Water pump house automatic optimization method and system based on big data

Abstract

The invention discloses a water pump house automatic optimization method and system based on big data. The method comprises the steps of data collection, knowledge graph construction, water pump house performance prediction model training and optimization algorithm design. The invention relates to the technical field of industrial automation optimization, in particular to a water pump room automation optimization method and system based on big data. According to the scheme, mixed features are extracted in combination with time sequence, frequency domain transformation and a water pump physical model, a time sequence knowledge graph is constructed, and the water pump room automation optimization method based on big data is obtained through feature screening and minority class fault data enhancement; the conversion from unstructured big data to structured knowledge is realized, and the feature basis of subsequent prediction is tamped; then a hybrid neural network model is trained, the energy consumption and fault probability of the water pump house are jointly predicted, the prediction precision is improved through multi-dimensional verification, and the model interpretability is also enhanced; and finally, constructing a multi-objective optimization framework, optimizing configuration parameters of the pump set, simulating robustness of an uncertain scene evaluation scheme, and dynamically adjusting parameters to enable an optimization strategy to better meet actual operation requirements.

Description

Technical Field

[0001] This invention relates to the field of industrial automation optimization technology, specifically to a method and system for the automation optimization of water pump stations based on big data. Background Technology

[0002] Under the general trend of industrial automation and intelligent development, intelligent operation and maintenance of various infrastructures has become a key focus of industry development. As a core power facility in water conservancy, construction, and industrial production, the operating efficiency, energy consumption control, and fault early warning capabilities of pump stations directly affect the stability and energy-saving effect of the overall system. The rapid development of technologies such as big data, machine learning, and knowledge graphs provides technical support for the automation optimization of pump stations. The industry's demand for data-driven status perception, accurate prediction, and intelligent configuration optimization of pump stations is increasingly urgent. Achieving energy conservation and consumption reduction, reducing downtime, and improving operation and maintenance efficiency through technological means has become the core vision for the automation development of pump stations.

[0003] Current technologies related to pump station automation still face numerous challenges. At the data level, the processing of pump station operational data largely remains at the stage of unstructured analysis. The correlation between data and equipment entities and fault types is weak, and the problem of unbalanced fault data distribution has not been effectively addressed, hindering efficient relational reasoning and knowledge transformation. Analysis often employs a single model, making it difficult to simultaneously capture local features and global sequence dependencies in time-series data. Cross-modal interaction feature mining is insufficient, and model interpretability is poor, resulting in unreliable and inaccurate predictions. At the configuration optimization level, optimization often focuses on a single objective, failing to consider both energy consumption control and equipment maintenance costs. Furthermore, the optimization process does not take into account various uncertainties in actual operation, leading to insufficient robustness of optimization schemes. Hyperparameter adjustments lack scientific basis and dynamic mechanisms, making it difficult to adapt to the actual operating scenarios of pump stations. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an automated optimization method and system for pumping stations based on big data. Addressing the problems in existing pumping station data analysis, such as unstructured time-series data, lack of physical principle support for feature extraction, weak data-equipment entity relationships hindering effective relational reasoning, and uneven distribution of fault data with inaccurate key feature selection, this solution extracts hybrid features from preprocessed data using time-series analysis, frequency domain transformation, and pump physical models. It constructs a time-series knowledge graph integrating features, relationships, entities, and timestamps, and uses graph embedding algorithms to generate low-dimensional node representations. Then, it dynamically selects key features using mutual information and lasso regression, and combines adversarial learning to enhance minority class fault data, thus achieving the transformation of unstructured big data into dynamic structured knowledge. Furthermore, existing pumping station performance prediction methods often employ single models, making it difficult to simultaneously capture local features and global sequence dependencies in time-series data. Cross-modal interaction feature mining is insufficient, and model interpretability is poor, relying solely on simple indicators for verification. To address the issues of insufficient prediction accuracy and robustness, this solution trains a hybrid neural network model integrating convolutional neural networks and transformers. Utilizing a multi-head attention mechanism, it focuses on key time points and cross-modal interaction information to jointly predict real-time energy consumption and failure probability in pumping stations, quantifying the contribution of each feature. Furthermore, existing pumping station configuration optimizations are often single-objective, failing to adequately balance energy consumption and maintenance costs. The optimization process often neglects various uncertainties in actual operation, leading to insufficient robustness. Additionally, the lack of scientific basis for hyperparameter adjustment results in poor applicability in real-world scenarios. This solution constructs a multi-objective optimization framework without dominated ranking, combining differential evolution algorithms to optimize pumping speed, maintenance intervals, and other configuration parameters, balancing total energy consumption and maintenance costs due to failures. Monte Carlo methods, combined with low-difference sequences and Latin hypercube sampling, are used to generate uncertain scenarios to evaluate the robustness and sensitivity of the optimization scheme. Meta-learning dynamically adjusts hyperparameters and sets a backoff mechanism, achieving accurate multi-objective optimization of pumping station configuration.

[0005] The technical solution adopted in this invention is as follows: an automated optimization method for water pump stations based on big data, which includes the following steps:

[0006] Step S1: Data acquisition. Multiple sensors are installed at key locations in the pump room to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. Potential fault modes are initially identified by combining the autoencoder model.

[0007] Step S2: Construct a knowledge graph, extract hybrid features from the preprocessed data, capture relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then construct a time series knowledge graph, associate the extracted features with various entities and process them using graph embedding algorithms, and finally perform feature enhancement and select key features;

[0008] Step S3: Train the pump house performance prediction model. Train a hybrid neural network model combining convolutional neural networks and converters, using a multi-head attention mechanism to focus on key information, and predict the pump house's operating status, including real-time energy consumption and failure probability. Then, validate the model. Value analysis of feature contribution;

[0009] Step S4: Optimize the algorithm design, formulate a multi-objective optimization framework, use the differential evolution algorithm to optimize the pump group configuration parameters, balance the total energy consumption and maintenance-related costs, and then use the Monte Carlo method to generate uncertainty scenarios in combination with relevant sequences to evaluate the robustness and sensitivity of the optimization scheme.

[0010] Further, in step S1, the data acquisition specifically includes the following steps:

[0011] Step S11: Construct a sensor network by installing multiple sensors at key locations in the pump house, including the pump body, pipelines, motors, and control cabinet, to collect multimodal data.

[0012] Step S12: Data preprocessing. The collected data is normalized, linearly interpolated and filled, and anomaly detection is performed using a sliding window. Potential fault modes are initially identified by combining the autoencoder model.

[0013] Furthermore, in step S2, the construction of the knowledge graph specifically includes the following steps:

[0014] Step S21: Extract high-level features. High-level features are extracted from the preprocessed data. Time-series analysis and frequency domain transformation are used to capture the pump's vibration modes, flow fluctuations, and energy consumption trends. Specifically, the time-series analysis involves using… The model fitting residuals are transformed into a fast Fourier transform in the frequency domain, and a hybrid feature is generated by combining the pump physical model with the Bernoulli equation.

[0015] Step S22: Construct a knowledge graph for the pump house, associating extracted features with entities, including: pump components and fault types; use a graph embedding algorithm. Generate low-dimensional representations to support semantic queries, relational reasoning, and anomaly propagation analysis, transforming big data into dynamic structured knowledge;

[0016] Step S23: Feature enhancement, using mutual information and Regression Dynamic Selection feature, Furthermore, it enhances minority class fault data by generating adversarial networks, improves dataset balance, and introduces adversarial learning to simulate rare scenarios.

[0017] Furthermore, in step S3, the training of the pump house performance prediction model specifically includes the following steps:

[0018] Step S31: Train the hybrid neural network model, combining Extracting temporal local features, To handle global sequence dependencies, a multi-head attention mechanism is used to focus on key time points and cross-modal interactions, enabling the prediction of pump station operating status, including real-time energy consumption and failure probability; multi-modal learning and self-attention are integrated.

[0019] Step S32: Model validation, using SHAP values ​​to analyze feature contributions, combined with cross-validation and adversarial example testing.

[0020] Furthermore, in step S4, the optimization algorithm design specifically includes the following steps:

[0021] Step S41: Develop a multi-objective optimization framework, use differential evolution algorithm and predict uncertainty to optimize pump set configuration parameters, including speed, maintenance interval, balance energy consumption and efficiency, introduce hybrid evolution strategy and uncertainty guidance, and optimize objectives;

[0022] Step S42: Simulate and optimize the scenario, using the Monte Carlo method combined with In a scenario of uncertain sequence generation, the robustness and sensitivity of the optimization scheme are evaluated, and meta-learning is used to adjust the hyperparameters to maximize the expected utility.

[0023] The automated optimization system for pumping stations based on big data provided by this invention includes a data acquisition module, a knowledge graph construction module, a pumping station performance prediction model training module, and an optimization algorithm design module.

[0024] The data acquisition module installs multiple sensors at key locations in the pump house to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. The potential fault modes are initially identified by combining the autoencoder model, and the data is sent to the knowledge graph construction module.

[0025] The knowledge graph construction module receives data sent by the data acquisition module, extracts mixed features from the preprocessed data, captures relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then constructs a time series knowledge graph, associates the extracted features with various entities and processes them using graph embedding algorithms, finally performs feature enhancement, filters key features, and sends the data to the pump house performance prediction model training module.

[0026] The module for training the pump house performance prediction model receives data from the knowledge graph construction module, trains a hybrid neural network model combining convolutional neural networks and a transducer, and uses a multi-head attention mechanism to focus on key information to predict the pump house's operating status, including real-time energy consumption and failure probability. The model is then validated. The value analysis features contribute and the data is sent to the optimization algorithm design module;

[0027] The optimization algorithm design module receives data sent by the training pump house performance prediction model module, formulates a multi-objective optimization framework, uses the differential evolution algorithm to optimize the pump group configuration parameters, balances total energy consumption and maintenance-related costs, and then uses the Monte Carlo method combined with relevant sequences to generate uncertainty scenarios to evaluate the robustness and sensitivity of the optimization scheme.

[0028] The beneficial effects achieved by the present invention using the above solution are as follows:

[0029] (1) In the existing pump room data analysis, the operating data is mostly unstructured time series data, the feature extraction lacks physical principle support, and the data has a weak connection with equipment entities, making it difficult to carry out effective relationship reasoning. At the same time, the fault data is unevenly distributed and the key feature selection lacks accuracy. This solution extracts hybrid features from the preprocessed data by combining time series analysis, frequency domain transformation and pump physical model, constructs a time series knowledge graph that integrates features, relationships, entities and timestamps, and uses graph embedding algorithm to generate low-dimensional node representations. Then, it uses mutual information and lasso regression to dynamically select key features and combines adversarial learning to enhance minority class fault data. This realizes the transformation of unstructured big data into dynamic structured knowledge, strengthens the semantic association and relationship reasoning ability between features and equipment entities, accurately selects high-value features and improves the balance of the dataset, and lays a structured and high-quality feature foundation for subsequent performance prediction.

[0030] (2) In view of the problems that existing pump house performance prediction mostly adopts a single model, which makes it difficult to capture the local features and global sequence dependencies of time series data at the same time, the cross-modal interaction feature mining is insufficient, and the model interpretability is poor and the prediction accuracy and robustness are insufficient due to the reliance on simple index verification, this solution trains a hybrid neural network model that integrates convolutional neural network and converter, and uses a multi-head attention mechanism to focus on key time points and cross-modal interaction information to achieve joint prediction of real-time energy consumption and failure probability of pump house. The contribution of each feature is quantitatively analyzed, and the model is verified in multiple dimensions by combining cross-validation and adversarial sample testing. This effectively captures the local and global features of time series data, fully explores the interaction value of multimodal data, and greatly improves the accuracy and precision of pump house operation status prediction. At the same time, the contribution of features to the prediction results is quantified, enhancing the interpretability of the model. Multi-dimensional verification further ensures the robustness and effectiveness of the model in practical applications.

[0031] (3) In view of the fact that the existing pump configuration optimization of water pumping stations is mostly single-objective optimization, which fails to fully balance energy consumption and maintenance costs, and the optimization process does not consider various uncertain factors in actual operation, resulting in insufficient robustness of the optimization scheme, and the lack of scientific basis for hyperparameter adjustment makes the scheme less applicable in actual scenarios, this scheme constructs a multi-objective optimization framework without dominant sorting, combines differential evolution algorithm to optimize configuration parameters such as pump speed and maintenance interval, balances total energy consumption and maintenance costs caused by failure, and then uses Monte Carlo method combined with low difference sequence and Latin hypercube sampling to generate uncertain scenarios to evaluate the robustness and sensitivity of the optimization scheme. Through meta-learning, the hyperparameters are dynamically adjusted and a backoff mechanism is set to realize multi-objective accurate optimization of pump configuration, fully taking into account the actual needs of energy consumption control and equipment maintenance, effectively improving the robustness and anti-interference ability of the optimization scheme in complex uncertain scenarios, and the dynamic hyperparameter adjustment and backoff mechanism ensure the maximization of the expected utility of the optimization scheme, making the optimization strategy more in line with the actual operation needs of the water pumping station. Attached Figure Description

[0032] Figure 1 A schematic diagram illustrating the automated optimization method for water pump stations based on big data provided by this invention;

[0033] Figure 2 A schematic diagram of the automated optimization system for pumping stations based on big data provided by the present invention;

[0034] Figure 3 This is a schematic diagram of step S1;

[0035] Figure 4 This is a schematic diagram of step S2;

[0036] Figure 5 This is a schematic diagram of step S3;

[0037] Figure 6 This is a schematic diagram of step S4.

[0038] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0041] Example 1, see Figure 1 The present invention provides an automated optimization method for water pumping stations based on big data, which includes the following steps:

[0042] Step S1: Data acquisition. Multiple sensors are installed at key locations in the pump room to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. Potential fault modes are initially identified by combining the autoencoder model.

[0043] Step S2: Construct a knowledge graph, extract hybrid features from the preprocessed data, capture relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then construct a time series knowledge graph, associate the extracted features with various entities and process them using graph embedding algorithms, and finally perform feature enhancement and select key features;

[0044] Step S3: Train the pump house performance prediction model. Train a hybrid neural network model combining convolutional neural networks and converters, using a multi-head attention mechanism to focus on key information, and predict the pump house's operating status, including real-time energy consumption and failure probability. Then, validate the model. Value analysis of feature contribution;

[0045] Step S4: Optimize the algorithm design, formulate a multi-objective optimization framework, use the differential evolution algorithm to optimize the pump group configuration parameters, balance the total energy consumption and maintenance-related costs, and then use the Monte Carlo method to generate uncertainty scenarios in combination with relevant sequences to evaluate the robustness and sensitivity of the optimization scheme.

[0046] Example 2, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S1, the data acquisition specifically includes the following steps:

[0047] Step S11: Construct a sensor network by installing multiple sensors at key locations in the pump house, including the pump body, pipelines, motors, and control cabinet, to collect multimodal data, including pump speed, vibration frequency, inlet and outlet pressure, flow rate, ambient temperature, humidity, current, and voltage. Simultaneously, collect historical energy consumption and fault records for the pump house. Fault records are represented by binary labels of 0 or 1, as shown below:

[0048] ;

[0049] in, Indicates time The aggregated data vector, with the dimension being the number of sensors. ; Indicates the first The raw readings of each sensor; Indicates sensor The calibration weights, with a preset range of [value missing]. ; The random noise term follows a Gaussian distribution. , Variance estimation using historical data; This represents the denoised data, obtained by minimizing the difference between the original data and the mean of the sliding window. The squared error plus Norm gradient regularization term Optimize usage Optimizer, initial learning rate Iterate until the error is less than the threshold. Or a maximum of 50 iterations;

[0050] Step S12: Data preprocessing. A sliding window is used to normalize, linearly interpolate, and detect anomalies in the collected data. Combined with an autoencoder model, potential fault modes are initially identified, as shown below:

[0051] ;

[0052] in, Representing data points Abnormal scores, range greater than Considered abnormal; express Function used to normalize scores to ; The reconstruction error is represented by the Euclidean norm. Indicates the scaling factor for the abnormal threshold; This represents the variational autoencoder function, introducing the slope. of Activation function and The weight matrix is ​​used for encoding and decoding, with the dimension being the input dimension. , Hidden unit, used during initialization method; and This is the bias vector; the loss function is the mean squared error, with a weight of 0.001. The optimizer minimizes until convergence, and the learning rate... The batch size is 32.

[0053] Example 3, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S2, the construction of the knowledge graph specifically includes the following steps:

[0054] Step S21: Extract high-level features. High-level features are extracted from the preprocessed data. Time-series analysis and frequency domain transformation are used to capture the pump's vibration modes, flow fluctuations, and energy consumption trends. Specifically, the time-series analysis involves using… The model fitting residuals are transformed into Fast Fourier Transform in the frequency domain, and combined with the pump physics model of Bernoulli's equation to generate mixed features, as shown below:

[0055] ;

[0056] in, Indicates the first A mixed feature vector; Represents the time series The fast Fourier transform is used to extract the frequency domain components; This represents element-wise multiplication; This represents the physical prior vector, with a length consistent with the number of frequency domain components obtained after the time series undergoes a Fast Fourier Transform. Each term is a dimensionless weight used to describe the relative sensitivity and contribution of the corresponding frequency band to pump performance indicators; specifically... Each weight maps the pump's efficiency characteristics, material damping, structural modal frequencies, and the inherent vibration response of the impeller and bearings onto the frequency spectrum. Indicates the physical integral weight; This indicates the rate of change of pump speed after adjustment. The pressure function is fitted using a quadratic polynomial. Indicates the baseline pump efficiency coefficient. Indicates the wear penalty factor. The vibration integral is used to calculate the cumulative wear index; the integral uses... Similar rules, time window Seconds, with a step size of 0.1 seconds;

[0057] Step S22: Construct a knowledge graph for the pump house, associating extracted features with entities, including: pump components and fault types; use a graph embedding algorithm. Generate low-dimensional representations that support semantic queries, relational reasoning, and anomaly propagation analysis, transforming big data into dynamic structured knowledge, as shown below:

[0058] ;

[0059] in, Represents a node The embedding vector; This indicates that the loss is minimized through partial random walks and negative sampling; This represents the Euclidean distance loss for the embedding of adjacent nodes; express Norm regularization; Represents the time-series smoothing term; Representing a temporal knowledge graph, quadruples Representing features respectively ,relation ,entity timestamp Optimize usage Optimizer, learning rate 150 iterations ;

[0060] Step S23: Feature enhancement, using mutual information and Regression Dynamic Selection feature, Furthermore, it enhances minority class fault data through generative adversarial networks, improves dataset balance, and introduces adversarial learning to simulate rare scenarios, as shown below:

[0061] ;

[0062] in, This represents the enhanced feature weight vector. Representing the feature weight vector, through Minimize mean square error plus Regularization, Represents the feature regularization coefficient. The number of samples; The target label is a binary label indicating whether the pump room is faulty. The characteristic matrix; Indicates the first The mutual information between features and targets is used for pre-screening, with a threshold of [value missing]. ; Represents the target random variable. Represents a characteristic random variable, This represents the joint probability distribution of the target and its features. Indicate target The marginal probability distribution is equal to the label. The number of samples divided by the total number of samples, Representation of features The marginal probability distribution is equal to the feature falling within the lattice. Divide the number of samples by the total number of samples; optimize using coordinate descent, iterating until convergence to a residual less than 1 / 2. ;choose of feature, This represents the natural logarithm function.

[0063] By performing the above operations, this solution addresses the problems in existing pump house data analysis, such as the prevalence of unstructured time-series data, lack of physical principle support for feature extraction, weak correlation between data and equipment entities making effective relationship reasoning difficult, and unbalanced distribution of fault data and lack of precision in key feature selection. This solution extracts hybrid features from preprocessed data using time-series analysis, frequency domain transformation, and pump physical models. It constructs a time-series knowledge graph integrating features, relationships, entities, and timestamps, and uses graph embedding algorithms to generate low-dimensional node representations. Then, it dynamically selects key features using mutual information and lasso regression, and enhances minority class fault data through adversarial learning. This achieves the transformation of unstructured big data into dynamic structured knowledge, strengthens the semantic association and relationship reasoning capabilities between features and equipment entities, accurately selects high-value features, and improves the balance of the dataset, laying a structured and high-quality feature foundation for subsequent performance prediction.

[0064] Example 4, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S3, the training of the pump house performance prediction model specifically includes the following steps:

[0065] Step S31: Train the hybrid neural network model, combining Extracting temporal local features, To handle global sequence dependencies, a multi-head attention mechanism is used to focus on key time points and cross-modal interactions, enabling the prediction of the pump station's operating status, including real-time energy consumption and failure probability. Multimodal learning and self-attention are integrated, as shown below:

[0066] ;

[0067] in, This represents the predicted output vector, specifically in the form of: [energy consumption, failure probability]; Represents the normalized exponential function, This represents a 6-layer encoder, based on the input feature matrix. Perform convolution operations to obtain convolutional features. ,use The kernel size is 3, and the channel size is 64. This indicates the attention of the bulls. Indicates the number of heads of attention; For scaling dot product attention: ; , , These represent the input triples of the attention mechanism: query, key, and value matrix. Indicates the transpose symbol; Indicates the first The output of each attention head; For the first The object can learn the projection matrix, projecting the input onto the corresponding... dimensional subspace, with dimension 1 , express The model's hidden dimensions; This represents the projection matrix used for linear transformation after multi-head concatenation; the training loss is the cross-entropy loss, and the optimizer is... The learning rate is 0.0005, the batch size is 64, and the process continues until the validation accuracy is greater than 96%.

[0068] Step S32: Model validation, using SHAP values ​​to analyze feature contributions, combined with cross-validation and adversarial example testing, as shown below:

[0069] ;

[0070] in, Indicates the first Features value; For Monte Carlo sampling numbers; and Representing subsets The number of elements and the complete set of features The total number of elements; Representation of the complete set of features Remove the first Any subset after a feature ; Distinguishing the union symbol; For feature subset The value function, For using subsets The predicted vector; Express the expectation; calculate the expectation for each sample. Aggregates to global importance; threshold It is considered a key feature.

[0071] By performing the above operations, this solution addresses the problems of existing pump house performance prediction methods that often rely on single models, making it difficult to simultaneously capture local features and global sequence dependencies in time-series data, insufficient cross-modal interaction feature mining, poor model interpretability, and inadequate prediction accuracy and robustness due to reliance on simple indicators for verification. This solution trains a hybrid neural network model that integrates convolutional neural networks and transformers, leveraging a multi-head attention mechanism to focus on key time points and cross-modal interaction information. This enables joint prediction of real-time energy consumption and failure probability in pump houses, quantifies the contribution of each feature, and conducts multi-dimensional verification of the model using cross-validation and adversarial example testing. This effectively captures local and global features of time-series data, fully explores the interaction value of multi-modal data, significantly improves the accuracy and precision of pump house operation status prediction, quantifies the contribution of features to the prediction results, enhances model interpretability, and further ensures the robustness and effectiveness of the model in practical applications through multi-dimensional verification.

[0072] Example 5, see Figure 1 and Figure 6 This embodiment is based on the above embodiment. In step S4, the optimization algorithm design specifically includes the following steps:

[0073] Step S41: Develop a multi-objective optimization framework, using differential evolution algorithm and predictive uncertainty to optimize pump set configuration parameters, including speed, maintenance interval, balanced energy consumption, and efficiency. Introduce a hybrid evolutionary strategy and uncertainty guidance. The optimization objectives are as follows:

[0074] ;

[0075] in, express Multi-objective functions, using Approximate frontier; and The target weight can be represented by... Method adjustment; Total energy consumption, This represents the probability of failure. Indicates the first In a generational population, each individual is normalized to The parameter vector has a size of 200; Differential Evolution Pairs Generate the next generation The process depends on the hyperparameter: scaling factor. scaling factor ; Mutation expression: , For the mutation vector, , and Represents three distinct individual vectors randomly selected from the current population; the population parameter boundary is... Iterate until convergence, with an upper limit of 300 iterations and a population diversity threshold of 0.01.

[0076] Step S42: Simulate and optimize the scenario, using the Monte Carlo method combined with In a scenario with uncertain sequence generation, the robustness and sensitivity of the optimization scheme are evaluated, and meta-learning is used to adjust the hyperparameters to maximize the expected utility, as shown below:

[0077] ;

[0078] in, The overall robustness score needs to be minimized; It expresses expectations for the scenario; This represents the variance of the target value across the set of simulated scenarios; This represents the simulated risk coefficient; Indicates at confidence level The following conditions apply to the risk value, for the worst-case scenario. Proportional scenes; Indicate the confidence level, considering the worst-case scenario. Scene mean; To represent a simulated scenario, use Low-difference sequences are sampled from a multidimensional normal distribution , The vector representing the mean of the input distribution. Represents one-dimensional standard deviation, combined with Latin hypercube sampling. Improve uniformity; generate 5000 scenes, calculate statistics, if If so, then use meta-optimization to fall back to S41 to adjust the weights.

[0079] By performing the above operations, this solution addresses the problems of existing pump station configuration optimizations being mostly single-objective optimizations that fail to adequately balance energy consumption and maintenance costs, lack of robustness due to insufficient consideration of various uncertainties in actual operation, and poor applicability of the optimization schemes due to the lack of scientific basis for hyperparameter adjustment. This solution constructs a multi-objective optimization framework without dominated sorting, combines differential evolution algorithms to optimize configuration parameters such as pump speed and maintenance intervals, balancing total energy consumption and maintenance costs caused by failures. Then, it utilizes the Monte Carlo method combined with low-difference sequences and Latin hypercube sampling to generate uncertain scenarios, evaluating the robustness and sensitivity of the optimization scheme. Through meta-learning, it dynamically adjusts hyperparameters and sets a backoff mechanism, achieving multi-objective and precise optimization of pump station configuration. This fully considers the actual needs of energy consumption control and equipment maintenance, effectively improving the robustness and anti-interference ability of the optimization scheme in complex and uncertain scenarios. The dynamic hyperparameter adjustment and backoff mechanism ensure the maximization of the expected utility of the optimization scheme, making the optimization strategy more aligned with the actual operating needs of the pump station.

[0080] Example 6, see Figure 2 This embodiment is based on the above embodiments. The automated optimization system for water pumping stations based on big data provided by the present invention includes a data acquisition module, a knowledge graph construction module, a water pumping station performance prediction model training module, and an optimization algorithm design module.

[0081] The data acquisition module installs multiple sensors at key locations in the pump house to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. The potential fault modes are initially identified by combining the autoencoder model, and the data is sent to the knowledge graph construction module.

[0082] The knowledge graph construction module receives data sent by the data acquisition module, extracts mixed features from the preprocessed data, captures relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then constructs a time series knowledge graph, associates the extracted features with various entities and processes them using graph embedding algorithms, finally performs feature enhancement, filters key features, and sends the data to the pump house performance prediction model training module.

[0083] The module for training the pump house performance prediction model receives data from the knowledge graph construction module, trains a hybrid neural network model combining convolutional neural networks and a transducer, and uses a multi-head attention mechanism to focus on key information to predict the pump house's operating status, including real-time energy consumption and failure probability. The model is then validated. The value analysis features contribute and the data is sent to the optimization algorithm design module;

[0084] The optimization algorithm design module receives data sent by the training pump house performance prediction model module, formulates a multi-objective optimization framework, uses the differential evolution algorithm to optimize the pump group configuration parameters, balances total energy consumption and maintenance-related costs, and then uses the Monte Carlo method combined with relevant sequences to generate uncertainty scenarios to evaluate the robustness and sensitivity of the optimization scheme.

[0085] It should be noted that, in this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0086] 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.

[0087] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for automated optimization of pump stations based on big data, characterized by: The method includes the following steps: Step S1: Data acquisition. Multiple sensors are installed at key locations in the pump room to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. Potential fault modes are initially identified by combining the autoencoder model. Step S2: Construct a knowledge graph, extract hybrid features from the preprocessed data, capture relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then construct a time series knowledge graph, associate the extracted features with various entities and process them using graph embedding algorithms, and finally perform feature enhancement and select key features; Step S3: Train the pump house performance prediction model. Train a hybrid neural network model combining convolutional neural networks and converters, using a multi-head attention mechanism to focus on key information, and predict the pump house's operating status, including real-time energy consumption and failure probability. Then, validate the model. Value analysis of feature contribution; Step S4: Optimize the algorithm design, formulate a multi-objective optimization framework, use the differential evolution algorithm to optimize the pump group configuration parameters, balance the total energy consumption and maintenance-related costs, and then use the Monte Carlo method to generate uncertainty scenarios in combination with relevant sequences to evaluate the robustness and sensitivity of the optimization scheme.

2. The automated optimization method for water pumping stations based on big data according to claim 1, characterized in that: In step S2, the construction of the knowledge graph specifically includes the following steps: Step S21: Extract high-level features. High-level features are extracted from the preprocessed data. Time-series analysis and frequency domain transformation are used to capture the pump's vibration modes, flow fluctuations, and energy consumption trends. Specifically, the time-series analysis involves using… The model fitting residuals are transformed into a fast Fourier transform in the frequency domain, and a hybrid feature is generated by combining the pump physical model with the Bernoulli equation. Step S22: Construct a knowledge graph for the pump house, associating extracted features with entities, including: pump components and fault types; use a graph embedding algorithm. Generate low-dimensional representations to support semantic queries, relational reasoning, and anomaly propagation analysis, transforming big data into dynamic structured knowledge; Step S23: Feature enhancement, using mutual information and Regression Dynamic Selection feature, Furthermore, it enhances minority class fault data by generating adversarial networks, improves dataset balance, and introduces adversarial learning to simulate rare scenarios.

3. The automated optimization method for water pumping stations based on big data according to claim 1, characterized in that: In step S3, the training of the pump house performance prediction model specifically includes the following steps: Step S31: Train the hybrid neural network model, combining Extracting temporal local features, To handle global sequence dependencies, a multi-head attention mechanism is used to focus on key time points and cross-modal interactions, enabling the prediction of pump station operating status, including real-time energy consumption and failure probability; multi-modal learning and self-attention are integrated. Step S32: Model validation, using SHAP values ​​to analyze feature contributions, combined with cross-validation and adversarial example testing.

4. The automated optimization method for water pumping stations based on big data according to claim 1, characterized in that: In step S4, the optimization algorithm design specifically includes the following steps: Step S41: Develop a multi-objective optimization framework, use differential evolution algorithm and predict uncertainty to optimize pump set configuration parameters, including speed, maintenance interval, balance energy consumption and efficiency, introduce hybrid evolution strategy and uncertainty guidance, and optimize objectives; Step S42: Simulate and optimize the scenario, using the Monte Carlo method combined with In a scenario of uncertain sequence generation, the robustness and sensitivity of the optimization scheme are evaluated, and meta-learning is used to adjust the hyperparameters to maximize the expected utility.

5. The automated optimization method for water pumping stations based on big data according to claim 1, characterized in that: In step S1, the data acquisition specifically includes the following steps: Step S11: Construct a sensor network by installing multiple sensors at key locations in the pump house, including the pump body, pipelines, motors, and control cabinet, to collect multimodal data. Step S12: Data preprocessing. The collected data is normalized, linearly interpolated and filled, and anomaly detection is performed using a sliding window. Potential fault modes are initially identified by combining the autoencoder model.

6. A big data-based automated optimization system for pumping stations, used to implement the big data-based automated optimization method for pumping stations as described in any one of claims 1-5, characterized in that: It includes a data acquisition module, a knowledge graph construction module, a pump house performance prediction model training module, and an optimization algorithm design module.

7. The automated optimization system for water pumping stations based on big data according to claim 6, characterized in that: The data acquisition module installs multiple sensors at key locations in the pump house to collect various multimodal data, as well as historical energy consumption and fault records. The collected data is denoised, and then normalized linear interpolation and anomaly detection are performed using a sliding window. The potential fault modes are initially identified by combining the autoencoder model, and the data is sent to the knowledge graph construction module. The knowledge graph construction module receives data sent by the data acquisition module, extracts mixed features from the preprocessed data, captures relevant patterns and trends through time series analysis, frequency domain transformation and physical model, then constructs a time series knowledge graph, associates the extracted features with various entities and processes them using graph embedding algorithms, finally performs feature enhancement, filters key features, and sends the data to the pump house performance prediction model training module. The module for training the pump house performance prediction model receives data from the knowledge graph construction module, trains a hybrid neural network model combining convolutional neural networks and a transducer, and uses a multi-head attention mechanism to focus on key information to predict the pump house's operating status, including real-time energy consumption and failure probability. The model is then validated. The value analysis features contribute and the data is sent to the optimization algorithm design module; The optimization algorithm design module receives data sent by the training pump house performance prediction model module, formulates a multi-objective optimization framework, uses the differential evolution algorithm to optimize the pump group configuration parameters, balances total energy consumption and maintenance-related costs, and then uses the Monte Carlo method combined with relevant sequences to generate uncertainty scenarios to evaluate the robustness and sensitivity of the optimization scheme.

Images