Regional water affair intelligent early warning system and method based on cloud computing

Abstract

The invention discloses a regional water affair intelligent early warning system and method based on cloud computing, and particularly relates to the technical field of water affair management.Water affair data are collected to generate an operation data set and upload the operation data set to a cloud end, and a pre-trained regional water affair prediction model deployed in the cloud end is utilized to perform prediction precision analysis; identifying whether the model has an early sign of insufficient dynamic precision; if the models have defects, generating characteristic indexes through dynamic disturbance testing and environment change response analysis, and dividing the models into high-fitness or low-fitness types; the low-fitness model is optimized and then re-applied, finally, the high-fitness or optimized model is combined with a preset early warning mechanism, an early warning signal is sent out in real time under the triggering condition, and the intellectualization and reliability of regional water affair management are improved; according to the method, through dynamic disturbance testing and environment change response analysis, the adaptive capacity of the model in a complex environment is comprehensively evaluated, and it is ensured that the regional water affair prediction model can accurately reflect dynamic changes.

Description

Technical Field

[0001] The present invention relates to the technical field of water management, and more specifically, to a cloud computing-based intelligent early warning system and method for regional water management. Background Art

[0002] With the acceleration of global climate change and urbanization, regional water management faces increasingly complex challenges, including frequent flood disasters, aggravated water pollution, and imbalance between water resource supply and demand. These problems not only pose great pressure on the ecological environment but also have a profound impact on regional economy and residents' lives.

[0003] Traditional water management and early warning methods usually rely on single sensors or simple rule models and are difficult to effectively cope with complex changes in a dynamic environment. For example, in the event of sudden heavy rain or industrial pollution discharge, traditional methods may fail to issue early warning signals in a timely manner due to insufficient data processing capabilities or lack of dynamic adaptability, missing key response opportunities. In addition, with the explosive growth of water system data volume, the limitations of traditional technologies in data collection, processing, and analysis are further manifested.

[0004] In recent years, the rapid development of cloud computing, big data, and artificial intelligence technologies has provided new solutions for regional water management. By using a cloud computing platform, real-time processing and efficient storage of massive water data can be achieved, and through intelligent prediction models, the operating state of the water system can be accurately predicted. However, the adaptability of existing intelligent prediction models in a dynamic environment may have problems with insufficient applicability. Especially when facing external environmental disturbances and emergencies, their prediction accuracy and reliability may significantly decline. Therefore, there is an urgent need for a cloud computing-based intelligent regional water early warning method to address the challenges of water prediction and early warning in a dynamic environment. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A cloud computing-based intelligent early warning method for regional water management, comprising the following steps:

[0007] Obtain the collected data of regional water management, obtain the operating data set of regional water management and upload it to the cloud;

[0008] Obtain the pre-trained regional water management prediction model currently in use in the cloud, and conduct prediction accuracy analysis to preliminarily identify whether there are early signs of insufficient dynamic accuracy in the pre-trained regional water management prediction model currently in use.

[0009] If there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use, further feature analysis is carried out to explore the sensitivity of the model to dynamic disturbances and its response to environmental changes. Then, based on the results of the feature analysis, the pre-trained regional water service prediction model currently in use is classified as a high-fitness model or a low-fitness model.

[0010] When the pre-trained regional water service prediction model currently in use is classified as a low-fitness model, the model is updated and optimized, and the optimized regional water service prediction model is applied.

[0011] According to the prediction results of the high-fitness model or the optimized regional water service prediction model, and a preset warning mechanism, a warning signal is issued when the prediction results activate the warning mechanism.

[0012] In a preferred embodiment, performing prediction accuracy analysis means:

[0013] Divide a fixed time window into multiple consecutive sub-windows with the same time length. Run the current prediction model in the cloud, input the collected real-time data into the model, and each sub-window generates a prediction result. Record the actual prediction values of the model and the true monitoring values at the corresponding time points, construct a comparison dataset of the predictions and true values for each sub-window, calculate the absolute value of the difference between the actual prediction values of the model and the true monitoring values at the corresponding time points respectively, and then compare all the absolute values within the same sub-window with a preset deviation threshold. Mark the time points with absolute values greater than or equal to the deviation threshold as abnormal time points, and count the total number of abnormal time points within the same sub-window.

[0014] In a preferred embodiment, initially identifying whether there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use means:

[0015] Calculate the average value of abnormal points and the standard deviation of abnormal points for the total number of abnormal time points in all sub-windows, then compare the average value of abnormal points with a preset reference average value, and compare the standard deviation of abnormal points with a preset reference standard deviation. If the average value of abnormal points is less than or equal to the preset reference average and the standard deviation of abnormal points is less than or equal to the preset reference standard deviation, a normal signal is generated. If the average value of abnormal points is not less than or equal to the preset reference average and the standard deviation of abnormal points is not less than or equal to the preset reference standard deviation, an abnormal signal is generated. When the abnormal signal is generated, it indicates that there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use.

[0016] In a preferred embodiment, exploring the sensitivity of the model to dynamic disturbances and its response to environmental changes means:

[0017] Perform dynamic perturbation testing on the pre-trained regional water service prediction model for the current application, generate a perturbation sensitivity index that reflects the degree of change in the output result of the model when the input data is perturbed, and measures the sensitivity of the model to dynamic perturbations, and generate an environmental change response index that reflects the response ability of the model to external environmental changes by combining the amplitude of dynamic environmental changes and the accuracy of the model prediction results.

[0018] In a preferred embodiment, the acquisition logic of the perturbation sensitivity index is as follows:

[0019] Generate a set of basic input data for input into the model for prediction to obtain a basic prediction result. Perturb the input data to generate several sets of perturbed data sets D 1 , D 2 ,..., D M , where M represents the number of sets of perturbed data sets, i.e., the number of perturbation times. The perturbation formula is:

[0020] D j (i) = D(i) + δ j (i); D(i) is the i-th sample of the basic input, and δ j (t) is the amount of perturbation applied to the i-th sample at the j-th perturbation, usually a small random number (normal distribution or uniform distribution), and D j (i) represents the i-th sample after the j-th perturbation;

[0021] Input the basic data set and each set of perturbed data sets D j into the model respectively to obtain the corresponding predicted outputs. The predicted output of the basic data set is marked as The predicted outputs of each set of perturbed data sets D j are marked as Then substitute into the perturbation sensitivity index calculation formula:

[0022] represents the i-th predicted value after the j-th perturbation, represents the i-th predicted value without perturbation, N represents the number of test samples during dynamic perturbation testing, and PSI represents the perturbation sensitivity index.

[0023] In a preferred embodiment, the acquisition logic of the environmental change response index is as follows:

[0024] Obtain environmental factor data related to regional water services, determine the time series of environmental factor X, record the environmental factor value X(t) at each time point, and for the environmental factor time series, calculate the change amplitude between adjacent time points: ΔX(t) = |X(t) - X(t - 1)|; X(t) represents the environmental factor value at the t-th time point, X(t - 1) represents the environmental factor value at the (t - 1)-th time point, and ΔX(t) represents the environmental change amplitude at the t-th time point;

[0025] Collect the prediction results and true observations of the model, and calculate the prediction error at each time point: E(t) = |Y predicted (t) - Y actual (t)|; E(t) represents the prediction error at the t-th time point, Y predicted (t) represents the prediction result of the model at the t-th time point, Y actual (t) represents the true observation at the t-th time point;

[0026] The formula for calculating the environmental change response index is:

[0027] T represents the total number of time points, ∈ represents a preset constant, α and β are both preset non-zero proportionality coefficients, γ is a preset non-linear amplification factor with a value greater than one, PJ represents the average value of all environmental factor data, and ECRI represents the environmental change response index.

[0028] In a preferred embodiment, classifying the currently applied pre-trained regional water service prediction model as a high fitness model or a low fitness model based on the result of feature analysis means:

[0029] Obtain the environmental change response index and perturbation sensitivity index of the currently applied pre-trained regional water service prediction model, and use them together as the input data of fuzzy logic. Use the adaptation type of the currently applied pre-trained regional water service prediction model as the output data of fuzzy logic. Perform fuzzy processing on the input variables, convert the values of the input variables into fuzzy sets, perform fuzzy processing on the output variables, convert the output variables into fuzzy sets, formulate fuzzy rules to describe the adaptation type fitness under different combinations of data types, and infer the adaptation type of the currently applied pre-trained regional water service prediction model through the fuzzy rules with the fuzzy input variables.

[0030] In a preferred embodiment, a cloud computing-based intelligent early warning system for regional water services includes:

[0031] A data acquisition module that obtains the acquisition data of regional water services, obtains the operation data set of regional water services and uploads it to the cloud;

[0032] The prediction accuracy analysis module obtains the pre-trained regional water service prediction model of the current application deployed in the cloud, and conducts prediction accuracy analysis to preliminarily identify whether there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model of the current application;

[0033] The feature extraction module, if there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model of the current application, conducts further feature analysis to explore the sensitivity of the model to dynamic perturbations and the degree of response to environmental changes;

[0034] The adaptation type classification module classifies the pre-trained regional water service prediction model of the current application into a high-fitness model or a low-fitness model based on the results of feature analysis;

[0035] The optimization application module, when the pre-trained regional water service prediction model of the current application is classified as a low-fitness model, updates and optimizes the model, and applies the optimized regional water service prediction model;

[0036] The early warning module issues an early warning signal when the prediction result activates the early warning mechanism according to the prediction results of the high-fitness model or the optimized regional water service prediction model and the preset early warning mechanism.

[0037] The technical effects and advantages of the present invention:

[0038] Through dynamic perturbation testing and environmental change response analysis, the present invention comprehensively evaluates the adaptability of the model in a complex environment, ensures that the regional water service prediction model can accurately reflect dynamic changes. When the prediction result activates the early warning mechanism, the system can quickly issue an early warning signal, providing sufficient response time for management departments and users, effectively reducing the losses caused by emergencies. By obtaining the perturbation sensitivity index and the environmental change response index, the present invention can deeply analyze the dynamic performance of the model and optimize the low-fitness model. The optimized model has significantly improved prediction accuracy in a dynamic environment and can better cope with emergencies such as floods and pollution.

[0039] The present invention establishes a monitoring and adjustment mechanism for model accuracy for perturbation factors in a dynamic environment, significantly improving the stability of the system in a complex environment. Even in the case of fluctuating data quality or drastic changes in environmental conditions, the system can still operate efficiently. By using the cloud computing platform for data storage and processing, the present invention effectively reduces the data processing cost of the regional water service system. Through intelligent prediction and early warning, the management efficiency is greatly improved, and the need for manual intervention is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings;

[0041] Figure 1 This is the schematic diagram of a cloud - based intelligent early warning method for regional water services in the present invention.

[0042] Figure 2 This is the schematic diagram of a cloud - based intelligent early warning system for regional water services in the present invention. Specific embodiments

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0044] Refer to Figure 1 - Figure 2 The following embodiments are obtained:

[0045] Embodiment 1: A cloud - based intelligent early warning method for regional water services includes the following steps:

[0046] Obtain the collected data of regional water services, obtain the operation dataset of regional water services and upload it to the cloud; use Internet of Things sensors, monitoring devices, etc. to collect regional water service data (such as water level, water quality, flow rate, rainfall, etc.) in real time, clean, denoise and pre - process the collected raw data to form a standardized operation dataset of regional water services, and upload the operation dataset to the cloud to provide input data for subsequent model analysis. Ensure that the data source for model analysis is timely and comprehensive, covering the dynamic characteristics of regional water services. Uploading the data to the cloud can make full use of cloud computing resources to improve the efficiency of data storage and calculation.

[0047] Obtain the pre - trained regional water service prediction model currently in use deployed in the cloud, and conduct prediction accuracy analysis to preliminarily identify whether there are early signs of insufficient dynamic accuracy in the pre - trained regional water service prediction model currently in use; extract the pre - trained regional water service prediction model deployed in the cloud, confirm its current adaptation environment and configuration parameters, input the real - time operation data into the model, generate prediction results using a fixed time window, calculate the difference between the model prediction value and the true value, analyze the distribution of prediction errors and outliers, and compare the average value and standard deviation of the outliers with the preset reference value to preliminarily determine whether there are early signs of insufficient dynamic accuracy in the model, ensure the applicability of the prediction model, and judge whether its accuracy in the current dynamic environment meets the early warning requirements, and discover the possible deficiencies of the model in advance to provide a basis for subsequent optimization.

[0048] If there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use, further feature analysis is carried out to explore the sensitivity of the model to dynamic perturbations and its response to environmental changes. Then, based on the results of the feature analysis, the pre-trained regional water service prediction model currently in use is classified as a high-fitness model or a low-fitness model. Through dynamic perturbation testing, the sensitivity of the model to input perturbations is evaluated to generate a perturbation sensitivity index. Combining with changes in dynamic environmental factors, the response ability of the model to external changes is analyzed to generate an environmental change response index. By synthesizing these two feature indicators, the adaptability of the model in dynamic change scenarios is evaluated, the specific reasons and performance characteristics of the model's deficiencies (such as excessive sensitivity or response lag) are deeply understood, and the impact of the dynamic environment on the model performance is quantified, which can also provide a specific direction for adaptive optimization.

[0049] When the pre-trained regional water service prediction model currently in use is classified as a low-fitness model, the model is updated and optimized, and the optimized regional water service prediction model is applied. Using the fuzzy logic method, with the perturbation sensitivity index and the environmental change response index as inputs and the model adaptation type as the output, fuzzy processing and rule reasoning are carried out. According to the analysis results, the model is classified as a high-fitness model (good adaptability) or a low-fitness model (insufficient adaptability), automatically classifying the adaptability level of the model to ensure the consistency and objectivity of the analysis results, providing a basis for formulating targeted strategies (such as continued use or optimization) for different types of models. For the models classified as low-fitness, they are re-optimized (such as adjusting parameters, incremental training, or reconstruction), the optimized models are redeployed, and real-time data is used to verify their adaptability and prediction accuracy, improving the dynamic environment adaptability of the models, ensuring their effectiveness in the current application scenario, and reducing false alarms or missed alarms caused by insufficient model adaptability.

[0050] According to the prediction results of the high-fitness model or the optimized regional water service prediction model, and the preset early warning mechanism, an early warning signal is issued when the prediction results activate the early warning mechanism. Using the high-fitness model or the optimized model, the operating state of the regional water service is predicted in real time, the threshold of the early warning mechanism is set, and an early warning signal is automatically generated when the prediction results exceed the threshold. The early warning information is transmitted to relevant departments or users to assist in decision-making and response measures, realizing real-time monitoring and early warning of the regional water service, reducing losses caused by emergencies (such as floods or pollution), and ensuring the efficiency and safety of regional water service management.

[0051] Performing prediction accuracy analysis means: dividing a fixed time window into multiple consecutive sub - windows of the same time length, running the current prediction model in the cloud, inputting the collected real - time data into the model, generating prediction results for each sub - window, recording the actual prediction values of the model and the true monitoring values at the corresponding time points, constructing a comparison dataset of predictions and true values for each sub - window, calculating the absolute value of the difference between the actual prediction value of the model and the true monitoring value at the corresponding time point respectively, then comparing all the absolute values within the same sub - window with a preset deviation threshold, marking the time points where the absolute value is greater than or equal to the deviation threshold as abnormal time points, and counting the total number of abnormal time points within the same sub - window.

[0052] Primarily identifying whether there are early signs of insufficient dynamic accuracy in the pre - trained regional water service prediction model currently in use means: calculating the average value and standard deviation of abnormal points for the total number of abnormal time points in all sub - windows, then comparing the average value of abnormal points with a preset reference average value, and comparing the standard deviation of abnormal points with a preset reference standard deviation. If the average value of abnormal points is less than or equal to the preset reference average and the standard deviation of abnormal points is less than or equal to the preset reference standard deviation, a normal signal is generated. If the average value of abnormal points is not less than or equal to the preset reference average and the standard deviation of abnormal points is not less than or equal to the preset reference standard deviation, an abnormal signal is generated. When the abnormal signal is generated, it indicates that there are early signs of insufficient dynamic accuracy in the pre - trained regional water service prediction model currently in use.

[0053] Dividing a fixed time window into multiple consecutive sub - windows can refine the performance evaluation of the model, clarify the prediction performance of the model at different time periods, calculate the absolute value of the prediction error for each time point, and compare it with the preset deviation threshold, which can effectively mark the abnormal time points with large prediction deviations. By counting the number of abnormal time points, it is possible to quickly discover the moments when the model may fail in a dynamic environment, give early warnings of potential problems, count the total number of abnormal time points within each sub - window, and construct a comparison dataset of predictions and true values at the sub - window level, which helps to form a data - driven model diagnosis method. By counting the number of abnormal points in each sub - window, generating the average value and standard deviation of abnormal points, and comparing them with the preset reference values, it is possible to quickly determine whether there are early signs of insufficient dynamic accuracy in the model. This judgment provides a clear basis for subsequent model exploration and avoids blind adjustment.

[0054] Exploring the sensitivity of the model to dynamic disturbances and the degree of response to environmental changes means:

[0055] Perform dynamic perturbation testing on the pre-trained regional water service prediction model currently in use, generate a perturbation sensitivity index that reflects the degree of change in the output result of the model when the input data is perturbed, and measures the sensitivity of the model to dynamic perturbations, and generate an environmental change response index that combines the amplitude of dynamic environmental changes and the accuracy of the model prediction results, reflecting the response ability of the model to external environmental changes. By simulating the perturbation of the input data and evaluating the amplitude of change in the model output result, the stability and robustness of the model in a dynamic input environment can be identified. The environmental change response analysis combines the dynamic change amplitude of environmental factors and the model prediction error to measure the adaptability of the model to external environmental changes. These two analyses can help to comprehensively understand the adaptation performance of the model in complex and dynamic scenarios, rather than being limited to the accuracy in static scenarios.

[0056] Dynamic perturbation testing: It can reveal the sensitivity of the model to small changes in input data (such as noise, sensor errors), and determine whether there are problems of being overly sensitive or insensitive to perturbations.

[0057] Environmental change response index: By analyzing the response effect of the model when environmental factors change drastically (such as sudden rainfall, sharp increase in flow rate), the dynamic change of its prediction accuracy is evaluated. These tests can locate potential weaknesses of the model and provide directions for subsequent optimization. Exploring the impacts of dynamic perturbations and environmental changes can ensure that the prediction performance of the model is more stable in dynamic scenarios, thereby enhancing the warning reliability and accuracy of the water service system. When the external environment changes drastically (such as sudden floods or pollution), the optimized model can issue warning signals more quickly and accurately, reducing the losses caused by emergencies.

[0058] The acquisition logic of the perturbation sensitivity index is as follows:

[0059] Obtain the currently running regional water service prediction model from the cloud, including its model architecture, training parameters, and historical data records. Verify whether the model is the latest version, and confirm whether the pre-trained environment and dataset are consistent with the current application environment. Collect real-time water service data in the current application area, including key variables such as water level, water quality, flow velocity, rainfall, etc., and generate a dataset containing multiple time periods for the actual operation analysis of the model. The perturbation sensitivity index is used to measure the degree of change in the output result of the prediction model when small changes (perturbations) occur in the input data, reflecting the sensitivity and adaptability of the model to perturbations.

[0060] Generate a set of basic input data for inputting into the model for prediction to obtain basic prediction results. Apply perturbations to the input data to generate several sets of perturbation datasets D 1 , D 2 ,..., D M , where M represents the number of sets of perturbation datasets, i.e., the number of perturbation times. The perturbation formula is:

[0061] D j (i) = D(i) + δ j (i); D(i) is the i-th sample of the base input, and δ j (i) is the amount of perturbation applied to the i-th sample at the j-th perturbation, usually a small random number (normal distribution or uniform distribution), and D j (i) represents the i-th sample after the j-th perturbation; Reasons for using the normal distribution as the amount of perturbation: Naturalness and real-world simulation: In many practical scenarios, environmental factors or data acquisition errors often follow a normal distribution. Using the amount of perturbation generated by the normal distribution can be closer to the random perturbation situation in the real world. Combining centrality and randomness: The normal distribution is characterized by most of the amounts of perturbation being concentrated near the mean and fewer extreme values. This characteristic can simulate the actual scenario where small perturbations are dominant and large perturbations occur occasionally.

[0062] Reasons for using the uniform distribution as the amount of perturbation: Uniformity and unbiasedness: The uniform distribution is characterized by each possible amount of perturbation having an equal probability, which can simulate completely unknown or randomly perturbed situations without a preferred direction. Range controllability: The range of the amount of perturbation generated by the uniform distribution is fixed, which is convenient for strictly constraining the perturbation intensity and avoiding excessive perturbation from affecting the analysis results. Testing the robustness of the model: The uniform distribution can provide a relatively uniform perturbation distribution, which is convenient for verifying the stability of the model under all possible perturbation situations.

[0063] Input the base dataset and each group of perturbed datasets D j into the model respectively to obtain the corresponding predicted outputs. The predicted output of the base dataset is marked as Each group of perturbed datasets D j 's predicted output is marked as Then substitute it into the perturbation sensitivity index calculation formula:

[0064] represents the i-th predicted value after the j-th perturbation, represents the i-th predicted value without perturbation, N represents the number of test samples during dynamic perturbation testing, and PSI represents the perturbation sensitivity index.

[0065] Perturbation generation: By applying small perturbations to the input data, simulate the behavior of the model when facing environmental changes or data noise; Prediction change analysis: Calculate the changes between the basic prediction results and the perturbed prediction results, capture the degree of response of the model to small input changes, average the prediction changes in all perturbation cases to generate a unified perturbation sensitivity index, and quantify the overall sensitivity of the model; A higher value of the perturbation sensitivity index indicates that the model is very sensitive to input perturbations and may have a risk of insufficient adaptability; A lower value of the perturbation sensitivity index indicates that the model has strong robustness and adaptability to perturbations; In this way, the adaptability of the model in a dynamic environment can be objectively evaluated, and it can also provide a direction for model optimization.

[0066] The acquisition logic of the environmental change response index is as follows:

[0067] Obtain environmental factor data related to regional water services such as rainfall, water level changes, water quality parameters, etc. Generally, use the data corresponding to the most drastic environmental factor. The more drastic the change, the easier it is to explore the response degree of the model. Determine the time series of environmental factor X, record the environmental factor value X(t) at each time point, and calculate the change amplitude between adjacent time points for the environmental factor time series: ΔX(t) = |X(t) - X(t - 1)|; X(t) represents the environmental factor value at the t-th time point, X(t - 1) represents the environmental factor value at the (t - 1)-th time point, and ΔX(t) represents the environmental change amplitude at the t-th time point;

[0068] Collect the prediction results and true observed values of the model, and calculate the prediction error at each time point: E(t) = |Y predicted (t) - Y actual (t)|; E(t) represents the prediction error at the t-th time point, Y predicted (t) represents the prediction result of the model at the t-th time point, Y actual (t) represents the true observed value at the t-th time point;

[0069] The formula for calculating the environmental change response index is:

[0070] T represents the total number of time points, ∈ represents a preset constant, α and β are both preset non-zero proportional coefficients, γ is a preset non-linear amplification factor with a value greater than one to enhance the sensitivity to large error points, PJ represents the average value of all environmental factor data, and ECRI represents the environmental change response index. α * ΔX(t) 2 Emphasize the weight effect when the change amplitude of the environmental factor is large, and square to amplify the influence of dynamic changes. Emphasize the relative change ratio, which reflects the suddenness of environmental changes and amplifies scenarios with small but sudden changes. Environmental factor change assessment: Quantify the change trend of the dynamic environment by calculating the environmental change amplitude between adjacent time points. Association between prediction error and environmental change: Combine the prediction error with the environmental change amplitude to highlight the impact of environmental changes on model prediction. Exponential normalization calculation: Eliminate the absolute magnitude difference through normalization so that the exponential value can directly reflect the adaptability of the model. Higher exponential value: Indicates that when the environmental change is large, the model prediction error increases significantly and the adaptability is poor. Lower exponential value: Indicates that the model can better maintain the prediction accuracy in a dynamic environment and has strong adaptability. Through the environmental change response index, the performance of the model in dynamic environmental changes can be comprehensively evaluated, providing an important basis for model optimization in a dynamic environment.

[0071] Dividing the currently applied pre-trained regional water service prediction model into a high-adaptability model or a low-adaptability model based on the results of feature analysis means:

[0072] Obtain the environmental change response index and the perturbation sensitivity index of the currently applied pre-trained regional water service prediction model, and use them together as the input data for fuzzy logic. Use the adaptation type of the currently applied pre-trained regional water service prediction model as the output data of fuzzy logic. Fuzzify the input variables, convert the values of the input variables into fuzzy sets, fuzzify the output variables, convert the output variables into fuzzy sets, formulate fuzzy rules to describe the adaptation type fitness under different combinations of data types, and infer the adaptation type of the currently applied pre-trained regional water service prediction model through the fuzzy rules for the fuzzified input variables.

[0073] According to the feature analysis results, obtain the dynamic perturbation sensitivity index and the response ability index to environmental changes of the model respectively. These two feature indexes jointly reflect the adaptation ability of the model in a dynamic change scenario. Use the dynamic perturbation sensitivity index and the environmental change response ability index as the input variables of the fuzzy logic system. Fuzzify the input variables, map the specific numerical values into fuzzy sets, such as fuzzy categories like "high", "medium", "low", etc. For example, the sensitivity index can be divided into "low sensitivity", "medium sensitivity", "high sensitivity". The response ability index can be divided into "low response ability", "medium response ability", "high response ability".

[0074] The output variable is the adaptation type of the model, such as "high-adaptability model" or "low-adaptability model". Fuzzify the output variable and convert the adaptation type into a fuzzy set. For example, "high adaptability", "medium adaptability", "low adaptability".

[0075] According to the combination of characteristic indicators, fuzzy rules are designed to describe the adaptation type fitness under different combinations. For example: when the sensitivity indicator is "high sensitivity" and the response ability indicator is "low response ability", the model may have a low adaptability and should be classified as a "low fitness model". When the sensitivity indicator is "low sensitivity" and the response ability indicator is "high response ability", the model may have a high adaptability and should be classified as a "high fitness model". When both sensitivity and response ability are "medium", the model adaptability may be at a medium level.

[0076] Using fuzzy logic rules, the fuzzified input variables are inferred through the rules to calculate the fitness of each adaptation type. During the inference process, the input indicator combination corresponding to each rule will generate a membership degree of an adaptation type.

[0077] The fuzzified result after inference is defuzzified to convert it into a clear adaptation type result. The final adaptation type of the output model is judged whether it is a "high fitness model" or a "low fitness model". If the model is classified as a "high fitness model", it means that it can adapt to dynamic change scenarios and no further optimization is required. If the model is classified as a "low fitness model", it needs to enter the optimization process to retrain or adjust its parameters.

[0078] Fuzzy logic can handle the uncertainty of characteristic indicators. Through the fuzzification and inference processes, an accurate classification of the model adaptation type is achieved. After clearly classifying the low fitness models, targeted optimization or replacement can be carried out to improve the prediction performance. Through the classification of adaptation types, the effectiveness of the regional water service prediction model in dynamic change scenarios is ensured, and the accuracy and timeliness of the early warning mechanism are guaranteed.

[0079] Embodiment 2: A regional water service intelligent early warning system based on cloud computing, comprising:

[0080] A data acquisition module, which acquires the acquisition data of the regional water service, obtains the operation data set of the regional water service and uploads it to the cloud;

[0081] A prediction accuracy analysis module, which acquires the pre-trained regional water service prediction model currently in use deployed in the cloud and conducts a prediction accuracy analysis to preliminarily identify whether there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use;

[0082] A feature extraction module, if there are early signs of insufficient dynamic accuracy in the pre-trained regional water service prediction model currently in use, then conducts further feature analysis to explore the sensitivity of the model to dynamic disturbances and the degree of response to environmental changes;

[0083] An adaptation type classification module divides the pre-trained regional water service prediction model of the current application into a high-fitness model or a low-fitness model based on the results of feature analysis;

[0084] An optimization application module updates and optimizes the model when the pre-trained regional water service prediction model of the current application is classified as a low-fitness model, and applies the optimized regional water service prediction model;

[0085] An early warning module issues an early warning signal when the prediction result activates the early warning mechanism according to the prediction result of the high-fitness model or the optimized regional water service prediction model, and the preset early warning mechanism.

[0086] The early warning module is based on two core inputs: the prediction results of the high-fitness model or the optimized regional water service prediction model, which provide predictions of future trends in the regional water service status. The preset early warning mechanism, including thresholds and rules, is used to determine whether the current prediction result reaches the early warning trigger condition.

[0087] Steps for constructing the early warning mechanism: Early warning threshold setting: Set the early warning thresholds for different scenarios according to historical data and the specific requirements of the regional water service. For example: When the water level exceeds a certain set value, it may trigger a flood early warning. When the concentration of water quality pollutants exceeds the standard, it may trigger a pollution early warning.

[0088] Multi-level early warning rules: Set multi-level early warnings (such as mild, moderate, severe) according to the severity and impact range of the prediction results, and each level corresponds to different response strategies. For example: Mild early warning: Prompt to pay attention, no immediate measures are required. Moderate early warning: Requires preliminary intervention by relevant departments. Severe early warning: Immediately initiate an emergency response.

[0089] Trigger logic of the early warning signal:

[0090] Compare the prediction result with the preset threshold: Compare the model prediction result with the threshold of the early warning mechanism one by one to determine whether the early warning condition is met. If the prediction result exceeds the early warning threshold, the corresponding early warning signal is triggered. If the prediction result is within the safe range, the early warning signal is not triggered. Trend analysis: Not only consider a single prediction result, but also analyze the change trend of the result. For example, when the water level gradually rises and approaches the threshold, an early warning signal may be triggered in advance.

[0091] Generation of the early warning signal: When the prediction result activates the early warning mechanism, the system will generate the corresponding early warning signal, which includes the following information: Early warning type (such as flood, pollution, water supply shortage, etc.). Early warning level (such as mild, moderate, severe). Early warning range (such as the geographical area affected). Time information (such as when it is expected to reach the critical state). The early warning signal is sent to relevant departments and users through various methods, such as SMS notification, email, push on the real-time monitoring platform, etc.

[0092] Triggering of emergency response: According to the level and type of early warning signals, the early warning module can also trigger the emergency response module to initiate corresponding management and intervention measures. For example, during a flood warning, relevant departments are notified to turn on drainage facilities. During a pollution warning, water quality purification equipment is dispatched or the backup water source is switched.

[0093] Based on a highly adaptable or optimized model, the early warning module ensures the accuracy and timeliness of early warning signals, providing sufficient response time for relevant departments. Different levels of early warning signals correspond to different response strategies, which can optimize resource allocation, reduce unnecessary interventions and losses, improve the intelligent level of regional water management through an accurate early warning mechanism, enhance the ability to respond to emergencies, and reduce potential disaster risks. The early warning module not only issues signals but also helps the management department quickly formulate scientific response measures.

[0094] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0095] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0096] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0097] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0098] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A regional water affairs intelligent early warning method based on cloud computing, characterized in that: The following steps are involved: Acquire the collected data of regional water affairs, obtain the operational data set of regional water affairs and upload it to the cloud, obtain the pre-trained regional water affairs prediction model currently applied and deployed in the cloud, and perform prediction accuracy analysis to preliminarily identify whether the pre-trained regional water affairs prediction model currently applied has early signs of insufficient dynamic accuracy; If the currently used pre-trained regional water affairs forecasting model shows early signs of insufficient dynamic accuracy, further feature analysis is conducted to explore the model's sensitivity to dynamic disturbances and its responsiveness to environmental changes, and then the currently used pre-trained regional water affairs forecasting model is classified as a high fitness model or a low fitness model based on the results of the feature analysis; When the currently applied pre-trained regional water affairs prediction model is classified as a low fitness model, the model is updated and optimized, and the optimized regional water affairs prediction model is applied; According to the prediction results of the high-fitness model or the optimized regional water affairs prediction model, and the preset early warning mechanism, an early warning signal is issued when the prediction results activate the early warning mechanism.

2. According to the cloud computing-based regional water affairs intelligent early warning method of claim 1, it is characterized in that: To perform a forecast accuracy analysis means: A fixed time window is divided into multiple continuous sub-windows with the same time length, the current prediction model is run on the cloud, the collected real-time data is input into the model, each sub-window generates a prediction result, the actual prediction value of the model and the real monitoring value at the corresponding time point are recorded, and a comparison data set of the prediction and the real value of each sub-window is constructed. The absolute value of the difference between the actual prediction value of the model and the real monitoring value at the corresponding time point is calculated respectively, and then all absolute values ​​in the same sub-window are compared with the preset deviation threshold, and the time point with an absolute value greater than or equal to the deviation threshold is marked as an abnormal time point, and the total number of abnormal time points in the same sub-window is counted.

3. According to the cloud computing-based regional water affairs intelligent early warning method of claim 2, it is characterized in that: Early signs of insufficient dynamic accuracy in the currently used pre-trained regional water forecasting models are: The total number of abnormal time points in all sub-windows is used to calculate the average value and standard deviation of the abnormal points, and then the average value of the abnormal points is compared with the preset benchmark average value, and the standard deviation of the abnormal points is compared with the preset benchmark standard deviation. If the average value of the abnormal points is less than or equal to the preset benchmark average and the standard deviation of the abnormal points is less than or equal to the preset benchmark standard deviation, a normal signal is generated. If the average value of the abnormal points is less than or equal to the preset benchmark average and the standard deviation of the abnormal points is less than or equal to the preset benchmark standard deviation, an abnormal signal is generated. When the abnormal signal is generated, it indicates that the pre-trained regional water affairs prediction model currently being applied has early signs of insufficient dynamic accuracy.

4. According to the cloud computing-based regional water affairs intelligent early warning method of claim 3, it is characterized in that: Exploring the model's sensitivity to dynamic disturbances and its responsiveness to environmental changes refers to: A dynamic disturbance test is conducted on the currently used pre-trained regional water affairs prediction model to generate a disturbance sensitivity index that reflects the degree of change in the model's output results when the input data is disturbed, which measures the model's sensitivity to dynamic disturbances. In addition, an environmental change response index is generated by combining the amplitude of dynamic environmental changes and the accuracy of the model's prediction results to reflect the model's responsiveness to external environmental changes.

5. According to the cloud computing-based regional water affairs intelligent early warning method of claim 4, it is characterized in that: The logic for obtaining the disturbance sensitivity index is: Generate a set of basic input data for input into the model for prediction, obtain basic prediction results, apply perturbations to the input data, and generate several sets of perturbation data sets D1, D2, ..., D M , M represents the number of groups of perturbation data sets, i.e., the number of perturbations, and the perturbation formula is: D j (i)=D(i)+δ j (i); D(i) is the i-th sample of the basic input, δ j (i) is the disturbance amount applied to the i-th sample by the j-th disturbance, D j (i) represents the i-th sample after the j-th disturbance; The basic data set and each group of perturbation data sets D j Input the model respectively to obtain the corresponding prediction output. The prediction output of the basic data set is marked as Each set of perturbation data sets D j The predicted output of Then substitute it into the disturbance sensitivity index calculation formula: represents the i-th predicted value after the j-th disturbance, represents the i-th predicted value when there is no disturbance, N represents the number of test samples during dynamic disturbance test, and PSI represents the disturbance sensitivity index.

6. The method for intelligent early warning of regional water affairs based on cloud computing according to claim 5 is characterized in that: The logic for obtaining the environmental change response index is: Obtain environmental factor data related to regional water affairs, determine the time series of environmental factor X, record the environmental factor value X(t) at each time point, and calculate the change range between adjacent time points for the environmental factor time series: ΔX(t) = |X(t)-X(t-1)|; X(t) represents the environmental factor value at the t-th time point, X(t-1) represents the environmental factor value at the t-1th time point, and ΔX(t) represents the environmental change range at the t-th time point; Collect the model's prediction results and true observations, and calculate the prediction error at each time point: E(t) = |Y predicted (t)-Y actual (t)|; E(t) represents the prediction error at the tth time point, Y predicted (t) represents the prediction result of the model at the tth time point, Y actual (t) represents the true observation value at the tth time point; The formula for calculating the environmental change response index is: T represents the total number of time points, ∈ represents a preset constant, α and β are both preset non-zero proportional coefficients, γ is a preset nonlinear amplification factor, and its value is greater than one, PJ represents the average value of all environmental factor data, and ECRI represents the environmental change response index.

7. The method for intelligent early warning of regional water affairs based on cloud computing according to claim 6 is characterized in that: Based on the results of feature analysis, the currently applied pre-trained regional water affairs prediction model is divided into a high fitness model or a low fitness model, which means: The environmental change response index and disturbance sensitivity index of the currently applied pre-trained regional water affairs prediction model are obtained and used as the input data of the fuzzy logic together. The adaptation type of the currently applied pre-trained regional water affairs prediction model is used as the output data of the fuzzy logic. The input variables are fuzzified and the values ​​of the input variables are converted into fuzzy sets. The output variables are fuzzified and the output variables are converted into fuzzy sets. Fuzzy rules are formulated to describe the adaptability of the adaptation type under different data type combinations. The fuzzified input variables are inferred through fuzzy rules to obtain the adaptation type of the currently applied pre-trained regional water affairs prediction model.

8. A regional water affairs intelligent early warning system based on cloud computing, implemented based on a regional water affairs intelligent early warning method based on cloud computing according to any one of claims 1 to 7, characterized in that: include: The data collection module obtains the collected data of regional water affairs, obtains the operating data set of regional water affairs and uploads it to the cloud; The prediction accuracy analysis module obtains the pre-trained regional water affairs prediction model currently applied and deployed in the cloud, and performs prediction accuracy analysis to preliminarily identify whether the pre-trained regional water affairs prediction model currently applied has early signs of insufficient dynamic accuracy; Feature extraction module: If the currently applied pre-trained regional water forecasting model shows early signs of insufficient dynamic accuracy, further feature analysis is performed to explore the model's sensitivity to dynamic disturbances and its responsiveness to environmental changes; The adaptation type classification module divides the currently applied pre-trained regional water affairs prediction model into a high fitness model or a low fitness model based on the result of feature analysis; The optimization application module updates and optimizes the model when the pre-trained regional water affairs prediction model currently applied is classified as a low fitness model, and applies the optimized regional water affairs prediction model; The early warning module, based on the prediction results of the high fitness model or the optimized regional water affairs prediction model and the preset early warning mechanism, sends out an early warning signal when the prediction results activate the early warning mechanism.

Images