Water flow management optimization method and system based on machine learning

Abstract

The invention discloses a water flow management optimization method and system based on machine learning. The method comprises the steps of data management, water flow characteristic optimization, frequency shift flow direction modeling and water flow management optimization. The invention relates to the technical field of water flow metering management, adopts an integral intelligent algorithm flow combining feature optimization, flow direction modeling and optimization management, obtains qualitative and quantitative information of water flow through feature optimization and flow direction modeling, and comprehensively improves the quality and intelligence of water flow metering and management according to the intelligent process of optimization management. The overall usability is also improved; performing water flow characteristic optimization by adopting a multi-convolution integrated and improved one-dimensional convolutional neural network; carrying out frequency shift flow direction modeling by adopting an improved time sequence prediction network combining Doppler effect analysis and frequency domain analysis; and carrying out water flow comprehensive management optimization by adopting a reinforcement learning dynamic water flow distribution adjustment method combining physical constraints and a hydraulic equation.

Description

Technical Field

[0001] The present invention relates to the technical field of water flow measurement management, and in particular to a water flow management optimization method and system based on machine learning. Background Art

[0002] The water flow management optimization method and system based on machine learning aims to optimize the distribution and regulation of water resources through intelligent algorithms and models. This method uses classical algorithms or machine learning technology to automatically analyze and predict the changing trend of water flow by real-time monitoring of data such as water flow, flow velocity and flow direction, and optimizes scheduling according to demand. Its main function is to improve the efficiency of water resource utilization, reduce waste, avoid resource shortages or surpluses caused by uneven water flow, and realize dynamic regulation to ensure that water flow distribution can quickly adapt to changes under different environmental conditions and ensure the continuous and stable supply of water resources. This intelligent optimization method can be widely used in agricultural irrigation, urban water supply, river management and other fields to help achieve more scientific and efficient water flow management.

[0003] However, in the existing water flow measurement and management process, there are existing water flow measurement methods. Due to the complex flow patterns in open channels and rivers and signal interference, it is difficult to measure the flow rate stably and accurately. The conventional flow meter method has technical problems in flow measurement, such as large data fluctuations and unstable signal quality;

[0004] Among the existing intelligent extraction methods of water flow characteristics, there are traditional classical algorithms, namely digital filtering algorithms and curve fitting algorithms. In actual measurements, since the channel may be narrow, the electromagnetic waves emitted by the radar will hit the shore. At the same time, there may be a large number of floating objects on the water surface, causing the signal amplitude to change continuously. The water flow may be turbulent, and the flow state is not good. These factors will lead to the identification of incorrect Doppler frequency shifts, and thus the surface flow velocity cannot be measured correctly. The most intuitive manifestation is that the measured flow velocity data will fluctuate over a large range. Due to the complexity and diversity of water flow measurement and management sites, there will be technical problems where filtering cannot be completed.

[0005] In the existing water flow frequency shift and flow direction modeling and prediction methods, there are technical problems such as large noise interference in water flow frequency shift signals, complex flow direction changes, and difficulty in fully capturing temporal and spatial characteristics; in the existing intelligent water flow management process, there are technical problems such as lack of physical constraints in water flow distribution schemes, difficulty in achieving precise adjustment of flow distribution, and poor dynamic adaptability. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for optimizing water flow management based on machine learning. In the existing water flow measurement and management process, the existing water flow measurement method has the problem that it is difficult to stably and accurately measure the flow rate due to the complex flow state and signal interference in open channels and rivers, and the conventional flow meter method has technical problems such as large data fluctuations and unstable signal quality in flow measurement. This solution creatively adopts an overall intelligent algorithm process combining feature optimization, flow direction modeling and optimization management, obtains qualitative and quantitative information of water flow through feature optimization and flow direction modeling, and comprehensively uses the intelligent process of optimization management to , which improves the quality and intelligence of water flow measurement and management, and also improves the overall usability; in the existing intelligent extraction methods of water flow characteristics, there are traditional classic algorithms, namely digital filtering algorithms and curve fitting algorithms. In actual measurements, since the channel may be narrow, the electromagnetic waves emitted by the radar will hit the shore. At the same time, there may be a large number of floating objects on the water surface, causing the signal amplitude to change continuously. The water flow may be turbulent, and the flow state is not good. These factors will lead to the identification of incorrect Doppler frequency shifts, and thus the surface flow velocity cannot be measured correctly. The most intuitive manifestation is that the measured flow velocity data will fluctuate over a large range. Due to the complexity and diversity of the water flow measurement and management sites, there will be filtering that cannot be completed. In order to solve the technical problem of the situation, this solution creatively adopts a one-dimensional convolutional neural network improved by multi-convolution integration to optimize the water flow characteristics. By adaptively improving and designing the structure of the one-dimensional convolutional network, the intelligence and efficiency of water flow feature extraction are improved, and the problem that the single Doppler frequency shift method is prone to recognition errors is solved. At the same time, the overall stability and reliability of the method are improved; in view of the technical problems that the water flow frequency shift signal noise interference is large, the flow direction changes are complex, and the spatiotemporal characteristics are difficult to fully capture in the existing water flow frequency shift flow direction modeling and prediction methods, this solution creatively adopts an improved time series prediction network that combines Doppler effect analysis and frequency domain analysis. By combining two classic modeling and analysis methods with spatiotemporal characteristics for combined prediction, the modeling and prediction effects of water flow frequency shift and flow direction are improved overall. In view of the technical problems in the existing intelligent water flow management process, such as the lack of physical constraints in the water flow distribution scheme, the difficulty in achieving precise adjustment of flow distribution and poor dynamic adaptability, this scheme creatively adopts a reinforcement learning dynamic water flow distribution adjustment method that combines physical constraints and hydraulic equations to optimize the comprehensive management of water flow. By combining the physical constraint model, the physical characteristics of the water flow management results are optimized. At the same time, through the improvement of the reward function and the integration of frequency shift flow direction information, the flexibility and accuracy of water flow management are improved as a whole.

[0007] The technical solution adopted by the present invention is as follows: The present invention provides a water flow management optimization method based on machine learning, the method comprising the following steps:

[0008] Step S1: data management;

[0009] Step S2: water flow characteristic optimization;

[0010] Step S3: frequency shift flow direction modeling;

[0011] Step S4: Water flow management optimization.

[0012] Further, in step S1, the data management is used to collect, store and preprocess the original data set required for water flow management, specifically to obtain water flow management original data through multi-source sensor data collection, and to obtain optimized water flow management data through data preprocessing;

[0013] The water flow management raw data includes water flow data, water level data, environmental support data, signal evaluation data, water flow pattern data and time-space data;

[0014] The data preprocessing includes data denoising, exception processing, data standardization and data smoothing operations;

[0015] The optimized water flow management data includes optimized water flow data, optimized water level data, optimized environmental support data, optimized signal evaluation data, optimized water flow state data and optimized time-space data.

[0016] Further, in step S2, the water flow characteristic optimization is used to extract and optimize the spatiotemporal characteristics of the water flow, specifically, based on the optimized water flow management data, a one-dimensional convolutional neural network improved by multi-convolution integration is used to optimize the water flow characteristics to obtain the spatiotemporal characteristic data of the water flow, including the following steps:

[0017] Step S21: constructing a water flow input layer, specifically, taking the optimized water flow management data as input data samples to construct the water flow input layer;

[0018] Step S22: constructing a one-dimensional convolution layer improved by multi-convolution integration, specifically, sequentially integrating the depthwise separable convolution layer, the hybrid dilated convolution layer, and the adaptive frequency domain pooling layer to obtain a one-dimensional convolution layer improved by multi-convolution integration;

[0019] The depthwise separable convolution layer specifically uses a depthwise separable convolution with a dilation factor of 2 and the number of channels is set to 64;

[0020] The hybrid dilated convolution layer specifically constructs a dilated convolution module with dilation factors alternating between [1, 3, 5], and introduces channel attention after each dilated convolution module to construct a hybrid dilated convolution layer;

[0021] The adaptive frequency domain pooling layer specifically adopts the adaptive frequency domain pooling layer to replace the fully connected layer in the standard one-dimensional convolution module, so as to retain the time series frequency information;

[0022] Step S23: constructing a branch output layer, specifically constructing a regression branch and a classification branch to predict the Doppler shift value and flow direction category;

[0023] The regression branch is used to output the Doppler frequency shift value and represent the time characteristics of the water flow;

[0024] The classification branch is used to output flow direction categories, including forward, reverse, interference and turbulent, and to represent the spatial characteristics of the water flow;

[0025] Step S24: constructing an optimized loss function, specifically constructing a mean square error loss and a cross entropy loss, performing weighted summation, and constructing the optimized loss function;

[0026] Step S25: water flow characteristic optimization model training, specifically, through the water flow input layer, the multi-convolution integrated improved one-dimensional convolution layer, the branch output layer and the optimized loss function, the water flow characteristic optimization model training is performed to obtain the water flow characteristic optimization model Model FLNX ;

[0027] Step S26: Optimizing water flow characteristics, specifically, using the water flow characteristic optimization model Model according to the optimized water flow management data FLNX , optimize the water flow characteristics and obtain the spatiotemporal characteristic data of water flow.

[0028] Further, in step S3, the frequency shift flow direction modeling is used to establish a flow direction prediction model based on water flow frequency shift to identify the change trend and flow direction of the water flow. Specifically, based on the spatiotemporal characteristic data of the water flow, an improved time series prediction network combining Doppler effect analysis and frequency domain analysis is used to perform frequency domain flow direction modeling to obtain frequency shift flow direction change prediction data, including the following steps:

[0029] Step S31: constructing a water flow spatiotemporal characteristic input layer, specifically, inputting the water flow spatiotemporal characteristic data as raw data for frequency shift flow direction modeling to construct the water flow spatiotemporal characteristic input layer;

[0030] Step S32: Doppler effect analysis, specifically, performing frequency shift value spectrum analysis modeling on the original data input for frequency shift flow direction modeling according to the Doppler effect to obtain Doppler frequency shift analysis data;

[0031] Step S33: frequency domain feature analysis, specifically, extracting and analyzing the frequency domain features of the original data input for frequency shift flow direction modeling through fast Fourier transform to obtain frequency domain feature data;

[0032] Step S34: constructing an improved time series prediction network, specifically, according to the Doppler frequency shift analysis data and the frequency domain feature data, through frequency domain attention optimization, multi-scale time series modeling, frequency domain long short-term memory network construction and modeling layer construction, to obtain the improved time series prediction network;

[0033] Step S35: frequency shift flow direction modeling, specifically, through the water flow spatiotemporal characteristic input layer, the Doppler effect analysis, the frequency domain characteristic analysis and the improved time series prediction network, model training and data prediction of frequency shift flow direction modeling are performed to obtain frequency shift flow direction change prediction data.

[0034] Further, in step S4, the water flow management optimization is used to comprehensively manage the distribution and regulation of water flow, specifically combining the water flow spatiotemporal characteristic data and the frequency shift flow direction change prediction data, using a reinforcement learning dynamic water flow distribution adjustment method combining physical constraints and hydraulic equations, to perform comprehensive water flow management optimization, and obtain water flow management distribution reference data, including the following steps:

[0035] Step S41: Modeling the water flow management problem, which is used to define the management target of the water flow, specifically, minimizing flow fluctuation and optimizing water flow distribution as the basic target model of the water flow management problem, and obtaining a water flow management target model;

[0036] Step S42: physical constraint modeling, specifically constructing mass conservation equations and hydraulic equations as physical constraint models of the model;

[0037] Step S43: dynamic water flow distribution reinforcement learning, specifically, constructing a standard reinforcement learning parameter group according to the water flow management target model, and improving the design of the reward function to obtain a dynamic water flow distribution reinforcement learning architecture;

[0038] The improved design of the reward function specifically refers to weighting according to the stability of water flow regulation and the stability of flow frequency shift to obtain a dynamic water flow distribution reward function, and performing reinforcement learning training according to the dynamic water flow distribution reward function;

[0039] Step S44: iterative training of water flow management optimization, specifically using a deep Q-value learning method to perform iterative optimization of water flow distribution and Q-value update to obtain an optimal model for water flow management optimization;

[0040] Step S45: water flow management, specifically, performing water flow management according to the water flow management optimization optimal model to obtain water flow management allocation reference data;

[0041] The water flow management allocation reference data specifically includes regional water flow allocation recommendations, flow adjustment strategies and comprehensive flow prediction reference data.

[0042] The present invention provides a water flow management optimization system based on machine learning, comprising a data management module, a feature processing module, a prediction optimization module and an output management module;

[0043] The data management module is used for data management, and obtains optimized water flow management data through data management, and sends the optimized water flow management data to the feature processing module;

[0044] The feature processing module is used for optimizing water flow characteristics, obtaining water flow spatiotemporal feature data through water flow characteristic optimization, and sending the water flow spatiotemporal feature data to the prediction optimization module;

[0045] The prediction optimization module is used for frequency shift flow direction modeling, obtains frequency shift flow direction change prediction data through frequency shift flow direction modeling, and sends the frequency shift flow direction change prediction data to the output management module;

[0046] The output management module is used for water flow management optimization, and water flow management allocation reference data is obtained through water flow management optimization.

[0047] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0048] (1) In the existing water flow measurement and management process, there are existing water flow measurement methods. Due to the complex flow patterns and signal interference in open channels and rivers, it is difficult to measure the flow rate stably and accurately. The conventional flow meter method has technical problems such as large data fluctuations and unstable signal quality in flow measurement. This solution creatively adopts an overall intelligent algorithm process that combines feature optimization, flow modeling and optimization management. Through feature optimization and flow modeling, qualitative and quantitative information of water flow is obtained, and the intelligent process of optimization management is comprehensively followed to improve the quality and intelligence of water flow measurement and management, and also improve the overall usability;

[0049] (2) In the existing intelligent extraction methods of water flow characteristics, there are traditional classical algorithms, namely digital filtering algorithms and curve fitting algorithms. In actual measurements, since the channel may be narrow, the electromagnetic waves emitted by the radar will hit the shore. At the same time, there may be a large number of floating objects on the water surface, causing the signal amplitude to change continuously. The water flow may be turbulent, and the flow state is not good. These factors will lead to the identification of incorrect Doppler frequency shifts, and thus the surface flow velocity cannot be measured correctly. The most intuitive manifestation is that the measured flow velocity data will fluctuate over a large range. Due to the complexity and diversity of the on-site water flow measurement and management, there will be technical problems in which filtering cannot be completed. This scheme creatively adopts a one-dimensional convolutional neural network improved by multi-convolution integration to optimize water flow characteristics. By adaptively improving and designing the structure of the one-dimensional convolutional network, the intelligence and efficiency of water flow feature extraction are improved. While solving the problem of easy identification errors in the single Doppler frequency shift method, the overall stability and reliability of the method are also improved.

[0050] (3) In view of the technical problems in the existing water flow frequency shift and flow direction modeling and prediction methods, such as large noise interference of water flow frequency shift signals, complex flow direction changes, and difficulty in fully capturing spatiotemporal characteristics, this scheme creatively adopts an improved time series prediction network that combines Doppler effect analysis and frequency domain analysis. By combining the two classic modeling and analysis methods with spatiotemporal characteristics for combined prediction, the overall modeling and prediction effect of water flow frequency shift and flow direction is improved.

[0051] (4) In view of the technical problems in the existing intelligent water flow management process, such as the lack of physical constraints in the water flow distribution scheme, the difficulty in achieving precise adjustment of flow distribution and poor dynamic adaptability, this scheme creatively adopts a reinforcement learning dynamic water flow distribution adjustment method that combines physical constraints and hydraulic equations to optimize the comprehensive management of water flow. By combining the physical constraint model, the physical characteristics of the water flow management results are optimized. At the same time, through the improvement of the reward function and the integration of frequency shift flow direction information, the overall flexibility and accuracy of water flow management are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of a water flow management optimization method based on machine learning provided by the present invention;

[0053] Figure 2 A schematic diagram of a water flow management optimization system based on machine learning provided by the present invention;

[0054] Figure 3 A schematic diagram of the process of optimizing the water flow characteristics in step S2;

[0055] Figure 4 A schematic diagram of the process of modeling the frequency shift flow direction in step S3;

[0056] Figure 5 Schematic diagram of the process for optimizing water flow management in step S4.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0060] Example 1, see Figure 1 The present invention provides a water flow management optimization method based on machine learning, which comprises the following steps:

[0061] Step S1: data management;

[0062] Step S2: water flow characteristic optimization;

[0063] Step S3: frequency shift flow direction modeling;

[0064] Step S4: Water flow management optimization.

[0065] By performing the above operations, in the existing water flow measurement and management process, there are existing water flow measurement methods. Due to the complex flow state in open channels and rivers and signal interference, it is difficult to measure the flow rate stably and accurately. The conventional flow meter method has technical problems such as large data fluctuations and unstable signal quality in flow measurement. This solution creatively adopts an overall intelligent algorithm process that combines feature optimization, flow modeling and optimization management. Through feature optimization and flow modeling, qualitative and quantitative information of water flow is obtained, and the intelligent process of optimization management is comprehensively followed to improve the quality and intelligence of water flow measurement to management, and also improve the overall usability.

[0066] Example 2, see Figure 1 and Figure 2 , in step S1, the data management is used to collect, store and preprocess the original data set required for water flow management, specifically, to obtain the original water flow management data through multi-source sensor data collection, and to obtain the optimized water flow management data through data preprocessing;

[0067] The water flow management raw data includes water flow data, water level data, environmental support data, signal evaluation data, water flow pattern data and time-space data;

[0068] The water flow data includes flow velocity and flow value measured by a radar flow meter;

[0069] The environmental support data includes meteorological condition data and hydrological condition data; the meteorological condition data specifically includes temperature, humidity and wind speed data, and the hydrological condition data specifically includes precipitation and river slope data;

[0070] The signal evaluation data includes signal strength, signal-to-noise ratio, and echo signal quality data generated by the radar flow meter;

[0071] The water flow state data includes water flow direction data, turbulence intensity data and flow velocity change data;

[0072] The temporal and spatial data include timestamp and geographic location reference information data;

[0073] The data preprocessing includes data denoising, exception processing, data standardization and data smoothing operations;

[0074] The data denoising includes signal filtering and wavelet transform denoising operations;

[0075] The abnormality processing specifically uses standard deviation statistics to identify abnormal values, and fills abnormal values ​​and missing values ​​through linear interpolation;

[0076] The data standardization specifically adopts the maximum and minimum standardization method to perform range standardization of numerical data;

[0077] The data smoothing operation specifically uses an exponential smoothing method to perform smoothing and optimization processing on the time series data;

[0078] The optimized water flow management data includes optimized water flow data, optimized water level data, optimized environmental support data, optimized signal evaluation data, optimized water flow state data and optimized time-space data.

[0079] Example 3, see Figure 1 , Figure 2 and Figure 3This embodiment is based on the above embodiment. In step S2, the water flow characteristic optimization is used to extract and optimize the spatiotemporal characteristics of the water flow. Specifically, based on the optimized water flow management data, a one-dimensional convolutional neural network improved by multi-convolution integration is used to optimize the water flow characteristics to obtain the spatiotemporal characteristic data of the water flow, including the following steps:

[0080] Step S21: constructing a water flow input layer, specifically, taking the optimized water flow management data as input data samples to construct the water flow input layer;

[0081] Step S22: constructing a one-dimensional convolution layer improved by multi-convolution integration, specifically, sequentially integrating the depthwise separable convolution layer, the hybrid dilated convolution layer, and the adaptive frequency domain pooling layer to obtain a one-dimensional convolution layer improved by multi-convolution integration;

[0082] The depthwise separable convolution layer specifically uses a depthwise separable convolution with a dilation factor of 2 and sets the number of channels to 64. The calculation formula is:

[0083] ;

[0084] Where y1 is the output of the depthwise separable convolutional layer, Conv 1D (·) is a one-dimensional convolution operation function, x is the input data sample, filters is the number of convolution kernels, used to indicate the number of channels, kernel_size is the kernel size, and dilation is the dilation factor;

[0085] The hybrid dilated convolution layer specifically constructs a dilated convolution module with dilation factors alternating between [1, 3, 5], and introduces channel attention after each dilated convolution module to construct a hybrid dilated convolution layer. The calculation formula is:

[0086] ;

[0087] In the formula, y att is the output of the mixed dilated convolutional layer, a L is the channel attention weight of the Lth hybrid dilated convolutional layer, L is the hybrid dilated convolutional layer index, and DConv 1D (·) is the dilated convolution operation function, y L-1 is the output of the convolution layer of layer L-1, filters is the number of convolution kernels, used to indicate the number of channels, kernel_size is the kernel size, and dilation is the dilation factor;

[0088] The adaptive frequency domain pooling layer specifically uses the adaptive frequency domain pooling layer to replace the fully connected layer in the standard one-dimensional convolution module to retain the time series frequency information. The calculation formula is:

[0089] ;

[0090] In the formula, y freq is the adaptive frequency domain pooling output feature, AFPool(·) is the adaptive frequency domain pooling operation function, y att is the output of the mixed dilated convolutional layer;

[0091] Step S23: constructing a branch output layer, specifically constructing a regression branch and a classification branch to predict the Doppler shift value and flow direction category;

[0092] The regression branch is used to output the Doppler frequency shift value and represent the time characteristics of the water flow. The calculation formula is:

[0093] ;

[0094] In the formula, y reg is the output of the regression branch, which is used to output the flow time characteristics. Dense(·) is the fully connected layer operation function. freq is the adaptive frequency domain pooling output feature, units is the number of output units;

[0095] The classification branch is used to output the flow direction categories, including forward, reverse, interference and turbulence, and represents the spatial characteristics of the water flow. The calculation formula is:

[0096] ;

[0097] In the formula, y class is the classification branch output, used to output the flow category, Dense(·) is the fully connected layer operation function, y freq is the adaptive frequency domain pooling output feature, units is the number of output units, activation is the activation type, and softmax is the softmax classifier function identifier;

[0098] Step S24: construct an optimized loss function, specifically constructing a mean square error loss and a cross entropy loss, performing weighted summation, and constructing the optimized loss function. The calculation formula is:

[0099] L total =b1L reg +b2L class ;

[0100] Where, L total is the optimization loss function, b1 is the regression loss weight, b2 is the classification loss weight, L reg is the regression loss function, specifically the mean square error function, L class It is a classification loss function, specifically using the cross entropy loss function;

[0101] Step S25: water flow characteristic optimization model training, specifically, through the water flow input layer, the multi-convolution integrated improved one-dimensional convolution layer, the branch output layer and the optimized loss function, the water flow characteristic optimization model training is performed to obtain the water flow characteristic optimization model Model FLNX ;

[0102] Table 1 is a reference parameter table of the water flow characteristic optimization model. As shown in the table, preferably, the value of the regression loss weight b1 is 0.7, and the value of the classification loss weight b2 is 0.3;

[0103] Table 1 Reference parameters of water flow characteristic optimization model

[0104]

[0105] Step S26: Optimizing water flow characteristics, specifically, using the water flow characteristic optimization model Model according to the optimized water flow management data FLNX , optimize the water flow characteristics and obtain the spatiotemporal characteristic data of water flow.

[0106] By performing the above operations, in the existing intelligent extraction methods of water flow characteristics, there are traditional classical algorithms, namely digital filtering algorithms and curve fitting algorithms. In actual measurements, since the channel may be narrow, the electromagnetic waves emitted by the radar will hit the shore. At the same time, there may be a large number of floating objects on the water surface, causing the signal amplitude to change continuously. The water flow may be turbulent, and the flow state is not good. These factors will lead to the identification of incorrect Doppler frequency shifts, thereby failing to correctly measure the surface velocity. The most intuitive manifestation is that the measured velocity data will fluctuate over a large range. Due to the complexity and diversity of the on-site water flow measurement and management, there will be technical problems in which filtering cannot be completed. This scheme creatively adopts a one-dimensional convolutional neural network improved by multi-convolution integration to optimize water flow characteristics. By adaptively improving and designing the structure of the one-dimensional convolutional network, the intelligence and efficiency of water flow feature extraction are improved, and the problem of easy recognition errors in the single Doppler frequency shift method is solved, while also improving the overall stability and reliability of the method.

[0107] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the frequency shift flow direction modeling is used to establish a flow direction prediction model based on water flow frequency shift to identify the change trend and flow direction of the water flow. Specifically, based on the spatiotemporal characteristic data of the water flow, an improved time series prediction network combining Doppler effect analysis and frequency domain analysis is used to perform frequency domain flow direction modeling to obtain frequency shift flow direction change prediction data, including the following steps:

[0108] Step S31: constructing a water flow spatiotemporal characteristic input layer, specifically, inputting the water flow spatiotemporal characteristic data as raw data for frequency shift flow direction modeling to construct the water flow spatiotemporal characteristic input layer;

[0109] Step S32: Doppler effect analysis, specifically, based on the Doppler effect, frequency shift value spectrum analysis modeling is performed on the original data input for frequency shift flow direction modeling to obtain Doppler frequency shift analysis data, and the calculation formula is:

[0110] ;

[0111] In the formula, f shift is the Doppler shift analysis data, v flow is the water velocity, c is the sound wave propagation speed, f ref is a reference frequency value, specifically, a regression branch output in the spatiotemporal characteristic data of water flow;

[0112] Step S33: frequency domain feature analysis, specifically, extracting and analyzing the frequency domain features of the original data input for the frequency shift flow direction modeling by fast Fourier transform to obtain frequency domain feature data, and the calculation formula is:

[0113] ;

[0114] Where, X freq is the frequency domain feature data, FTT(·) is the fast Fourier transform function, X S / T It is the spatiotemporal characteristic data of water flow converted into frequency domain characteristics;

[0115] Step S34: constructing an improved time series prediction network, specifically, according to the Doppler frequency shift analysis data and the frequency domain feature data, through frequency domain attention optimization, multi-scale time series modeling, frequency domain long short-term memory network construction and modeling layer construction, to obtain the improved time series prediction network;

[0116] The frequency domain attention optimization is used to learn the weight of the frequency domain features to enhance the model's ability to process the frequency domain features. Specifically, the frequency domain attention optimization is performed through the frequency domain feature attention mechanism. The calculation formula is:

[0117] ;

[0118] In the formula, A freq is the frequency domain attention weight, sig(·) is the S-type activation function, W freq is the frequency domain weight matrix, X freq is the frequency domain feature data;

[0119] The multi-scale time series modeling is used to model the flow trend change of water flow, specifically by constructing a multi-scale convolution layer to perform splicing of frequency domain features. The calculation formula is:

[0120] ;

[0121] Where Y M / S is the multi-scale temporal modeling feature output, Conv(·) is the convolution operation function, X S / T is the spatiotemporal characteristic data of water flow converted into frequency domain features, k1 is the convolution kernel size of the first layer of multi-scale convolution, and m is the total number of convolution layers of multi-scale convolution;

[0122] The frequency domain long short-term memory network is constructed to add frequency domain features on the basis of the standard long short-term memory network, specifically to construct a standard long short-term memory neural network and perform frequency domain feature fusion on the hidden state to construct the frequency domain long short-term memory network;

[0123] The modeling layer is constructed to output the frequency shift and direction change category of the water flow, specifically to construct a cross entropy loss and a multi-task learning branch to perform modeling layer construction;

[0124] Step S35: frequency shift flow direction modeling, specifically, through the water flow spatiotemporal characteristic input layer, the Doppler effect analysis, the frequency domain characteristic analysis and the improved time series prediction network, model training and data prediction of frequency shift flow direction modeling are performed to obtain frequency shift flow direction change prediction data.

[0125] By performing the above operations, in order to address the technical problems in the existing water flow frequency shift and flow direction modeling and prediction methods, such as large noise interference of water flow frequency shift signals, complex flow direction changes, and difficulty in fully capturing spatiotemporal characteristics, this solution creatively adopts an improved time series prediction network that combines Doppler effect analysis and frequency domain analysis. By combining the two classic modeling and analysis methods with spatiotemporal characteristics for combined prediction, the overall modeling and prediction effect of water flow frequency shift and flow direction is improved.

[0126] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the water flow management optimization is used to comprehensively manage the distribution and regulation of water flow. Specifically, the water flow spatiotemporal characteristic data and the frequency shift flow direction change prediction data are combined, and a reinforcement learning dynamic water flow distribution adjustment method combining physical constraints and hydraulic equations is used to optimize the water flow comprehensive management and obtain water flow management distribution reference data, including the following steps:

[0127] Step S41: Modeling of water flow management problem, used to define the management target of water flow, specifically, minimizing flow fluctuation and optimizing water flow distribution as the basic target model of water flow management problem, and obtaining the water flow management target model, the calculation formula is:

[0128] ;

[0129] Where minJ is the minimization optimization objective function, n is the total number of water flow areas, i is the water flow area index, Q i (t) is the actual water flow in the ith area at time t, Q target,i (t) is the optimized target water flow; f shift,i (t) is the actual frequency shift value of the ith region at time t, f shift,i (t-1) is the actual frequency shift value of the ith region at time t-1;

[0130] Step S42: physical constraint modeling, specifically constructing mass conservation equations and hydraulic equations as physical constraint models of the model;

[0131] Step S43: dynamic water flow distribution reinforcement learning, specifically, constructing a standard reinforcement learning parameter group according to the water flow management target model, and improving the design of the reward function to obtain a dynamic water flow distribution reinforcement learning architecture;

[0132] The improved design of the reward function specifically refers to weighting according to the stability of water flow regulation and the stability of flow frequency shift to obtain a dynamic water flow distribution reward function, and performing reinforcement learning training according to the dynamic water flow distribution reward function;

[0133] The calculation formula of the dynamic water flow distribution reward function is:

[0134] ;

[0135] Where R(t) is the dynamic water flow distribution reward function, is the water flow reward weight, n is the total number of water flow areas, i is the water flow area index, Q i (t) is the actual water flow in the ith area at time t, Q target,i (t) is the optimized target water flow rate, is the frequency shift value reward weight, f shift,i (t) is the actual frequency shift value of the ith region at time t, f shift,i (t-1) is the actual frequency shift value of the ith region at time t-1;

[0136] Step S44: iterative training of water flow management optimization, specifically using a deep Q-value learning method to perform iterative optimization of water flow distribution and Q-value update to obtain an optimal model for water flow management optimization;

[0137] Step S45: water flow management, specifically, performing water flow management according to the water flow management optimization optimal model to obtain water flow management allocation reference data;

[0138] The water flow management allocation reference data specifically includes regional water flow allocation recommendations, flow adjustment strategies and comprehensive flow prediction reference data.

[0139] By executing the above operations, in order to address the technical problems in the existing intelligent water flow management process, such as the lack of physical constraints in the water flow distribution scheme, the difficulty in achieving precise adjustment of flow distribution and poor dynamic adaptability, this scheme creatively adopts a reinforcement learning dynamic water flow distribution adjustment method that combines physical constraints and hydraulic equations to optimize the comprehensive management of water flow. By combining the physical constraint model, the physical characteristics of the water flow management results are optimized. At the same time, through the improvement of the reward function and the integration of frequency shift flow direction information, the overall flexibility and accuracy of water flow management are improved.

[0140] Example 6, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides a water flow management optimization system based on machine learning, including a data management module, a feature processing module, a prediction optimization module and an output management module;

[0141] The data management module is used for data management, and obtains optimized water flow management data through data management, and sends the optimized water flow management data to the feature processing module;

[0142] The feature processing module is used for optimizing water flow characteristics, obtaining water flow spatiotemporal feature data through water flow characteristic optimization, and sending the water flow spatiotemporal feature data to the prediction optimization module;

[0143] The prediction optimization module is used for frequency shift flow direction modeling, obtains frequency shift flow direction change prediction data through frequency shift flow direction modeling, and sends the frequency shift flow direction change prediction data to the output management module;

[0144] The output management module is used for water flow management optimization, and water flow management allocation reference data is obtained through water flow management optimization.

[0145] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0146] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0147] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A water flow management optimization method based on machine learning, characterized in that: The method comprises the following steps: Step S1: data management, obtaining optimized water flow management data; Step S2: Optimizing water flow characteristics, using a one-dimensional convolutional neural network improved by multi-convolution integration to optimize water flow characteristics and obtain spatiotemporal characteristic data of water flow; The multi-convolution integrated improved one-dimensional convolution layer sequentially integrates the depth-separable convolution layer, the hybrid dilated convolution layer and the adaptive frequency domain pooling layer to obtain a multi-convolution integrated improved one-dimensional convolution layer; The depthwise separable convolution layer uses a depthwise separable convolution with a dilation factor of 2, and sets the number of channels to 64; the hybrid dilated convolution layer constructs a dilated convolution module with dilation factors alternating between [1, 3, 5], and introduces channel attention after each dilated convolution module to construct a hybrid dilated convolution layer; the adaptive frequency domain pooling layer uses an adaptive frequency domain pooling layer to replace the fully connected layer in the standard one-dimensional convolution module to retain time series frequency information; Step S3: frequency shift flow direction modeling, using an improved time series prediction network that combines Doppler effect analysis and frequency domain analysis to perform frequency domain flow direction modeling and obtain frequency shift flow direction change prediction data; Step S4: Water flow management optimization, using a reinforcement learning dynamic water flow allocation adjustment method that combines physical constraints and hydraulic equations to perform comprehensive water flow management optimization and obtain water flow management allocation reference data.

2. The method for optimizing water flow management based on machine learning according to claim 1, characterized in that: In step S1, the data management is used to collect, store and preprocess the original data set required for water flow management, specifically to obtain the original water flow management data through multi-source sensor data collection, and to obtain optimized water flow management data through data preprocessing; The water flow management raw data includes water flow data, water level data, environmental support data, signal evaluation data, water flow pattern data and time-space data; The data preprocessing includes data denoising, exception processing, data standardization and data smoothing operations.

3. The method for optimizing water flow management based on machine learning according to claim 2, characterized in that: In step S2, the water flow characteristic optimization is used to extract and optimize the spatiotemporal characteristics of the water flow. Specifically, based on the optimized water flow management data, a one-dimensional convolutional neural network improved by multi-convolution integration is used to optimize the water flow characteristics to obtain the spatiotemporal characteristic data of the water flow, including the following steps: Step S21: constructing a water flow input layer, specifically, taking the optimized water flow management data as input data samples to construct the water flow input layer; Step S22: constructing a one-dimensional convolution layer improved by multi-convolution integration, specifically, sequentially integrating the depthwise separable convolution layer, the hybrid dilated convolution layer, and the adaptive frequency domain pooling layer to obtain a one-dimensional convolution layer improved by multi-convolution integration; The depthwise separable convolution layer specifically uses a depthwise separable convolution with a dilation factor of 2 and the number of channels is set to 64; The hybrid dilated convolution layer specifically constructs a dilated convolution module with dilation factors alternating between [1, 3, 5], and introduces channel attention after each dilated convolution module to construct a hybrid dilated convolution layer; The adaptive frequency domain pooling layer specifically adopts the adaptive frequency domain pooling layer to replace the fully connected layer in the standard one-dimensional convolution module, so as to retain the time series frequency information; Step S23: constructing a branch output layer, specifically constructing a regression branch and a classification branch to predict the Doppler shift value and flow direction category; The regression branch is used to output the Doppler frequency shift value and represent the time characteristics of the water flow; The classification branch is used to output flow direction categories, including forward, reverse, interference and turbulent, and to represent the spatial characteristics of the water flow; Step S24: constructing an optimized loss function, specifically constructing a mean square error loss and a cross entropy loss, performing weighted summation, and constructing the optimized loss function; Step S25: water flow characteristic optimization model training, specifically, through the water flow input layer, the multi-convolution integrated improved one-dimensional convolution layer, the branch output layer and the optimized loss function, the water flow characteristic optimization model training is performed to obtain the water flow characteristic optimization model Model FLNX ; Step S26: Optimizing water flow characteristics, specifically, using the water flow characteristic optimization model Model according to the optimized water flow management data FLNX , optimize the water flow characteristics and obtain the spatiotemporal characteristic data of water flow.

4. The method for optimizing water flow management based on machine learning according to claim 3, characterized in that: In step S3, the frequency shift flow direction modeling is used to establish a flow direction prediction model based on water flow frequency shift to identify the change trend and flow direction of the water flow. Specifically, based on the spatiotemporal characteristic data of the water flow, an improved time series prediction network combining Doppler effect analysis and frequency domain analysis is used to perform frequency domain flow direction modeling to obtain frequency shift flow direction change prediction data, including the following steps: Step S31: constructing a water flow spatiotemporal characteristic input layer, specifically, inputting the water flow spatiotemporal characteristic data as raw data for frequency shift flow direction modeling to construct the water flow spatiotemporal characteristic input layer; Step S32: Doppler effect analysis, specifically, performing frequency shift value spectrum analysis modeling on the original data input for frequency shift flow direction modeling according to the Doppler effect to obtain Doppler frequency shift analysis data; Step S33: frequency domain feature analysis, specifically, extracting and analyzing the frequency domain features of the original data input for frequency shift flow direction modeling through fast Fourier transform to obtain frequency domain feature data; Step S34: constructing an improved time series prediction network, specifically, according to the Doppler frequency shift analysis data and the frequency domain feature data, through frequency domain attention optimization, multi-scale time series modeling, frequency domain long short-term memory network construction and modeling layer construction, to obtain the improved time series prediction network; Step S35: frequency shift flow direction modeling, specifically, through the water flow spatiotemporal characteristic input layer, the Doppler effect analysis, the frequency domain characteristic analysis and the improved time series prediction network, model training and data prediction of frequency shift flow direction modeling are performed to obtain frequency shift flow direction change prediction data.

5. The method for optimizing water flow management based on machine learning according to claim 4, characterized in that: In step S4, the water flow management optimization is used to comprehensively manage the distribution and regulation of water flow. Specifically, the water flow spatiotemporal characteristic data and the frequency shift flow direction change prediction data are combined, and a reinforcement learning dynamic water flow distribution adjustment method combining physical constraints and hydraulic equations is adopted to perform comprehensive water flow management optimization and obtain water flow management distribution reference data.

6. The method for optimizing water flow management based on machine learning according to claim 5, characterized in that: In step S4, the water flow management optimization comprises the following steps: Step S41: Modeling the water flow management problem, which is used to define the management target of the water flow, specifically, minimizing flow fluctuation and optimizing water flow distribution as the basic target model of the water flow management problem, and obtaining a water flow management target model; Step S42: physical constraint modeling, specifically constructing mass conservation equations and hydraulic equations as physical constraint models of the model; Step S43: dynamic water flow distribution reinforcement learning, specifically, constructing a standard reinforcement learning parameter group according to the water flow management target model, and improving the design of the reward function to obtain a dynamic water flow distribution reinforcement learning architecture; The improved design of the reward function specifically refers to weighting according to the stability of water flow regulation and the stability of flow frequency shift to obtain a dynamic water flow distribution reward function, and performing reinforcement learning training according to the dynamic water flow distribution reward function; The calculation formula of the dynamic water flow distribution reward function is: ； Where R(t) is the dynamic water flow distribution reward function, is the water flow reward weight, n is the total number of water flow areas, i is the water flow area index, Q i (t) is the actual water flow in the ith area at time t, Q target,i (t) is the optimized target water flow rate, is the frequency shift value reward weight, f shift,i (t) is the actual frequency shift value of the ith region at time t, f shift,i (t-1) is the actual frequency shift value of the ith region at time t-1; Step S44: iterative training of water flow management optimization, specifically using a deep Q-value learning method to perform iterative optimization of water flow distribution and Q-value update to obtain an optimal model for water flow management optimization; Step S45: water flow management, specifically, performing water flow management according to the water flow management optimization optimal model to obtain water flow management allocation reference data.

7. The method for optimizing water flow management based on machine learning according to claim 6, characterized in that: In step S45, the water flow management allocation reference data specifically includes regional water flow allocation recommendations, flow adjustment strategies and flow direction flow comprehensive prediction reference data.

8. A water flow management optimization system based on machine learning, used to implement a water flow management optimization method based on machine learning as described in any one of claims 1 to 7, characterized in that: It includes data management module, feature processing module, prediction optimization module and output management module.

9. The water flow management optimization system based on machine learning according to claim 8, characterized in that: The data management module is used for data management, and obtains optimized water flow management data through data management, and sends the optimized water flow management data to the feature processing module; The feature processing module is used for optimizing water flow characteristics, obtaining water flow spatiotemporal feature data through water flow characteristic optimization, and sending the water flow spatiotemporal feature data to the prediction optimization module; The prediction optimization module is used for frequency shift flow direction modeling, obtains frequency shift flow direction change prediction data through frequency shift flow direction modeling, and sends the frequency shift flow direction change prediction data to the output management module; The output management module is used for water flow management optimization, and water flow management allocation reference data is obtained through water flow management optimization.

Images