An intelligent accounting method for industrial water consumption

Abstract

The application discloses an intelligent accounting method for industrial water consumption, and relates to the technical field of industrial water consumption monitoring and intelligent accounting, and comprises the following steps: step one: receiving and analyzing water consumption accounting requirements, simultaneously acquiring historical data and real-time data from a historical data warehouse and a real-time data acquisition system, respectively, pre-processing, and generating a pre-processed data set. The intelligent accounting method for industrial water consumption realizes real-time self-adaptive adjustment of time-varying factors in industrial water consumption accounting by adopting multi-level data fusion, an improved deep learning network, data sliding window online learning, feedback closed-loop control and a hierarchical incremental updating strategy, significantly improves the prediction accuracy and system response speed of nonlinear water consumption behavior, and thus effectively overcomes the deficiency of static offline training of an accounting model in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial water consumption monitoring and intelligent accounting, in particular to an industrial water consumption intelligent accounting method. BACKGROUND

[0002] In the existing industrial water consumption accounting method, due to the complexity of the industrial scene, such as multiple devices, multiple links, and multiple water types, the system needs to process multi-source heterogeneous data from sensors, production logs, environmental monitoring devices, etc. However, the existing technology has the following key problems, that is, the industrial water scene has strong time-varying characteristics, such as production plan adjustment, device start-stop, seasonal change, and the existing accounting model is mostly static or based on historical data offline training, lacking real-time adaptive ability. For example, the nonlinear relationship of cooling water consumption at high and low load of the device cannot be accurately captured by the linear model, resulting in cumulative water consumption prediction deviation. SUMMARY

[0003] (I) Technical problems solved

[0004] In view of the deficiencies of the prior art, the present application provides an industrial water consumption intelligent accounting method, which solves the problem that the existing accounting model is mostly static or based on historical data offline training, lacking real-time adaptive ability, resulting in cumulative water consumption prediction deviation.

[0005] (II) Technical solutions

[0006] In order to achieve the above purpose, the present application is realized by the following technical solutions: an industrial water consumption intelligent accounting method, comprising the following steps:

[0007] Step 1: receiving and analyzing water accounting requirements, and simultaneously obtaining historical data and real-time data from a historical data warehouse and a real-time data acquisition system respectively; preprocessing the obtained historical data and real-time data respectively, including data cleaning, outlier rejection, missing data compensation and multi-source data time synchronization, to generate a preprocessed data set;

[0008] Step 2: selecting or constructing an accounting model containing a baseline offline training module and a real-time adaptive adjustment module from a preconfigured dynamic accounting model library; using the accounting model to preliminarily predict the preprocessed data set to obtain a preliminary water consumption calculation value;

[0009] Step 3: comparing the preliminary calculation value with the preset standard value or real-time monitoring result to determine the prediction deviation; when the prediction deviation exceeds the predetermined tolerance, starting an online incremental learning mechanism to real-time adjust the key parameters of the accounting model to capture the nonlinear change of cooling water at high and low load;

[0010] Step four: re-calculate the pre-processed data set based on the updated accounting model, generate the corrected industrial water consumption; output the corrected industrial water consumption, and store the latest model parameters in the dynamic model library for subsequent adaptive optimization.

[0011] Preferably, the accounting model consists of two main parts:

[0012] Baseline offline training module for establishing a preliminary prediction model using historical data;

[0013] Real-time adaptive adjustment module for dynamically correcting the preliminary prediction model according to online feedback data, thereby achieving real-time response to equipment start-stop, production plan adjustment and seasonal changes in industrial production.

[0014] Preferably, in the data preprocessing process of step one, a multi-level data fusion algorithm is introduced, which includes:

[0015] Multi-channel data synchronization processing;

[0016] Feature reconstruction module for feature extraction and reconstruction of heterogeneous data from sensors, production logs and environmental monitoring devices;

[0017] Noise suppression unit for reducing data interference, thereby providing higher quality input data for subsequent accounting models.

[0018] Preferably, the real-time adaptive adjustment module uses an improved deep learning network, which includes:

[0019] Online prediction submodule based on long short-term memory network (LSTM) for preliminary capture of time series changes;

[0020] Adaptive fuzzy logic correction submodule for real-time fuzzy processing and correction of prediction errors to achieve accurate modeling of cooling water nonlinear behavior;

[0021] The process is as follows: obtain the current water consumption data of the industrial site from the real-time data acquisition system, and preprocess the data including cleaning, standardization and time series construction to form a time series data set suitable for deep learning model input; and input to the online prediction submodule based on long short-term memory network (LSTM); the LSTM module uses its time series memory capability to capture the time-varying characteristics of industrial water consumption, producing a preliminary prediction value that reflects the dynamic changes in the current and near-term production process; compare the preliminary prediction value output by the LSTM module with the actual water consumption data collected in real time to calculate the prediction error, which is used to measure the deviation between the model prediction and the actual situation on site;

[0022] The calculated prediction error is passed as input to an adaptive fuzzy logic correction submodule; the submodule first fuzzifies the prediction error, mapping the error value to a fuzzy membership degree, for soft decision-making on the error, and the fuzzy membership degree includes "low error", "moderate error", and "high error"; according to a preset adaptive fuzzy rule (the rule can be dynamically adjusted according to historical feedback data), a corresponding correction factor or correction amount is determined to form a real-time correction value; the correction value obtained by the adaptive fuzzy logic correction submodule is combined with the preliminary prediction value of the LSTM module to generate a corrected final prediction result;

[0023] The final prediction result takes into account both time series characteristics and real-time error correction, and achieves immediate response to factors such as equipment start-stop, production plan adjustment, and seasonal changes;

[0024] The corrected prediction result is compared with subsequent real-time collected data to form new error feedback; using this feedback information, the weight parameters of the LSTM module in the deep learning network and the fuzzy logic correction rule are updated online, achieving adaptive optimization of the model.

[0025] Preferably, the online incremental learning mechanism uses a data sliding window strategy, combined with a hybrid historical and real-time data learning method, to update the parameters of the accounting model in stages and at different granularities, thereby ensuring the continuous stability and real-time adaptability of the model prediction in the context of dynamic changes in industrial production environments;

[0026] The process of the online incremental learning mechanism updating the parameters of the accounting model in stages and at different granularities includes: continuously collecting industrial water-related data from various real-time data sources (such as sensors, production logs, and environmental monitoring equipment); preprocessing the collected real-time data, including cleaning, outlier removal, missing data filling, and time synchronization, to ensure data quality and timeliness; at the same time, extracting historical data similar to the current industrial production environment from the historical data warehouse and preprocessing it in the same way; then, according to the set time range or data volume, a data sliding window is established, which contains real-time data in the latest period of time; the window continuously slides forward, and when new data enters, old data is removed accordingly, ensuring that the model training is always based on the latest data dynamic changes; mixing the real-time data in the sliding window with the historical data, and selecting the historical data that best matches the current situation according to production plan, equipment status, and seasonal factor indicators; forming a hybrid data set that retains the long-term statistical characteristics obtained during initial model training, and reflects the latest industrial water dynamics;

[0027] The mixed data set is used to coarsely adjust the parameters of the overall model, update the global prediction model, and capture the overall trend of water consumption. Sub-models are constructed for each water consumption category, and the parameters of each sub-model are independently adjusted to more accurately reflect the nonlinear time-varying characteristics of each category. An online incremental learning algorithm (such as online gradient descent or other online optimization methods) is used to update the parameters at different levels in a granular manner. A smaller learning rate is used for the global model parameters to ensure overall stability. A larger adjustment step is used for key sub-models or local parameters to quickly respond to real-time data fluctuations. The update process is performed gradually within each time window to ensure that the model always follows the latest data changes. The updated model is validated in real time within the sliding window by comparing the model prediction values with the actual collected data to calculate the prediction error. If the error still exceeds the preset tolerance, further refinement adjustments are triggered until the model prediction accuracy meets the requirements. The validation process forms a closed-loop feedback mechanism to provide a basis for the next parameter update. After online incremental learning and feedback adjustment, the final updated model parameters are stored in the dynamic model library. The next accounting task will directly call the latest model parameters to achieve continuous self-adaptive optimization of industrial water consumption accounting.

[0028] Preferably, the intelligent industrial water consumption accounting method establishes a feedback closed-loop control mechanism. After the model output prediction value is obtained, it is compared with the actual collected data in real time to form a dynamic feedback signal. This signal is used to automatically trigger the adjustment of the model parameters, and a feedback regulator is used to implement the hierarchical regulation of the prediction error of each water consumption type.

[0029] The process of the feedback closed-loop control mechanism is as follows:

[0030] The calculated prediction error is input into a feedback signal generator. The error is classified into slight, moderate, and severe error categories based on the error size and distribution. A dynamic feedback signal is generated, which contains error level information for different water consumption types and clearly reflects the deviation between each prediction component and the actual value.

[0031] When the dynamic feedback signal shows that the prediction error of a water consumption type exceeds the preset tolerance, the feedback regulator automatically activates the online incremental learning mechanism. Based on the error level information in the feedback signal, the system selects the appropriate adjustment strategy to automatically fine-tune the key parameters in the model, thereby reducing the difference between the prediction value and the actual data.

[0032] Preferably, the real-time self-adaptive adjustment module implements a hierarchical incremental update strategy. Different categories of water consumption are respectively associated with independent real-time prediction sub-models. Each sub-model independently updates the parameters online, and the global prediction performance is optimized through inter-layer coordination mechanisms.

[0033] The process of the real-time self-adaptive adjustment module implementing a hierarchical incremental update strategy is as follows:

[0034] From the pre-processed real-time data, according to the production process and on-site monitoring information, the data is automatically identified and separated according to different water categories, including cooling water, process water and auxiliary water. The data of each category enters the corresponding sub-data channel separately to ensure that the data received by each sub-model is targeted and highly relevant. For each water category, an independent real-time prediction sub-model is constructed. Each sub-model uses an online learning algorithm suitable for the characteristics of the category to capture the time series dynamic characteristics of the category. Each sub-model implements an independent online parameter update mechanism to continuously adjust its weight through a data sliding window strategy, ensuring that the prediction value is dynamically updated with real-time data.

[0035] Then the hierarchical incremental update mechanism is implemented:

[0036] The first layer (local update): each real-time prediction sub-model independently performs online parameter update using the separated data, introduces the latest data into the model training process, and completes the incremental update of the local model.

[0037] The second layer (global collaboration): after the local update is completed, an inter-layer collaboration mechanism is introduced. The prediction results of each sub-model are fused through a global integration module to generate a global prediction value of the overall industrial water consumption. The inter-layer collaboration mechanism uses dynamic weighted average, decision fusion or neural network fusion strategy to evaluate and correct the prediction results of each sub-model to ensure the optimization of the overall prediction performance.

[0038] The global prediction result is compared with the actual collected data in real time to calculate the overall prediction error. Then, according to the error analysis result, adjustment information is transmitted to each sub-model and the global integration module through feedback signals, so that the local sub-models are further refined and adjusted based on the global collaborative feedback while maintaining independent update.

[0039] This closed-loop feedback mechanism ensures that the sub-models of different water types can "communicate with each other", fully considering the mutual influence between categories during adjustment, and improving the overall prediction accuracy.

[0040] After hierarchical incremental update and feedback closed-loop adjustment, the updated parameters of each sub-model are stored in the dynamic model library as the basis for subsequent accounting tasks. The global integration module outputs the final industrial water consumption prediction result after collaborative correction, and transmits the result to the upper decision and monitoring system to realize the continuous optimization of the global prediction performance.

[0041] Preferably, the industrial water consumption intelligent accounting system comprises:

[0042] A data acquisition unit is configured to acquire industrial water-related data in real time from multiple data sources (sensors, production logs, and environmental monitoring devices);

[0043] A data preprocessing unit is configured to clean, fuse, time synchronize, and extract features from the acquired multi-source data;

[0044] A dynamic accounting model unit is configured to preliminarily predict the industrial water consumption based on the pre-set dynamic accounting model and the preprocessed data;

[0045] A real-time adaptive updating unit is configured to realize real-time adjustment of the accounting model parameters based on online feedback;

[0046] A feedback closed-loop control unit is configured to generate a feedback signal by comparing the model prediction result with the actual data, and trigger model updating;

[0047] A result output unit is configured to transmit the corrected industrial water consumption accounting result to a management server and archive it.

[0048] Preferably, the real-time adaptive updating unit comprises:

[0049] A deep neural network adjustment module is configured to predict errors of the model based on the online acquired data;

[0050] A data sliding window online learning module is configured to update the model parameters in batches and time periods;

[0051] The two modules cooperatively constitute a real-time error compensation system to realize rapid response to data fluctuations and dynamic optimization of the model.

[0052] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, all or part of the steps of the intelligent industrial water consumption accounting method can be implemented.

[0053] (Three) beneficial effects

[0054] The present application provides an intelligent industrial water consumption accounting method. It has the following beneficial effects:

[0055] (I) The intelligent industrial water consumption accounting method realizes real-time adaptive adjustment of time-varying factors in industrial water consumption accounting by using multi-level data fusion, improved deep learning network, data sliding window online learning, feedback closed-loop control, and hierarchical incremental updating strategy, significantly improves the prediction accuracy and system response speed of nonlinear water consumption behavior, and effectively overcomes the shortcomings of static offline training of the accounting model in the prior art.

[0056] (II) The industrial water consumption intelligent accounting method realizes independent online prediction and hierarchical incremental update of different types of water, and realizes optimization of global prediction results through interlayer coordination mechanism, so that the system can realize real-time self-adaptation and accurate accounting of various types of water consumption when dealing with industrial field device start-stop, production plan adjustment and seasonal changes. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is a flowchart of the industrial water consumption intelligent accounting method of the present application.

[0058] Figure 2 It is a framework diagram of the industrial water consumption intelligent accounting system of the present application. DETAILED DESCRIPTION

[0059] 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 a part of the embodiments of the present application, not all the 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.

[0060] Embodiment 1, please refer to Figure 1 The present application provides a technical solution: an industrial water consumption intelligent accounting method, comprising the following steps:

[0061] Step 1: Receive and analyze water accounting requirements, and obtain historical data and real-time data from historical data warehouse and real-time data acquisition system respectively; preprocess the obtained historical data and real-time data, including data cleaning, outlier rejection, missing data compensation and multi-source data time synchronization, to generate a preprocessed data set;

[0062] Step 2: Select or construct an accounting model containing a baseline offline training module and a real-time self-adaptive adjustment module from a pre-set dynamic accounting model library; use the accounting model to preliminarily predict the preprocessed data set and obtain a preliminary water consumption calculation value;

[0063] Step 3: Compare the preliminary calculation value with the pre-set standard value or real-time monitoring result to determine the prediction deviation; when the prediction deviation exceeds the predetermined tolerance, start the online incremental learning mechanism to adjust the key parameters of the accounting model in real time to capture the nonlinear changes of cooling water in high load and low load states;

[0064] Step 4: Recalculate the preprocessed data set based on the updated accounting model to generate the corrected industrial water consumption; output the corrected industrial water consumption, and store the latest model parameters in the dynamic model library for subsequent adaptive optimization.

[0065] The accounting model consists of two main parts:

[0066] Baseline offline training module for establishing a preliminary prediction model using historical data;

[0067] Real-time adaptive adjustment module for dynamically correcting the preliminary prediction model according to online feedback data, thereby achieving real-time response to equipment start-stop, production plan adjustment and seasonal changes in industrial production;

[0068] The implementation process of the baseline offline training module for establishing a preliminary prediction model is as follows:

[0069] Extract historical water consumption data related to the industrial site from the industrial water historical data warehouse, including various production logs, equipment operation records and previously collected sensor data; clean the collected historical data, eliminate outliers and error data; perform missing value filling and time synchronization processing to ensure data continuity and time consistency; through data standardization and normalization processing, eliminate the dimensional differences between different data sources; based on historical data, extract key features related to industrial water consumption (such as daily water consumption, equipment operating status, production cycle, seasonal changes, etc.); segment or cluster the data to construct a feature data set that meets the prediction requirements, and label the data to provide a basis for subsequent training;

[0070] According to the preprocessed feature data set, select a suitable prediction model (such as a regression model, a neural network model or a decision tree model); use offline training algorithms to optimize the model parameters and build a preliminary industrial water consumption prediction model to capture long-term trends and patterns in historical data; use cross-validation and leave-one-out methods to evaluate the accuracy of the trained model; based on the evaluation results, optimize the model parameters and structure to ensure that the model's fitting effect on historical data meets the expected requirements;

[0071] Store the verified and optimized baseline prediction model and its parameters in the dynamic model library; the baseline model serves as the initial reference for the subsequent real-time adaptive adjustment module, providing a solid starting point for real-time accounting and online incremental learning;

[0072] Through the baseline offline training module, a preliminary industrial water consumption prediction model can be established using historical data, providing a reliable foundation for subsequent real-time adaptive mechanisms, thereby effectively solving the problem of insufficient real-time adaptability caused by static and offline training of existing accounting models;

[0073] The real-time self-adaptive adjustment module dynamically corrects the preliminary prediction model according to online feedback data: the system continuously collects the latest industrial water actual data from various real-time data sources (such as field sensors, production logs, environmental monitoring equipment, etc.); the collected data is preprocessed (such as data cleaning, time synchronization, etc.) to ensure the accuracy and timeliness of the feedback data; the processed real-time feedback data is integrated with the output data of the preliminary prediction model to provide a basis for subsequent error analysis; the predicted water consumption value of the preliminary prediction model is compared with the actually collected feedback data to calculate the prediction error of each key period and water consumption category; a predetermined error tolerance is set, and when the error exceeds the tolerance, it is considered that the current model cannot accurately reflect the dynamic changes on site and needs to be corrected;

[0074] When the prediction error is detected to exceed the tolerance, the online incremental learning mechanism is automatically triggered; a data sliding window strategy is used to integrate the real-time feedback data in the latest period of time with part of the historical data to form a mixed data set for model updating; this step ensures that the model correction can capture the dynamic effects caused by equipment start-stop, production plan adjustment and seasonal changes in real time according to the latest production situation; online learning algorithms (such as online gradient descent, online recursive neural network update or other incremental learning methods suitable for real-time data processing) are used to update the key parameters of the preliminary prediction model;

[0075] In the parameter updating process, a phased and granular strategy is adopted:

[0076] Global parameter adjustment stage: first, the model as a whole is coarsely adjusted to ensure the correct capture of the overall trend;

[0077] Local refinement update stage: fine-tuning of parameters is performed for different water types (such as cooling water, process water, etc.) to ensure accurate response to nonlinear and time-varying characteristics;

[0078] The updated model predicts the current data again and compares it with the real-time feedback data; the model parameters are further adjusted according to the new error value until the prediction error is within the predetermined tolerance, forming a continuous closed-loop feedback correction process; this process ensures that the system can continuously adapt to changes in the industrial site and achieve real-time response; the updated model parameters obtained through online incremental learning are stored in the dynamic model library as the initial model for subsequent accounting tasks; the updated corrected prediction value is output for real-time guidance of industrial water consumption management and scheduling, and the model update log is recorded for subsequent performance evaluation and further optimization; through the real-time self-adaptive adjustment module, the preliminary prediction model can be dynamically corrected based on online feedback data, thereby achieving real-time response and accurate prediction of water consumption fluctuations caused by factors such as equipment start-stop, production plan adjustment and seasonal changes in the industrial site.

[0079] In the data preprocessing process of step one, a multi-level data fusion algorithm is introduced, which includes: multi-channel data synchronous processing;

[0080] A feature reconstruction module is configured to extract and reconstruct features from heterogeneous data from sensors, production logs and environmental monitoring devices.

[0081] A noise suppression unit is configured to reduce data interference and provide higher quality input data for subsequent accounting models.

[0082] The real-time adaptive adjustment module uses an improved deep learning network, which includes:

[0083] An online prediction submodule based on a long short-term memory network (LSTM) is configured to preliminarily capture time series changes.

[0084] An adaptive fuzzy logic correction submodule is configured to perform real-time fuzzy processing and correction on prediction errors to achieve accurate modeling of the nonlinear behavior of cooling water.

[0085] The process is as follows: current water consumption data in the industrial field is obtained from a real-time data acquisition system, and the data is preprocessed, including cleaning, standardization and time series construction, to form a time series data set suitable for input into a deep learning model; and input into an online prediction submodule based on a long short-term memory network (LSTM); the LSTM module uses its time series memory capability to capture the time-varying characteristics of industrial water consumption, producing a preliminary prediction value that reflects the dynamic changes in the current and near-term production process; the preliminary prediction value output by the LSTM module is compared with the actual water consumption data collected in real time to calculate the prediction error, which is used to measure the degree of deviation between the model prediction and the actual field conditions.

[0086] The calculated prediction error is passed as input to an adaptive fuzzy logic correction submodule; this submodule first performs fuzzy processing on the prediction error, mapping the error value to a fuzzy membership degree for soft decision-making on the error, and the fuzzy membership degree includes "low error", "moderate error" and "high error"; according to pre-set adaptive fuzzy rules (rules can be dynamically adjusted according to historical feedback data), a corresponding correction factor or correction amount is determined to form a real-time correction value; the correction value obtained from the adaptive fuzzy logic correction submodule is combined with the preliminary prediction value from the LSTM module to generate a corrected final prediction result.

[0087] It needs to be further explained that in the specific implementation process, the final prediction result comprehensively considers the time series characteristics and real-time error correction, realizes the immediate response to factors such as equipment start-stop, production plan adjustment and seasonal change; compare the corrected prediction result with the subsequent real-time acquisition data to form a new error feedback; use this feedback information to update the weight parameters of the LSTM module in the deep learning network and the fuzzy logic correction rule online, realize the adaptive optimization of the model; the updated model parameters and correction rules are stored in the dynamic model library to continue to play a role in subsequent prediction;

[0088] Through the above process, the online prediction sub-module based on long short-term memory network (LSTM) in the improved deep learning network is responsible for preliminary capturing the time series change of industrial water, and the adaptive fuzzy logic correction sub-module is responsible for real-time fuzzy processing and correction of prediction error. The two work together to realize real-time adaptive response to the dynamic fluctuation of water consumption caused by complex factors such as equipment start-stop, production plan adjustment and seasonal change in industrial production.

[0089] The online incremental learning mechanism adopts a data sliding window strategy, combines a hybrid history and real-time data learning method, and updates the parameters of the accounting model in stages and granularities, so as to ensure the continuous stability and real-time adaptability of the model prediction in the background of dynamic changes in industrial production environment;

[0090] The process of the online incremental learning mechanism updating the parameters of the accounting model in stages and granularities includes: continuously collecting industrial water related data from various real-time data sources (such as sensors, production logs, environmental monitoring equipment); preprocessing the collected real-time data, including cleaning, outlier removal, missing data filling and time synchronization, to ensure data quality and timeliness; at the same time, extract historical data similar to the current industrial production environment from the historical data warehouse and preprocess it in the same way; then according to the set time range or data volume, establish a data sliding window, which contains real-time data in the latest period of time; the window continuously slides forward, when new data enters, the old data moves out accordingly, ensuring that the model training is always based on the dynamic changes of the latest data; mix the real-time data in the sliding window with the historical data, select the historical data that best matches the current situation according to the production plan, equipment state and seasonal factor indicators; form a hybrid data set that retains the long-term statistical characteristics obtained during the initial training of the model and reflects the latest industrial water dynamics;

[0091] The overall model parameters are coarsely adjusted using a mixed data set, the global prediction model is updated, and the overall trend of water consumption is captured. Sub-models are constructed for each water consumption category, and the respective parameters are independently adjusted to more accurately reflect the nonlinear time-varying characteristics of each category. An online incremental learning algorithm (such as online gradient descent or other online optimization methods) is used to update the parameters at different levels in a granular manner. A smaller learning rate is used for the global model parameters to ensure overall stability. A larger adjustment step is used for key sub-models or local parameters to quickly respond to real-time data fluctuations. The update process is performed gradually within each time window to ensure that the model always follows the latest data changes. The updated model is verified in real time within the sliding window by comparing the model prediction values with the actual collected data to calculate the prediction error. If the error still exceeds the preset tolerance, further refinement adjustments are triggered until the model prediction accuracy meets the requirements. The verification process forms a closed-loop feedback mechanism to provide a basis for the next parameter update. After online incremental learning and feedback adjustment, the final updated model parameters are stored in the dynamic model library. The next accounting task will directly call the latest model parameters to achieve continuous self-adaptive optimization of industrial water consumption accounting.

[0092] Through the above steps, the online incremental learning mechanism can use the data sliding window strategy and mixed historical and real-time data learning method to achieve parameter update at different stages and granularities for the accounting model, effectively solving the problem of static offline training and lack of real-time self-adaptive capability of traditional models.

[0093] The industrial water consumption intelligent accounting method establishes a feedback closed-loop control mechanism, compares the model output prediction value with the actual collected data in real time after the prediction value is output, forms a dynamic feedback signal, and uses the signal to automatically trigger the adjustment of the model parameters and achieve hierarchical regulation of the prediction error of each water consumption type through a feedback regulator.

[0094] The process of the feedback closed-loop control mechanism is as follows:

[0095] The calculated prediction error is input into a feedback signal generator, which classifies the error according to its size and distribution, including slight, moderate, and severe errors. A dynamic feedback signal is generated, which contains error level information for different water consumption types, clearly reflecting the deviation between each prediction component and the actual value.

[0096] When the dynamic feedback signal shows that the prediction error of a water consumption type exceeds the preset tolerance, the feedback regulator automatically activates the online incremental learning mechanism. The system selects the corresponding adjustment strategy based on the error level information in the feedback signal to automatically fine-tune the key parameters in the model, which is used to reduce the difference between the prediction value and the actual data.

[0097] It needs to be further explained that in the specific implementation process, the feedback regulator adopts a hierarchical regulation strategy for different water types, that is, different correction amplitudes and learning rates are set according to different error levels to realize hierarchical regulation of the global model and the local sub-model; the updated model parameters are immediately put into the prediction of the next cycle, and the new prediction result is compared with the actual data on site again, thereby forming a continuous closed-loop feedback process to continuously adaptively optimize the model performance;

[0098] After closed-loop regulation, the updated model parameters and the corresponding feedback signal data are stored in the dynamic model library as the basis for subsequent prediction and optimization; the system regularly evaluates the feedback closed-loop regulation effect and further optimizes the regulation algorithm according to long-term feedback to ensure that the model always maintains high adaptability to the dynamic changes of the industrial production site;

[0099] Through the above steps, the feedback closed-loop control mechanism can compare the model prediction value with the actual data on site in real time, form a dynamic feedback signal, and automatically trigger hierarchical regulation of the model parameters, thereby realizing accurate compensation and real-time adaptive adjustment of the prediction error of each water type.

[0100] The real-time adaptive adjustment module implements a hierarchical incremental update strategy, and different categories of water are respectively corresponding to independent real-time prediction sub-models. Each sub-model independently performs online parameter update, and realizes optimization of the global prediction performance through interlayer coordination mechanism;

[0101] The process of implementing the hierarchical incremental update strategy by the real-time adaptive adjustment module:

[0102] From the preprocessed real-time data, according to the production process and on-site monitoring information, the data is automatically identified and shunted according to different water categories, including cooling water, process water and auxiliary water. The data of each category is separately input into the corresponding sub-data channel to ensure that the data received by each sub-model is targeted and highly relevant. For each water category, an independent real-time prediction sub-model is constructed. Each sub-model uses an online learning algorithm suitable for the characteristics of the category to capture the time series dynamic characteristics of the category. Each sub-model realizes independent online parameter update mechanism, and continuously adjusts its weight through data sliding window strategy to ensure that the prediction value is dynamically updated with real-time data;

[0103] Then the hierarchical incremental update mechanism is implemented:

[0104] First layer (local update): each real-time prediction sub-model independently performs online parameter update using the shunted data, introduces the latest data into the model training process, and completes the incremental update of the local model;

[0105] The second layer (global coordination) introduces an inter-layer coordination mechanism after the local update is completed. The prediction results of each sub-model are fused by a global integration module to generate a global prediction value of the overall industrial water consumption. The inter-layer coordination mechanism adopts a dynamic weighted average, decision fusion or neural network fusion strategy to evaluate and correct the prediction results of each sub-model, so as to ensure the optimization of the overall prediction performance.

[0106] The global prediction result is compared with the actual collected data in real time to calculate the overall prediction error. Then, according to the error analysis result, adjustment information is transmitted to each sub-model and the global integration module through the feedback signal, so that the local sub-models can be further refined and adjusted based on the global coordination feedback while maintaining independent update.

[0107] This closed-loop feedback mechanism ensures that the sub-models of different water consumption types can "communicate with each other", fully considering the mutual influence between categories during adjustment, and improving the overall prediction accuracy.

[0108] After the hierarchical incremental update and feedback closed-loop adjustment, the updated parameters of each sub-model are stored in the dynamic model library as the basis for subsequent accounting tasks. The global integration module outputs the final industrial water consumption prediction result after coordination correction, and transmits the result to the upper layer decision and monitoring system to realize the continuous optimization of the global prediction performance.

[0109] Through the above process, independent online prediction and hierarchical incremental update of different categories of water are realized, and the global prediction result is optimized through the inter-layer coordination mechanism, so as to ensure that the system can adapt in real time and accurately account for the water consumption of each category when dealing with the start and stop of industrial site equipment, production plan adjustment and seasonal changes.

[0110] In example 2, please refer to Figure 2 The industrial water consumption intelligent accounting system comprises:

[0111] A data acquisition unit is configured to acquire industrial water consumption related data from multiple data sources (sensors, production logs, and environmental monitoring devices) in real time.

[0112] A data preprocessing unit is configured to clean, fuse, time synchronize, and extract features from the collected multi-source data.

[0113] A dynamic accounting model unit is configured to preliminarily predict the water consumption based on the preprocessed data according to a preconfigured dynamic accounting model.

[0114] A real-time adaptive update unit is configured to realize real-time adjustment of the accounting model parameters based on online feedback.

[0115] A feedback closed-loop control unit is configured to generate a feedback signal by comparing the model prediction result with the actual data, and trigger model update.

[0116] The result output unit transmits the corrected industrial water consumption accounting result to the management server and archives it.

[0117] The real-time adaptive updating unit comprises:

[0118] The deep neural network adjustment module is configured to perform error prediction on the model according to the online collected data.

[0119] The data sliding window online learning module is configured to update the model parameters in batches and in time periods.

[0120] The two modules cooperatively constitute a real-time error compensation system to realize fast response to data fluctuation and dynamic optimization of the model.

[0121] A computer readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, all or part of the steps of the intelligent industrial water consumption accounting method can be implemented.

[0122] The above technology realizes real-time adaptive adjustment of time-varying factors in industrial water consumption accounting by using multi-level data fusion, improved deep learning network, data sliding window online learning, feedback closed-loop control and hierarchical incremental updating strategy, significantly improves the prediction accuracy and system response speed of nonlinear water consumption behavior, and effectively overcomes the shortcomings of static offline training of the accounting model in the prior art.

[0123] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0124] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent calculation of industrial water consumption, characterized in that, Includes the following steps: Step 1: Receive and parse the water usage accounting requirements, and simultaneously obtain historical data and real-time data from the historical data warehouse and the real-time data acquisition system, respectively, perform preprocessing on each, and generate a preprocessed data set; Step 2: Select or construct an accounting model from the pre-set dynamic accounting model library that includes a baseline offline training module and a real-time adaptive adjustment module; use the accounting model to make preliminary predictions on the preprocessed data set to obtain preliminary water consumption calculation values; Step 3: Compare the preliminary water consumption calculation with the preset standard value or real-time monitoring results to determine the prediction deviation; when the prediction deviation exceeds the predetermined tolerance, activate the online incremental learning mechanism to adjust the key parameters of the accounting model in real time to capture the nonlinear changes of cooling water under high load and low load conditions. Step 4: Recalculate the preprocessed data set based on the updated accounting model to generate the corrected industrial water consumption; The corrected industrial water consumption is output, and the latest model parameters are stored in the dynamic accounting model library for subsequent adaptive optimization. The real-time adaptive adjustment module employs an improved deep learning network, which includes: An online prediction submodule based on the Long Short-Term Memory (LSTM) network is used to initially capture temporal changes; The adaptive fuzzy logic correction submodule is used to perform real-time fuzzification processing and correction of prediction errors; The real-time adaptive adjustment module operates as follows: It acquires current water usage data from the industrial site via a real-time data acquisition system, preprocesses the data to form a time-series dataset, and inputs it into an online prediction submodule based on a Long Short-Term Memory (LSTM) network. The online prediction submodule utilizes its time-series memory capability to capture the time-varying characteristics of industrial water consumption, generating a preliminary prediction value. The output preliminary prediction value is compared with the actual water usage data acquired in real-time, and the prediction error is calculated. This prediction error is then passed as input to an adaptive fuzzy logic correction submodule. This submodule first fuzzifies the prediction error, mapping the error value to fuzzy membership degrees, which include low, medium, and high errors. Based on preset adaptive fuzzy rules, it determines the corresponding correction factor or correction amount, forming a real-time correction value. Finally, the correction value is combined with the preliminary prediction value from the online prediction submodule to generate the corrected final prediction result.

2. The intelligent calculation method for industrial water consumption according to claim 1, characterized in that: The accounting model consists of two main parts: The baseline offline training module is used to build a preliminary prediction model using historical data; The real-time adaptive adjustment module is used to dynamically correct the preliminary prediction model based on online feedback data, thereby enabling real-time response to equipment start-up and shutdown, production plan adjustments and seasonal changes in industrial production.

3. The intelligent calculation method for industrial water consumption according to claim 2, characterized in that: In the data preprocessing process of step one, a multi-level data fusion algorithm is introduced, which includes: Multi-channel data synchronization processing; The feature reconstruction module is used to extract and reconstruct features from heterogeneous data from sensors, production logs, and environmental monitoring equipment. The noise suppression unit is used to reduce data interference, thereby providing higher quality input data for subsequent accounting models.

4. The intelligent calculation method for industrial water consumption according to claim 3, characterized in that: The online incremental learning mechanism adopts a data sliding window strategy, combining a hybrid historical and real-time data learning method to update the parameters of the accounting model; The online incremental learning mechanism updates the parameters of the accounting model in stages and at different granularities, including: continuously collecting industrial water-related data from various real-time data sources; preprocessing the collected real-time data, and simultaneously extracting historical data similar to the current industrial production environment from the historical data warehouse and performing the same preprocessing; then establishing a data sliding window based on a set time range or data volume, which contains real-time data within the latest time period; mixing the real-time data and historical data within the sliding window, and selecting the historical data that best matches the current situation based on production plans, equipment status, and seasonal factors; thus forming a mixed dataset. The overall model parameters are coarsely tuned using a mixed dataset to update the global prediction model and capture the overall trend of water consumption. Sub-models are constructed for each water consumption category, and their parameters are adjusted independently. An online incremental learning algorithm is used to update the parameters at different levels in a granular manner. The updated model is validated in real time within a sliding window, and the prediction error is calculated by comparing the model's predicted values ​​with the actual collected data. After online incremental learning and feedback adjustment, the finally updated model parameters are stored in a dynamic accounting model library.

5. The intelligent calculation method for industrial water consumption according to claim 4, characterized in that: The intelligent calculation method for industrial water consumption establishes a feedback closed-loop control mechanism. After the model outputs the predicted value, it compares it with the actual data collected on site in real time to form a dynamic feedback signal. This signal is used to automatically trigger the adjustment of model parameters and realize the hierarchical control of the prediction error of each water consumption type through the feedback regulator. The feedback closed-loop control mechanism works as follows: The calculated prediction error is input into the feedback signal generator, and the error is classified according to its magnitude and distribution, including slight, moderate and severe errors; a dynamic feedback signal is generated. When the dynamic feedback signal shows that the prediction error of water use type exceeds the preset tolerance, the feedback regulator automatically activates the online incremental learning mechanism. The system selects the corresponding adjustment strategy based on the error level information in the feedback signal and automatically fine-tunes the key parameters in the model to narrow the gap between the predicted value and the real-time data.

6. The intelligent calculation method for industrial water consumption according to claim 5, characterized in that: The real-time adaptive adjustment module implements a hierarchical incremental update strategy, which assigns different categories of water use to independent real-time prediction sub-models. Each sub-model updates its parameters online independently, and optimizes the global prediction performance through an inter-layer collaboration mechanism. The process by which the real-time adaptive adjustment module implements the hierarchical incremental update strategy: From the pre-processed real-time data, the data is automatically identified and diverted according to different water use categories based on production process and on-site monitoring information. These categories include cooling water, process water, and auxiliary water. Data for each category is entered into a corresponding sub-data channel. For each water use category, an independent real-time prediction sub-model is constructed. Each sub-model uses an online learning algorithm suitable for the characteristics of that category to capture the temporal dynamic features of that category. Subsequently, a tiered incremental update mechanism was implemented: First layer: Each real-time prediction sub-model independently performs online parameter updates using its own split data, introducing the latest data into the model training process to complete the incremental update of the local model; The second layer: After the local update is completed, an inter-layer collaboration mechanism is introduced. The prediction results of each sub-model are fused through the global integration module to generate a global prediction value of the overall industrial water consumption. This inter-layer collaboration mechanism uses dynamic weighted averaging, decision fusion or neural network fusion strategies to evaluate and correct the prediction results of each sub-model. The global prediction results are compared with the actual data collected on site in real time to calculate the overall prediction error. Then, based on the error analysis results, adjustment information is transmitted to each sub-model and the global integration module through feedback signals, so that the local sub-models can be further refined and adjusted based on the global collaborative feedback while maintaining independent updates. After hierarchical incremental updates and feedback closed-loop adjustments, each sub-model stores its updated parameters in the dynamic accounting model library as the basis for subsequent accounting tasks; the global integration module outputs the final industrial water consumption prediction results after collaborative correction and transmits the results to the upper-level decision-making and monitoring system.

7. An intelligent industrial water consumption calculation system, characterized in that, The method for intelligent calculation of industrial water consumption according to any one of claims 1 to 6 includes: The data acquisition unit is used to collect industrial water-related data from multiple data sources in real time. The data preprocessing unit cleans, merges, synchronizes time, and extracts features from the collected multi-source data. The dynamic accounting model unit performs preliminary water consumption prediction based on the pre-set dynamic accounting model and pre-processed data. The real-time adaptive update unit adjusts the parameters of the accounting model in real time based on online feedback; The feedback closed-loop control unit compares the model prediction results with real-time data and generates a feedback signal to trigger model updates. The result output unit transmits the corrected industrial water consumption calculation results to the management server and archives them.

8. The intelligent industrial water consumption calculation system according to claim 7, characterized in that: The real-time adaptive update unit includes: The deep neural network adjustment module is used to predict the error of the model based on online collected data; The online learning module with a sliding data window is used to update model parameters in batches and time periods. The two modules work together to form a real-time error compensation system, enabling rapid response to data fluctuations and dynamic optimization of the model.

9. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, enables the implementation of the intelligent calculation method for industrial water consumption as described in any one of claims 1 to 6.

Images