An artificial intelligence-based regional water demand calculation method

Abstract

The present application relates to the technical field of rainfall prediction, and particularly relates to a regional water demand calculation method based on artificial intelligence. The method comprises the following steps: firstly, obtaining basic meteorological data such as temperature, humidity, rainfall and historical water consumption data of a target region; then, extracting the humidity change trajectory and performing differential analysis with the theoretical trajectory to obtain the air humidity hysteresis characteristic; then, segmenting the rainfall time series, calculating the increase / decrease amplitude and rainfall interval of each segment, and forming rainfall characteristic data; subsequently, combining the air hysteresis humidity data, rainfall characteristic data and historical water consumption data to construct a regional water demand prediction model and output a prediction value; finally, correcting the prediction value according to the dynamic change of temperature to obtain the final regional water demand which is more in line with the actual conditions. The present application realizes intelligent prediction of regional water demand with high precision, strong environmental adaptability and more in line with the real evapotranspiration law by fusing humidity hysteresis, rainfall stage characteristics and temperature dynamic regulation.

Description

Technical Field

[0001] This invention relates to the field of rainfall forecasting technology, and in particular to an artificial intelligence-based method for calculating regional water demand. Background Technology

[0002] Accurate calculation of regional water demand is a crucial foundation for various fields, including agricultural irrigation management, water resource allocation, and urban green space maintenance. It plays a key role in improving water resource utilization efficiency and ensuring crop growth and the ecological environment. With increasing uncertainty in climate change and precipitation distribution, relying solely on historical water use experience or traditional statistical methods for water demand forecasting is no longer sufficient to meet the needs of refined management.

[0003] Existing technologies typically employ calculation methods based on average meteorological data and linear empirical formulas, considering only macro-meteorological factors such as temperature and rainfall, while ignoring the effects of air humidity lag, micro-topographical differences, and the impact of rainfall segmentation on soil moisture dynamics. This leads to deviations in prediction results in areas with short-term meteorological changes or complex terrain, failing to fully reflect the actual water demand of the region. Summary of the Invention

[0004] Therefore, it is necessary to provide an artificial intelligence-based method for calculating regional water demand in order to solve at least one of the aforementioned technical problems.

[0005] To achieve the above objectives, an artificial intelligence-based method for calculating regional water demand is provided, the method comprising the following steps: Step S1: Obtain basic meteorological data and historical water use data for the target area, wherein the basic meteorological data includes at least temperature, humidity and rainfall; Step S2: By analyzing the humidity change trajectory of the target area, and based on the difference between the humidity change trajectory and the preset theoretical change trajectory, the air humidity lag of the target area is analyzed to obtain the air humidity lag data of the target area. Step S3: Divide the rainfall into segments according to the time series, and calculate the increase or decrease in rainfall in each segment and the duration of rainfall interval between adjacent segments to obtain the rainfall characteristic data of the target area; Step S4: Construct a regional water demand prediction model based on air humidity data, rainfall characteristic data, and historical water use data; use the regional water demand prediction model to predict the water demand of the target area and obtain the predicted regional water demand value. Step S5: Adjust the predicted regional water demand by dynamically adjusting the temperature to obtain the adjusted regional water demand.

[0006] The present invention has the following beneficial effects: I. By constructing a multi-level analysis system encompassing humidity trajectory difference, residual sequence generation, micro-topographic thermal inertia calculation, and spatial matching, the correlation between air humidity changes and actual environmental characteristics is refined. By smoothing the humidity residual sequence in time segments, calculating correlation curves, and identifying the points of maximum correlation, the humidity response lag time can be accurately obtained across multiple time periods and scales. Simultaneously, by utilizing digital elevation data to obtain topographic slope and aspect, and combining this with land cover type to calculate the thermal inertia of grid cells, the lag time not only reflects meteorological factors but also comprehensively embodies micro-topographic thermophysical differences. This multi-dimensional coupling processing effectively avoids the problem of coarse lag descriptions caused by traditional methods relying solely on meteorological sequences, significantly improving the accuracy and spatial continuity of air humidity data.

[0007] Second, this invention uses air humidity retention data, rainfall characteristic data, and historical water use data as model inputs, enabling the prediction model to simultaneously capture important hydrological characteristics such as "the lagging effect of air humidity," "segmented increases and decreases in rainfall," and "the periodicity of water replenishment driven by rainfall intervals." Traditional water demand prediction typically relies on static indicators of rainfall and temperature and humidity, lacking in-depth consideration of rainfall variation, intervals between rainfall events, and the humidity retention effect, making it prone to prediction bias under rapidly changing meteorological conditions. This scheme, by introducing deeply structured meteorological segmentation characteristics and lag response information, enables the model to dynamically identify trends of rapid accumulation or dissipation of regional moisture, thereby significantly improving the prediction model's adaptability to different climate stages, different topographical areas, and different water use patterns, and enhancing the stability and interpretability of the prediction results.

[0008] Third, the temperature regulation steps of this invention not only consider the hourly temperature variation and diurnal temperature range, but also dynamically enhance or weaken the preliminary predicted values ​​based on the temperature change trend (heating or cooling phase), achieving intelligent correction of surface water loss. The dynamic control method of increasing the rate during the heating phase and decreasing it during the cooling phase makes the predicted results more consistent with the temporal differences in actual evapotranspiration rates, overcoming the shortcomings of traditional schemes that have coarse temperature corrections and difficulty in reflecting continuous temperature changes. By constructing a mapping relationship between temperature influence and predicted values, high prediction accuracy can be maintained under different seasons and different daily variation intensities, ensuring that the final regional water demand results are more practical, applicable in the field, and of greater regulatory significance. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the steps of an artificial intelligence-based method for calculating regional water demand. Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2. Figure 3This is a schematic diagram of the module architecture of a regional water demand calculation method based on artificial intelligence according to this application; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 3 A method for calculating regional water demand based on artificial intelligence, the method comprising the following steps: Step S1: Obtain basic meteorological data and historical water use data for the target area, wherein the basic meteorological data includes at least temperature, humidity and rainfall; In one embodiment, a meteorological monitoring unit is first deployed in the target area, including basic equipment such as temperature sensors, humidity sensors, and rainfall sensors. The temperature sensor is used to collect air temperature data in the area, with a sampling frequency set to once every 10 minutes; the humidity sensor is used to obtain relative humidity information, with a sampling frequency consistent with that of the temperature sensor; the rainfall sensor is used to monitor precipitation and rainfall duration in real time, with a sampling interval that can be set to once per hour or dynamically adjusted according to rainfall events.

[0014] The data acquisition module formats and processes the collected temperature, humidity, and rainfall data, and stores them uniformly according to timestamps to generate a basic meteorological data set for the target area. Simultaneously, historical water use data is obtained from the regional water management platform, including water consumption records for various water types (such as domestic water, agricultural irrigation water, and industrial water) at different time periods. Historical water use data can be summarized by day, week, or month for long-term trend analysis.

[0015] To ensure data integrity and accuracy, preliminary preprocessing is performed on the collected basic meteorological data and historical water usage data, including outlier detection, missing value compensation, and unit standardization. For example, abnormal temperature or humidity values ​​that occur during sensor acquisition are corrected by interpolation using data from nearby time points; missing or erroneous entries in historical water usage records can be filled in using data from surrounding water meters.

[0016] After data processing is completed, basic meteorological data and historical water use data are linked and stored to form a comprehensive dataset with time labels and spatial locations. This dataset can be directly used for subsequent regional water demand calculation and water resource scheduling analysis.

[0017] Step S2: By analyzing the humidity change trajectory of the target area, and based on the difference between the humidity change trajectory and the preset theoretical change trajectory, the air humidity lag of the target area is analyzed to obtain the air humidity lag data of the target area. In one embodiment, air humidity data of the target area is first acquired in real time based on a sensor network, and a humidity time series is generated according to a fixed sampling period (e.g., every 5 minutes or every 10 minutes). The time series data records the humidity changes at different sensor locations, and missing values ​​between samples can be filled in using an interpolation algorithm to form a continuous humidity change trajectory.

[0018] Subsequently, the data processing module compares the humidity change trajectory of the target area with a pre-established theoretical humidity change trajectory. The theoretical humidity change trajectory is usually generated based on meteorological models or historical statistical data, reflecting the ideal humidity response under specific environmental conditions. By calculating the humidity difference at each time step, a humidity deviation sequence for each sampling point is obtained.

[0019] To further analyze the lag effect of air humidity, the humidity deviation sequence is correlated with the time axis, and the characteristics of humidity response lag are identified through time lag analysis. For example, a sliding window is used to calculate the time delay required for the humidity deviation to reach its peak, or the cumulative response time of the deviation over a continuous time period is calculated.

[0020] Based on the above analysis, air humidity retention data for the target area is generated, including information such as humidity response delay time, humidity deviation amplitude, and hysteresis response trend for each monitoring point. Simultaneously, the data from each point are aggregated to form a regional air humidity retention data matrix, used to describe the spatial distribution characteristics of the humidity response throughout the target area.

[0021] Step S3: Divide the rainfall into segments according to the time series, and calculate the increase or decrease in rainfall in each segment and the duration of rainfall interval between adjacent segments to obtain the rainfall characteristic data of the target area; In one embodiment, real-time rainfall data for the target area is first acquired based on meteorological monitoring stations or on-site rain gauges, and a rainfall time series is generated according to a fixed sampling period (e.g., every 5 minutes or every 10 minutes). The time series records the continuous changes in rainfall over time, and missing values ​​between samples can be processed using linear interpolation or curve fitting methods to ensure the continuity of the sequence.

[0022] Subsequently, the data processing module divides the rainfall time series into several time periods, the length of which can be set according to application requirements (e.g., 10 minutes, 30 minutes, or 1 hour). Within each time period, the increase or decrease in rainfall is statistically analyzed, specifically including: the change in total rainfall within the time period, the average rainfall intensity, and the instantaneous peak rainfall.

[0023] Further calculations are performed on the rainfall intervals between adjacent time periods, i.e., the time interval between the end of one time period and the beginning of the next with no rainfall or low-intensity rainfall, to reflect the intermittent nature of rainfall. By combining the increase or decrease in rainfall with the rainfall intervals, rainfall characteristic data for the target area are generated, including rainfall amount, rainfall intensity, rainfall trend, and interval duration for each time period.

[0024] Step S4: Construct a regional water demand prediction model based on air humidity data, rainfall characteristic data, and historical water use data; use the regional water demand prediction model to predict the water demand of the target area and obtain the predicted regional water demand value. In one embodiment, the target area air humidity data obtained in step S2, the rainfall characteristic data generated in step S3, and the historical water use data of the target area are first used as input data. The air humidity data includes the humidity lag, lag time, and humidity change trend at each measuring point; the rainfall characteristic data includes the rainfall amount, rainfall intensity, increase / decrease, and rainfall interval duration for each time period; and the historical water use data includes each irrigation cycle, total water use, and water use distribution characteristics.

[0025] Based on the aforementioned multi-source data, a regional water demand prediction model can be constructed using machine learning or statistical regression methods, such as Support Vector Regression (SVR), Random Forest (RF), Multiple Linear Regression, or Long Short-Term Memory (LSTM) networks, to capture the combined impact of air humidity lag, rainfall variations, and historical water consumption on future water demand. During the model training phase, supervised learning is performed by inputting historical data and corresponding actual water demand values ​​to automatically determine the weights of each input feature and the prediction parameters.

[0026] After constructing the regional water demand prediction model, the model is used to predict water demand for the target area in the future. The specific steps include: inputting air humidity data, rainfall characteristic data, and historical water use data into the model according to the current prediction time period; the model outputs the predicted water demand value for the corresponding time period, forming a regional water demand prediction curve. The prediction results can be quantified by hour, day, or irrigation cycle to guide precision irrigation and water resource allocation.

[0027] In another embodiment, reference can be made to Figure 3 First, a multi-source data acquisition and fusion module is used to acquire air humidity data, rainfall characteristic data, and historical water use data for the target area. The air humidity data includes humidity lag, lag time, and humidity change trends at each measuring point; the rainfall characteristic data includes rainfall amount, intensity, increase / decrease, and interval duration for each time period; and the historical water use data covers irrigation cycles, total water use, and water distribution characteristics. This data is acquired and fused through a distributed data stream API gateway and ETL architecture, similar to the multi-source fusion logic of meteorological measurement data, solar thermal radiation data, historical irrigation data, and crop growth data.

[0028] Next, we proceed to the data preprocessing and feature engineering stage. We perform physical dimension unification, information supplementation, and outlier identification on the above multi-source data. At the same time, we perform spatiotemporal feature encoding and multi-source feature fusion. Then, we achieve feature engineering output through feature selection and recoding (such as one-hot encoding) to provide regular feature data for subsequent modeling. This is consistent with the logic of physical dimension processing and spatiotemporal feature encoding in the figure.

[0029] Subsequently, in the deep learning prediction model module, the processed air humidity, rainfall characteristics, and historical water use data are used as input feature sequences. On the one hand, spatial feature extraction is performed (using a graph attention network GAT to focus on the spatial correlation of humidity lag and rainfall intensity at each measuring point), and on the other hand, temporal feature extraction is performed (using a multi-scale attention LSTM to capture the time dependence of humidity change trends, rainfall interval duration, and historical water use cycles). Then, through a multi-feature fusion module (integrating external meteorological cross-features and user water use pattern features) and an uncertainty quantification module (combining Monte Carlo Dropout and Bayesian neural networks for prediction uncertainty analysis), the deep construction of the model's feature layer is completed.

[0030] After model construction, the model enters the training and optimization stage. Supervised learning is adopted, and historical data and corresponding actual water demand values ​​are used as labels to automatically determine the weights and prediction parameters of each input feature (air humidity retention, rainfall characteristics, and historical water use characteristics). At the same time, model performance is improved by combining model compression (such as knowledge distillation), hyperparameter optimization (such as Bayesian optimization), and transfer learning (such as cross-regional water use pattern transfer). This is consistent with the multi-strategy logic of model training and optimization shown in the figure.

[0031] The prediction results are then post-processed, including outputting water resource demand levels, predictive maintenance (such as model drift detection), and generating interpretable reports (such as generating reports on the contribution of each feature to the water demand prediction) for the regional water demand prediction values ​​output by the model, thereby realizing multi-dimensional value mining of the prediction results.

[0032] Finally, through system deployment and real-time prediction, the trained model is deployed to a microservice architecture, supporting real-time prediction through IoT services, inference services, and API gateways. The model is input into the target area's current time period's air humidity data, rainfall characteristics data, and historical water use data, outputting the corresponding time period's predicted water demand value, forming a regional water demand prediction curve (quantified by hour, day, or irrigation cycle), thus guiding precision irrigation and water resource allocation. Simultaneously, leveraging a data-driven feedback mechanism, based on feature learning and model updates (e.g., online learning to update the model), continuous optimization loops (e.g., iteratively optimizing the model by combining real-time water use feedback data), and model monitoring and alarms (e.g., monitoring prediction errors to trigger model retraining alarms), the long-term predictive performance of the model is ensured. At the attention mechanism level, both graph attention calculations for spatial feature extraction (GAT's Query, Key, Value attention weight calculation) and temporal inter-attention for temporal feature extraction (TPA's FFN-Query, Scale, weight normalization) provide interpretable attention mechanism support for the feature interactions in water demand prediction.

[0033] Step S5: Adjust the predicted regional water demand by dynamically adjusting the temperature to obtain the adjusted regional water demand.

[0034] In one embodiment, temperature data of the target area is first acquired, which can be collected in real time through weather stations, environmental sensors, or satellite remote sensing. The temperature data includes air temperature, soil surface temperature, and nighttime minimum temperature, and the time series characteristics of temperature changes are recorded.

[0035] The predicted regional water demand obtained in step S4 is dynamically adjusted based on temperature data. Specifically, this involves: performing a correlation analysis between the predicted regional water demand and temperature data; and proportionally adjusting the water demand based on the impact of temperature on evaporation, crop transpiration, and soil moisture loss. For example, when the real-time temperature is higher than the historical average temperature, the adjustment coefficient for the predicted water demand is increased; when the temperature is lower than the historical average, the predicted water demand is appropriately reduced.

[0036] The adjustment process can employ empirical formulas, regression models, or dynamic weighting methods. Sensitivity coefficients to water requirement based on temperature are set according to different crop types and soil properties, enabling tiered and regional dynamic adjustments. The adjusted regional water requirement maintains temporal continuity, avoiding jumps in predicted values ​​due to abrupt temperature changes, while ensuring the rationality of water requirement proportions across sub-regions.

[0037] After adjustment, the final regional water demand, dynamically adjusted for temperature, is output, providing a precise reference for irrigation scheduling, intelligent pump control, and optimized water resource management. This embodiment improves the accuracy and practicality of regional water demand prediction by introducing dynamic temperature adjustment, ensuring that irrigation volume can still be rationally arranged even under fluctuating temperatures.

[0038] As an example of the present invention, reference is made to Figure 2 As shown, step S2 in this example includes: Step S21: Extract humidity observation data of the target area from the basic meteorological data within a predetermined time interval; Step S22: Perform time series processing and noise filtering on the humidity observation data to generate a continuous humidity change trajectory; Step S23: Perform hourly difference analysis between the humidity change trajectory of the target area and the pre-set theoretical humidity change trajectory to obtain the humidity residual sequence; Step S24: Analyze the air humidity lag in the target area using the humidity residual sequence and the micro-topographic thermal inertia analysis of the target area to obtain the air humidity lag data of the target area.

[0039] In one embodiment, multi-source environmental data such as humidity, temperature, and rainfall are acquired through basic meteorological monitoring devices in the target area, with humidity observation data being the key monitoring target. The basic meteorological monitoring devices include automatic weather stations, UAV environmental monitoring modules, or satellite remote sensing sensors. Humidity observation data is collected every 5 to 30 minutes, with a data accuracy of ±1% RH. The humidity observation data sequence is organized according to timestamp order, missing or abnormal data is marked, and linear interpolation or spline interpolation methods are used to supplement it to ensure the integrity of the humidity sequence. Subsequently, the preliminary humidity sequence is processed by sliding window averaging or low-pass filtering to remove sensor noise and sudden fluctuations, while retaining the true trend of humidity change. Data with abrupt changes exceeding ±5% RH are marked or smoothed to obtain a continuous and reliable humidity change trajectory.

[0040] Next, the processed humidity change trajectory is time-by-time differencingd with the preset theoretical humidity change trajectory to generate a humidity residual sequence. The theoretical humidity change trajectory can be fitted to historical meteorological observation data or generated based on crop water requirement models. The humidity deviation at each time point is calculated through time-by-time differencing, and the mean, variance, and maximum deviation are statistically analyzed to form a humidity residual sequence. Abnormal fluctuations are also marked to provide basic data for subsequent lag analysis. Subsequently, the lag of the humidity response is analyzed by combining the micro-topographic parameters and surface thermal inertia characteristics of the target area. Specifically, slope, aspect, soil type, and surface heat capacity parameters are obtained, and the humidity residual sequence is coupled with these parameters to calculate the lag time and lag amplitude at each monitoring point, thereby generating an air humidity dataset for the target area, providing reliable input for regional water demand prediction.

[0041] In the process of regional water demand forecasting, atmospheric humidity data, rainfall characteristics of the target area, and historical water use data are input into the constructed regional water demand forecasting model. Rainfall characteristics are obtained by segmenting rainfall into time series segments, calculating the increase or decrease in rainfall in each segment, and the duration of rainfall intervals between adjacent segments. The forecasting model outputs a preliminary water demand forecast for the target area, which reflects a reasonable water demand based on current meteorological conditions and historical water use. To further improve forecast accuracy, the preliminary forecast is dynamically adjusted using real-time temperature data of the target area. The dynamic temperature adjustment strategy adjusts the forecast by increasing or decreasing the value according to a set ratio based on the magnitude and trend of the current temperature deviating from the reference temperature, resulting in the adjusted final regional water demand, thus providing a scientific basis for irrigation regulation or water resource management.

[0042] Preferably, the method for obtaining the micro-topography thermal inertia of the target area in step S24 includes: Acquire digital elevation data within the target area, wherein the digital elevation data contains several grid cells; Micro-topographic features of the target area are extracted based on digital elevation data, and land cover type of the target area is analyzed based on the topographic features to obtain land cover type data; Based on land cover type data, the surface thermophysical parameters of different cover types are determined, and the surface thermal inertia value of each grid cell is calculated based on digital elevation data and surface thermophysical parameters to obtain the micro-topographic thermal inertia of the target area.

[0043] In one embodiment, digital elevation data (DEM) of the target area is acquired. This data is obtained through methods such as satellite remote sensing, UAV aerial mapping, or ground laser scanning. The DEM data is stored in the form of grid cells, with each grid cell recording its corresponding elevation value. The grid resolution can be set from 1 meter to 10 meters depending on the size of the area. Subsequently, micro-topographic features of the target area are extracted based on the DEM data, including parameters such as slope, aspect, local undulation, and concavity / convexity. At the same time, the terrain is classified according to the land cover in the area, such as bare soil, grassland, farmland, building-covered areas, or water-covered areas, generating land cover type data.

[0044] Next, based on land cover type data, corresponding surface thermophysical parameters are assigned to different cover types, including specific heat capacity, thermal conductivity, density, and thermal inertia coefficient. These parameters can be obtained from standard geographic information databases, historical measurement data, or laboratory calibration results. The digital elevation data of each grid cell is coupled with its corresponding surface thermophysical parameters to obtain the surface thermal inertia value of each grid cell, forming a micro-topographic thermal inertia dataset. This dataset can reflect the response speed and heat storage capacity of different locations within the region to changes in ambient temperature, thus providing a basic input for subsequent air humidity lag analysis.

[0045] Preferably, the calculation of the surface thermal inertia value of each grid cell based on digital elevation data and surface thermophysical parameters includes: The center point location and boundary area of ​​each grid cell are determined based on digital elevation data, and the corresponding digital elevation data is extracted to obtain the local terrain height. Calculate the slope and aspect of each grid cell based on the local terrain height of each grid cell; The theoretical surface thermal inertia value of each grid cell is calculated using surface thermophysical parameters. The grid cell thermal energy is then adjusted based on the slope and aspect of the grid cell to obtain the surface thermal inertia value.

[0046] In one embodiment, digital elevation data (DEM) of the target area is acquired. This data can be obtained through satellite remote sensing, aerial photogrammetry, or lidar, with the resolution selected according to actual application requirements, for example, from 1 meter to 30 meters. The target area is divided into several regular grid cells, and the boundary area and center point location of each grid cell are determined by the grid division scheme. Subsequently, the elevation value of the center point of each grid cell and the elevation information within the grid coverage area are extracted, and the local terrain height of the grid cell is determined by statistical or weighted averaging methods.

[0047] Next, the slope and aspect are calculated based on the local topographic elevation of each grid cell. Slope represents the angle of inclination of the grid cell surface relative to the horizontal plane, and aspect represents the azimuth of the slope. Specifically, this is achieved by calculating the local slope and slope direction within the neighborhood of each grid cell using the center point and adjacent elevation points to obtain accurate slope and aspect information. These topographic parameters reflect the tendency of the surface to receive and radiate heat, providing a basis for adjusting the surface thermal inertia value.

[0048] Subsequently, the theoretical surface thermal inertia value was calculated for each grid cell using surface thermophysical parameters. These parameters include soil heat capacity, thermal conductivity, soil type, surface cover properties, and moisture content. The theoretical thermal inertia was calculated based on the surface thermophysical parameters of each grid cell, and slope and aspect information were used as correction factors to adjust for heat transfer efficiency and absorption capacity, taking into account the influence of topography on heat distribution, thus obtaining the final surface thermal inertia value. This surface thermal inertia value can be used for subsequent regional air humidity lag analysis, soil moisture response prediction, and regional water demand prediction.

[0049] Preferably, step S24 includes: Based on the target area, the corresponding humidity residual sequence is spatially matched with the micro-topographic thermal inertia value of each grid cell to obtain the spatially matched grid cell; Time delay analysis is performed on the humidity residual sequence within each grid cell to calculate the lag time of the humidity response relative to the theoretical humidity change; Based on the thermal inertia of micro-topography, the humidity lag time of each spatially matched grid cell is weighted and adjusted, and the weighted grid cell lag time is aggregated regionally to obtain the air humidity data of the target area.

[0050] In one embodiment, the target area is divided into several grid cells, and each grid cell has its corresponding micro-topographic thermal inertia value. The pre-obtained humidity residual sequence is spatially matched with each grid cell. Specifically, by comparing the geographic coordinates of the humidity residual data with the center point and coverage area of ​​the grid cells, the location of the grid cell corresponding to the humidity residual sequence at each moment is determined, realizing the spatial correspondence between the humidity residual sequence and the terrain grid, thereby obtaining a spatially matched set of grid cells.

[0051] Subsequently, time delay analysis was performed on the humidity residual sequence within each spatially matched grid cell. Based on the trend of humidity residual changes over time, the lag time of the humidity response relative to the theoretical humidity change was calculated, i.e., the delay in the response of humidity change to the actual observed air humidity value in the grid cell was measured. Time delay analysis can be achieved through methods such as sequence cross-correlation, peak response time identification, or sliding window response analysis to obtain the humidity lag characteristics of each grid cell.

[0052] After obtaining the initial lag time, the humidity lag time of each grid cell is weighted and adjusted using the micro-topographic thermal inertia. Specifically, grid cells with higher micro-topographic thermal inertia respond more slowly to humidity changes, and their lag time needs to be appropriately increased; grid cells with lower micro-topographic thermal inertia respond more quickly, and their lag time needs to be appropriately decreased. This weighting process makes the humidity lag time of each grid cell more closely match the terrain and thermophysical characteristics.

[0053] Finally, the weighted adjusted lag times of each grid cell are aggregated regionally. This aggregation process involves averaging or weighted averaging the lag times of all grid cells across the entire region, generating regional air hygroscopic data. The resulting air hygroscopic data can be used for further applications such as regional water demand forecasting, crop irrigation regulation, and microclimate analysis, providing a scientific basis for agricultural and water resource management.

[0054] Preferably, time delay analysis is performed on the humidity residual sequence within each grid cell to calculate the lag time of the humidity response relative to the theoretical humidity change, including: The humidity residual sequence of the grid cell is segmented according to a predetermined time interval to generate a continuous humidity residual subsequence; Each humidity residual subsequence is smoothed, and the time correlation function or cross-correlation coefficient between the smoothed humidity residual subsequence and the corresponding theoretical humidity change sequence is calculated to obtain the correlation curve. Identify the time offset corresponding to the maximum correlation point on the correlation curve, where the time offset is the lag time of the grid cell's humidity response relative to the theoretical humidity change.

[0055] In one embodiment, the humidity residual sequence of each grid cell is first segmented according to a predetermined time interval to obtain a continuous humidity residual subsequence. Each subsequence contains humidity residual data within a fixed time length for subsequent time-series analysis and delay calculation. During the segmentation process, an appropriate time length can be selected according to the specific monitoring frequency and data acquisition interval, such as one subsequence per hour or per minute, to ensure that the sequence reflects local humidity changes while also maintaining calculation accuracy.

[0056] Subsequently, each humidity residual subsequence is smoothed to eliminate the interference of measurement noise and local abrupt changes on the analysis results. Smoothing methods such as moving average, weighted smoothing, or low-pass filtering can be used to make the overall trend of the subsequence more continuous and stable. The smoothed humidity residual subsequence is then compared with the corresponding theoretical humidity change sequence. By calculating the time correlation function or cross-correlation coefficient, the correlation curve between the humidity response and the theoretical humidity change is obtained. The correlation curve describes the degree of matching between the humidity residual sequence and the theoretical humidity change at different time offsets, providing a basis for determining the lag time.

[0057] On the generated correlation curve, the point of maximum correlation and its corresponding time offset are identified. This time offset reflects the degree of lag in the humidity response of the grid cell relative to the theoretical humidity change, i.e., the humidity response lag time of the grid cell. By calculating the humidity response lag time of each grid cell separately, spatially distributed lag characteristic data can be obtained, providing basic information for subsequent weighted adjustments based on micro-topographic thermal inertia and the generation of regional air humidity data.

[0058] Preferred methods for identifying the most relevant points include: By sampling the correlation curve, a discrete correlation coefficient sequence is obtained; Identify all local maxima in the correlation coefficient sequence and record their corresponding time offsets and correlation coefficient values; Local maxima with correlation coefficients below a set threshold are removed, and the point with the highest correlation coefficient among the remaining local maxima is selected as the maximum correlation point, and its corresponding time offset is recorded. The time offset is used as the lag time of the grid cell's humidity response relative to the theoretical humidity change.

[0059] In one embodiment, the correlation curve of each grid cell is sampled to generate a discrete correlation coefficient sequence. The sampling interval can be set according to the temporal resolution of the humidity data to ensure that the discrete sequence can fully reflect the changing trend of the correlation curve. Subsequently, in the obtained discrete correlation coefficient sequence, all local maxima are identified, that is, points in each local neighborhood where the correlation coefficient is greater than that of the surrounding points, and the time offset and correlation coefficient value corresponding to each maxima are recorded.

[0060] To eliminate insignificant maxima caused by interference and noise, the identified local maxima points need to be screened. Specifically, a correlation coefficient threshold is set, and maxima points with correlation coefficients below this threshold are discarded, retaining only candidate points with high correlation and strong reliability. The remaining candidate points are then compared in terms of their correlation coefficients, and the point with the highest correlation coefficient is selected as the maximum correlation point. The time offset of the maximum correlation point represents the optimal time delay of the humidity response relative to the theoretical humidity change. This time offset can be used for subsequent lag time calculations and weighted adjustments of the grid cells.

[0061] Preferably, the setting of the predetermined time interval includes: setting the time interval for collecting humidity observation data to be within the range of 10 seconds to 30 minutes, wherein the time interval can be expressed in units of seconds, minutes or hours, and a fixed interval or an adaptive interval can be selected according to the rate of change of air humidity in the target area.

[0062] In one embodiment, to effectively monitor humidity changes in the target area, the time interval for collecting humidity observation data needs to be reasonably set. Specifically, the time interval for collecting humidity observation data can be set within the range of 10 seconds to 30 minutes to balance data accuracy and collection efficiency. The time interval can be expressed in seconds, minutes, or hours, flexibly selected according to actual needs. For areas with rapid changes in air humidity, a shorter fixed collection interval can be prioritized to ensure that humidity fluctuations are fully captured; for areas with slow changes in air humidity, a longer fixed interval can be selected, or the collection interval can be adaptively adjusted according to the real-time humidity change rate to achieve dynamic optimization of data collection. Through the above settings, a continuous, effective data sequence that conforms to the regional humidity characteristics can be obtained, providing a reliable basis for subsequent humidity residual analysis and lag time calculation.

[0063] Preferably, step S5 includes the following steps: Step S51: Collect temperature dynamic change data of the target area during the forecast period based on the regional water demand forecast value. The temperature dynamic change data includes hourly temperature change amplitude, temperature rise or fall trend and diurnal temperature difference parameter. Step S52: Calculate the impact of temperature change on surface water loss based on the hourly temperature change range and diurnal temperature difference parameters; Step S53: Map the influencing factors and the predicted regional water demand to obtain the preliminary water demand correction results after temperature adjustment; Step S54: Determine whether the area is in a heating or cooling phase by observing the temperature change trend, and adjust the preliminary water demand correction result based on the heating or cooling phase to obtain the adjusted regional water demand.

[0064] In one embodiment, temperature dynamics data of the target area during the forecast period are obtained based on the predicted regional water demand. This temperature dynamics data includes hourly temperature variation amplitude, temperature rise or fall trend, and diurnal temperature range parameters, collected from meteorological observation stations or meteorological simulation data in the target area. Based on this, the hourly temperature variation amplitude and diurnal temperature range parameters are combined to calculate the impact of temperature changes on surface water loss. Specifically, by analyzing the impact of temperature fluctuations on soil evaporation rate, vegetation transpiration, and surface runoff, the increase or decrease effect of hourly temperature changes on regional water loss is quantified, thereby obtaining the magnitude of the temperature change impact.

[0065] Subsequently, the impact of temperature changes is mapped to the predicted regional water demand, yielding a preliminary correction to the water demand after temperature adjustment. During the mapping process, the predicted regional water demand is dynamically adjusted proportionally based on the enhancing or diminishing effect of temperature on water loss, forming a preliminary adjustment value. Further, by analyzing temperature change trends, it is determined whether the current forecast period is in a warming or cooling phase, and a secondary adjustment is made to the preliminary water demand correction based on this phase. Specifically, during a warming phase, the water demand correction value is appropriately increased based on the rate and magnitude of temperature increase; during a cooling phase, the water demand correction value is appropriately decreased based on the magnitude of temperature decrease, thus obtaining the final adjusted regional water demand.

[0066] Preferably, determining whether a temperature is in a heating or cooling phase based on temperature change trends includes: When the temperature change trend is in the rising stage, the preliminary water demand correction result is amplified according to the temperature rise rate per unit time. When the temperature change trend is in the cooling phase, the preliminary water demand correction result is reduced based on the rate of temperature decrease per unit time.

[0067] In one embodiment, hourly temperature change data of the target area is analyzed for trends. The temperature change trend is determined by calculating the temperature difference and rate of change at consecutive time points. If the rate of temperature change over a consecutive period is positive and exceeds a set threshold, it is determined to be a warming phase; if the rate of temperature change is negative and exceeds the set threshold, it is determined to be a cooling phase. During the warming phase, the preliminary water demand correction result is amplified, i.e., the water demand correction value is increased proportionally according to the temperature rise and rate of increase to compensate for the increase in soil evaporation and plant transpiration caused by the temperature rise. During the cooling phase, the preliminary water demand correction result is reduced, i.e., the water demand correction value is appropriately reduced according to the temperature drop and rate of decrease to reflect the reduced water evaporation effect caused by the temperature drop. Through dynamic adjustment of the warming and cooling phases, temperature-sensitive correction of regional water demand is achieved, thereby obtaining more accurate adjusted water demand data.

[0068] Specifically, when the temperature rise is within the first range (e.g., 0.1–0.5℃ / hour), the preliminary correction result will be increased by 3%–5%; when the temperature rise is within the second range (e.g., 0.5–1.0℃ / hour), the preliminary correction result will be increased by 5%–10%; and when the temperature rise exceeds the third range (e.g., above 1.0℃ / hour), the preliminary correction result will be increased by 10%–15%. When the temperature fall is within the first range (e.g., 0.1–0.5℃ / hour), the preliminary correction result will be decreased by 2%–4%; when the temperature fall is within the second range (e.g., 0.5–1.0℃ / hour), the preliminary correction result will be decreased by 4%–8%; and when the temperature fall exceeds the third range (e.g., above 1.0℃ / hour), the preliminary correction result will be decreased by 8%–12%.

[0069] Of particular importance is the construction of a regional water demand prediction model based on air humidity data, rainfall characteristic data, and historical water use data, including: By integrating air humidity retention data, rainfall characteristic data, and historical water use data, a sample dataset is obtained. By dividing the sample dataset into a model training set and a model test set; The spatial dependencies of the model training set are extracted using a graph convolutional network algorithm to obtain the first spatial feature dataset; The second spatial feature dataset is obtained by applying regional differential weights to the first spatial feature dataset using an attention mechanism graph network. Construct a multi-layered LSTM architecture, which includes a bottom-layer LSTM, a middle-layer LSTM, and a high-layer LSTM. A pre-model of regional water demand was obtained by training a model on a second spatial feature dataset using a multi-level LSTM architecture. Based on the model test set, the regional water demand prediction pre-model is optimized and iterated to generate a regional water demand prediction model.

[0070] In one embodiment, the air humidity lag data, rainfall characteristic data, and historical water use data are first organized into a unified data structure. Specifically, data from different sources are fused according to the same timestamp, spatial grid number, and data fields to form a sample dataset containing information on the degree of environmental humidity lag, segmented rainfall amplitude, rainfall interval, and corresponding historical water use records. To ensure the effectiveness of model training, the integrated sample dataset is divided into a model training set and a model test set according to a preset ratio. The training set is used for model parameter training, and the test set is used for model performance verification and subsequent iterative optimization.

[0071] After the dataset is prepared, spatial relationship modeling is performed on the training set data based on the graph convolutional network algorithm. Specifically, using the grid cells of the target region as nodes, an adjacency structure is constructed to reflect the humidity propagation, rainfall impact range, and historical water use similarity between neighboring grids. The graph convolutional network runs on this structure to extract the spatial dependencies between different grid cells, thereby generating a first spatial feature dataset that reflects the spatial interaction characteristics within the region.

[0072] After obtaining the first spatial feature dataset, an attention-based graphical network is further introduced to differentiate the importance of spatial features in different regions. Specifically, the attention network assigns different weights to different regions based on their sensitivity to historical water consumption changes, differences in air humidity response, and the impact of rainfall on water demand. This weighted processing forms the second spatial feature dataset, which highlights spatial units that have a more significant impact on water demand prediction, thereby improving the learning efficiency and prediction accuracy of subsequent models.

[0073] Subsequently, the system constructs a time series model composed of a multi-layered LSTM (Long Short-Term Memory) architecture. This multi-layered architecture includes a bottom-layer LSTM, a middle-layer LSTM, and a top-layer LSTM. The bottom-layer LSTM extracts water demand fluctuation features on short time scales, the middle-layer LSTM identifies seasonal or periodic changes, and the top-layer LSTM integrates data trends over long time spans. By inputting the second spatial feature dataset into this multi-layered LSTM architecture, the model extracts deep temporal features of water demand layer by layer and iteratively trains on the training set, gradually converging the parameters to ultimately form a pre-model for regional water demand prediction.

[0074] Finally, the performance of the regional water demand prediction pre-model was validated using a model test set, including prediction error analysis, spatial consistency verification, and temporal trend matching evaluation. When the model's performance on the test set did not meet expectations, the graph convolution module, attention module, and multi-layer LSTM module were jointly optimized based on the error distribution, and the model's prediction effect was continuously improved through multiple iterations. The model, after iterative training, ultimately forms a regional water demand prediction model, used to reliably predict the future water demand of the target area, providing a basis for water resource allocation.

[0075] Of particular importance, after constructing the regional water demand prediction model, it also includes: Construct a model interpretability framework, which is built using SHAP values; Counterfactual analysis of the regional water demand prediction model was conducted using a model interpretation framework to obtain sensitivity data of the prediction results; An interpretability report of the regional water demand prediction model is generated based on the sensitivity data of the prediction results.

[0076] In one embodiment, after constructing the regional water demand prediction model, the model is further extended for interpretability to improve the transparency and understandability of the prediction results. Specifically, an interpretable framework for the regional water demand prediction model is first constructed. In this framework, the SHAP value is used as the core of the interpretation mechanism. By performing perturbation analysis on the input features of the prediction model, the contribution of air humidity data, rainfall characteristic data, and historical water use data in the prediction process is quantified. Each input feature of a sample is evaluated item by item, the corresponding SHAP value is calculated, and this value is stored as a feature importance weight in the interpretable framework for subsequent analysis.

[0077] After constructing the model's interpretability framework, counterfactual analysis was further performed on the regional water demand prediction model based on SHAP values. Specifically, for key input variables of the prediction model, such as the degree of humidity lag, the magnitude of rainfall increases or decreases, the duration of rainfall intervals, or historical water consumption fluctuations, hypothetical fine-tuning scenarios were generated for each variable, including various combinations of increasing, decreasing, and remaining unchanged input features. Subsequently, these scenarios were input into the regional water demand prediction model for inference calculations, and difference analysis was performed on the prediction results under each scenario to identify which input features are highly sensitive in the prediction results. The magnitude of the impact of these core features and the corresponding prediction trends were recorded as sensitivity data of the prediction results, serving as a key output for model interpretability.

[0078] After obtaining the sensitivity data, an interpretability report for the regional water demand prediction model is automatically generated based on this data. The report includes information such as the ranking of the impact of each input feature, regional spatial difference analysis, the sensitivity of the prediction results to changes in humidity lag, the impact path of changes in rainfall characteristics on water demand results, and the driving role of historical water use patterns on the prediction model. The report also includes key scenario examples from counterfactual analysis, such as changes in water demand predictions when rainfall intervals shorten and the magnitude of changes in model predictions when air humidity increases. This information is presented comprehensively through charts, text descriptions, and a list of important features, enabling users to intuitively understand the basis of the model's predictions and the sensitive variables.

[0079] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0080] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for calculating regional water demand based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Obtain basic meteorological data and historical water use data for the target area, wherein the basic meteorological data includes at least temperature, humidity and rainfall; Step S2: By analyzing the humidity change trajectory of the target area, and based on the difference between the humidity change trajectory and the preset theoretical change trajectory, the air humidity lag of the target area is analyzed to obtain the air humidity lag data of the target area; wherein, step S2 includes the following steps: Step S21: Extract humidity observation data of the target area within a predetermined time interval from the basic meteorological data; Step S22: Perform time series processing and noise filtering on the humidity observation data to generate a continuous humidity change trajectory; Step S23: Perform hourly difference analysis between the humidity change trajectory of the target area and the pre-set theoretical humidity change trajectory to obtain the humidity residual sequence; Step S24: Analyze the air humidity hysteresis of the target area using the humidity residual sequence and the micro-topographic thermal inertia analysis to obtain the air humidity hysteresis data of the target area; wherein, step S24 includes: Based on the target area, the corresponding humidity residual sequence is spatially matched with the micro-topographic thermal inertia value of each grid cell to obtain the spatially matched grid cell; Time delay analysis is performed on the humidity residual sequence within each grid cell to calculate the lag time of the humidity response relative to the theoretical humidity change; Based on the thermal inertia of micro-topography, the humidity lag time of each spatially matched grid cell is weighted and adjusted, and the weighted grid cell lag times are aggregated regionally to obtain the air humidity data of the target area. Specifically, time delay analysis is performed on the humidity residual sequence within each grid cell to calculate the lag time of the humidity response relative to the theoretical humidity change, including: The humidity residual sequence of the grid cell is segmented according to a predetermined time interval to generate a continuous humidity residual subsequence; Each humidity residual subsequence is smoothed, and the time correlation function or cross-correlation coefficient between the smoothed humidity residual subsequence and the corresponding theoretical humidity change sequence is calculated to obtain the correlation curve. The time offset corresponding to the maximum correlation point is identified on the correlation curve, where the time offset is the lag time of the grid cell's humidity response relative to the theoretical humidity change; the method for identifying the maximum correlation point includes: By sampling the correlation curve, a discrete correlation coefficient sequence is obtained; Identify all local maxima in the correlation coefficient sequence and record their corresponding time offsets and correlation coefficient values; Local maxima with correlation coefficients below a set threshold are removed, and the point with the highest correlation coefficient among the remaining local maxima is selected as the maximum correlation point, and its corresponding time offset is recorded. The time offset is used as the lag time of the grid cell's humidity response relative to the theoretical humidity change; Step S3: Divide the rainfall into segments according to the time series, and calculate the increase or decrease in rainfall in each segment and the duration of rainfall interval between adjacent segments to obtain the rainfall characteristic data of the target area; Step S4: Construct a regional water demand prediction model based on air humidity data, rainfall characteristic data, and historical water use data; use the regional water demand prediction model to predict the water demand of the target area and obtain the predicted regional water demand value. Step S5: Adjust the predicted regional water demand by dynamically adjusting the temperature to obtain the adjusted regional water demand.

2. The regional water demand calculation method based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the micro-topography thermal inertia of the target area in step S24 includes: Acquire digital elevation data within the target area, wherein the digital elevation data contains several grid cells; Micro-topographic features of the target area are extracted based on digital elevation data, and land cover type of the target area is analyzed based on the topographic features to obtain land cover type data; Based on land cover type data, the surface thermophysical parameters of different cover types are determined, and the surface thermal inertia value of each grid cell is calculated based on digital elevation data and surface thermophysical parameters to obtain the micro-topographic thermal inertia of the target area.

3. The regional water demand calculation method based on artificial intelligence according to claim 2, characterized in that, The calculation of the surface thermal inertia value for each grid cell based on digital elevation data and surface thermophysical parameters includes: The center point location and boundary area of ​​each grid cell are determined based on digital elevation data, and the corresponding digital elevation data is extracted to obtain the local terrain height. Calculate the slope and aspect of each grid cell based on the local terrain height of each grid cell; The theoretical surface thermal inertia value of each grid cell is calculated using surface thermophysical parameters. The grid cell thermal energy is then adjusted based on the slope and aspect of the grid cell to obtain the surface thermal inertia value.

4. The regional water demand calculation method based on artificial intelligence according to claim 1, characterized in that, The preset time interval setting includes setting the time interval for collecting humidity observation data to be within the range of 10 seconds to 30 minutes. The time interval can be expressed in units of seconds, minutes or hours, and a fixed interval or an adaptive interval can be selected according to the rate of change of air humidity in the target area.

5. The regional water demand calculation method based on artificial intelligence according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Collect temperature dynamic change data of the target area during the forecast period based on the regional water demand forecast value. The temperature dynamic change data includes hourly temperature change amplitude, temperature rise or fall trend and diurnal temperature difference parameter. Step S52: Calculate the impact of temperature change on surface water loss based on the hourly temperature change range and diurnal temperature difference parameters; Step S53: Map the influencing factors and the predicted regional water demand to obtain the preliminary water demand correction results after temperature adjustment; Step S54: Determine whether the area is in a heating or cooling phase by observing the temperature change trend, and adjust the preliminary water demand correction result based on the heating or cooling phase to obtain the adjusted regional water demand.

6. The regional water demand calculation method based on artificial intelligence according to claim 5, characterized in that, Determining whether a temperature is in a warming or cooling phase based on temperature change trends includes: When the temperature change trend is in the rising stage, the preliminary water demand correction result is amplified according to the temperature rise rate per unit time. When the temperature change trend is in the cooling phase, the preliminary water demand correction result is reduced based on the rate of temperature decrease per unit time.

Images