A regional multi-department water demand prediction method and system based on cross-department historical information

Abstract

The application discloses a regional multi-department water demand prediction method and system based on cross-department historical information, which comprises the following steps: determining a basic prediction unit according to a target region, a target department and a prediction period, and acquiring historical water demand and external driving data of the target region which are consistent with the prediction scale; for each basic prediction unit, constructing a target department's own historical water demand feature, an external driving feature and a cross-department historical information feature; respectively establishing a cross-department water demand prediction model and an independent water demand prediction model, comparing the verification errors of the two models for the same target department, and performing verification gate selection to determine the target water demand prediction model; and generating the final water demand prediction value of the corresponding basic prediction unit by using the target water demand prediction model. The application can use the historical water demand information of different target departments in a region for regional multi-department water demand prediction, and select the corresponding model according to the verification error, thereby improving the accuracy and stability of the regional multi-department water demand prediction.

Description

Technical Field

[0001] This invention relates to the field of water resources management technology, specifically to a method and system for predicting regional multi-sectoral water demand based on cross-sectoral historical information. Background Technology

[0002] Reliable water demand forecasts are essential for regional water resource allocation, supply-demand balance analysis, drought contingency planning, and water network scheduling. With the combined effects of climate change, urbanization, industrial restructuring, food security requirements, and ecological restoration goals, regional water demand is no longer simply a matter of total volume change. Instead, it exhibits a structural characteristic of co-evolution across multiple sectors, including domestic, agricultural, industrial, energy-related, and ecological water use. Existing water demand forecasting methods often focus on total water demand or individual water-using sectors, employing techniques such as trend extrapolation, seasonal decomposition, statistical regression, machine learning, or deep learning. While these methods can characterize the changing patterns of the target water demand sequence to some extent, they still have limitations in serving regional water resource allocation and water network scheduling. Different sectors are typically modeled independently before being aggregated, neglecting the interconnected relationships formed historically by shared economic activities, agricultural production, industrial structure, land use, ecological governance, and water use policies. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method for predicting regional multi-sector water demand based on cross-sectoral historical information. By introducing the lagged water demand characteristics of other sectors, the lagged total water demand characteristics of the region, and the lagged water demand ratio characteristics of each sector, the historical linkage relationship between water demand of multiple sectors is characterized, thereby improving the accuracy of regional multi-sector water demand prediction.

[0004] Technical solution: The present invention provides a method for predicting regional multi-sectoral water demand based on cross-sectoral historical information, comprising the following steps:

[0005] Based on the water resources management objectives of the target area, determine the target sector, forecast scale, and forecast period; obtain historical water demand data for the target area consistent with the forecast scale, as well as external driving data; and determine basic forecast units based on the target area, target sector, and forecast period according to the forecast scale.

[0006] For the basic forecasting unit, based on the historical water demand of the target area before the forecast period and external driving data, the historical water demand characteristics of the target sector itself, the external driving characteristics, and the cross-sectoral historical information characteristics are constructed.

[0007] For each prediction scale, the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics are combined with the historical water demand corresponding to the basic prediction unit to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; the training sample set is used to establish a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively;

[0008] Using the aforementioned validation sample set, the validation errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector within the same target area are calculated respectively. Based on the validation errors, validation gating is selected to determine the target water demand prediction model corresponding to the target sector within each target area. The target water demand prediction model is then used to generate the final water demand prediction value for the corresponding basic prediction unit.

[0009] Furthermore, the target sector's own historical water demand characteristics include at least one of the following: time trend characteristics, target sector's lagged water demand, lagged rolling mean, lagged rolling maximum value, lagged rolling minimum value, and lagged rolling standard deviation; the external driving characteristics include at least one of the following: population size characteristics, economic development characteristics, industrial structure characteristics, agricultural production characteristics, and ecological governance characteristics.

[0010] Furthermore, the time trend feature is the predicted period minus the starting period corresponding to the historical water demand plus one; the target sector's lagged water demand is the water demand of the target sector in one or more historical periods before the predicted period, consistent with the current prediction scale; the lagged rolling mean is the average of the historical water demand of multiple consecutive target areas before the predicted period; the lagged rolling maximum value is the maximum historical water demand of the target sector within the rolling time window; the lagged rolling minimum value is the minimum historical water demand of the target sector within the rolling time window; and the lagged rolling standard deviation is the degree of dispersion of the target sector's historical water demand relative to the lagged rolling mean within the rolling time window.

[0011] The population size category features include population size; the economic development category features include regional GDP and industrial added value; the industrial structure category features include industrial structure; the agricultural production category features include agricultural output value, grain sown area, cultivated land area and agricultural production level; and the ecological governance category features include ecological governance intensity.

[0012] Furthermore, the cross-departmental historical information features include at least one of the following: lagging water demand features of other departments, lagging total water demand features of the region, and lagging water demand ratio features of each water-demanding department; the lagging water demand features of other departments consist of the water demand of other water-demanding departments in the same region, excluding the target department, in one or more historical periods prior to the prediction period; the lagging total water demand features of the region are the sum of the water demand of all water-demanding departments in the same region in the corresponding historical period; the lagging water demand ratio features of each water-demanding department are the ratio of the water demand of each water-demanding department in the corresponding historical period to the total water demand of the region in that historical period.

[0013] Furthermore, the cross-sectoral water demand prediction model and the independent water demand prediction model are established using the same type of regression prediction model. The regression prediction model includes one of the following: linear regression model, support vector regression model, random forest regression model, gradient boosting regression model, extreme gradient boosting model, neural network regression model, and time series regression model. The cross-sectoral water demand prediction model takes the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics as inputs. The independent water demand prediction model takes the target sector's own historical water demand characteristics and external driving characteristics as inputs. Both the cross-sectoral water demand prediction model and the independent water demand prediction model use the historical water demand corresponding to the basic prediction unit as the training target value and output the water demand prediction value of the basic prediction unit.

[0014] Furthermore, the verification gating selection based on the verification error includes: when the verification error of the cross-departmental water demand prediction model is less than the verification error of the independent water demand prediction model, selecting the cross-departmental water demand prediction model as the target water demand prediction model corresponding to the target department; when the verification error of the cross-departmental water demand prediction model is greater than or equal to the verification error of the independent water demand prediction model, selecting the independent water demand prediction model as the target water demand prediction model corresponding to the target department.

[0015] Based on the same inventive concept, the present invention also provides a regional multi-sectoral water demand prediction system based on cross-sectoral historical information, comprising:

[0016] The data acquisition unit is used to determine the target sector, forecast scale, and forecast period based on the water resources management objectives of the target area; acquire historical water demand of the target area consistent with the forecast scale, as well as external driving data; and determine the basic forecast unit according to the forecast scale, based on the target area, target sector, and forecast period.

[0017] The feature construction unit is used to construct the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics based on the historical water demand of the target area before the prediction period and external driving data for the basic prediction unit.

[0018] The model building unit is used to combine the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics with the historical water demand corresponding to the basic prediction unit for each prediction scale to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; and the training sample set is used to build a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively.

[0019] The model selection and prediction unit is used to calculate the verification errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector in the same target area using the verification sample set, and to perform verification gating selection based on the verification errors to determine the target water demand prediction model corresponding to the target sector in each target area; and to generate the final water demand prediction value of the corresponding basic prediction unit using the target water demand prediction model.

[0020] Based on the same inventive concept, the present invention also provides an electronic device, comprising:

[0021] Memory, used to store computer programs;

[0022] A processor is used to execute the computer program to implement the regional multi-sector water demand prediction method based on cross-sectoral historical information.

[0023] Based on the same inventive concept, the present invention also provides a non-volatile storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the regional multi-sector water demand prediction method based on cross-sectoral historical information.

[0024] Based on the same inventive concept, the present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the aforementioned regional multi-sector water demand prediction method based on cross-sectoral historical information.

[0025] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are as follows: (1) The present invention transforms the historical linkage relationship between water demand of multiple sectors into computable features such as the lagging water demand characteristics of other sectors, the lagging total water demand characteristics of the region, and the water demand ratio characteristics of lagging sectors, thereby improving the problem of insufficient utilization of cross-sectoral historical information when each water demand sector makes independent predictions; (2) The present invention uses the target region, target sector, and prediction period to jointly determine the basic prediction unit, providing structured prediction results for regional multi-sectoral water demand analysis and subsequent water resource management; (3) The present invention adopts a cross-sectoral water demand prediction model based on a regression prediction model, which can utilize the nonlinear and interactive relationships in the structured features to improve the accuracy and applicability of regional multi-sectoral water demand prediction; (4) The present invention uses a verification gating mechanism to compare the verification errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the same target sector combination in the same region using the same error evaluation index, and selects the corresponding prediction model accordingly; when cross-sectoral historical information cannot reduce the verification error, the independent water demand prediction model is selected, thereby avoiding interference of invalid cross-sectoral historical information on the prediction results and improving the accuracy and stability of regional multi-sectoral water demand prediction. Attached Figure Description

[0026] Figure 1 This is a flowchart of the regional multi-sector water demand prediction method based on cross-sectoral historical information of the present invention.

[0027] Figure 2 This is a schematic diagram illustrating the input features of the cross-sectoral water demand prediction model of the present invention;

[0028] Figure 3 This is a comparison chart of the water demand prediction effects of different prediction models in the embodiments of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited to the embodiments described.

[0030] This invention provides a regional multi-sector water demand prediction method that can transform the historical linkages between water demand of multiple sectors in a region into computable predictable features and output structured water demand prediction results, thereby improving the accuracy of water demand prediction, configuration applicability, and uncertainty expression capabilities.

[0031] Example 1:

[0032] The present invention discloses a method for predicting regional multi-sectoral water demand based on cross-sectoral historical information, comprising the following steps:

[0033] Based on the water resources management objectives of the target area, determine the target sector, forecast scale, and forecast period; obtain historical water demand data for the target area consistent with the forecast scale, as well as external driving data; and determine basic forecast units based on the target area, target sector, and forecast period according to the forecast scale.

[0034] For the basic forecasting unit, based on the historical water demand of the target area before the forecast period and external driving data, the historical water demand characteristics of the target sector itself, the external driving characteristics, and the cross-sectoral historical information characteristics are constructed.

[0035] For example, the target sector's own historical water demand characteristics include at least one of the following: time trend characteristics, target sector's lagged water demand, lagged rolling mean, lagged rolling maximum value, lagged rolling minimum value, and lagged rolling standard deviation; the external driving characteristics include at least one of the following: population size characteristics, economic development characteristics, industrial structure characteristics, agricultural production characteristics, and ecological governance characteristics.

[0036] For example, the time trend feature is the predicted period minus the starting period corresponding to the historical water demand plus one; the target sector's lagged water demand is the water demand of the target sector in one or more historical periods before the predicted period, consistent with the current prediction scale; the lagged rolling mean is the average of the historical water demand of multiple consecutive target areas before the predicted period; the lagged rolling maximum value is the maximum historical water demand of the target sector within the rolling time window; the lagged rolling minimum value is the minimum historical water demand of the target sector within the rolling time window; and the lagged rolling standard deviation is the degree of dispersion of the historical water demand of the target sector within the rolling time window relative to the lagged rolling mean.

[0037] The population size category features include population size; the economic development category features include regional GDP and industrial added value; the industrial structure category features include industrial structure; the agricultural production category features include agricultural output value, grain sown area, cultivated land area and agricultural production level; and the ecological governance category features include ecological governance intensity.

[0038] For example, the cross-departmental historical information features include at least one of the following: lagging water demand features of other departments, lagging total water demand features of the region, and lagging water demand ratio features of each water-demanding department; the lagging water demand features of other departments consist of the water demand of other water-demanding departments in the same region other than the target department in one or more historical periods before the prediction period; the lagging total water demand feature of the region is the sum of the water demand of all water-demanding departments in the same region in the corresponding historical period; the lagging water demand ratio feature of each water-demanding department is the ratio of the water demand of each water-demanding department in the corresponding historical period to the total water demand of the region in that historical period.

[0039] For each prediction scale, the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics are combined with the historical water demand corresponding to the basic prediction unit to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; the training sample set is used to establish a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively;

[0040] For example, the cross-sectoral water demand prediction model and the independent water demand prediction model are established using the same type of regression prediction model. The regression prediction model includes one of the following: linear regression model, support vector regression model, random forest regression model, gradient boosting regression model, extreme gradient boosting model, neural network regression model, and time series regression model. The cross-sectoral water demand prediction model takes the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics as inputs. The independent water demand prediction model takes the target sector's own historical water demand characteristics and external driving characteristics as inputs. Both the cross-sectoral water demand prediction model and the independent water demand prediction model take the historical water demand corresponding to the basic prediction unit as the training target value and output the water demand prediction value of the basic prediction unit.

[0041] Using the aforementioned validation sample set, the validation errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector within the same target area are calculated respectively. Based on the validation errors, validation gating is selected to determine the target water demand prediction model corresponding to the target sector within each target area. The target water demand prediction model is then used to generate the final water demand prediction value for the corresponding basic prediction unit.

[0042] For example, the verification gating selection based on the verification error includes: when the verification error of the cross-departmental water demand prediction model is less than the verification error of the independent water demand prediction model, selecting the cross-departmental water demand prediction model as the target water demand prediction model corresponding to the target department; when the verification error of the cross-departmental water demand prediction model is greater than or equal to the verification error of the independent water demand prediction model, selecting the independent water demand prediction model as the target water demand prediction model corresponding to the target department.

[0043] like Figure 1 As shown, the method for predicting regional multi-sectoral water demand based on cross-sectoral historical information according to the present invention includes the following steps:

[0044] S1. Obtain the water resource management objectives for the target area, and determine the target sector, forecast scale, and forecast period based on the water resource management objectives for the target area; obtain the historical water demand corresponding to the target area with the same forecast scale, as well as external driving data; and combine the target area, target sector, and forecast period to form a basic forecast unit according to the forecast scale.

[0045] Specifically, the target area, target sector, forecast scale, and forecast period are first determined. The target area can be one or more of the following: a city area, a river basin, an irrigation district, etc. The target sector is determined based on the water resource management objectives of the target area. These objectives can be one or more of the following: domestic, industrial, agricultural, and ecological. Based on domestic water resource management objectives, domestic water demand sectors can be identified; based on industrial water resource management objectives, industrial water demand sectors can be identified, which can be further subdivided into energy-intensive and strategic industrial water demand sectors and general industrial water demand sectors based on industrial production needs; based on agricultural water resource management objectives, agricultural water demand sectors can be identified, which can be further subdivided into grain irrigation water demand sectors and other agricultural water demand sectors based on agricultural production needs; and based on ecological water resource management objectives, ecological environment water demand sectors can be identified. The prediction scale includes one or more of the following: annual, monthly, daily, and hourly. For each prediction scale, historical water demand data and external driving data of the target area consistent with the prediction scale are obtained, and the time unit corresponding to the prediction scale is used as the prediction period. Each prediction period corresponding to the annual, monthly, daily, and hourly scales is a year, a month, a date, and an hour interval, respectively.

[0046] Specifically, the domestic water demand sector is used to characterize the water demand related to residential life and public services; the energy-intensive and strategic industrial water demand sector is used to characterize the water demand of industrial sectors with high energy consumption intensity and industrial production within the scope of strategic industrial management in the target area; the general industrial water demand sector is used to characterize the water demand of other industrial production besides energy-intensive and strategic industries; the grain irrigation water demand sector is used to characterize the agricultural irrigation water demand aimed at grain crop production and food security; the other agricultural water demand sector is used to characterize the water demand for irrigation of cash crops, forestry, animal husbandry, fishery and other agricultural production besides grain irrigation; and the ecological environment water demand sector is used to characterize the water demand related to the maintenance, ecological restoration and ecological replenishment of ecosystems such as rivers, lakes, and wetlands.

[0047] Specifically, the external driving data includes at least one of the following categories: population size data, which represents the population size of the target region; economic development data, which represents the level of economic development, including at least one of the target region's GDP and industrial added value; industrial structure data, which represents the characteristics of the industrial structure, including at least one of the output value ratios of the primary, secondary, and tertiary industries; agricultural production data, which represents the status of agricultural production, including at least one of agricultural output value, grain sown area, cultivated land area, and agricultural production level; and ecological governance data, which represents the status of ecological governance, including the intensity of ecological governance.

[0048] Specifically, for each forecast scale, the forecast period is determined using the corresponding time unit, and the target region, target department, and forecast period are combined to form a basic forecast unit. The forecast scale is used to determine the time unit of the forecast period; for example, at a monthly scale, each forecast period corresponds to one month. At the same forecast scale, each basic forecast unit corresponds to one target region, one target department, and one forecast period. For the target region... Target Department and forecast period The corresponding basic prediction unit is represented as For historical periods, the historical water demand corresponding to this basic forecasting unit is denoted as... For the period to be predicted, the predicted water demand output by the model is denoted as... Basic prediction units corresponding to different prediction scales are constructed separately and used for water demand prediction at their respective prediction scales.

[0049] S2. For each basic prediction unit under each prediction scale, obtain the target sector corresponding to the basic prediction unit, and construct prediction features based on the historical water demand data of the target area that is available before the corresponding prediction period and is consistent with the current prediction scale, as well as external driving data. The prediction features include the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics.

[0050] For example, for any basic prediction unit Identify target departments And based on the target area During the forecast period Previously available historical water demand data for the target area, consistent with the current forecast scale, were used to construct the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics.

[0051] Specifically, the historical water demand characteristics of the target sector are constructed based on the historical water demand data of the target sector prior to the forecast period. The historical water demand characteristics of the target sector include at least one of the following: time trend characteristics, target sector lagged water demand, lagged rolling mean, lagged rolling maximum, lagged rolling minimum, and lagged rolling standard deviation.

[0052] The time trend feature is used to characterize the long-term trend of historical water demand for the target sector over time. Starting from the initial period of the historical water demand, it is continuously counted using the time unit corresponding to the current forecast scale. The time trend feature value of the initial period is set to 1, and then increases by 1 for each subsequent period. Let the initial period corresponding to the historical water demand be... Forecast period The corresponding time trend characteristics are:

[0053] (1)

[0054] in, Indicates the forecast period Corresponding time trend characteristics;

[0055] The lagged water demand of the target sector refers to the water demand of the target sector in one or more historical periods prior to the forecast period and consistent with the current forecast scale. It is used to characterize the short-term inertia and temporal continuity of the target sector's water demand. Let the number of lagged periods be... ,and If it is a positive integer, then the target area is... Internal target department During the forecast period The corresponding number The water demand with hysteresis is:

[0056] (2)

[0057] in, Indicates the target area Internal target department During the period to be predicted The corresponding number Lag demand; Indicates the target area Internal target department In historical periods The water demand.

[0058] The lagged rolling mean is the target department's continuous mean before the forecast period. The average historical water demand of each target area is used to characterize the recent overall water demand level of the target sector. Among them, The length of the scrolling time window, and If the integer is positive, then the lagging rolling mean is:

[0059] (5)

[0060] in, Indicates the target area Internal target department In historical periods The lagged rolling mean, Indicates the target area Internal target department In historical periods The water demand.

[0061] The lagging rolling maximum value is the maximum historical water demand of the target sector within the rolling time window, used to characterize the recent high water demand state of the target sector, and its calculation formula is:

[0062] (6)

[0063] in, Represents the basic prediction unit The corresponding maximum value of the lag scroll.

[0064] The lagging rolling minimum value is the minimum historical water demand of the target sector within the rolling time window, used to characterize the recent low water demand state of the target sector, and its calculation formula is:

[0065] (7)

[0066] in, Represents the basic prediction unit The corresponding minimum value of delayed rolling.

[0067] The lag rolling standard deviation represents the dispersion of the target sector's historical water demand relative to the lag rolling mean within the rolling time window. It is used to characterize the recent fluctuations in the target sector's water demand, and its calculation formula is as follows:

[0068] (8)

[0069] in, Indicates the target area Internal target department During the period to be predicted The corresponding lag rolling standard deviation.

[0070] For one or more lag orders Calculate the first number corresponding to the target department respectively. Lag demand; for one or more rolling time window lengths The lagged rolling mean, lagged rolling maximum, lagged rolling minimum, and lagged rolling standard deviation are calculated for each corresponding window. The time trend characteristics, the lagged water demand of each target sector, and the lagged rolling statistical characteristics corresponding to each rolling time window are combined to form the historical water demand characteristics of the target sector itself.

[0071] Specifically, the external driving feature is based on the target region. During the forecast period The water demand forecast is constructed using external driving data acquired prior to the start of the forecast, consistent with the current forecast scale. These external driving features include at least one of the following: population size features, economic development features, industrial structure features, agricultural production features, and ecological governance features. Specifically, the population size features, constructed based on the population size data, characterize the regional population size and its changing state; the economic development features, constructed based on the economic development data, characterize the regional economic development level and industrial economic development status; the industrial structure features, constructed based on the industrial structure data, characterize the composition of different industries in the regional economy; the agricultural production features, constructed based on the agricultural production data, characterize the regional agricultural production scale, agricultural production conditions, and food production status; and the ecological governance features, constructed based on the ecological governance data, characterize the regional ecological governance and ecological water use management status.

[0072] For example, the population size category features include population size, the economic development category features include regional GDP and industrial added value, the industrial structure category features include industrial structure, the agricultural production category features include agricultural output value, grain sown area, cultivated land area and agricultural production level, and the ecological governance category features include ecological governance intensity.

[0073] Set target area During the forecast period The external driving data acquired before the water demand forecasting began totaled [amount missing]. Item, target area During the forecast period The corresponding number External driving characteristic value Then the target region r during the prediction period The corresponding external driving feature vector is:

[0074] (9)

[0075] in, Indicates the target area During the forecast period The corresponding external driving feature vector; Indicates the target area During the forecast period The corresponding number Externally driven eigenvalues Each external driving feature is arranged in a pre-defined fixed order to form an external driving feature vector. This fixed order refers to the fixed position of each external driving feature within the feature vector, not a ranking based on the numerical value of the features. For the same prediction scale, the external driving feature data items and their positions within the feature vector remain consistent throughout model training, validation, and water demand prediction. For different target sectors within the same region and time period, their basic prediction units use the same external driving feature vector corresponding to that region in the corresponding time period. For basic prediction units using the same water demand prediction model (cross-sectoral water demand prediction model or independent water demand prediction model) at the same prediction scale, the data items, arrangement order, and dimensions of the external driving features remain consistent; the same arrangement order of external driving features is used during model training and water demand prediction. For different basic prediction units belonging to the same region, using the same prediction scale, and corresponding to the same prediction time period, the same external driving feature vector corresponding to that region in the corresponding prediction time period is used. For example, for a target region... Any target department within Basic prediction unit The corresponding external driving feature vectors are all .

[0076] Specifically, the cross-sectoral historical information features are constructed based on the historical water demand of each water-demanding sector within the same region before the forecast period, consistent with the current forecast scale. The cross-sectoral historical information features include at least one of the following: lagging water demand features of other sectors, regional lagging total water demand features, and lagging water demand percentage features of each water-demanding sector. Specifically, the lagging water demand features of other sectors consist of the water demand of other water-demanding sectors within the same region (excluding the target sector) for one or more historical periods before the forecast period, used to characterize the historical water demand status and its temporal changes of other water-demanding sectors. The regional lagging total water demand feature is the sum of the water demand of all water-demanding sectors within the same region in the corresponding historical period, used to characterize the overall water demand level of the region in the corresponding historical period. The lagging water demand percentage feature of each water-demanding sector is the ratio of the water demand of each water-demanding sector in the corresponding historical period to the total regional water demand in that historical period, used to characterize the historical composition of each water-demanding sector in the total regional water demand, such as... Figure 2 As shown.

[0077] Set target area The water demand sector within the region is grouped as follows The target department is Any other water-requiring sector The predicted period is The number of lag periods is .in, .

[0078] Target Department The corresponding number Other sectors' lagging water demand characteristics are due to the target area Internal satisfaction and Various water-consuming sectors during historical periods The composition of water demand is expressed as follows:

[0079] (10)

[0080] in, Indicates the target department The corresponding number Other sectors are lagging behind in water demand characteristics; Indicates the target area Any other water-demanding sector In historical periods The water demand.

[0081] To ensure that the lagging water demand characteristics of other sectors in different basic forecasting units have a definite arrangement, the target sector is excluded from the forecasting unit according to a fixed order of water demand sectors. Other water-consuming sectors besides those in historical periods The water demand is arranged in order.

[0082] Specifically, the target area In historical periods The regional lagging total water demand characteristics are as follows:

[0083] (11)

[0084] in, Indicates the target area All water-consuming sectors within the historical period The sum of water demand, i.e., the regional lagged total water demand characteristics; Indicates the target area The water-demanding departments within the region are grouped together; Indicates the target area Any other water-demanding sector In historical periods The water demand.

[0085] Specifically, target area Any other water-demanding sector In historical periods The proportion of lagging water demand in the sector is:

[0086] (12)

[0087] in, Indicates the target area Any other water-demanding sector In historical periods The proportion of water demand in certain sectors is lagging behind.

[0088] Target Department The corresponding number The characteristics of the lagging water demand ratio of different sectors in the target area Various water-consuming sectors within the country during historical periods The lagging water demand percentages of different sectors are arranged in a predetermined order of water demand, and are represented as follows:

[0089] (13)

[0090] in, Indicates the target area The various water-requiring departments are arranged in a fixed order. Indicates the target area The number of domestic water-related sectors; Representing the target area Domestic water demand sector In historical periods The proportion of lagging water demand in certain sectors; Indicates the target department The corresponding number Characteristics of the lagging water demand ratio in different sectors.

[0091] Using one or more lag periods, the corresponding lagged water demand characteristics of other sectors, the regional lagged total water demand characteristics, and the sectoral lagged water demand proportion characteristics are calculated, and then combined according to a fixed characteristic order to form a basic prediction unit. Corresponding cross-departmental historical information characteristics:

[0092] (14)

[0093] in, Represents the basic prediction unit Corresponding cross-departmental historical information characteristics; This represents the set of lag periods used; This indicates that the splicing is performed according to a predetermined feature order, and is an existing splicing function.

[0094] S3. For each forecast scale, for any basic forecasting unit with historical water demand... The input features corresponding to the basic prediction unit, namely the target sector's own historical water demand features, external driving features, and cross-sectoral historical information features, are combined with historical water demand to form a historical sample. The historical sample is divided into a training sample set and a validation sample set; the training sample set is used to establish a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively.

[0095] Specifically, for any basic prediction unit with historical water demand, the prediction feature corresponding to the basic prediction unit is used as the model input, and its corresponding known historical water demand is used as the training target value to form a historical sample; the prediction feature includes the target department's own historical water demand feature, external driving feature, and cross-departmental historical information feature.

[0096] Specifically, the input features are used to provide information related to historical water demand to the water demand prediction model (cross-sector water demand prediction model or independent water demand prediction model), and the known historical water demand is used to compare with the water demand prediction value output by the model during the model training process, and adjust the model parameters accordingly.

[0097] The input features of the cross-sectoral water demand forecasting model consist of basic forecasting units. The corresponding target department's own historical water demand characteristics, external driving characteristics, and cross-departmental historical information characteristics are concatenated to obtain:

[0098] (15)

[0099] in, Represents the basic prediction unit Corresponding input features for cross-sectoral water demand forecasting models; This indicates the target department's own historical water demand characteristics; Represents the externally driven feature vector; Indicates cross-departmental historical information characteristics; This indicates that the parts are assembled according to a predetermined sequence of features.

[0100] The input features of the independent water demand prediction model are derived from the same basic prediction unit. The corresponding target department's historical water demand characteristics and external driving characteristics are combined to obtain:

[0101] (16)

[0102] in, Represents the basic prediction unit The corresponding input features of the independent water demand forecasting model. The independent water demand forecasting model does not use cross-sectoral historical information features, and is used to obtain water demand forecasting results using only the target sector's own historical water demand information and external driving information.

[0103] Input features into the cross-sectoral water demand forecasting model Input features of independent water demand prediction models The basic prediction unit is used as input to the corresponding model. The corresponding known water demand is used as the training target value for the corresponding model. The input features of each model and the corresponding training target value together constitute the complete historical sample of the corresponding model.

[0104] Specifically, for each prediction scale, the historical samples are divided into a training sample set and a validation sample set according to the chronological order of the corresponding time periods and a preset ratio. The preset ratio is determined based on the total number of historical samples and the sample size required for model training and validation gating. The sum of the proportions of the training sample set and the validation sample set is 100%, and the proportion of the training sample set is greater than that of the validation sample set. During the sample partitioning process, the input features and training target values ​​of the same historical sample are assigned to the same sample set. For the same basic prediction unit, the cross-sectoral water demand prediction model and the independent water demand prediction model use the same sample set partitioning result to ensure that the two models are trained and compared based on the same basic prediction unit. The training sample set is used to determine the model structure and model parameters of the cross-sectoral water demand prediction model and the independent water demand prediction model; the validation sample set is used in S4 to calculate the validation error of the two models for the target sector combination in the same region and to perform validation gating selection; after completing the validation gating selection, the selected prediction model is used to generate the water demand prediction values ​​for each basic prediction unit.

[0105] Specifically, the cross-sectoral water demand forecasting model and the independent water demand forecasting model are established using the same type of regression forecasting model. The regression forecasting model is used to learn the mapping relationship between the input features corresponding to the basic forecasting unit and historical water demand, and outputs the water demand forecast value corresponding to the basic forecasting unit based on the input features. The regression forecasting model includes one of the following: linear regression model, support vector regression model, random forest regression model, gradient boosting regression model, extreme gradient boosting model, neural network regression model, and time series regression model. For the same forecasting scale, the cross-sectoral water demand forecasting model and the independent water demand forecasting model use the same type of regression forecasting model, so that the main difference between the two models lies in whether or not cross-sectoral historical information features are used. In one embodiment, both the cross-sectoral water demand forecasting model and the independent water demand forecasting model are established using the extreme gradient boosting algorithm XGBoost. The regression forecasting model established based on the extreme gradient boosting algorithm XGBoost includes multiple sequentially established regression trees. Each regression tree divides the basic prediction unit into corresponding leaf nodes based on one or more feature values ​​in the input features and the corresponding segmentation conditions, and the leaf node outputs a value; the values ​​output by each regression tree are accumulated to obtain the water demand prediction value of the corresponding basic prediction unit.

[0106] Specifically, the cross-sectoral water demand prediction model and the independent water demand prediction model are trained using training sample sets. During model training, the input features from the training samples are input into the corresponding regression prediction model, which generates water demand prediction values. The model parameters are adjusted based on the difference between the water demand prediction values ​​and the known water demand corresponding to the training samples, so that the water demand prediction values ​​generated by the model gradually approach the known water demand.

[0107] For cross-sector water demand forecasting models, the training samples are set as basic forecasting units. The model input is a combination of the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics. The known water demand corresponding to this basic prediction unit is then used. The model parameters of the cross-sectoral water demand forecasting model are determined through model training, using the target values ​​as training objectives. The trained cross-sectoral water demand forecasting model is represented as follows:

[0108] (17)

[0109] in, This indicates that the cross-sectoral water demand forecasting model targets the basic forecasting unit. Output water demand forecast; This represents a cross-sectoral water demand forecasting model; This refers to the input features formed by a combination of the target department's own historical water demand characteristics, external driving characteristics, and cross-departmental historical information characteristics. This represents the parameters of the cross-sectoral water demand prediction model determined through the training sample set.

[0110] For independent water demand forecasting models, the combination of the target sector's own historical water demand characteristics and external driving characteristics corresponding to the same basic forecasting unit in the training sample set is used as the model input, and the known water demand corresponding to the basic forecasting unit is used as the input. The model parameters of the independent water demand prediction model are determined through model training, using the target values ​​as training objectives. The trained independent water demand prediction model is represented as follows:

[0111] (18)

[0112] in, This indicates that the independent water demand prediction model is for the basic prediction unit. Output water demand forecast; This represents an independent water demand forecasting model; This represents the input characteristics formed by the combination of the target department's own historical water demand characteristics and external driving characteristics; This represents the parameters of the independent water demand prediction model determined through the training sample set.

[0113] After the model is trained, the input features corresponding to the basic prediction unit to be predicted are input into the corresponding model to obtain the water demand prediction value corresponding to the basic prediction unit.

[0114] For example, when the cross-sectoral water demand prediction model and the independent water demand prediction model are built using the Extreme Gradient Boosting (XGBoost) algorithm, the model parameters include the number of regression trees, the regression tree structure, feature segmentation conditions, and leaf node output values. In this case, the cross-sectoral water demand prediction model is represented as:

[0115] (19)

[0116] The independent water demand prediction model is expressed as:

[0117] (20)

[0118] in, and These represent the number of regression trees included in the cross-sector water demand forecasting model and the independent water demand forecasting model, respectively. Represents a regression tree index; In the cross-sector water demand forecasting model, the first... A regression tree outputs numerical values ​​based on input features; In the independent water demand prediction model, the first... Each regression tree outputs a numerical value based on the input features. The sum of the values ​​output by each regression tree is the water demand prediction generated by the corresponding model.

[0119] The cross-sectoral water demand forecasting model and the independent water demand forecasting model use the same basic forecasting unit, training target value, sample set partitioning method, and regression forecasting model type. The two models are trained separately, and their respective model parameters are determined. The difference between the two models' inputs is that the cross-sectoral water demand forecasting model's input includes cross-sectoral historical information features, while the independent water demand forecasting model's input does not. Therefore, under the condition that other modeling conditions are consistent, the validation errors when using cross-sectoral historical information features and when not using them can be compared.

[0120] S4. Using the verification sample set, calculate the verification error of the cross-departmental water demand prediction model and the independent water demand prediction model for the target department combination in the same region, and select the verification gate based on the verification error to determine the cross-departmental water demand prediction model or the independent water demand prediction model corresponding to the target department combination in each region as the target water demand prediction model; use the target water demand prediction model to generate the final water demand prediction value of the corresponding basic prediction unit.

[0121] Specifically, for each basic prediction unit in the validation sample set The historical water demand characteristics of the target sector corresponding to the basic prediction unit, its external driving characteristics, and cross-sectoral historical information characteristics are input into the cross-sectoral water demand prediction model to obtain the cross-sectoral water demand prediction value. The historical water demand characteristics and external driving characteristics of the target sector corresponding to the same basic prediction unit are input into the independent water demand prediction model to obtain the independent water demand prediction value. Each basic prediction unit in the validation sample set has a known water demand. The water demand predictions generated by the two models are compared with the known water demand corresponding to the same basic prediction unit, and the validation errors of the two models are calculated. The validation error characterizes the degree of deviation between the model's predicted water demand and the known water demand; the smaller the validation error, the closer the model's predicted water demand on the validation sample set is to the known water demand. The validation gating uses a pre-determined error evaluation index to calculate the validation errors of the two models. The error evaluation index includes one of the following: root mean square error, mean absolute error, mean absolute percentage error, and weighted average absolute percentage error. For target sector combinations within the same region under the same prediction scale, the cross-sector water demand prediction model and the independent water demand prediction model use the same validation samples and error evaluation index.

[0122] In one implementation, the root mean square error (RMSE) is used as the error evaluation index for verifying the gating selection. Let the target region be... and target department The set of prediction time periods in the validation sample set is If the set of prediction time periods contains at least one prediction time period, then the water demand prediction model... Target area and target department The verification error is:

[0123] (twenty one)

[0124] in, Indicates water demand prediction model Target area and target department The verification error; ,in, This represents a cross-sectoral water demand forecasting model. This represents an independent water demand forecasting model; Indicates the target area and target department The set of prediction time periods in the validation sample set; This indicates the number of prediction periods included in the prediction period set; Represents the basic prediction unit The corresponding known water demand; Represents the prediction model The water demand forecast output by this basic forecasting unit uses the same unit of measurement as the known water demand.

[0125] The validation error of the cross-sectoral water demand prediction model is calculated using equation (21). Validation error of independent water demand prediction model When the validation error of the cross-sectoral water demand forecasting model is less than that of the independent water demand forecasting model, it indicates that, under the current validation sample and error evaluation index, the model incorporating cross-sectoral historical information features has a smaller validation error. Therefore, the cross-sectoral water demand forecasting model is selected as the target area. and target department The corresponding target water demand prediction model. When the validation error of the cross-sectoral water demand prediction model is greater than or equal to the validation error of the independent water demand prediction model, it indicates that, under the current validation sample and error evaluation index, the model containing cross-sectoral historical information features has not achieved a smaller validation error. Therefore, the independent water demand prediction model is selected as the target area. and target department The corresponding target water demand prediction model.

[0126] Set target area and target department The corresponding gate value is Then the gate selection rule is verified as follows:

[0127] (twenty two)

[0128] in, Indicates targeting area and target department Select a cross-sectoral water demand forecasting model; Indicates targeting area and target department Choose an independent water demand prediction model.

[0129] target area and target department The corresponding target water demand prediction model is:

[0130] (twenty three)

[0131] in, This indicates that after verifying the gating selection, the target area is... and target department A defined target water demand prediction model; This represents the cross-sectoral water demand prediction model established in step S3; This represents the independent water demand prediction model established in step S3.

[0132] For each prediction scale, the above-mentioned validation gating selection is implemented using the validation sample set corresponding to that prediction scale, resulting in the gating value and target water demand prediction model for each region's target sector combination at that prediction scale. The validation sample set is used only to calculate the validation error and determine the gating value. Through the above validation gating selection, for each region's target sector combination, a prediction model with a smaller validation error on the validation sample set is selected for subsequent generation of the final water demand prediction value corresponding to that region's target sector combination.

[0133] After completing the verification gating selection, for any basic prediction unit When the target area and target department Corresponding gate value When the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics corresponding to the basic prediction unit are input into the cross-sectoral water demand prediction model; when Then, the historical water demand characteristics and external driving characteristics of the target sector corresponding to the basic prediction unit are input into the independent water demand prediction model to obtain the final water demand prediction value:

[0134] (twenty four)

[0135] in, Indicates the target area Internal target department During the forecast period The final water demand forecast; and These represent the water demand forecasts generated by the cross-sectoral water demand forecasting model and the independent water demand forecasting model, respectively.

[0136] For each forecast scale, the final water demand forecast for each basic forecast unit, i.e., the target sector within the region, is obtained during the forecast period.

[0137] Example 2:

[0138] A regional multi-sector water demand forecasting system based on cross-sectoral historical information includes:

[0139] The data acquisition unit is used to determine the target sector, forecast scale, and forecast period based on the water resources management objectives of the target area; acquire historical water demand of the target area consistent with the forecast scale, as well as external driving data; and determine the basic forecast unit according to the forecast scale, based on the target area, target sector, and forecast period.

[0140] The feature construction unit is used to construct the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics based on the historical water demand of the target area before the prediction period and external driving data for the basic prediction unit.

[0141] The model building unit is used to combine the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics with the historical water demand corresponding to the basic prediction unit for each prediction scale to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; and the training sample set is used to build a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively.

[0142] The model selection and prediction unit is used to calculate the verification errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector in the same target area using the verification sample set, and to perform verification gating selection based on the verification errors to determine the target water demand prediction model corresponding to the target sector in each target area; and to generate the final water demand prediction value of the corresponding basic prediction unit using the target water demand prediction model.

[0143] Example 3:

[0144] An electronic device, comprising:

[0145] Memory, used to store computer programs;

[0146] A processor is used to execute the computer program to implement the regional multi-sector water demand prediction method based on cross-sectoral historical information.

[0147] Example 4:

[0148] A non-volatile storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the regional multi-sector water demand prediction method based on cross-sectoral historical information.

[0149] Example 5:

[0150] A computer program product includes a computer program / instructions that, when executed by a processor, implement the aforementioned regional multi-sector water demand forecasting method based on cross-sectoral historical information.

[0151] Example verification:

[0152] This embodiment uses a multi-sectoral monthly water demand forecasting method for a city in Zhejiang Province as an example to illustrate the specific implementation process and forecasting results of the regional multi-sectoral water demand forecasting method based on cross-sectoral historical information described in this invention. In this embodiment, the forecasting scale is monthly, and each forecasting period corresponds to one month.

[0153] Based on the city's water resource management objectives, water-demanding sectors are categorized into domestic water demand, energy-intensive and strategic industrial water demand, general industrial water demand, grain irrigation water demand, other agricultural water demand, and ecological environment water demand. Historical monthly water demand data for each sector from January 1971 to December 2010 are obtained, along with external driving data such as regional population size, GDP, industrial structure, industrial added value, agricultural output, grain sown area, cultivated land area, and ecological governance intensity for the corresponding periods, forming a regional multi-sectoral monthly water demand forecast dataset.

[0154] In this embodiment, at the monthly scale, target regions, target sectors, and months are combined to form basic forecasting units. Each basic forecasting unit corresponds to a target region, a target sector, and a month. For each basic forecasting unit, its corresponding water-demand sector is determined as the target sector, and based on the data available before the corresponding forecast month, the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics are constructed.

[0155] The target sector's own historical water demand characteristics include time trend characteristics, lagging water demand, lagging rolling mean, lagging rolling maximum, lagging rolling minimum, and lagging rolling standard deviation. Cross-sectoral historical information characteristics include lagging water demand characteristics of other sectors, regional lagging total water demand characteristics, and the proportion of lagging water demand in each sector. The target sector's lagging water demand, lagging rolling statistical characteristics, and cross-sectoral historical information characteristics are all constructed from historical water demand data prior to the forecast month, without using water demand data for the forecast month and beyond.

[0156] For basic prediction units with known water demand, their corresponding input features are associated with the known water demand to form historical samples. Based on the chronological order of the months corresponding to each historical sample, the historical samples from January 1971 to December 1990 are divided into a training sample set, and the historical samples from January 1991 to December 2002 are divided into a validation sample set. The training sample set is used to build cross-sectoral water demand prediction models and independent water demand prediction models, respectively; the validation sample set is used to calculate the validation error of the two models for the same target sector combination within the same region and to perform validation gating selection.

[0157] This embodiment employs the Extreme Gradient Boosting (XGBoost) algorithm to establish both a cross-sectoral water demand prediction model and an independent water demand prediction model. The cross-sectoral water demand prediction model uses the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics as inputs; the independent water demand prediction model uses the target sector's own historical water demand characteristics and external driving characteristics as inputs, without using cross-sectoral historical information characteristics. Both models use the same basic prediction unit, training target values, and sample set partitioning results, and are trained using their respective training sample sets.

[0158] This embodiment uses root mean square error (RMSE) as the error evaluation metric for the validation gating selection. For each target sector combination within a region, the RMSE of the cross-sectoral water demand forecasting model and the independent water demand forecasting model are calculated on the validation sample set. The cross-sectoral water demand forecasting model is selected when its validation error is less than that of the independent water demand forecasting model; the independent water demand forecasting model is selected when its validation error is greater than or equal to that of the independent water demand forecasting model. After completing the validation gating selection, the final water demand forecast value for the corresponding basic forecasting unit is generated using the selected forecasting model.

[0159] To compare the prediction results of different models, this embodiment uses a seasonal mean model, a Holt-Winters model, an LSTM model, and an independent water demand prediction model as comparison models. The seasonal mean model, Holt-Winters model, and LSTM model are only used for comparing prediction results and do not participate in the validation gating selection. Figure 3 As shown, the root mean square errors of the seasonal mean model, Holt-Winters model, LSTM model, independent water demand forecasting model, and cross-sectoral water demand forecasting model are 178.2 × 10⁻⁶. 4 m 3 133.17×10 4 m 3 166.07×10 4 m 3 85.89×10 4 m 3 and 84.35×10 4 m 3The weighted average absolute percentage error (MASE) of the cross-sectoral water demand forecasting model was 10.81%. Compared with the Holt-Winters model and the LSTM model, the root mean square error (RMSE) of the cross-sectoral water demand forecasting model was reduced by 36.7% and 49.2%, respectively. Under the sample division, model parameters, and error evaluation index settings of this embodiment, after incorporating the lagged water demand characteristics of other sectors, the lagged total water demand characteristics of the region, and the lagged water demand proportion characteristics of each sector, the RMSSE of the cross-sectoral water demand forecasting model was reduced by 1.8% compared with the independent water demand forecasting model; further, after adopting the validation gating selection based on the RMSSE criterion, the RMSSE was further reduced to 80.79 × 10⁻⁶. 4 m 3 .

[0160] Using July 2011 as the month to be predicted, and based on data available before July 2011, historical water demand characteristics, external driving characteristics, and cross-departmental historical information characteristics of the target departments corresponding to each basic prediction unit were constructed. These characteristics were then input into the target water demand prediction model determined by the verification gating selection to generate the final water demand prediction values ​​for each target department in the city in July 2011. During the model extrapolation prediction process, the model parameters and gating results remained unchanged, and actual water demand data from July 2011 and thereafter were not used. Table 1 lists the final water demand prediction values ​​for each target department in July 2011.

[0161] Table 1. Final water demand forecasts for target sectors within a city in Zhejiang Province in July 2011.

[0162]

[0163] The above results demonstrate that the method of the present invention can use historical water demand information of different water demand sectors in the same region for regional multi-sector water demand prediction, and reduce the prediction error of regional multi-sector water demand prediction results under the adopted error evaluation index by verifying the selection of a water demand prediction model that matches the target sector combination in each region.

Claims

1. A method for predicting regional multi-sectoral water demand based on cross-sectoral historical information, characterized in that, Includes the following steps: The target sector, forecast scale, and forecast period are determined based on the water resources management objectives of the target area. Obtain historical water demand data for the target area consistent with the forecast scale, as well as external driving data; determine basic forecasting units based on the target area, target sector, and forecast period according to the forecast scale. For the basic forecasting unit, based on the historical water demand of the target area before the forecast period and external driving data, the historical water demand characteristics of the target sector itself, the external driving characteristics, and the cross-sectoral historical information characteristics are constructed. For each prediction scale, the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics are combined with the historical water demand corresponding to the basic prediction unit to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; the training sample set is used to establish a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively; Using the aforementioned validation sample set, the validation errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector within the same target area are calculated respectively. Based on the validation errors, validation gating is selected to determine the target water demand prediction model corresponding to the target sector within each target area. The target water demand prediction model is then used to generate the final water demand prediction value for the corresponding basic prediction unit.

2. The regional multi-sectoral water demand prediction method based on cross-sectoral historical information according to claim 1, characterized in that, The target sector's own historical water demand characteristics include at least one of the following: time trend characteristics, target sector's lagged water demand, lagged rolling mean, lagged rolling maximum, lagged rolling minimum, and lagged rolling standard deviation; the external driving characteristics include at least one of the following: population size characteristics, economic development characteristics, industrial structure characteristics, agricultural production characteristics, and ecological governance characteristics.

3. The regional multi-sectoral water demand prediction method based on cross-sectoral historical information according to claim 1, characterized in that, The time trend feature is the predicted period minus the starting period corresponding to the historical water demand plus one; the target sector's lagged water demand is the water demand of the target sector in one or more historical periods before the predicted period, consistent with the current prediction scale; the lagged rolling mean is the average of the historical water demand of multiple consecutive target areas before the predicted period; the lagged rolling maximum value is the maximum historical water demand of the target sector within the rolling time window; the lagged rolling minimum value is the minimum historical water demand of the target sector within the rolling time window; the lagged rolling standard deviation is the degree of dispersion of the historical water demand of the target sector within the rolling time window relative to the lagged rolling mean; The population size category features include population size; the economic development category features include regional GDP and industrial added value; the industrial structure category features include industrial structure; the agricultural production category features include agricultural output value, grain sown area, cultivated land area and agricultural production level; and the ecological governance category features include ecological governance intensity.

4. The regional multi-sectoral water demand prediction method based on cross-sectoral historical information according to claim 1, characterized in that, The cross-departmental historical information features include at least one of the following: lagging water demand features of other departments, lagging total water demand features of the region, and lagging water demand ratio features of each water-demanding department; the lagging water demand features of other departments consist of the water demand of other water-demanding departments in the same region other than the target department in one or more historical periods before the prediction period; the lagging total water demand features of the region are the sum of the water demand of all water-demanding departments in the same region in the corresponding historical period; the lagging water demand ratio features of each water-demanding department are the ratio of the water demand of each water-demanding department in the corresponding historical period to the total water demand of the region in that historical period.

5. The regional multi-sectoral water demand prediction method based on cross-sectoral historical information according to claim 1, characterized in that, The cross-sectoral water demand prediction model and the independent water demand prediction model are established using the same type of regression prediction model. The regression prediction model includes one of the following: linear regression model, support vector regression model, random forest regression model, gradient boosting regression model, extreme gradient boosting model, neural network regression model, and time series regression model. The cross-sectoral water demand prediction model takes the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics as inputs. The independent water demand prediction model takes the target sector's own historical water demand characteristics and external driving characteristics as inputs. Both the cross-sectoral water demand prediction model and the independent water demand prediction model use the historical water demand corresponding to the basic prediction unit as the training target value and output the water demand prediction value of the basic prediction unit.

6. The regional multi-sectoral water demand prediction method based on cross-sectoral historical information according to claim 1, characterized in that, The verification gating selection based on the verification error includes: when the verification error of the cross-departmental water demand prediction model is less than the verification error of the independent water demand prediction model, selecting the cross-departmental water demand prediction model as the target water demand prediction model corresponding to the target department; when the verification error of the cross-departmental water demand prediction model is greater than or equal to the verification error of the independent water demand prediction model, selecting the independent water demand prediction model as the target water demand prediction model corresponding to the target department.

7. A regional multi-sectoral water demand prediction system based on cross-sectoral historical information, characterized in that, include: The data acquisition unit is used to determine the target sector, forecast scale, and forecast period based on the water resources management objectives of the target area; Obtain historical water demand data for the target area consistent with the forecast scale, as well as external driving data; determine basic forecasting units based on the target area, target sector, and forecast period according to the forecast scale. The feature construction unit is used to construct the target sector's own historical water demand characteristics, external driving characteristics, and cross-sectoral historical information characteristics based on the historical water demand of the target area before the prediction period and external driving data for the basic prediction unit. The model building unit is used to combine the historical water demand characteristics of the target sector itself, external driving characteristics, and cross-sectoral historical information characteristics with the historical water demand corresponding to the basic prediction unit for each prediction scale to form a historical sample; the historical sample is divided into a training sample set and a validation sample set; and the training sample set is used to build a cross-sectoral water demand prediction model and an independent water demand prediction model, respectively. The model selection and prediction unit is used to calculate the verification errors of the cross-sectoral water demand prediction model and the independent water demand prediction model for the target sector in the same target area using the verification sample set, and to perform verification gating selection based on the verification errors to determine the target water demand prediction model corresponding to the target sector in each target area; and to generate the final water demand prediction value of the corresponding basic prediction unit using the target water demand prediction model.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the regional multi-sector water demand forecasting method based on cross-sectoral historical information as described in any one of claims 1-6.

9. A non-volatile storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the regional multi-sector water demand prediction method based on cross-sectoral historical information as described in any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the regional multi-sector water demand prediction method based on cross-sectoral historical information as described in any one of claims 1-6.

Images