Industrial water quota Bayesian network uncertainty modeling and correcting method

Abstract

The invention discloses an industrial water quota Bayesian network uncertainty modeling and correcting method, and relates to the technical field of industrial water. Comprising the following steps: acquiring historical data and real-time monitoring data of industrial water quota to obtain an industrial water data set; the obtained industrial water data set is preprocessed; constructing a Bayesian network topology by adopting a mixed learning algorithm; training parameters in the Bayesian network by adopting a Markov chain Monte Carlo method to obtain trained network parameters; based on the Bayesian network, outputting probability distribution of the water consumption quota; the node state of the Bayesian network is updated through real-time data assimilation, and the water demand is dynamically predicted. The Bayesian network is adopted to integrate historical data and real-time monitoring information, the uncertainty of water consumption quota is quantified, model parameters are dynamically corrected through probabilistic reasoning, the method adapts to industrial process changes, prediction errors are reduced, and the robustness and reliability of water saving measures are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial water, and particularly relates to an industrial water quota Bayesian network uncertainty modeling and correction method. BACKGROUND

[0002] Industrial water quota refers to the limited value of water consumption of a water user in a certain period. Water quota is an important tool for water saving and water resource management, and is widely used in water planning, water resource demonstration, water taking permission, planned water management, water saving evaluation, water saving carrier construction, water saving management, progressive price increase for exceeding quota, water saving supervision and inspection, etc.

[0003] As a management tool, industrial water quota plays an important role in water saving and water resource management, but it also has many limitations in practical application, mainly in the following aspects:

[0004] (1) Quota is usually based on historical data or specific technical conditions, and it is difficult to dynamically reflect the improvement of water efficiency brought by technological progress, process innovation or production scale change;

[0005] (2) The production process and water demand of different industries (such as steel, textile, food processing) or even different enterprises in the same industry are significantly different, and a unified quota may not accurately match the actual demand;

[0006] (3) The scientificity of quota depends highly on the accuracy and integrity of water monitoring data, but there are problems such as insufficient metering facilities and false data reporting in many areas or enterprises.

[0007] Bayesian network is a graphical network based on probability reasoning. Simply put, it is composed of nodes and edges. Nodes represent random variables, which can be various factors that may affect safety, such as equipment status, personnel behavior, etc.; edges represent the conditional dependence relationship between variables. Through this network structure, we can intuitively present the mutual influence between various factors, and then use the methods of probability theory and graph theory for reasoning and analysis.

[0008] Traditional safety risk assessment methods often have difficulty in fully considering the complex interactions between various factors. Bayesian network model can easily integrate multiple factors affecting safety for analysis. Bayesian network model can clearly show the relationship between these factors and accurately assess the overall safety risk level.

[0009] Traditional industrial water quota mostly adopts static indicators or linear regression models, which are difficult to quantify the uncertainty caused by data missing, process differences and environmental changes. Quota revision relies on manual experience or periodic adjustment, which lags behind actual production changes and easily causes unreasonable water resource allocation or high compliance costs. Existing methods cannot dynamically respond to fluctuations in water efficiency due to the lack of systematic integration of multi-source data (such as real-time monitoring, weather, and equipment status).

[0010] Therefore, the industrial water quota Bayesian network uncertainty modeling and revision method is proposed to solve the problems existing in the prior art, which is a problem that the person skilled in the art urgently needs to solve. SUMMARY

[0011] Therefore, the industrial water quota Bayesian network uncertainty modeling and revision method is provided, which integrates historical data and real-time monitoring information using Bayesian networks, quantifies the uncertainty of water quota, dynamically revises model parameters through probabilistic reasoning, adapts to industrial process changes, reduces prediction errors, and improves the robustness and reliability of water-saving measures.

[0012] To achieve the above purpose, the technical scheme is adopted as follows:

[0013] An industrial water quota Bayesian network uncertainty modeling and revision method comprises the following steps:

[0014] S1. Data acquisition: acquire historical data and real-time monitoring data of industrial water quota to obtain an industrial water data set;

[0015] S2. Data preprocessing: preprocess the acquired industrial water data set;

[0016] S3. Bayesian network structure learning: adopt a hybrid learning algorithm to construct a Bayesian network topology;

[0017] S4. Network training: train the parameters in the Bayesian network using the Markov Chain Monte Carlo method to obtain trained network parameters;

[0018] S5. Uncertainty quantification: based on the Bayesian network, output the probability distribution of the water quota;

[0019] S6. Dynamic revision: update the Bayesian network node state through real-time data assimilation to dynamically predict water demand.

[0020] Optionally, the industrial water data set includes water intake data, circulating water rate data, product output data, temperature data, and humidity data.

[0021] Optionally, the acquired industrial water data set is preprocessed by data cleaning, missing value filling, data transformation, and anomaly detection in S2.

[0022] Optionally, the PC algorithm based on domain knowledge constraints in S3 and the greedy search construct the Bayesian network topology.

[0023] Optionally, the specific content of the PC algorithm based on domain knowledge constraints is:

[0024] Input: conditional mutual information between calculated variables; use industrial significance level alpha = 0.01 for independence test; retain strong relationships determined by domain knowledge when edge orientation;

[0025] Output: initial network structure obtained by the PC algorithm based on domain knowledge constraints;

[0026] The specific content of the greedy search is:

[0027] Input: initial network structure obtained by the PC algorithm based on domain knowledge constraints, industrial water data set, and domain knowledge constraints, including prohibited reverse causality and mandatory key connection;

[0028] Output: optimized Bayesian network structure and weights and confidence of each edge.

[0029] Optionally, S4 also includes dividing the preprocessed industrial water data set into training set, test set and validation set according to 8:1:1.

[0030] Optionally, the specific content of S5 based on Bayesian network to output the probability distribution of water quota is:

[0031]

[0032] Wherein, Q is the target quota variable, E is the observed evidence variable, and Pa is the parent node set.

[0033] Optionally, the specific content of S6 for updating the Bayesian network node state through real-time data assimilation and dynamically predicting water demand and proposing a quota correction factor is:

[0034]

[0035] Wherein, Q 实际 is the measured water consumption, and theta is the network parameter. When B > B threshold, the quota revision is triggered.

[0036] According to the above technical solution, compared with the prior art, the present application provides an industrial water quota Bayesian network uncertainty modeling and correction method, which has the following beneficial effects:

[0037] (1) The application integrates historical data and real-time monitoring information by using a Bayesian network, quantifies the uncertainty of water consumption quota, dynamically corrects model parameters through probabilistic reasoning, adapts to changes in industrial processes, reduces prediction errors, and improves the robustness and reliability of water-saving measures;

[0038] (2) The application constructs a Bayesian network based on mixed uncertainty, improves physical interpretability, couples quota output and water-saving cost function, and supports enterprises to balance compliance and economy. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0040] Figure 1 A flow chart of an industrial water quota Bayesian network uncertainty modeling and correction method is provided. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0042] Referring to Figure 1 The application discloses an industrial water quota Bayesian network uncertainty modeling and correction method, comprising the following steps:

[0043] S1. Data acquisition: acquire historical data and real-time monitoring data of industrial water quota, and obtain an industrial water data set;

[0044] S2. Data preprocessing: preprocessing the acquired industrial water data set;

[0045] S3. Bayesian network structure learning: constructing a Bayesian network topology by using a hybrid learning algorithm;

[0046] S4. Network training: training the parameters in the Bayesian network by using a Markov chain Monte Carlo method, and obtaining trained network parameters;

[0047] S5. Uncertainty quantification: outputting the probability distribution of water consumption quota based on the Bayesian network;

[0048] S6. Dynamic correction: updating the Bayesian network node state through real-time data assimilation to dynamically predict water demand.

[0049] Further, the industrial water data set includes water intake data, circulating water rate data, product output data, temperature data, humidity data.

[0050] Further, the industrial water data set obtained in S2 is preprocessed by data cleaning, missing value filling, data transformation and outlier detection.

[0051] Further, the Bayesian network topology is constructed in S3 based on the PC algorithm and greedy search of domain knowledge constraints.

[0052] Further, the specific content of the PC algorithm with domain knowledge constraints is:

[0053] Input: conditional mutual information between calculated variables; use industrial significance level α = 0.01 for independence test; retain strong relationships determined by domain knowledge when edge orientation;

[0054] Output: initial network structure obtained by the PC algorithm with domain knowledge constraints;

[0055] The specific content of the greedy search is:

[0056] Input: initial network structure obtained by the PC algorithm with domain knowledge constraints, industrial water data set, and domain knowledge constraints, including prohibited reverse causality and mandatory key connection;

[0057] Output: optimized Bayesian network structure and weights and confidence of each edge.

[0058] Further, S4 also includes dividing the preprocessed industrial water data set into training set, test set and validation set according to 8:1:1.

[0059] Further, the specific content of S5 based on Bayesian network to output the probability distribution of water quota is:

[0060]

[0061] Where, Q is the target quota variable, E is the observed evidence variable, and Pa is the parent node set.

[0062] Further, in S6, the Bayesian network node state is updated through real-time data assimilation to dynamically predict water demand, and the specific content of the quota correction factor is:

[0063]

[0064] Where, Q 实际To measure the water, θ is the network parameters, when B > B threshold trigger revision of the quota.

[0065] In one specific embodiment, including the following:

[0066] A chemical company as an example:

[0067] Data input: collect nearly 3 years of water data (water intake, circulating water rate, product output), weather data (temperature, humidity), equipment operation log;

[0068] Network training: after the initialization of Bayesian network, through Markov chain Monte Carlo iteration 100,000 times to converge, get the cooling tower efficiency on the sensitivity of water consumption is 0.23; 0.23 (95% HDI [0.18, 0.28]), namely efficiency every 10%, water consumption decreased by 2.3%;

[0069] Output prediction distribution: mean E 预测 = 69.5 m3 / h, standard deviation σ = 5.2 m3 / h, 95% confidence interval for [59.7, 79.1] m 3 / h;

[0070] When the measured water consumption Q 实际 = 125, the abnormal degree of quantitative calculation Z-score:

[0071]

[0072] Generally |z| > 3 is abnormal, far beyond the normal threshold;

[0073] Dynamic correction: measured water consumption in summer high temperature period is more than predicted, trigger correction factor α:

[0074]

[0075] α = 1.8 (> threshold 1.5), when α > 1.5, the system performs the following actions:

[0076] Parameter re-estimation: assume that the temperature node distribution should be updated from N(28, 2) to N(32, 3), retraining to get new parameters θ 新 ;

[0077] The ratio of the expected value of the new and old distribution:

[0078]

[0079] Therefore, the quota is raised by 15% (from 10.0 to 11.5 m 3 / ton);

[0080] Short-term measures are to start the standby water cooling system (increase 20% heat exchange area); long-term measures are to suggest installing spray cooling device.

[0081] The various embodiments described in this specification are presented by way of example, and each embodiment is presented with the intention of providing additional details into the disclosure of the application. Each embodiment can be combined with one or more other embodiments concerning different aspects of the application. Nothing reported is dedicated to any single embodiment or set of embodiments, except individual sections titled "Embodiment" which follow.

[0082] The above description of disclosed embodiments is intended to be illustrative and not restrictive. Many embodiments will be apparent to those of skill in the art upon reading this disclosure. The scope of the application should, therefore, be determined not with reference to the above description, but instead should be determined with reference to the appended claims, along with their full scope of equivalents. The disclosure of all articles and references referred to herein are incorporated by reference in their entirety.

Claims

1. A method for modeling and correcting uncertainties in Bayesian networks for industrial water quotas, characterized in that, Includes the following steps: S1. Data Acquisition: Acquire historical and real-time monitoring data of industrial water consumption quotas to obtain an industrial water consumption dataset; S2. Data Preprocessing: Preprocess the acquired industrial water dataset; S3. Bayesian Network Structure Learning: Bayesian network topology is constructed using a hybrid learning algorithm; S4. Network Training: The parameters in the Bayesian network are trained using the Markov chain Monte Carlo method to obtain the trained network parameters; S5. Uncertainty Quantification: Based on Bayesian networks, output the probability distribution of water consumption quotas; S6. Dynamic Correction: Update the state of Bayesian network nodes through real-time data assimilation to dynamically predict water demand.

2. The method for modeling and correcting uncertainties in Bayesian networks for industrial water quotas according to claim 1, characterized in that, The industrial water dataset includes water intake data, circulating water rate data, product output data, temperature data, and humidity data.

3. The method for modeling and correcting uncertainty in Bayesian networks for industrial water quotas according to claim 1, characterized in that, In S2, the acquired industrial water dataset undergoes preprocessing including data cleaning, missing value imputation, data transformation, and outlier detection.

4. The method for modeling and correcting uncertainties in Bayesian networks for industrial water quotas according to claim 1, characterized in that, In S3, a PC algorithm based on domain knowledge constraints and a greedy search are used to construct a Bayesian network topology.

5. The method for modeling and correcting uncertainties in Bayesian networks for industrial water quotas according to claim 4, characterized in that, The specific content of the domain knowledge-constrained PC algorithm is as follows: Input: Calculate conditional mutual information between variables; perform independence tests using an industrial significance level of α = 0.01; preserve strong relationships determined by domain knowledge during edge orientation; Output: The initial network structure obtained by the domain knowledge-constrained PC algorithm; The specific content of the greedy search is: Input: The initial network structure obtained by the domain knowledge-constrained PC algorithm, the industrial water dataset, and the domain knowledge constraints, which include prohibiting reverse causality and enforcing key connections; Output: The optimized Bayesian network structure and the weights and confidence scores of each edge.

6. The method for modeling and correcting uncertainties in Bayesian networks for industrial water quotas according to claim 1, characterized in that, S4 also includes dividing the pretreated industrial water dataset into training, testing, and validation sets in an 8:1:1 ratio.

7. The method for modeling and correcting uncertainty in Bayesian networks for industrial water quotas according to claim 1, characterized in that, The specific content of the probability distribution of water consumption quota output based on Bayesian network in S5 is as follows: Where Q is the target quota variable, E is the observed evidence variable, and Pa is the set of parent nodes.

8. The method for modeling and correcting uncertainty in Bayesian networks for industrial water quotas according to claim 1, characterized in that, S6 updates the state of Bayesian network nodes through real-time data assimilation to dynamically predict water demand, and proposes a quota correction factor as follows: Among them, Q 实际 The actual water consumption is represented by θ, which is a network parameter. Quota revision is triggered when B > B threshold.

Images