Agricultural machine well based on electricity separation and abnormality identification water metering method

Abstract

The present application relates to the field of electricity water metering, and discloses an agricultural machine well electricity water metering method based on electricity separation and abnormality identification, which comprises the following steps: sequentially completing multi-source water electricity time sequence collection preprocessing and load feature library building, and implementing multi-model parallel agricultural irrigation electricity intelligent decoupling, water electricity abnormality self-adaptive identification repair, multi-factor dynamic water conversion coefficient modeling, water quantity measurement and multi-dimensional precision verification whole-process metering. The present application solves the triple defects of traditional metering mechanism black box, data distortion and conversion static state, and improves the precision of agricultural machine well groundwater metering.

Description

Technical Field

[0001] This invention relates to the field of water metering by electricity conversion, and in particular to a method for water metering by electricity conversion in agricultural wells based on electricity separation and anomaly identification. Background Technology

[0002] Agricultural groundwater irrigation is a key target for water use management. Traditional direct-measurement water meters are prone to clogging and corrosion in areas with high sediment, iron and manganese content. In addition, rural wells are scattered and have high maintenance costs. Electricity-based water metering has become the mainstream alternative metering method due to its low cost and resistance to water quality interference. The current industry-standard electricity-to-water conversion scheme has two major shortcomings. First, in southern agricultural areas, irrigation pumps, fertigation equipment, supplemental lighting, and household electricity consumption are often mixed. Existing NILM and simple machine learning decomposition models can only roughly break down the load, failing to fully exploit the impact of pump start-up and shutdown, cyclical timing, and time-frequency differences. This makes it difficult to accurately separate pure irrigation power consumption, and the mixed electricity consumption directly leads to a distortion in the water volume conversion base. Second, existing technologies mostly use static conversion coefficients across the entire domain, without coupling dynamic operating conditions such as seasonal fluctuations in groundwater levels, transformer loads, and pump aging. At the same time, high temperatures in the field, lightning strikes, and equipment outages can easily generate dirty data such as missing breakpoints, sudden spikes, and system offsets. Traditional linear interpolation and fixed threshold discrimination methods cannot achieve adaptive repair and anomaly identification. The superposition of multiple errors leads to large deviations in irrigation water volume calculation, making it difficult to support accurate collection of water resource taxes and refined groundwater supervision.

[0003] Existing technologies only address single aspects of load separation or coefficient correction, lacking an integrated closed-loop metering solution encompassing "electricity decoupling, anomaly identification, data repair, and dynamic water conversion." This results in poor model generalization and insufficient adaptability to the diverse mixed electricity consumption scenarios in Guangdong's agriculture. Conventional load separation algorithms only capture short-term power fluctuations, lacking the ability to mine long-term irrigation cycle correlations, easily misclassifying domestic and agricultural auxiliary electricity consumption as irrigation load. Data quality control relies on manual screening and simple interpolation, leading to significant deviations from the actual evolution of hydropower over long periods due to missing data. Furthermore, the water conversion coefficient remains fixed year-round, ignoring the electromechanical conversion efficiency fluctuations caused by water level changes during wet and dry seasons. The entire metering process suffers from three major technical bottlenecks: a black box mechanism, data distortion, and static conversion, failing to meet the practical needs of high-precision monitoring of agricultural groundwater extraction.

[0004] This patented solution addresses the aforementioned deficiencies by employing an electric-to-water metering method for agricultural wells that separates electrical charge and identifies anomalies. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method for measuring water volume in agricultural wells by electricity conversion based on electricity separation and anomaly identification.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for measuring water volume in agricultural wells by electricity conversion based on electricity separation and anomaly identification, comprising the following steps: Multi-source high-frequency hydropower time-series data were collected and preprocessed from sample wells monitored for hydropower conversion, and a multi-scenario load characteristic library was constructed. Based on a multi-scenario load feature library, a model parallel method is used to intelligently separate agricultural irrigation power consumption to obtain time-series data of pure irrigation power consumption. The time series data of pure irrigation power consumption is used to identify hydropower time series anomalies, and multi-algorithm adaptive interpolation is used to repair the anomalies. Based on the standard pure irrigation power consumption time series dataset, a multi-factor dynamic conversion coefficient model is constructed to dynamically correct the water conversion coefficient and obtain the dynamic electricity-to-water conversion coefficient. Based on the dynamic electro-water conversion coefficient, the irrigation water volume of the monitoring sample wells was calculated using electro-water conversion, and the accuracy of the calculation results was verified in multiple dimensions.

[0007] Furthermore, in a preferred embodiment of the present invention, the step of collecting and preprocessing multi-source high-frequency hydropower time-series data from the hydropower monitoring sample wells and constructing a multi-scenario load feature library specifically includes: Obtain sample wells for monitoring water quality using electricity conversion and use them as data collection points; synchronously deploy metering and data acquisition terminals on all sample wells for monitoring water quality using electricity conversion. The metering and data acquisition terminals include the main smart meter for the distribution area, the independent sub-meters for well pumps, and the branch metering meters; Obtain the central processing terminal, connect all metering and acquisition terminals, set the acquisition time period of the metering and acquisition terminals in the central processing terminal, and control all metering and acquisition terminals to acquire multi-source high-frequency hydropower time-series data in real time according to the acquisition time period. By introducing historical data networks, auxiliary business data of multi-source high-frequency hydropower time series data are retrieved, and combined with multi-source high-frequency hydropower time series data, an original multi-source time series dataset is constructed. At the same time, data standardization preprocessing of the original multi-source time series dataset is performed to obtain the target multi-source time series dataset. The target multi-source time series dataset is divided into three categories of electricity consumption scenario sample subsets according to the electricity consumption structure: pure irrigation scenario sample subset, irrigation combined with farmers' life scenario sample subset, and other scenario sample subsets. For each type of electricity consumption scenario sample subset, multi-scenario load features are extracted, including time-domain features, frequency-domain features, time-frequency joint features, and equipment behavior features. The multi-scenario load features are then integrated into an electrical equipment feature vector in the central processing terminal. By combining the feature vectors of all electrical equipment and sample subsets of three types of electricity consumption scenarios, a multi-scenario load feature library is constructed.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of intelligently separating agricultural irrigation power consumption based on a multi-scenario load feature library and using a model parallel method to obtain time-series data of pure irrigation power consumption specifically involves: Four types of data processing models are deployed in parallel in the central processing terminal, including the NILM non-intrusive load identification model, the TCN-BiLSTM-Attention deep decomposition model, the ANN artificial neural network model, and the SVM support vector machine model. The multi-scenario load feature library is connected with four types of data processing models. In the NILM non-intrusive load identification model, the feature data of the multi-scenario load feature library is used to detect the start and stop events of electrical equipment. The time domain and frequency domain feature vectors are matched to realize the initial split of mixed loads and output coarsely separated agricultural irrigation power consumption data. The coarsely separated agricultural irrigation electricity data includes the electricity consumption of agricultural irrigation, domestic electricity, and other agricultural electricity. By using the TCN temporal convolutional layer in the TCN-BiLSTM-Attention deep decomposition model, local high-frequency feature data of the multi-scenario load feature library is extracted. Then, the BiLSTM bidirectional long short-term memory layer is used to mine the long-term temporal dependence of the irrigation load daily cycle and seasonal cycle of the feature data of the multi-scenario load feature library. Finally, the Attention mechanism is used to add temporal noise and output a standard pure irrigation power time series. Simultaneously, the feature vectors of electrical equipment are input into the ANN artificial neural network model and the SVM support vector machine model to calculate the proportion coefficient of agricultural irrigation electricity consumption. A unified evaluation index is introduced to analyze the proportion coefficient of agricultural irrigation electricity consumption, coarsely separated agricultural irrigation electricity consumption data and standard pure irrigation power time series, and to measure the electricity separation accuracy, scenario generalization ability and computing efficiency of the four types of models. A scoring mechanism is set up to evaluate the accuracy of power separation, the ability to generalize scenarios, and the computational efficiency. The model with the highest score is labeled as the optimal model, and the output data of the optimal model is output and marked as time series data of pure irrigation power consumption.

[0009] Furthermore, in a preferred embodiment of the present invention, the step of identifying hydropower time-series anomalies in pure irrigation power consumption data and performing multi-algorithm adaptive interpolation repair based on the anomalies specifically includes: Within the central processing terminal, a rule library for determining anomalies in hydropower data is introduced. This rule library categorizes abnormal hydropower data into four types: missing data due to equipment disconnection, sudden spikes caused by lightning strikes or equipment malfunctions, systematic offset data caused by meter aging and drift, and seasonal deviation data where the load time series does not match the crop irrigation cycle. The sliding window dynamic threshold judgment rule is built into the hydropower data anomaly judgment rule base. The pure irrigation power consumption time series data is introduced into the hydropower data anomaly judgment rule base. Combined with the sliding window dynamic threshold judgment rule, the pure irrigation power consumption time series data is traversed to judge hydropower data anomalies. Threshold classification was performed on four types of abnormal hydropower data to obtain three abnormality level labels: mild transient noise, moderate continuous missing data, and severe system offset. Abnormality level labels were also applied to the time series data of pure irrigation power consumption. By retrieving time-series data of pure irrigation power consumption under different anomaly level labels from historical data networks, an interpolation and repair algorithm is used to generate a standard pure irrigation power consumption time-series dataset.

[0010] Furthermore, in a preferred embodiment of the present invention, the step of constructing a multi-factor dynamic conversion coefficient model based on a standard pure irrigation power consumption time-series dataset for dynamically correcting the water conversion coefficient to obtain a dynamic electricity-to-water conversion coefficient specifically involves: In the central processing terminal, a multi-factor dynamic conversion coefficient model is constructed. The construction method of the multi-factor dynamic conversion coefficient model is to acquire all the multi-dimensional operating condition driving factors of the monitoring sample wells with electric water conversion in real time through the central processing terminal. Using the standard pure irrigation power consumption time series dataset as the model training benchmark sample, and combining multi-dimensional operating condition driving factors, a multi-factor weighted regression calculation framework is built to quantify the influence weight of multi-dimensional operating condition driving factors on the water pump hydro-electricity conversion efficiency, and obtain a multi-factor dynamic conversion coefficient model. The multi-factor dynamic conversion coefficient model is used to output the dynamic electricity-to-water conversion coefficient. The update cycle of the multi-factor dynamic conversion coefficient model is preset. When the update cycle of the multi-factor dynamic conversion coefficient model is reached, the multi-dimensional operating condition driving factors are re-acquired, and the influence weight of the multi-dimensional operating condition driving factors on the water pump water-electricity conversion efficiency is updated. The dynamic electricity-to-water conversion coefficient is output in real time.

[0011] Furthermore, in a preferred embodiment of the present invention, the step of calculating the irrigation water volume of the monitoring sample wells based on the dynamic electro-water conversion coefficient, and verifying the accuracy of the calculation results in multiple dimensions, specifically includes: In the central processing terminal, a standard time step is preset. Based on the standard time step, the dynamic electricity-to-water conversion coefficient and the standard pure irrigation power consumption time series dataset are multiplied step by step to output the well groundwater irrigation extraction volume under different time steps, which is calibrated as the initial irrigation water volume calculation result. Reacquire water monitoring sample wells using electrostatic precipitation, mark them as independent monitoring sample wells, and perform three-segment accuracy verification based on the independent monitoring sample wells; The three-stage accuracy verification includes accuracy verification of four types of data processing models, accuracy verification of standard pure irrigation power consumption time series dataset, and accuracy verification of dynamic electricity-to-water conversion coefficient. If all three precision checks meet the preset values, the initial irrigation water volume calculation result will be marked as the target irrigation water volume calculation result. If the accuracy verification does not meet the preset value, the multi-scenario load feature library, hydropower anomaly judgment threshold and multi-factor weighted regression weight are updated in reverse iteration, and the power separation, data repair and dynamic coefficient calculation process is re-executed until the target irrigation water volume measurement result is output.

[0012] This invention addresses the technical deficiencies in the background technology and offers the following advantages: it sequentially completes the preprocessing of multi-source hydropower time-series acquisition and load feature database construction, and implements intelligent decoupling of agricultural irrigation power generation across multiple models, adaptive identification and repair of hydropower anomalies, multi-factor dynamic water conversion coefficient modeling, water volume calculation, and multi-dimensional accuracy verification throughout the entire metering process. This invention overcomes the triple defects of traditional metering mechanisms—black box approach, data distortion, and static conversion—and improves the metering accuracy of groundwater extraction from agricultural wells. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0014] Figure 1 A flowchart of an electric water metering method for agricultural wells based on power separation and anomaly identification is shown. Figure 2 A flowchart is shown illustrating the method for outputting dynamic electro-water conversion coefficients and calculating irrigation water volume of sample wells using electro-water conversion. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1A flowchart illustrating a method for metering water volume in agricultural wells based on electricity separation and anomaly identification is shown, including the following steps: Multi-source high-frequency hydropower time-series data were collected and preprocessed from sample wells monitored for hydropower conversion, and a multi-scenario load characteristic library was constructed. Based on a multi-scenario load feature library, a model parallel method is used to intelligently separate agricultural irrigation power consumption to obtain time-series data of pure irrigation power consumption. The time series data of pure irrigation power consumption is used to identify hydropower time series anomalies, and multi-algorithm adaptive interpolation is used to repair the anomalies. Based on the standard pure irrigation power consumption time series dataset, a multi-factor dynamic conversion coefficient model is constructed to dynamically correct the water conversion coefficient and obtain the dynamic electricity-to-water conversion coefficient. Based on the dynamic electro-water conversion coefficient, the irrigation water volume of the monitoring sample wells was calculated using electro-water conversion, and the accuracy of the calculation results was verified in multiple dimensions.

[0018] Furthermore, in a preferred embodiment of the present invention, the step of collecting and preprocessing multi-source high-frequency hydropower time-series data from the hydropower monitoring sample wells and constructing a multi-scenario load feature library specifically includes: Obtain sample wells for monitoring water quality using electricity conversion and use them as data collection points; synchronously deploy metering and data acquisition terminals on all sample wells for monitoring water quality using electricity conversion. The metering and data acquisition terminals include the main smart meter for the distribution area, the independent sub-meters for well pumps, and the branch metering meters; Obtain the central processing terminal, connect all metering and acquisition terminals, set the acquisition time period of the metering and acquisition terminals in the central processing terminal, and control all metering and acquisition terminals to acquire multi-source high-frequency hydropower time-series data in real time according to the acquisition time period. By introducing historical data networks, auxiliary business data of multi-source high-frequency hydropower time series data are retrieved, and combined with multi-source high-frequency hydropower time series data, an original multi-source time series dataset is constructed. At the same time, data standardization preprocessing of the original multi-source time series dataset is performed to obtain the target multi-source time series dataset. The target multi-source time series dataset is divided into three categories of electricity consumption scenario sample subsets according to the electricity consumption structure: pure irrigation scenario sample subset, irrigation combined with farmers' life scenario sample subset, and other scenario sample subsets. For each type of electricity consumption scenario sample subset, multi-scenario load features are extracted, including time-domain features, frequency-domain features, time-frequency joint features, and equipment behavior features. The multi-scenario load features are then integrated into an electrical equipment feature vector in the central processing terminal. By combining the feature vectors of all electrical equipment and sample subsets of three types of electricity consumption scenarios, a multi-scenario load feature library is constructed.

[0019] It should be noted that standardized monitoring wells were selected as data collection points, and three layers of metering equipment were deployed simultaneously. These included a main meter for collecting mixed total load, independent sub-meters for wells collecting pure irrigation electricity values, and separate meters for domestic, water and fertilizer, and supplemental lighting electricity consumption, achieving synchronous comparison and collection of total load and various sub-loads. This three-level synchronous metering architecture—main meter + independent irrigation sub-meter + various miscellaneous electricity sub-meters—is specifically adapted to scenarios with multiple overlapping loads, avoiding the inability to verify accuracy due to the lack of true value comparison after load separation. Secondly, a unified minute-level high-frequency data collection cycle was configured, and all meters, water level, and flow sensors were centrally managed to synchronously collect high-frequency time-series data on power, electricity consumption, groundwater level, and pipeline flow. This achieved timestamp-aligned synchronous collection across all devices and connected to a historical business database to retrieve auxiliary information such as meteorological data, crop planting data, irrigation regulations, pump parameters, and transformer load. This data was then fused with the high-frequency time-series data of water and electricity to form a raw dataset, which underwent standardization processing, including unifying units, time intervals, and formats, and removing garbled, duplicate, and missing timestamp data. Finally, samples were stored independently for each scenario, including: pure irrigation with only well pumping, irrigation plus household use, and irrigation plus water and fertilizer / supplementary lighting / secondary pumping for other agricultural production scenarios. Because the load superposition patterns vary greatly across different scenarios, training the model with a unified dataset is prone to poor scenario generalization. Scenario-based segmentation allows for the targeted extraction of load features specific to each scenario, improving the model's decomposition accuracy under different mixed power consumption scenarios. Finally, standardized feature vectors were extracted, including time-domain (start-stop abrupt changes, peak values, periods), frequency-domain (Fourier transform motor harmonics), time-frequency joint (wavelet transform), and behavioral features (irrigation cycle, equipment start-stop patterns). The purpose is to provide conditions for subsequent modeling.

[0020] Furthermore, in a preferred embodiment of the present invention, the step of intelligently separating agricultural irrigation power consumption based on a multi-scenario load feature library and using a model parallel method to obtain time-series data of pure irrigation power consumption specifically involves: Four types of data processing models are deployed in parallel in the central processing terminal, including the NILM non-intrusive load identification model, the TCN-BiLSTM-Attention deep decomposition model, the ANN artificial neural network model, and the SVM support vector machine model. The multi-scenario load feature library is connected with four types of data processing models. In the NILM non-intrusive load identification model, the feature data of the multi-scenario load feature library is used to detect the start and stop events of electrical equipment. The time domain and frequency domain feature vectors are matched to realize the initial split of mixed loads and output coarsely separated agricultural irrigation power consumption data. The coarsely separated agricultural irrigation electricity data includes the electricity consumption of agricultural irrigation, domestic electricity, and other agricultural electricity. By using the TCN temporal convolutional layer in the TCN-BiLSTM-Attention deep decomposition model, local high-frequency feature data of the multi-scenario load feature library is extracted. Then, the BiLSTM bidirectional long short-term memory layer is used to mine the long-term temporal dependence of the irrigation load daily cycle and seasonal cycle of the feature data of the multi-scenario load feature library. Finally, the Attention mechanism is used to add temporal noise and output a standard pure irrigation power time series. Simultaneously, the feature vectors of electrical equipment are input into the ANN artificial neural network model and the SVM support vector machine model to calculate the proportion coefficient of agricultural irrigation electricity consumption. A unified evaluation index is introduced to analyze the proportion coefficient of agricultural irrigation electricity consumption, coarsely separated agricultural irrigation electricity consumption data and standard pure irrigation power time series, and to measure the electricity separation accuracy, scenario generalization ability and computing efficiency of the four types of models. A scoring mechanism is set up to evaluate the accuracy of power separation, the ability to generalize scenarios, and the computational efficiency. The model with the highest score is labeled as the optimal model, and the output data of the optimal model is output and marked as time series data of pure irrigation power consumption.

[0021] It should be noted that four differentiated algorithm models are simultaneously built on a unified central processing terminal. These four models are adapted to different data scales and electricity consumption scenarios. The NILM non-intrusive load identification model serves as the basic control model, the TCN-BiLSTM-Attention model is the main high-precision deep learning model, and ANN and SVM are lightweight auxiliary models for small sample sizes. The four models are operated in parallel to improve computational efficiency. Specifically, a multi-scenario load feature library is connected to the four data processing models. Based on the time-frequency characteristics of water pumps, household appliances, and agricultural auxiliary equipment, the power jump events of equipment start-stop are identified. The total mixed electricity consumption is initially divided into three sub-items: irrigation, domestic, and other agricultural electricity consumption. This is then combined with the TCN-BiLSTM-Attention deep decomposition model. The TCN layer captures unique local high-frequency features such as instantaneous impacts during pump start-up and shutdown, and short-term power fluctuations. BiLSTM bidirectionally traverses the entire time series to uncover daily and seasonal cyclical electricity consumption patterns for irrigation. An Attention mechanism amplifies the feature weights of irrigation periods and suppresses irrelevant noise from daily life and lighting, ultimately outputting a pure irrigation power time series free from extraneous power interference. These three functions integrate short-term local features with long-term time series patterns, filtering interference from multiple dimensions. Standardized feature vectors are fed into ANN and SVM, and the proportion of irrigation electricity to total load is calculated through nonlinear fitting and the optimal classification hyperplane. RMSE, MAE, MAPE, and R² indices are uniformly used to simultaneously quantify the separation error, cross-scenario adaptability, and computation time of the four models, reducing the computational requirements of the equipment. Finally, weighted scoring rules are set for accuracy, generalization ability, and computational efficiency, automatically calculating the comprehensive score of the four models. The output of the model with the highest score is selected as the final pure irrigation power time series and sent to the subsequent anomaly identification and repair process.

[0022] Furthermore, in a preferred embodiment of the present invention, the step of identifying hydropower time-series anomalies in pure irrigation power consumption data and performing multi-algorithm adaptive interpolation repair based on the anomalies specifically includes: Within the central processing terminal, a rule library for determining anomalies in hydropower data is introduced. This rule library categorizes abnormal hydropower data into four types: missing data due to equipment disconnection, sudden spikes caused by lightning strikes or equipment malfunctions, systematic offset data caused by meter aging and drift, and seasonal deviation data where the load time series does not match the crop irrigation cycle. The sliding window dynamic threshold judgment rule is built into the hydropower data anomaly judgment rule base. The pure irrigation power consumption time series data is introduced into the hydropower data anomaly judgment rule base. Combined with the sliding window dynamic threshold judgment rule, the pure irrigation power consumption time series data is traversed to judge hydropower data anomalies. Threshold classification was performed on four types of abnormal hydropower data to obtain three abnormality level labels: mild transient noise, moderate continuous missing data, and severe system offset. Abnormality level labels were also applied to the time series data of pure irrigation power consumption. By retrieving time-series data of pure irrigation power consumption under different anomaly level labels from historical data networks, an interpolation and repair algorithm is used to generate a standard pure irrigation power consumption time-series dataset.

[0023] It should be noted that the central processing terminal has a built-in standardized anomaly rule knowledge base that records all fault types, classifying distorted data into four fixed anomaly types: breakpoint missing, sudden spikes, systematic offset, and seasonal deviation. Simultaneously, the rule knowledge base incorporates a sliding window dynamic threshold discrimination logic. The output pure irrigation power consumption time series is imported into the rule base segment by segment using a sliding window to traverse the time series. Node judgment is performed based on the specific thresholds for each of the four anomaly types to determine if data anomalies exist. If so, a tiered threshold is set based on the distortion duration, deviation magnitude, and duration period of the four anomalies. All identified anomalies are classified into three levels: mild transient noise, moderate continuous missing, and severe systematic offset. A tiered label is automatically assigned to the corresponding anomaly nodes in the time series, distinguishing between easily repairable, moderately repairable, and difficult-to-repair dirty data. Finally, the preset hierarchical repair strategy in the historical data network is retrieved: for mild transient noise and single-point short missing data, linear interpolation is used; for moderate continuous missing data, MICE multi-chain interpolation is used; for severe long time series and large-area missing data, LSTM generative time series reconstruction model is used; the corresponding algorithm is automatically matched to complete the repair, and completely invalid and unrepairable data is removed, and a cleaned and distortion-free standard pure irrigation power consumption time series dataset is output.

[0024] Figure 2 The flowchart illustrates the method for outputting dynamic electro-hydraulic coefficients and calculating irrigation water volume from monitored sample wells using electro-hydraulic coefficients, including the following steps: S202: Based on the standard pure irrigation power consumption time series dataset, a multi-factor dynamic conversion coefficient model is constructed to dynamically correct the water conversion coefficient and obtain the dynamic electricity-to-water conversion coefficient. S204: Based on the dynamic electro-water conversion coefficient, the irrigation water volume of the monitoring sample wells is calculated using electro-water conversion, and the accuracy of the calculation results is verified in multiple dimensions.

[0025] Furthermore, in a preferred embodiment of the present invention, the step of constructing a multi-factor dynamic conversion coefficient model based on a standard pure irrigation power consumption time-series dataset for dynamically correcting the water conversion coefficient to obtain a dynamic electricity-to-water conversion coefficient specifically involves: In the central processing terminal, a multi-factor dynamic conversion coefficient model is constructed. The construction method of the multi-factor dynamic conversion coefficient model is to acquire all the multi-dimensional operating condition driving factors of the monitoring sample wells with electric water conversion in real time through the central processing terminal. Using the standard pure irrigation power consumption time series dataset as the model training benchmark sample, and combining multi-dimensional operating condition driving factors, a multi-factor weighted regression calculation framework is built to quantify the influence weight of multi-dimensional operating condition driving factors on the water pump hydro-electricity conversion efficiency, and obtain a multi-factor dynamic conversion coefficient model. The multi-factor dynamic conversion coefficient model is used to output the dynamic electricity-to-water conversion coefficient. The update cycle of the multi-factor dynamic conversion coefficient model is preset. When the update cycle of the multi-factor dynamic conversion coefficient model is reached, the multi-dimensional operating condition driving factors are re-acquired, and the influence weight of the multi-dimensional operating condition driving factors on the water pump water-electricity conversion efficiency is updated. The dynamic electricity-to-water conversion coefficient is output in real time.

[0026] It should be noted that real-time acquisition of multi-dimensional operating condition driving factors for all monitoring sample wells using the electricity-to-water conversion method is crucial. This includes real-time groundwater level depth, transformer load rate, ambient temperature, seasonal irrigation factors, pump rated power, head, and equipment service life. Traditional electricity-to-water conversion methods only use a single variable, electricity consumption, to calculate fixed coefficients, completely ignoring key variables that affect pump efficiency, such as groundwater depth, pump aging, and seasonal load. The same set of coefficients exhibits significant errors under different seasons and with varying pump types. This new method introduces multi-dimensional coupling factors, including groundwater dynamics, equipment operating conditions, and seasonal agricultural activities, establishing a multi-coupled input system of water, electricity, equipment, and environment. Simultaneously, a multi-factor dynamic conversion coefficient model is constructed. This model uses a standard pure irrigation power consumption time-series dataset as the training benchmark sample, combined with multi-dimensional operating condition driving factors, to build a multi-factor weighted regression calculation framework. Through regression fitting, the contribution weight of each factor to the water-to-electricity conversion efficiency is calculated, completing model training. After training, the model can output corresponding dynamic conversion coefficients based on real-time operating conditions. Because fixed coefficients are essentially averaged weights, they cannot distinguish the magnitude of the effects of each factor under different operating conditions. Weighted regression can quantify the influence of each factor, accurately depicting the incremental patterns of pumping power consumption when water levels drop, pumps age, or loads increase. Finally, a preset update cycle is used. At each update node, the terminal automatically re-collects the latest groundwater level, load, seasonal agricultural activities, and pump wear data, refits the weights of each factor, updates the model parameters, and continuously outputs dynamic water conversion coefficients adapted to the current operating conditions. Because groundwater depth fluctuates significantly seasonally, crop irrigation cycles change with the growth period, and pumps continuously age, fixed coefficients, if not updated throughout the year, will accumulate calculation errors. Periodically iterated weights can continuously correct the conversion standard according to changes in operating conditions, achieving dynamic adaptation.

[0027] Furthermore, in a preferred embodiment of the present invention, the step of calculating the irrigation water volume of the monitoring sample wells based on the dynamic electro-water conversion coefficient, and verifying the accuracy of the calculation results in multiple dimensions, specifically includes: In the central processing terminal, a standard time step is preset. Based on the standard time step, the dynamic electricity-to-water conversion coefficient and the standard pure irrigation power consumption time series dataset are multiplied step by step to output the well groundwater irrigation extraction volume under different time steps, which is calibrated as the initial irrigation water volume calculation result. Reacquire water monitoring sample wells using electrostatic precipitation, mark them as independent monitoring sample wells, and perform three-segment accuracy verification based on the independent monitoring sample wells; The three-stage accuracy verification includes accuracy verification of four types of data processing models, accuracy verification of standard pure irrigation power consumption time series dataset, and accuracy verification of dynamic electricity-to-water conversion coefficient. If all three precision checks meet the preset values, the initial irrigation water volume calculation result will be marked as the target irrigation water volume calculation result. If the accuracy verification does not meet the preset value, the multi-scenario load feature library, hydropower anomaly judgment threshold and multi-factor weighted regression weight are updated in reverse iteration, and the power separation, data repair and dynamic coefficient calculation process is re-executed until the target irrigation water volume measurement result is output.

[0028] It should be noted that by pre-setting multiple levels of standard statistical time steps, the dynamics of each time slice are multiplied by the electricity-to-water conversion factor and the standard pure irrigation power consumption for the corresponding time period, and the time-sharing, multi-level groundwater irrigation extraction volume is calculated separately, resulting in an initial water volume calculation result that has not undergone accuracy verification. This is because traditional electricity-to-water conversion only outputs the total monthly water volume and lacks multi-time-scale stratified measurement, which cannot support refined water use pattern analysis. The time-step time-series matching calculation enables refined water volume accounting for all time periods, matching the business needs of groundwater dynamic monitoring and crop growth stage water use management.

[0029] Independent monitoring wells not involved in model training and feature library construction were selected as the validation set, and a three-stage verification was conducted: the first stage verified the accuracy of separating agricultural irrigation electricity consumption for the four types of models, using the actual measured irrigation electricity consumption from the independent sub-meters of the wells as the true value; the second stage verified the data quality of the standard electricity dataset after anomaly identification and interpolation repair; the third stage verified the degree of matching between the water volume calculated by the multi-factor dynamic conversion coefficient and the actual flow rate measured by the standard water meter in the pipeline. The three-stage verification can accurately locate the error generation points in the entire process. Finally, the system pre-sets error thresholds for each stage. When the errors of the electricity separation, data repair, and dynamic conversion verification are all below the set thresholds, the calculation results of the entire process are deemed reliable, and the initial water volume result is used as the final standardized target irrigation water volume output. If any error in any of the three verification stages exceeds the standard, the system automatically reverses the source and optimizes the corresponding module parameters: if the power separation accuracy is insufficient, the multi-scenario load feature library is expanded and updated; if the data repair error exceeds the standard, the sliding window threshold for water and electricity anomaly judgment is adjusted and the interpolation model parameters are optimized; if the water volume deviation exceeds the standard, the multi-factor weighted regression weights are updated; the entire process of power separation, anomaly repair, and dynamic coefficient calculation is automatically restarted and iterated until all three verification stages meet the standards.

[0030] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for metering water volume in agricultural wells based on electricity separation and anomaly identification, characterized in that: Includes the following steps: Multi-source high-frequency hydropower time-series data were collected and preprocessed from sample wells monitored for hydropower conversion, and a multi-scenario load characteristic library was constructed. Based on a multi-scenario load feature library, a model parallel method is used to intelligently separate agricultural irrigation power consumption to obtain time-series data of pure irrigation power consumption. The time series data of pure irrigation power consumption is used to identify hydropower time series anomalies, and multi-algorithm adaptive interpolation is used to repair the anomalies. Based on the standard pure irrigation power consumption time series dataset, a multi-factor dynamic conversion coefficient model is constructed to dynamically correct the water conversion coefficient and obtain the dynamic electricity-to-water conversion coefficient. Based on the dynamic electro-water conversion coefficient, the irrigation water volume of the monitoring sample wells was calculated using electro-water conversion, and the accuracy of the calculation results was verified in multiple dimensions.

2. The method for measuring water volume in agricultural wells based on electricity separation and anomaly identification as described in claim 1, characterized in that, The process of acquiring and preprocessing multi-source high-frequency hydropower time-series data from sample wells monitored for hydropower conversion, and constructing a multi-scenario load feature library, specifically involves: Obtain sample wells for monitoring water quality using electricity conversion and use them as data collection points; synchronously deploy metering and data acquisition terminals on all sample wells for monitoring water quality using electricity conversion. The metering and data acquisition terminals include the main smart meter for the distribution area, the independent sub-meters for well pumps, and the branch metering meters; Obtain the central processing terminal, connect all metering and acquisition terminals, set the acquisition time period of the metering and acquisition terminals in the central processing terminal, and control all metering and acquisition terminals to acquire multi-source high-frequency hydropower time-series data in real time according to the acquisition time period. By introducing historical data networks, auxiliary business data of multi-source high-frequency hydropower time series data are retrieved, and combined with multi-source high-frequency hydropower time series data, an original multi-source time series dataset is constructed. At the same time, data standardization preprocessing of the original multi-source time series dataset is performed to obtain the target multi-source time series dataset. The target multi-source time series dataset is divided into three categories of electricity consumption scenario sample subsets according to the electricity consumption structure: pure irrigation scenario sample subset, irrigation combined with farmers' life scenario sample subset, and other scenario sample subsets. For each type of electricity consumption scenario sample subset, multi-scenario load features are extracted, including time-domain features, frequency-domain features, time-frequency joint features, and equipment behavior features. The multi-scenario load features are then integrated into an electrical equipment feature vector in the central processing terminal. By combining the feature vectors of all electrical equipment and sample subsets of three types of electricity consumption scenarios, a multi-scenario load feature library is constructed.

3. The method for measuring water volume in agricultural wells based on electricity separation and anomaly identification as described in claim 1, characterized in that, The method, based on a multi-scenario load feature library, employs a model-parallel approach to intelligently separate agricultural irrigation power consumption, obtaining time-series data on pure irrigation power consumption. Specifically: Four types of data processing models are deployed in parallel in the central processing terminal, including the NILM non-intrusive load identification model, the TCN-BiLSTM-Attention deep decomposition model, the ANN artificial neural network model, and the SVM support vector machine model. The multi-scenario load feature library is connected with four types of data processing models. In the NILM non-intrusive load identification model, the feature data of the multi-scenario load feature library is used to detect the start and stop events of electrical equipment. The time domain and frequency domain feature vectors are matched to realize the initial split of mixed loads and output coarsely separated agricultural irrigation power consumption data. The coarsely separated agricultural irrigation electricity data includes the electricity consumption of agricultural irrigation, domestic electricity, and other agricultural electricity. By using the TCN temporal convolutional layer in the TCN-BiLSTM-Attention deep decomposition model, local high-frequency feature data of the multi-scenario load feature library is extracted. Then, the BiLSTM bidirectional long short-term memory layer is used to mine the long-term temporal dependence of the irrigation load daily cycle and seasonal cycle of the feature data of the multi-scenario load feature library. Finally, the Attention mechanism is used to add temporal noise and output a standard pure irrigation power time series. Simultaneously, the feature vectors of electrical equipment are input into the ANN artificial neural network model and the SVM support vector machine model to calculate the proportion coefficient of agricultural irrigation electricity consumption. A unified evaluation index is introduced to analyze the proportion coefficient of agricultural irrigation electricity consumption, coarsely separated agricultural irrigation electricity consumption data and standard pure irrigation power time series, and to measure the electricity separation accuracy, scenario generalization ability and computing efficiency of the four types of models. A scoring mechanism is set up to evaluate the accuracy of power separation, the ability to generalize scenarios, and the computational efficiency. The model with the highest score is labeled as the optimal model, and the output data of the optimal model is output and marked as time series data of pure irrigation power consumption.

4. The method for metering water volume in agricultural wells based on electricity separation and anomaly identification as described in claim 1, characterized in that, The process of identifying hydropower time-series anomalies in pure irrigation power consumption data and performing multi-algorithm adaptive interpolation repair based on outliers specifically involves: Within the central processing terminal, a rule library for determining anomalies in hydropower data is introduced. This rule library categorizes abnormal hydropower data into four types: missing data due to equipment disconnection, sudden spikes caused by lightning strikes or equipment malfunctions, systematic offset data caused by meter aging and drift, and seasonal deviation data where the load time series does not match the crop irrigation cycle. The sliding window dynamic threshold judgment rule is built into the hydropower data anomaly judgment rule base. The pure irrigation power consumption time series data is introduced into the hydropower data anomaly judgment rule base. Combined with the sliding window dynamic threshold judgment rule, the pure irrigation power consumption time series data is traversed to judge hydropower data anomalies. Threshold classification was performed on four types of abnormal hydropower data to obtain three abnormality level labels: mild transient noise, moderate continuous missing data, and severe system offset. Abnormality level labels were also applied to the time series data of pure irrigation power consumption. By retrieving time-series data of pure irrigation power consumption under different anomaly level labels from historical data networks, an interpolation and repair algorithm is used to generate a standard pure irrigation power consumption time-series dataset.

5. The method for metering water volume in agricultural wells based on electricity separation and anomaly identification as described in claim 1, characterized in that, Based on the standard pure irrigation power consumption time series dataset, a multi-factor dynamic conversion coefficient model is constructed to dynamically correct the water conversion coefficient, resulting in a dynamic electricity-to-water conversion coefficient. Specifically: In the central processing terminal, a multi-factor dynamic conversion coefficient model is constructed. The construction method of the multi-factor dynamic conversion coefficient model is to acquire all the multi-dimensional operating condition driving factors of the monitoring sample wells with electric water conversion in real time through the central processing terminal. Using the standard pure irrigation power consumption time series dataset as the model training benchmark sample, and combining multi-dimensional operating condition driving factors, a multi-factor weighted regression calculation framework is built to quantify the influence weight of multi-dimensional operating condition driving factors on the water pump hydro-electricity conversion efficiency, and obtain a multi-factor dynamic conversion coefficient model. The multi-factor dynamic conversion coefficient model is used to output the dynamic electricity-to-water conversion coefficient. The update cycle of the multi-factor dynamic conversion coefficient model is preset. When the update cycle of the multi-factor dynamic conversion coefficient model is reached, the multi-dimensional operating condition driving factors are re-acquired, and the influence weight of the multi-dimensional operating condition driving factors on the water pump water-electricity conversion efficiency is updated. The dynamic electricity-to-water conversion coefficient is output in real time.

6. The method for metering water volume in agricultural wells based on electricity separation and anomaly identification as described in claim 1, characterized in that, The method involves calculating the irrigation water volume of monitoring sample wells based on a dynamic electro-water conversion factor, and then verifying the accuracy of the calculation results in multiple dimensions. Specifically: In the central processing terminal, a standard time step is preset. Based on the standard time step, the dynamic electricity-to-water conversion coefficient and the standard pure irrigation power consumption time series dataset are multiplied step by step to output the well groundwater irrigation extraction volume under different time steps, which is calibrated as the initial irrigation water volume calculation result. Reacquire water monitoring sample wells using electrostatic precipitation, mark them as independent monitoring sample wells, and perform three-segment accuracy verification based on the independent monitoring sample wells; The three-stage accuracy verification includes accuracy verification of four types of data processing models, accuracy verification of standard pure irrigation power consumption time series dataset, and accuracy verification of dynamic electricity-to-water conversion coefficient. If all three precision checks meet the preset values, the initial irrigation water volume calculation result will be marked as the target irrigation water volume calculation result. If the accuracy verification does not meet the preset value, the multi-scenario load feature library, hydropower anomaly judgment threshold and multi-factor weighted regression weight are updated in reverse iteration, and the power separation, data repair and dynamic coefficient calculation process is re-executed until the target irrigation water volume measurement result is output.

Images