, simulating the robustness of an uncertain scene evaluation scheme, and enabling a dynamic parameter adjustment and optimization strategy to be more fit with actual operation requirements.

Inventors

WANG CHUAN
LIU CHAO
ZHANG JUNMEI
TANG BO
ZHANG YONGXIAO
SHI KAI
LIU FENG
MI CHENGLIANG
HAN XIANMING
LI LONG
WANG MINGQIN
WANG ANQIN
Wang Dengzhen

Assignees

徐州中矿慧鼎通信科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260225

Claims (7)

1. The big data-based automatic optimization method for the water pump room is characterized by comprising the following steps of: step S1, data acquisition, namely installing a multi-sensor at a key position of a water pump room, acquiring various multi-mode data, historical energy consumption and fault records, denoising the acquired data, performing normalized linear interpolation filling and anomaly detection by using a sliding window, and primarily identifying a potential fault mode by combining a self-encoder model; S2, constructing a knowledge graph, extracting mixed features from the preprocessed data, capturing related modes and trends through time sequence analysis, frequency domain transformation and a physical model, then constructing the time sequence knowledge graph, associating the extracted features with various entities and processing by using a graph embedding algorithm, and finally carrying out feature enhancement and screening key features; Step S3, training a water pump room performance prediction model, training a mixed neural network model combining a convolutional neural network and a converter, focusing key information by utilizing a multi-head attention mechanism, predicting the running state of the water pump room including real-time energy consumption and fault probability, and then verifying the model, wherein the method comprises the steps of Analyzing the characteristic contribution by value; And S4, optimizing algorithm design, formulating a multi-objective optimization framework, optimizing pump set configuration parameters by using a differential evolution algorithm, balancing total energy consumption and maintenance related cost, generating an uncertainty scene by combining a related sequence by using a Monte Carlo method, and evaluating the robustness and sensitivity of an optimization scheme.
2. The big data-based water pump house automation optimization method of claim 1, wherein in step S2, the knowledge graph is constructed, and the method specifically comprises the following steps: Step S21, extracting high-level characteristics from the preprocessed data, capturing vibration modes, flow fluctuation and energy consumption trend of the pump by using time sequence analysis and frequency domain transformation, wherein the time sequence analysis is specifically implemented by using Model fitting residual errors, transforming a frequency domain into fast Fourier transform, and generating mixed characteristics by combining a pump physical model of a Bernoulli equation; Step S22, constructing a knowledge graph of the water pump house, and associating the extracted features with the entities, wherein the entities comprise a pump assembly and a fault type, and a graph embedding algorithm is used Generating a low-dimensional representation, supporting semantic query, relationship reasoning and exception propagation analysis, and converting big data into dynamic structural knowledge; step S23, feature enhancement, using mutual information and Regression dynamic selection The characteristic features of the method are that, And few fault data are enhanced through generating an countermeasure network, so that the balance of a data set is improved, and a countermeasure learning simulation rare scene is introduced.
3. The automated big data based water pump house optimization method of claim 1, wherein in step S3, the training water pump house performance prediction model specifically comprises the following steps: Step S31, training a hybrid neural network model and combining Extracting time sequence local characteristics, Processing global sequence dependence, focusing key time points and cross-modal interaction by using a multi-head attention mechanism, and realizing the prediction of the running state of a water pump house, including real-time energy consumption and fault probability; Step S32, model verification, wherein characteristic contribution is analyzed by using SHAP values, and cross verification and challenge sample testing are combined.
4. The automated big data based water pump house optimization method of claim 1, wherein in step S4, the optimization algorithm design specifically comprises the following steps: Step 41, formulating a multi-objective optimization framework, optimizing pump set configuration parameters including rotation speed, maintenance interval, balance energy consumption and efficiency by using a differential evolution algorithm and prediction uncertainty, introducing a hybrid evolution strategy and uncertainty guidance, and optimizing an objective; step S42, simulating an optimized scene, and combining the method by using a Monte Carlo method The sequence generates an uncertainty scene, evaluates the robustness and sensitivity of the optimization scheme, and adjusts the hyper-parameters using meta-learning to maximize the desired utility.
5. The big data based water pump house automation optimization method of claim 1, wherein in step S1, the data acquisition specifically comprises the following steps: S11, constructing a sensor network, wherein the sensor network comprises a pump body, a pipeline, a motor and a control cabinet, wherein a plurality of sensors are arranged at key positions of a water pump room, and multi-mode data are acquired; And step S12, preprocessing the data, carrying out normalization, linear interpolation filling and anomaly detection on the acquired data by using a sliding window, and primarily identifying a potential failure mode by combining a self-encoder model.
6. The big data-based water pump room automatic optimization system is used for realizing the big data-based water pump room automatic optimization method according to any one of claims 1-5 and is characterized by comprising a data acquisition module, a knowledge graph construction module, a training water pump room performance prediction model module and an optimization algorithm design module.
7. The big data-based water pump house automation optimization system of claim 6, wherein the data acquisition module is provided with a multi-sensor at a key position of the water pump house, acquires a plurality of multi-mode data, historical energy consumption and fault records, performs denoising processing on the acquired data, performs normalized linear interpolation filling and anomaly detection by using a sliding window, preliminarily identifies potential fault modes by combining a self-encoder model, and sends the data to the knowledge graph construction module; The system comprises a knowledge graph constructing module, a data acquisition module, a training water pump house performance prediction model module, a time sequence analysis module, a frequency domain transformation module, a physical model capturing related modes and trends, a time sequence knowledge graph constructing module, a graph embedding algorithm processing module and a data processing module, wherein the knowledge graph constructing module receives data sent by the data acquisition module, extracts mixed features from the preprocessed data, associates the extracted features with various entities, processes the extracted features by using the graph embedding algorithm, and finally carries out feature enhancement, screens key features and sends the data to the training water pump house performance prediction model module; The water pump house performance prediction model training module receives the data sent by the knowledge graph construction module, trains a mixed neural network model combining a convolutional neural network and a converter, the key information is focused by utilizing a multi-head attention mechanism, the running state of the water pump room including real-time energy consumption and fault probability is predicted, and then the model is verified, and the model is verified by The value analysis feature contribution is carried out, and data are sent to an optimization algorithm design module; The optimization algorithm design module receives data sent by the training water pump house performance prediction model module, formulates a multi-objective optimization framework, optimizes pump set configuration parameters by using a differential evolution algorithm, balances total energy consumption and maintenance related cost, generates an uncertainty scene by combining a related sequence by using a Monte Carlo method, and evaluates the robustness and sensitivity of an optimization scheme.

Description

Big data-based automatic optimization method and system for water pump room Technical Field The invention relates to the technical field of industrial automation optimization, in particular to a big data-based water pump house automation optimization method and system. Background Under the large trend of industrial automation and intelligent development, intelligent operation and maintenance of various infrastructures becomes an important industrial development point, and a water pump house is used as a core power facility in the fields of water conservancy, construction, industrial production and the like, so that the operation efficiency, energy consumption control and fault early warning capability of the intelligent operation and maintenance system directly influence the stability and energy saving effect of the whole system. The rapid development of technologies such as big data, machine learning, knowledge patterns and the like provides technical support for automatic optimization of the water pump room, the requirements of realizing data-driven state sensing, accurate prediction and intelligent configuration optimization of the water pump room in industry are increasingly urgent, the energy conservation and consumption reduction of the water pump room, fault shutdown reduction and operation and maintenance efficiency improvement are realized through technical means, and the method becomes a core prospect of automatic development of the water pump room. The related technology of the current water pump room automation still has a plurality of problems, in the data level, the processing of water pump room operation data stays in an unstructured analysis stage, the association of data with equipment entities and fault types is weak, the problem of unbalanced fault data distribution is not solved effectively, efficient relation reasoning and knowledge transformation cannot be achieved, single models are adopted for analysis, local characteristic and global sequence dependence of time sequence data are difficult to capture simultaneously, cross-modal interaction characteristic mining is insufficient, model interpretability is poor, reliability and accuracy of a prediction result are required to be improved, in the configuration optimization level, single targets are taken as optimization directions, energy consumption control and equipment maintenance cost cannot be considered, various uncertainty factors in actual operation are not considered in the optimization process, robustness of an optimization scheme is insufficient, scientific basis and dynamic mechanism are lacked in super-parameter adjustment, and the actual operation scene of the water pump room is difficult to adapt. Disclosure of Invention Aiming at the problems that in the prior art, operation data are mostly unstructured time sequence data, feature extraction lacks physical principle support, and effective relation reasoning is difficult to develop due to weak relevance between the data and equipment entities, meanwhile, fault data distribution is unbalanced, critical feature screening lacks precision, the method and the system are used for optimizing a mixed feature, a relation, entity and timestamp time sequence knowledge map is established by combining time sequence analysis, frequency domain transformation and a water pump physical model from preprocessing data, node low-dimensional representation is generated by utilizing a graph embedding algorithm, then critical features are dynamically screened by means of mutual information and lasso regression, and transformation from unstructured big data to dynamic structural knowledge is realized by combining anti-learning enhancement of few kinds of fault data, and aiming at the problems that in the prior art, the performance prediction of the water pump room is difficult to simultaneously capture local features and global sequence dependence of the time sequence data, the model has poor interpretability, the problem that the prediction precision and robustness are insufficient due to simple verification, the combined characteristic is optimized by means of the aid of simple verification, the combined prediction precision and the fact that the mixed feature is insufficient by means of simple verification, the mixed feature is fully optimized by means of the aid of the fuzzy feature, the optimal performance of the neural pump network, the optimal performance is fully optimized by means of the fact that the optimal performance of the mixed feature is insufficient in the time-domain transformation, and the water pump has low-cost, and the water pump performance is fully optimized by means of the fact that the optimal performance of the water pump has low-down-cost, and the performance of the water pump has low-level, and the performance of the water pump system, the scheme optimizes the configuration parameters such as the rotation speed of the pump set, the maintenance interval and the like by c