Intelligent adjustment multi-stream water meter control method based on multi-dimensional data prediction

Abstract

The invention discloses an intelligent adjustment multi-flow water meter control method based on multi-dimensional data prediction, and particularly relates to the technical field of fluid measurement and control. The method comprises the steps that firstly, multi-dimensional data of a multi-flow water meter is collected and subjected to structured storage, a water meter multi-dimensional data matrix is constructed, then the future flow is predicted through a multi-model fusion algorithm, and the multi-flow water meter multi-dimensional data matrix is obtained; outputting reliability indexes of estimated flow and a prediction result, generating an intelligent adjustment control instruction in combination with target flow, optimizing parameters in a multi-model fusion algorithm based on an intelligent adjustment result and quantitative prediction deviation, and triggering a next round of data acquisition according to an optimization result; through a data acquisition technology, multi-dimensional data of time, flow and pressure are combined, a multi-dimensional data matrix is constructed, and through a prediction model fusing trend extrapolation and periodic pattern recognition, preliminary estimation of future flow is realized, and timeliness of intelligent adjustment of a multi-stream water meter is improved.

Description

Technical Field

[0001] This invention relates to the field of fluid measurement and control technology, specifically to a smart adjustment multi-jet water meter control method based on multi-dimensional data prediction. Background Technology

[0002] With the increasing demand for refined water resource management, multi-jet water meters, as a metering device that can simultaneously measure the flow rate of multiple pipe branches or complex flow fields, are widely used in multi-user water use scenarios such as residential communities and industrial plants. They are not only used for accurate metering, but also for dynamic flow control through regulating valves to meet intelligent management needs such as stable water pressure, leakage prevention and control, and on-demand distribution.

[0003] Traditional water meter flow control methods often employ on / off control based on fixed thresholds or simple PID (proportional-integral-derivative) control, neglecting key influencing factors such as pipeline pressure and time cycles (e.g., weekday / weekend water usage patterns). Existing prediction models are mostly static parameter models, such as the ARIMA model with fixed coefficients, which remain unchanged after training and are therefore static parameter models. This characteristic gives them good interpretability in stable scenarios, thus enabling water meter flow prediction and control.

[0004] Existing water meter control methods have significant shortcomings when facing complex and ever-changing actual water usage conditions: they suffer from delayed response, intervening only after flow anomalies occur, failing to anticipate and prevent problems in advance; they lack adaptability, with fixed control parameters unable to accommodate different users and time periods; and they rely solely on flow rate for judgment, ignoring the rich information contained in key related dimensions such as pressure and time, resulting in low control accuracy and a high rate of error. Therefore, there is an urgent need for an intelligent flow control method capable of proactive prediction, adaptive learning, and comprehensive utilization of multi-dimensional information. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a smart adjustment multi-jet water meter control method based on multi-dimensional data prediction, comprising:

[0007] S1: Through data acquisition technology, multidimensional data of multi-jet water meters are collected in real time to obtain a multidimensional dataset for analyzing and controlling water flow.

[0008] S2: Water Meter Multidimensional Data Matrix Construction Module: Based on a preset time window, the collected multidimensional dataset is structured and stored according to the time series to construct a water meter multidimensional data matrix, and the stored data is updated according to the update trigger conditions;

[0009] S3: Based on the current multidimensional data of water meters and the multidimensional data matrix of water meters, predict future flow through a multi-model fusion algorithm, evaluate the prediction results, and output the estimated flow and the reliability index of the prediction results.

[0010] S4: Based on the output estimated flow and the reliability index of the prediction results, combined with the target flow, generate intelligent adjustment and control commands;

[0011] S5: After the preset time window ends, collect the actual traffic data within that period and compare it with the output estimated traffic to quantify the prediction deviation. At the same time, optimize the parameters in the multi-model fusion algorithm based on the prediction deviation, and trigger the next round of data collection based on the optimization results.

[0012] The technical effects and advantages of this invention are as follows:

[0013] 1. This invention combines multi-dimensional data such as time, flow rate, and pressure through data acquisition technology and constructs a multi-dimensional data matrix. By integrating a prediction model that combines "trend extrapolation" and "periodic pattern recognition", it achieves a preliminary estimate of future flow rate, breaking through the limitation of traditional methods that only respond to the current state, and at the same time improving the timeliness of intelligent adjustment of multi-jet water meters.

[0014] 2. By evaluating the prediction results, the generation of control commands not only depends on the predicted values ​​but is also strongly correlated with the reliability index of the prediction. This greatly reduces the risk of misoperation caused by short-term uncertainty of the model, improves the robustness of the system, and solves the problem of response lag in fixed parameter control when the prediction reliability is low.

[0015] 3. This invention optimizes the intelligent adjustment results and uses the real-time generated prediction error to dynamically update the key parameters inside the multi-model fusion algorithm model, enabling the prediction model to continuously evolve and adapt to the changes in water usage characteristics of different users and different seasons, fundamentally solving the problem of poor adaptability of traditional fixed parameter controllers. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0017] Figure 2 This is a schematic diagram of the method flow of the present invention.

[0018] Figure 3 This is a schematic diagram of the multi-model fusion algorithm of the present invention.

[0019] Figure 4 This is a schematic diagram of the optimization process of the multi-model fusion algorithm of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 As shown, the present invention provides an intelligent regulation multi-jets water meter control system based on multi-dimensional data prediction, including a water meter multi-dimensional data acquisition module, a water meter multi-dimensional data matrix construction module, a water meter flow prediction and evaluation module, an intelligent regulation control command generation module, and an intelligent regulation result optimization module.

[0022] The water meter multidimensional data acquisition module is connected to the water meter multidimensional data matrix construction module, the water meter flow prediction and evaluation module is connected to the water meter multidimensional data matrix construction module and the intelligent regulation control command generation module, respectively, and the intelligent regulation result optimization module is connected to the intelligent regulation control command generation module and the water meter multidimensional data acquisition module, respectively.

[0023] Water meter multidimensional data acquisition module: Through data acquisition technology, it collects multidimensional data of multi-jet water meters in real time, obtains multidimensional datasets for analyzing and controlling water flow, and transmits them to the water meter multidimensional data matrix construction module;

[0024] Water meter multidimensional data matrix construction module: Based on a preset time window, the multidimensional dataset is stored in a structured manner according to the time series, a water meter multidimensional data matrix is ​​constructed, the stored data is updated according to the update trigger conditions, and the water meter multidimensional data matrix is ​​transmitted to the water meter flow prediction and evaluation module.

[0025] Water meter flow prediction and evaluation module: Based on the current water meter multidimensional data and water meter multidimensional data matrix, it predicts the future flow through a multi-model fusion algorithm, evaluates the prediction results, and outputs the estimated flow and the reliability index of the prediction results to the intelligent regulation and control command generation module.

[0026] Intelligent regulation and control command generation module: Based on the output estimated flow and the reliability index of the prediction results, combined with the target flow, it generates intelligent regulation and control commands and transmits the intelligently regulated flow to the intelligent regulation result optimization module;

[0027] Intelligent adjustment result optimization module: Based on the intelligently adjusted flow rate, it compares it with the output estimated flow rate to quantify the prediction deviation. At the same time, it optimizes the parameters in the multi-model fusion algorithm based on the prediction deviation and triggers the next round of data collection based on the optimization results.

[0028] Please see Figure 2As shown, the intelligent regulation multi-jets water meter control method based on multi-dimensional data prediction includes: S1: Real-time acquisition of multi-dimensional data from multi-jets water meters using data acquisition technology to obtain a multi-dimensional dataset for analyzing and controlling water flow; S2: Water meter multi-dimensional data matrix construction module: Based on a preset time window, the acquired multi-dimensional dataset is structured and stored according to time series to construct a water meter multi-dimensional data matrix, and the stored data is updated according to update trigger conditions; S3: Based on the current water meter multi-dimensional data and the water meter multi-dimensional data matrix, a multi-model fusion algorithm is used to predict future flow, and the prediction results are evaluated to output the estimated flow and the reliability index of the prediction results; S4: Based on the output estimated flow and the reliability index of the prediction results, combined with the target flow, an intelligent regulation control command is generated; S5: After the preset time window ends, the actual flow data within that period is collected and compared with the output estimated flow to quantify the prediction deviation. Simultaneously, the parameters in the multi-model fusion algorithm are optimized based on the prediction deviation, and the next round of data acquisition is triggered based on the optimization results.

[0029] S1: Through data acquisition technology, multidimensional data of the multi-jet water meter is collected in real time to obtain a multidimensional dataset for analyzing and controlling water flow. The data acquisition technology includes flow sensor, pressure sensor and clock sensor, and collects flow rate Q(t), pipeline pressure P(t) and time t at a fixed sampling period Δt (e.g. 1s); to obtain the multidimensional dataset D(t) for analyzing and controlling water flow, D(t)=[Q(t),P(t),t];

[0030] In this embodiment, it should be specifically noted that the multi-jet water meter has built-in data acquisition technology including sensors such as the flow sensor for collecting flow rate, the pressure sensor for collecting pipeline pressure, and the clock sensor for identifying water usage periods (e.g., day / night, weekday / weekend).

[0031] S2: Water Meter Multidimensional Data Matrix Construction Module: Based on a preset time window, the collected multidimensional dataset is structured and stored according to time series to construct a water meter multidimensional data matrix, and the stored data is updated according to update trigger conditions, including the following steps:

[0032] S2.1: Based on a preset time window T, the flow rate and pipeline pressure in the multidimensional dataset are first preprocessed. The preprocessing includes outlier handling using the 3σ principle, missing value handling using interpolation, and normalization using Min-Max. Then, the flow rate and pipeline pressure in the preprocessed multidimensional dataset are stored in a structured manner according to the time series, and a water meter multidimensional data matrix M is constructed, with a dimension of (T / Δt)×3, where 3 represents the three components of time, flow rate, and pipeline pressure, and (T / Δt) represents the number of sampling points within the preset time window.

[0033] In this embodiment, it is necessary to specifically explain that the preset time window is used for modeling and analyzing the number of historical data points. Its value is determined according to the system memory and prediction requirements. For example, for data from the past 5 minutes, if Δt = 1s, then T = 300. The 3σ principle, interpolation method, and Min-Max normalization method in the preprocessing are all existing technologies and will not be described in detail here. The maximum and minimum values ​​in Min-Max normalization are based on the maximum and minimum values ​​of all sampling points. The interpolation method can be linear interpolation, mean interpolation, etc. Time data generally does not need to be standardized because it exists as a sequence feature.

[0034] S2.2: When a new set of data is collected, the historical set of data outside the preset time window is removed in a first-in-first-out (FIFO) manner, and the water meter multidimensional data matrix M is updated to obtain the updated water meter multidimensional data matrix M. g ;

[0035] Please see Figure 3 As shown, S3: Based on the current multidimensional data of water meters and the multidimensional data matrix of water meters, the future flow rate is predicted through a multi-model fusion algorithm. Simultaneously, the prediction results are evaluated, and the estimated flow rate and reliability indicators of the prediction results are output. This includes the following steps:

[0036] S3.1: The multi-model fusion algorithm includes a trend extrapolation sub-model and a periodic pattern recognition sub-model;

[0037] S3.2: Trend Extrapolation Sub-model: Based on the preprocessed flow rate Q in the current water meter multidimensional data. y (t), pipeline pressure P y (t) and the previous flow rate Q in the water meter multidimensional data matrix. y (t-1) and pipeline pressure P y (t-1), to obtain the trend prediction value Q of the water meter flow rate at the k-th sampling time. tr (t+k), Q tr (t+k)=a1×Q y (t)+a2×[Q y (t)-Q y (t-1)]×{1+a3×[P y (t)-P y (t-1)]}, a1 represents the weighting coefficient of the current flow, with a value range of (0,1), reflecting the contribution of the current flow to the trend; a2 represents the weighting coefficient of the flow difference, with a value range of (-1,1), reflecting the continuity of the flow change trend; a3 represents the correction coefficient of the pressure difference, with a value range of (-1,1), reflecting the impact of pressure fluctuations on the flow trend.

[0038] This embodiment specifically explains that the trend extrapolation sub-model in the multi-model fusion algorithm borrows from the use of trend and differencing in classic time series forecasting models (such as ARIMA). The core principle of the traditional ARIMA model is: ARIMA (Autoregressive Integral Moving Average) is a classic time series forecasting model, consisting of three parts: autoregression (AR), differencing (I), and moving average (MA). Autoregression (AR): the current value is linearly correlated with the historical value; differencing (I): the non-stationary series is differised d times to transform it into a stationary series, eliminating the non-stationarity caused by trends or periodicity; moving average (MA): the current value is linearly correlated with the historical error, capturing the random fluctuations of the data. This invention constructs a lightweight linear model, improves the traditional ARIMA by introducing a stress factor, and reduces computational overhead by optimizing parameters through a1, a2, and a3.

[0039] S3.3: Periodic Pattern Recognition Sub-model:

[0040] S3.3.1: Based on the current water meter multidimensional data D d The system extracts the hour feature h(t) (e.g., 0-23) and the rest day feature w(t) from the time dimension, where 0 represents a weekday and 1 represents a rest day. It also extracts the average pipeline pressure μ(P(t)) from the pipeline pressure dimension of the water meter multidimensional data, constructing a water meter multidimensional feature vector F(t), where F(t) = [h(t), w(t), μ(P(t)), Q(t)], and Q(t) represents the current flow rate. Then, it calculates the relationship between the water meter multidimensional feature vector F(t) and the historical time t in the water meter multidimensional data matrix. i eigenvector F(t) i The weighted distance d(t,t) i Select the n historical moments (t1, t2, ..., tn) with the smallest distance. n (e.g., n=4) ω j This represents the weight of the j-th feature of the water meter, and 4 represents the four multi-dimensional features of the water meter, such as h(t) weight ω1=0.4, w(t) weight ω2=0.2, μ(P(t)) weight ω3=0.2 and Q(t) weight ω4=0.2;

[0041] S3.3.2: For the traffic at the k-th sampling time (e.g., 10s) following the n historical moments with the smallest distance, introduce a time attention weight att(t,t) i Calculate the weighted average value as the periodic prediction value Q of the water meter flow rate. pa (t+k), , , t de The decay period is represented by Q(t) (e.g., 24 hours, to give higher weight to recent similar moments).i +k) represents t i The historical flow rate at subsequent k sampling times is shown. For example, at t1+10s, the historical flow record shows Q(t1+10s)=1.5m. 3 / h, ω i ω represents the weighted distance weight of the water meter. i =1 / d(t,t i );

[0042] In this embodiment, it should be specifically noted that rest days are based on statutory holidays, such as weekends and public holidays; the time period is characterized by the number of hours, and the work and rest characteristics are characterized by whether it is a rest day.

[0043] S3.4: The trend prediction value Q obtained from the trend extrapolation sub-model tr (t+k) and the periodic prediction value Q obtained from the periodic pattern recognition sub-model pa (t+k) is dynamically weighted and fused to obtain the estimated traffic Q. pr (t+k), the weight b is adaptively adjusted according to the error err of the two sub-models, Q pr (t+k)=b×Q tr (t+k)+(1-b)×Q pa (t+k), b=err pa / (err tr +err pa ), err tr and err pa These represent the average error of the trend extrapolation sub-model and the average error of the periodic pattern recognition sub-model, respectively. The average error of the trend extrapolation sub-model is calculated by taking n historical samples and considering historical time t. i The average of the absolute differences between the predicted traffic flow and the actual traffic flow at the corresponding time is obtained. Similarly, the average error of the periodic pattern recognition sub-model for n historical samples is obtained.

[0044] This embodiment specifically illustrates that by complementing short-term trends and cyclical patterns, using error-driven dynamic weight allocation, and deeply synergizing with prediction and evaluation reliability indicators, it achieves high-precision prediction of multi-jet water meters in complex water use scenarios, providing reliable support for intelligent control. It not only solves the scenario limitations of a single model, but also reduces the cost of manual intervention through an adaptive mechanism, ultimately achieving the technical effects of accurate prediction, stable adjustment, and strong adaptability, which is significantly better than traditional single-model or fixed-weight fusion schemes.

[0045] S3.5: Based on the most recent n estimated traffic Q pr The average error of (t+k) historical samples err pr The sum of flow rate and pipeline pressure fluctuation coefficient within a preset time window T, dsva The prediction results of the two sub-models, Q tr (t+k) and Q pa The deviation rate mc of (t+k) is used to obtain the reliability index C(t+k) of the prediction result, C(t+k) = c1 × [1 / 1 + (err)]. pr / Q L )]+c2×(1 / 1+ds va )+c3×(1 / 1+mc), Q L For flow range, k represents the prediction step size, and e represents a constant (very small, e.g., 10). -6 To avoid a denominator of 0, c1+c2+c3=1; the flow rate and pipeline pressure fluctuation coefficients are obtained by the ratio of the standard deviation to the average value of the flow rate and pipeline pressure, respectively.

[0046] In this embodiment, it should be specifically explained that the mean variance of flow rate and pipeline pressure refers to the average of the variances calculated for flow rate and pressure within a preset time window T.

[0047] S4: Based on the output estimated flow rate and the reliability index of the prediction results, combined with the target flow rate, generate intelligent adjustment and control commands, based on the output estimated flow rate Q. pr (t+k), and combine it with the target flow Q. ref By subtracting the values, we obtain the flow deviation ΔQ, where ΔQ = Q. pr (t+k)-Q ref After inversely normalizing the flow deviation ΔQ, a smart adjustment control command u(t) is generated using a PID control algorithm to achieve pre-adjustment of the flow. ΔQ f K represents the flow deviation after denormalization. p K i and K d K represents the proportional coefficient, integral coefficient, and differential coefficient, respectively. p =K p0 ×(1+k p ×(1-C(t+k))), K p0 and k p These represent the base proportional gain and the proportional adjustment gain (e.g., K). p0 =5, the preset baseline value, and k p =0.5 (used to control the proportional adjustment range), K i =K i0 ×(1+k i ×(1-C(t+k))), K i0 and k i These represent the base integral coefficient and the integral adjustment gain (e.g., K). i0=0.1, the preset baseline value, and k i =0.3 (used to control the integral adjustment amplitude), K d =K d0 ×(1-k d ×(1-C(t+k))), K d0 and k d These represent the fundamental differential coefficients and the differential adjustment gain (e.g., K). d0 =2, the preset baseline value, and k d =0.4, used to control the differential adjustment amplitude); the flow deviation ΔQ is obtained by inverse normalization. f ΔQ f =ΔQ×(Q max -Q min ), Q max and Q min These are the maximum and minimum values ​​of the flow rate based on all sampling points, respectively.

[0048] In this embodiment, it should be specifically noted that the target flow rate can be preset by the system administrator based on actual water demand and dynamically distributed according to the regional water balance strategy.

[0049] In this embodiment, it is necessary to specifically explain that when the reliability index C(t+k) is low (e.g., less than 0.6), the adjustment coefficient is increased to improve the response speed. When the reliability index C(t+k) is high (e.g., greater than or equal to 0.6), the adjustment coefficient is decreased to avoid overshoot. The derivative coefficient is positively correlated with the reliability index C(t+k). When the reliability index is low, the derivative coefficient is decreased to reduce the sensitivity to noise. When the reliability index is high, the derivative coefficient is increased to enhance the trend prediction capability.

[0050] This embodiment specifically explains how the improved ARIMA model, playing a role in short-term flow trend prediction, is integrated with the periodic pattern recognition model to provide a basis for PID control: real-time flow, pipeline pressure, and time are collected from multi-jet water meters to form a multi-dimensional data matrix of the water meters; the improved ARIMA model, combined with the current flow, flow difference, and pressure difference, outputs the flow trend prediction value for the k-th step in the future; the trend prediction result is dynamically weighted and fused with the periodic pattern recognition result to obtain the final flow prediction value; finally, based on the prediction value and confidence level, the PID parameters are dynamically adjusted to generate intelligent control commands, achieving precise pre-regulation of the flow.

[0051] Please see Figure 4As shown, S5: After the preset time window ends, collect the actual traffic data within that period and compare it with the output estimated traffic to quantify the prediction deviation. At the same time, optimize the parameters in the multi-model fusion algorithm based on the prediction deviation, and trigger the next round of data collection based on the optimization results. This includes the following steps:

[0052] S5.1: After the preset time window ends, collect the actual flow data Q for k uncontrolled prediction steps within that period. act (t+iΔt), and the estimated flow rate Q for the output k prediction steps. pr Compare (t+iΔt) and calculate the prediction bias ΔQ. u(t) , k represents the prediction step size, and Δt represents the sampling period;

[0053] In this embodiment, it should be specifically noted that the actual flow data for the uncontrolled k prediction steps involves periodically pausing the control action, with S4 not outputting any commands, the valve maintaining a fixed opening, and the natural flow rate Q being directly collected. act (t+iΔt) is used as the calibration benchmark and compared with the estimated flow rate.

[0054] S5.2: Based on prediction bias ΔQ u(t) The gradient descent algorithm is used to optimize the parameters θ in the multi-model fusion algorithm. k The optimization process involves parameters including the weighting coefficient a1 for the current flow, the weighting coefficient a2 for the flow difference, and the correction coefficient a3 for the pressure difference, resulting in the optimized parameter θ. k+1 θ k+1 =θ k -η×∇ θ L(ΔQ u(t) ), where η is the learning rate (which can be set according to operating conditions, with a typical value of 0.001~0.01), ∇ θ L(ΔQ u(t) ) represents the loss function L(ΔQ) obtained through the gradient descent algorithm. u(t) The gradient of L(ΔQ) with respect to parameter θ u(t) )=(ΔQ u(t) ) 2 If the prediction deviation ΔQ is consecutively m times (e.g., 3 to 5 times), u(t) When ≤ the corresponding threshold (e.g., 0.01m) 3 The output ( / h) indicates that the model accuracy has met the actual control requirements, and further optimization is unnecessary; the process can be terminated directly by outputting the optimized parameter θ. k+1 If the number of optimization iterations reaches the preset maximum (e.g., 500 times), but the prediction deviation still does not meet the condition, the optimization is forcibly terminated, and the prediction deviation ΔQ during the iteration process is output. u(t) The minimum corresponding parameter is used as the optimized parameter θ. k+1Finally, the next round of data collection is triggered based on the optimization results.

[0055] This embodiment specifically explains that by driving parameter iteration through prediction deviation feedback, the model can autonomously converge to the optimal parameter combination in different scenarios. Compared with fixed parameter models, prediction deviation provides high-precision input for PID control. At the same time, parameter optimization achieves dynamic improvement of model performance through iteration driven by prediction deviation, thus constructing a closed-loop intelligent mechanism of perception-prediction-optimization-control. This enables the multi-jet water meter control system to have self-learning, self-adaptation, and anti-interference capabilities, maintaining high-precision prediction and stable control in complex water use scenarios.

[0056] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0057] In conclusion, the above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be considered as such.

[0058] It is included within the scope of protection of this invention.

Claims

1. A method for intelligent regulation of multi-jet water meters based on multi-dimensional data prediction, characterized in that: include: S1: Through data acquisition technology, multidimensional data of multi-jet water meters are collected in real time to obtain a multidimensional dataset for analyzing and controlling water flow. S2: Water Meter Multidimensional Data Matrix Construction Module: Based on a preset time window, the collected multidimensional dataset is structured and stored according to the time series to construct a water meter multidimensional data matrix, and the stored data is updated according to the update trigger conditions; S3: Based on the current multidimensional data of water meters and the multidimensional data matrix of water meters, predict future flow through a multi-model fusion algorithm, evaluate the prediction results, and output the estimated flow and the reliability index of the prediction results. S4: Based on the output estimated flow and the reliability index of the prediction results, combined with the target flow, generate intelligent adjustment and control commands; S5: After the preset time window ends, collect the actual traffic data within that period and compare it with the output estimated traffic to quantify the prediction deviation. At the same time, optimize the parameters in the multi-model fusion algorithm based on the prediction deviation, and trigger the next round of data collection based on the optimization results.

2. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The data acquisition technology in S1 includes a flow sensor, a pressure sensor, and a clock sensor, which collects the flow rate Q(t), pipeline pressure P(t), and time t at a fixed sampling period Δt; and obtains a multidimensional dataset D(t) for analyzing and controlling the water flow, where D(t) = [Q(t), P(t), t].

3. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The multi-model fusion algorithm in S3 includes a trend extrapolation sub-model and a periodic pattern recognition sub-model.

4. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 3, characterized in that: The trend extrapolation sub-model in S3 is based on the preprocessed flow rate Q in the current multidimensional water meter data. y (t), pipeline pressure P y (t) and the previous flow rate Q in the water meter multidimensional data matrix. y (t-1) and pipeline pressure P y (t-1), to obtain the trend prediction value Q of the water meter flow rate at the k-th sampling time. tr (t+k), Q tr (t+k)=a1×Q y (t)+a2×[Q y (t)-Q y (t-1)]×{1+a3×[P y (t)-P y (t-1)]}, a1 represents the weighting coefficient of the current flow rate, with a value range of (0,1), a2 represents the weighting coefficient of the flow rate difference, with a value range of (-1,1), and a3 represents the correction coefficient of the pressure difference, with a value range of (-1,1).

5. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 3, characterized in that: The periodic pattern recognition sub-model in S3 includes: S3.3.1: Based on the current water meter multidimensional data D d The system extracts the hourly feature h(t) and the day / rest feature w(t) from the time dimension, where 0 represents a weekday and 1 represents a restday. It then extracts the average pipeline pressure μ(P(t)) from the pipeline pressure dimension of the water meter multidimensional data, constructing a multidimensional feature vector F(t) for the water meter: F(t) = [h(t), w(t), μ(P(t)), Q(t)], where Q(t) represents the current flow rate. Finally, it calculates the relationship between the multidimensional feature vector F(t) and the historical time t in the water meter multidimensional data matrix. i eigenvector F(t) i The weighted distance d(t,t) i Select the n historical moments (t1, t2, ..., tn) with the smallest distance. n ); S3.3.2: For the traffic flow at the k-th sampling time following the n historical times with the smallest distance, introduce a time attention weight att(t,t) i Calculate the weighted average as the periodic forecast value Q. pa (t+k), , , t de Q(t) represents the decay period. i +k) represents t i The flow rate at subsequent k sampling times in the historical time, ω i ω represents the weighted distance weight of the water meter. i =1 / d(t,t i ).

6. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The estimated flow in S3 is the trend prediction value Q obtained from the trend extrapolation submodel. tr (t+k) and the periodic prediction value Q obtained from the periodic pattern recognition sub-model pa (t+k) is dynamically weighted and fused to obtain the estimated traffic Q. pr (t+k), the weight b is adaptively adjusted according to the error err of the two sub-models, Q pr (t+k)=b×Q tr (t+k)+(1-b)×Q pa (t+k), b=err pa / (err tr +err pa ), err tr and err pa These represent the average error of the trend extrapolation sub-model and the average error of the periodic pattern recognition sub-model, respectively. The average error of the trend extrapolation sub-model is calculated by taking n historical samples at historical time t. i The average of the absolute difference between the predicted traffic flow and the actual traffic flow at the corresponding time is obtained. Similarly, the average error of the periodic pattern recognition sub-model for n historical samples is obtained.

7. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The reliability index of the prediction results in S3 is based on the most recent n estimated flows Q. pr The average error of (t+k) historical samples err pr The sum of flow rate and pipeline pressure fluctuation coefficient within a preset time window T, ds va The prediction results of the two sub-models, Q tr (t+k) and Q pa The deviation rate mc of (t+k) is used to obtain the reliability index C(t+k) of the prediction result, C(t+k) = c1 × (1 / 1 + err) pr )+c2×(1 / 1+ds va )+c3×(1 / 1+mc).

8. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The intelligent adjustment and control command in S4 is based on the estimated output flow rate Q. pr (t+k), and combine it with the target flow Q. ref By subtracting the values, we obtain the flow deviation ΔQ, where ΔQ = Q. pr (t+k)-Q ref After inversely normalizing the flow deviation ΔQ, the intelligent adjustment control command u(t) is generated through the PID control algorithm.

9. The intelligent adjustment multi-jet water meter control method based on multi-dimensional data prediction according to claim 1, characterized in that: The S5 implementation includes: S5.1: After the preset time window ends, collect the actual flow data Q for k uncontrolled prediction steps within that period. act (t+iΔt), and the estimated flow rate Q for the output k prediction steps. pr Compare (t+iΔt) and calculate the prediction bias ΔQ. u(t) , k represents the prediction step size, and Δt represents the sampling period; S5.2: Based on prediction bias ΔQ u(t) The gradient descent algorithm is used to optimize the parameters θ in the multi-model fusion algorithm. k The optimization process involves parameters including the weighting coefficient a1 for the current flow, the weighting coefficient a2 for the flow difference, and the correction coefficient a3 for the pressure difference, resulting in the optimized parameter θ. k+1 θ k+1 =θ k -η×∇ θ L(ΔQ u(t) ), where η is the learning rate, ∇ θ L(ΔQ u(t) ) represents the loss function L(ΔQ) obtained through the gradient descent algorithm. u(t) The gradient of L(ΔQ) with respect to parameter θ u(t) )=(ΔQ u(t) ) 2 If the prediction deviation ΔQ is m consecutive times u(t) When the value is less than or equal to the corresponding threshold, it indicates that the model accuracy has met the actual control requirements, and further optimization is unnecessary; the optimized parameters θ can be output directly and the process terminated. k+1 If the number of optimization iterations reaches the preset maximum, but the prediction deviation still does not meet the condition, the optimization is forcibly terminated, and the prediction deviation ΔQ during the iteration process is output. u(t) The minimum corresponding parameter is used as the optimized parameter θ. k+1 Finally, the next round of data collection is triggered based on the optimization results.

Images