CN111412959B - Flow online monitoring calculation method, monitor and monitoring system - Google Patents
Flow online monitoring calculation method, monitor and monitoring system Download PDFInfo
- Publication number
- CN111412959B CN111412959B CN202010353603.5A CN202010353603A CN111412959B CN 111412959 B CN111412959 B CN 111412959B CN 202010353603 A CN202010353603 A CN 202010353603A CN 111412959 B CN111412959 B CN 111412959B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- flow
- output
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01F—MEASURING VOLUME, VOLUME FLOW, MASS FLOW OR LIQUID LEVEL; METERING BY VOLUME
- G01F1/00—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow
- G01F1/66—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow by measuring frequency, phase shift or propagation time of electromagnetic or other waves, e.g. using ultrasonic flowmeters
- G01F1/663—Measuring the volume flow or mass flow of fluid or fluent solid material wherein the fluid passes through a meter in a continuous flow by measuring frequency, phase shift or propagation time of electromagnetic or other waves, e.g. using ultrasonic flowmeters by measuring Doppler frequency shift
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01P—MEASURING LINEAR OR ANGULAR SPEED, ACCELERATION, DECELERATION, OR SHOCK; INDICATING PRESENCE, ABSENCE, OR DIRECTION, OF MOVEMENT
- G01P5/00—Measuring speed of fluids, e.g. of air stream; Measuring speed of bodies relative to fluids, e.g. of ship, of aircraft
- G01P5/24—Measuring speed of fluids, e.g. of air stream; Measuring speed of bodies relative to fluids, e.g. of ship, of aircraft by measuring the direct influence of the streaming fluid on the properties of a detecting acoustical wave
- G01P5/241—Measuring speed of fluids, e.g. of air stream; Measuring speed of bodies relative to fluids, e.g. of ship, of aircraft by measuring the direct influence of the streaming fluid on the properties of a detecting acoustical wave by using reflection of acoustical waves, i.e. Doppler-effect
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention provides a flow online monitoring and calculating method, a monitor and a monitoring system, wherein the method comprises the following steps: step 1, data preprocessing and DCL model parameter initialization; step 2, selecting sample dimensions corresponding to the N H-ADCP cells; randomly dividing the initial population into 5 shares, and calculating each share corresponding to a model to obtain a model output value; taking the root mean square error of the output value and the measured value sample as a fitness function for optimizing, comparing the fitness, selecting the minimum value as an optimal index set of the iteration, and recording an optimal feature dimension set and a parameter set; recording the corresponding optimal feature dimension, the calculation model and the model parameters by using the selected replacement global optimal fitness value; if the requirements of precision and maximum iteration times cannot be met, gradually expanding the lower limit of feature dimension selection and starting the next iteration calculation; otherwise, outputting the optimal model and corresponding model parameters, giving out a final flow calculation value, and ending the loop iteration; and 3, updating the parameters.
Description
Technical Field
The invention belongs to the field of hydrological tests, and particularly relates to an on-line flow monitoring and calculating method, a monitor and a monitoring system.
Technical Field
In recent years, with the issuance of water resource management systems, river and lake growth systems and other systems, and the implementation of national key projects such as medium and small river hydrologic monitoring projects and hydrologic modernization capacity improvement, higher requirements are provided for the timeliness and the precision of hydrologic monitoring elements. Of the many hydrologic monitoring elements, the most important and interesting is the monitoring of flow. The traditional flow monitoring method mostly adopts a flow meter method, so that the workload is large and the real-time monitoring can not be realized. In recent years, with the development of sensor technology, various sensors based on acoustic, optical and electromagnetic methods are developed in a large number, and currently, the online flow monitoring is the most widely used H-ADCP online monitoring mode, as shown in fig. 1, the flow velocity measurement principle of the H-ADCP sensor is based on the doppler effect, in which an ultrasonic transducer emits ultrasonic waves into water, and the ultrasonic waves are scattered by a scatterer in the water to generate the doppler shift effect, which is related to the water velocity, so as to indirectly calculate the water velocity of different cell distances (different cells divided by a dotted line in the figure) of a water layer where the sensor is located. The power of the transducer is generally divided into 300kHz, 600kHz and 1200kHz, and the corresponding maximum detection distances are 110m, 50m and 12m respectively. The flow velocity cells can be flexibly arranged according to different test sections and precision requirements, the general diameter of a large river is 5m, the diameter of a medium river is 1m, the upper inclined dotted line and the lower inclined dotted line sent by the sensor in the figure are the actual divergence range of sound waves, and the maximum installation depth and the minimum installation depth of the instrument are determined by the range.
The H-ADCP determines the speed of the water flow relative to the instrument by sending sound wave signals to a water layer where the instrument is located and receiving and processing the signal intensity reflected by particles in the section water body, so that the absolute speed of the water flow is obtained. However, the hydrological test section is mostly affected by factors such as water conservancy projects, backwater jacking, tides and section erosion and deposition changes, and the water flow characteristics are complex and changeable, so that difficulties are brought to the H-ADCP local single-layer unit grid flow velocity estimation of section flow, and the H-ADCP flow online monitoring has the defects and disadvantages of poor modeling precision, difficulty in flow velocity characteristic selection, large data complexity and the like. Therefore, how to better improve the applicable river reach and monitoring accuracy of the H-ADCP is very urgent and needed.
At present, most of methods for calculating the flow rate by the H-ADCP are to establish the relation between the average flow rate of the cross section and the index flow rate, that is, to establish a calibration curve or a regression equation, so as to obtain the flow rate of the cross section according to the water passing area. Some scholars also develop research by adopting algorithms such as wavelet analysis and BP neural network. The existing flow calculation method based on the H-ADCP has the following problems:
(1) the calculation accuracy is not high: the existing calculation method mostly uses mathematical statistics or a generalized curve formula of vertical distribution of flow velocity as a core, but the method is only applicable to river reach with regular section or strong regularity of flow velocity distribution, and has great limitation in practical application, so that the calculation accuracy of the method is often not high.
(2) The problem of parameter selection: algorithms such as wavelet analysis and BP neural network cannot realize the output of the optimal cells, the calculation result often has larger uncertainty, and the model parameters cannot be automatically adjusted along with the increase of the number of samples or the change of the test section, so that the method cannot be popularized and applied in a large range.
Disclosure of Invention
The present invention has been made to solve the above problems, and an object of the present invention is to provide an online river flow monitoring calculation method, a monitoring device, and a monitoring system, which have high applicability, can effectively identify and extract a nonlinear characteristic between a flow velocity and a cross-sectional flow of an H-ADCP cell, and can actually improve accuracy and monitoring efficiency of online river flow monitoring.
In order to achieve the purpose, the invention adopts the following scheme:
< method >
The invention provides a flow online monitoring and calculating method, which is characterized by comprising the following steps:
Inputting monitored section data, H-ADCP cell flow velocity data and manual comparison data, and at least performing abnormal value screening, model input format sorting and data normalization preprocessing on the data;
initializing at least the number of H-ADCP cells, the cell size, the model rate periodicity, the sample number in the inspection period and the initial parameters of each submodel as DCL model parameters;
step 2, learning and feature fusion of input sample data features
Step 2-1, corresponding to N H-ADCP cells, selecting a sample (1/2) x N dimensions, if the dimension number is less than the number N of modelsmodelsIs then set to Nmodels(ii) a And, the PSO initial population is randomly divided into 5 parts, and each part corresponds to: the neural network model comprises a back propagation neural network model BP, an Elman neural network model ELMAN, a radial basis function neural network model RBF, a generalized regression neural network model GRNN and a support vector machine model SVM, wherein one of the 5 models is selected; are respectively marked as BPinput、ELMANinput、RBFinput、GRNNinputAnd SVMinputAnd respectively inputting the data into respective models for calculation to obtain model output value BPoutput、 ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutput;
Step 2-2, respectively sampling the respective output values BP of the 5 modelsoutput、ELMANoutput、RBFoutput、 GRNNoutputAnd SVMoutputFitness function BP taking root mean square error MSE of measured value samples as optimizationmes、ELMANmes、RBFmes、GRNNmesAnd SVMmesSelecting 5 MSEs with the minimum fitness values as the optimal index set MSE of the iteration by comparing the fitness degrees among different populationsmAnd simultaneously recording the corresponding optimal feature dimension set NmAnd a corresponding parameter set Pm(ii) a If it is the first iteration, then MSEmAs global optimum fitness value MSEbest;
Step 2-3, using MSE selected in step 2-2mReplacing global optimal fitness value MSEbestRecording the corresponding optimal feature dimension, calculation model and model parameters;
step 2-4 comparison of MSEbestWhether the iteration frequency of the model reaches the maximum value is judged;
step 2-5, if the precision requirement cannot be met and the maximum iteration number is not reached, gradually expanding the feature dimension to select the lower limit of 1/N according to the modes of [1/2,1/4, … and 1/N ], returning to the step 2-1, and starting the next iterative computation; otherwise, executing the next step operation;
step 2-6, outputting the optimal model and corresponding model parameters, giving out a final flow calculation value, and ending the loop iteration;
step 3, updating parameters
With the continuous increase of the flow monitoring data and samples, when the monitored flow data reaches a certain scale or when new characteristic samples of the flow data appear, returning to the step 1 to re-rate and update the model.
The flow online monitoring and calculating method provided by the invention can also have the following characteristics: in step 1, when the total output number of the H-ADCP cells exceeds 50, performing feature dimension reduction processing on initialized data; and (5) taking the data after the feature dimension reduction processing as input data of the step 2.
The flow online monitoring and calculating method provided by the invention can also have the following characteristics: in step 3, the new characteristic sample of the flow data is a sample in which a flow maximum, minimum or abrupt change occurs.
< monitor >
Further, the present invention provides an online flow rate monitor, comprising: the data acquisition processing part is used for acquiring monitored section data, H-ADCP cell flow rate data and manual comparison data, and at least carrying out abnormal value screening, model input format sorting and data normalization preprocessing on the data; the model initialization part is in communication connection with the data acquisition processing part and at least takes the number of the H-ADCP cells, the cell size, the model rate periodicity, the sample number of the inspection period and the initial parameters of each sub-model as the DCL model parameters to perform initialization processing; a feature learning fusion part, which is connected with the data acquisition processing part and the model initialization part in a communication way, firstly selects N H-ADCP cells corresponding to the sample (1/2) multiplied by N dimensions, if the dimension number is less than the model number NmodelsIs then set to Nmodels(ii) a And, the PSO initial population is randomly divided into 5 parts, and each part corresponds to: the neural network model comprises a back propagation neural network model BP, an Elman neural network model ELMAN, a radial basis function neural network model RBF, a generalized regression neural network model GRNN and a support vector machine model SVM, wherein one of the 5 models is selected; are respectively marked as BPinput、ELMANinput、RBFinput、GRNNinputAnd SVMinputAnd respectively inputting the data into respective models for calculation to obtain model output value BPoutput、ELMANoutput、RBFoutput、 GRNNoutputAnd SVMoutput(ii) a Then, the output value samples BP of the 5 models are respectivelyoutput、 ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutputFitness function BP taking root mean square error MSE of measured value samples as optimizationmes、ELMANmes、RBFmes、GRNNmesAnd SVMmesSelecting 5 MSEs with the minimum fitness values as the optimal index set MSE of the iteration by comparing the fitness degrees among different populationsmAnd simultaneously recording the corresponding optimal feature dimension set NmAnd a corresponding parameter set Pm;If it is the first iteration, then MSEmAs global optimum fitness value MSEbest(ii) a Reuse of selected MSEmReplacing global optimal fitness value MSEbestRecording the corresponding optimal feature dimension, calculation model and model parameters; compare MSEbestWhether the iteration frequency of the model reaches the maximum value is judged; if the precision requirement is not met and the maximum iteration number is not met, press [1/2,1/4, …,1/N]Gradually enlarging the feature dimension, selecting the lower limit of 1/n, and starting the next iterative computation; otherwise, outputting the optimal model and corresponding model parameters, giving out a final flow calculation value as monitoring flow data, and finishing the loop iteration; a parameter updating part which is connected with the data acquisition processing part, the model initialization part and the characteristic learning fusion part in a communication way, and when the monitored flow data reaches a certain scale or when a new characteristic sample of the flow data appears, the parameter updating part returns to the data acquisition processing part againRe-rating and updating the model; and the control part is in communication connection with the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part and controls the operation of the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part.
< monitoring System >
Furthermore, the present invention provides an online flow monitoring system, which is characterized by comprising: at least one on-line flow monitor described in < monitor >; and the monitoring terminal is in communication connection with the flow online monitor, receives flow data sent by the flow online monitor and displays the flow data for monitoring personnel to check.
Preferably, the online flow monitoring system provided by the invention can also have the following characteristics: the monitoring terminal displays the flow data monitored by each flow online monitor at the corresponding river reach in the topographic map in real time, and displays the historical flow data of the corresponding river reach according to the monitoring instructions of monitoring personnel.
Action and Effect of the invention
Compared with the prior art, the invention has the beneficial effects that:
(1) the method has better self-learning capability according to the change of the samples, and also has better characteristic learning capability and calculation precision when the number of the samples is less;
(2) the method can extract the optimal feature combination by global optimization, and ensure the calculation precision of the model while reducing the calculation complexity;
(3) the invention not only has higher simulation precision, but also can provide the optimal characteristic combination, and can effectively reduce the quantity of the representative vertical lines of the flow test, thereby reducing the working strength of the hydrological test;
(4) the invention can self-adaptively select the model and adjust the parameters according to the characteristics of the test river reach, thereby having better popularization and application values.
In conclusion, the flow online monitoring and calculating method, the monitoring instrument and the monitoring system provided by the invention are more intelligent and efficient, the accuracy and efficiency of river flow online monitoring are effectively improved, the working intensity is reduced, and the method and the system are very suitable for large-scale popularization and application.
Drawings
FIG. 1 is a schematic diagram of H-ADCP flow on-line monitoring;
FIG. 2 is a flow chart of a flow online monitoring and calculating method according to an embodiment of the present invention;
FIG. 3 is a diagram of the position of the Liluo lake hydrology station and a test river reach according to an embodiment of the present invention;
FIG. 4 is a comparison graph of predicted values and measured values of a DCL model (of the present invention) according to an embodiment of the present invention;
FIG. 5 is a comparison of typical model predictions involved in an embodiment of the present invention.
Detailed Description
The following describes in detail a specific embodiment of the flow online monitoring and calculating method according to the present invention with reference to the accompanying drawings.
< example one >
As shown in fig. 2, the flow online monitoring and calculating method provided in this embodiment includes the following steps:
(1) and (4) preprocessing data. The method mainly realizes data input, data processing and initialization of the DCL model parameters. The data input includes: testing section data, manual comparison data and H-ADCP unit grid flow velocity data in a hydrological station, and carrying out data preprocessing operations such as abnormal value screening, model input format sorting, data normalization processing and the like on the data. The initialization of the model parameters mainly comprises the following steps: the number of H-ADCP cells, the cell size, the number of samples of the model rate periodicity and inspection period, the initial parameters of each sub-model, etc. And (3) taking the processed data and the initial model parameters as the input of the step (2).
(2) And (5) reducing the dimension of the feature. The total output number of the H-ADCP cells can be set between 1 and 128 according to actual needs, the model calculation complexity is obviously increased along with the increase of the cells, and for the hydrological station flow monitoring data with a complex data structure (the total output number of the H-ADCP cells exceeds 50), feature dimension reduction processing needs to be carried out on the data input in the step (1); otherwise, skipping the step, and directly using the data generated in the step (1) as the input of the step (3). The dimensionality reduction method can be selected from a PCA (principal component analysis) method, a flow velocity distribution characteristic method, a traversal verification method and an empirical judgment method, the core effect is to reduce the complexity of data to the maximum extent on the basis of ensuring available information of a sample, and the model is defaulted to be the PCA method. And (4) reducing the dimension of the input data features (corresponding to the H-ADCP cells), reducing the complexity of the data structure, and generating a new structure containing original data information as the input of the step (3).
(3) And (4) deep feature learning. The part is the core of the whole model and mainly completes the learning of the input sample data characteristics and the characteristic fusion, namely the hidden intrinsic rule of the H-ADCP flow rate data is self-adaptively learned from the H-ADCP cell flow rate data. The specific implementation process is as follows:
first, to reduce the complexity of data, select samples (1/2) × N dimensions (corresponding to N H-ADCP cells), if the dimension number is less than the number of model NmodelsIs then set to Nmodels(ii) a Meanwhile, in order to reduce the computational complexity, the PSO initial population is randomly divided into 5 parts, and each part corresponds to one of 5 models, namely BP (back propagation neural network model), ELMAN (Elman neural network model), RBF (radial basis function neural network model), GRNN (generalized regression neural network model) and SVM (support vector machine model), and is respectively marked as BPinput、ELMANinput、RBFinput、GRNNinputAnd SVMinputAnd respectively inputting the data into respective models for calculation to obtain model output value BPoutput、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutput。
② the above 5 models (BP)output、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutput) The root Mean Square Error (MSE) of the respective output value sample and measured value sample is used as the optimized fitness function BPmes、ELMANmes、RBFmes、GRNNmesAnd SVMmesSelecting 5 MSEs with the minimum fitness values as the optimal index set MSE of the iteration by comparing the fitness degrees among different populationsmAnd simultaneously recording the corresponding optimal feature dimension set NmAnd a corresponding parameter set Pm. If it is the firstOne iteration, then MSEmAs global optimum fitness value MSEbest。
Thirdly, selecting the step two to obtain MSEmReplacing the global optimum fitness value MSEbestAnd recording the corresponding optimal feature dimension, calculation model and model parameters.
Fourthly, comparing MSEbestAnd if the set expected value is reached, the value can be set as the flow test precision value required by the measuring station. And meanwhile, judging whether the iteration times of the model reach the maximum value.
If the accuracy requirement is not met and the maximum iteration number is not met, gradually expanding the feature dimension and selecting the lower limit of 1/N according to the mode of [1/2,1/4, …,1/N ], and returning to the step I to start the next iteration calculation. Otherwise, executing the next model output operation.
Sixthly, outputting an optimal model and corresponding model parameters through iterative optimization of the steps, giving a final flow calculation value, and finishing the circular iteration.
(4) And updating and outputting model parameters. And (3) completing the establishment and calibration work of the DCL model based on the existing flow monitoring sample through the steps (1) to (3). With the increasing of the flow monitoring data and samples, when the data reaches a certain scale (flexibly set according to sampling intervals and station characteristics), especially when new characteristic samples such as a flow maximum value, a minimum value or a sudden change value occur, the step (1) needs to be returned again, and the model needs to be calibrated and updated again. Before re-rating and updating the model, the calculation of all input samples adopts the calculation of the optimal model.
And (3) evaluating a model:
using the most widely used deterministic coefficient R2And the root mean square error MSE as a model evaluation method, R2The larger the value and the smaller the MSE value, the better the model, and the calculation formula is as follows:
(1) deterministic coefficient R2
(2) Root mean square error MSE
In the formula: q (t) is the measured cross-sectional flow at time t,to evaluate the mean value of q (t) over the time period,t is the total time for the cross-sectional flow calculated by the model.
Calculation example:
the Luo lake hydrology station of the Shenzhen river midstream is selected as a model verification station (see FIG. 3), and the station is positioned at 200m upstream of the phoenix tree river junction in the Shenzhen river and the hong Shenzhen and is about 10km away from the Shenzhen river junction. The test section is a canalization section, the widest position of the water surface is about 40m, and the station is simultaneously influenced by tide and a junction jacking, so that the water level-flow relation is scattered, the flow direction is uncertain, and the positive and negative changes in the flow day are frequent. The station flow online monitoring equipment adopts a Channel Master type horizontal acoustic Doppler velocity profiler of a TRDI company, the working frequency is 600KHZ, and a fixed river Channel side mounting mode is adopted. The H-ADCP sampling interval is 10 minutes, the number of the cells is 40 (in fig. 2, the speed measuring cells from the left bank to the right bank of the station are 1# to 40 #), and the size of each cell is 1 m. The manual comparison is carried out by adopting an aerial ADCP.
Actually measured cross section flow sequences (158 measurements in total) and H-ADCP horizontal flow velocity sequences in the same period in 1-8 months in 2019 of the station are selected as research samples. And taking 100 measurement times of 1-5 months, including the first 10 measurement processes, as a model calibration sample, and taking 58 samples in the last 5 measurement processes as verification samples.
As can be seen from Table 1, the present invention (DCL model) better solves the problem of determining the imbalance between the complexity of data and the simulation accuracy than other models. In terms of data complexity, the DCL model can not only reduce the number of the feature cells from 40 to 11, but also reduce the number of the feature cells to 11An optimal combination of cells can be given. Through model calibration and cell flow rate characteristic identification, the optimal combination of the cell flow rates of the H-ADCP of the Luo lake station is {5,9,12,15,17,19,21,24,26,28,35 }. According to the combination, the first 5 cells and the last 5 cells are not selected, mainly because the flow velocity data of the partial cells are influenced by instrument blind areas, bank water flow disorder and the like, the relation is disordered and the usability is not strong, and the middle cell data has better correlation and can be extracted approximately at equal intervals to form a final characteristic sequence, which has high consistency with the instrument characteristics and the flow velocity distribution characteristics. In the aspect of simulation precision, the DCL model has the best simulation effect, the MSE is 36.75, and R is20.93 was reached, which was also the only one in all models that exceeded 0.9.
Comparing fig. 4 and fig. 5, it can be seen that the DCL model has better performance effect no matter the process fitting degree or the relative error, and for the aspect of maximum simulation, also 84.7m from the maximum point in all models3The maximum simulation result of other models is 46.48m3The/s is increased to 56.97m3/s。
TABLE 1 comparison of prediction results of different models
Summary of the application
Aiming at the defects of poor modeling precision, difficult flow rate characteristic selection, large data complexity and the like of the existing H-ADCP flow online monitoring, the invention provides a flow online monitoring method (corresponding to a DCL model). The Rou lake hydrological station affected by tide and complex in water level flow relation is selected as a research object, 158 flow comparison measurement data in 2019 in 1-8 months are adopted, and the applicability of the method is analyzed from multiple aspects such as advantages and disadvantages of various algorithms, applicability of dimensionality reduction data, model stability and the like. The results show that: the method has good self-learning capability, and has good flow velocity characteristic learning capability and flow prediction precision when the number of actually measured flow samples is small.
< example two >
In the second embodiment, an online flow monitoring system is provided, which includes at least one online flow monitor and a monitoring terminal.
The flow online monitor comprises a data acquisition processing part, a model initialization part, a feature learning fusion part, a parameter updating part and a control part. The data acquisition processing part is used for executing the content about the data processing in the steps (1) and (2) of the embodiment. The model initialization unit is used to execute the contents of the model initialization in step (1) of the embodiment. The feature learning fusion part is used for executing the content of the step (3) in the first embodiment. The parameter updating part is used for executing the content of the step (4) in the first embodiment. The control part is connected with the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part in a communication way and controls the operation of the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part.
The monitoring terminal is in communication connection with the flow online monitors, receives flow data sent by the flow online monitors, displays the flow data monitored by each flow online monitor in real time at a corresponding river reach in a topographic map and checks the flow data for monitoring personnel, displays monitoring information to the monitoring personnel visually, and further displays historical flow data of the corresponding river reach according to monitoring instructions of the monitoring personnel.
The above embodiments are merely illustrative of the technical solutions of the present invention. The flow on-line monitoring and calculating method, the monitoring instrument and the monitoring system according to the present invention are not limited to the above embodiments, but are subject to the scope defined by the claims. Any modification or supplement or equivalent replacement made by a person skilled in the art on the basis of this embodiment is within the scope of the invention as claimed in the claims.
Claims (6)
1. A flow online monitoring and calculating method is characterized by comprising the following steps:
step 1, data preprocessing and DCL model parameter initialization
Inputting monitored section data, H-ADCP cell flow velocity data and manual comparison data, and at least performing abnormal value screening, model input format sorting and data normalization preprocessing on the data;
initializing at least the number of H-ADCP cells, the cell size, the model rate periodicity, the sample number in the inspection period and the initial parameters of each submodel as DCL model parameters;
step 2, learning and feature fusion of input sample data features
Step 2-1, corresponding to N H-ADCP cells, selecting a sample (1/2) x N dimensions, if the dimension number is less than the number N of modelsmodelsIs then set to Nmodels(ii) a And, the PSO initial population is randomly divided into 5 parts, and each part corresponds to: the neural network model comprises a back propagation neural network model BP, an Elman neural network model ELMAN, a radial basis function neural network model RBF, a generalized regression neural network model GRNN and a support vector machine model SVM, wherein one of the 5 models is selected; are respectively marked as BPinput、ELMANinput、RBFinput、GRNNinputAnd SVMinputAnd respectively inputting the data into respective models for calculation to obtain model output value BPoutput、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutput;
Step 2-2, respectively sampling the respective output values BP of the 5 modelsoutput、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutputFitness function BP taking root mean square error MSE of measured value samples as optimizationmes、ELMANmes、RBFmes、GRNNmesAnd SVMmesSelecting 5 MSEs with the minimum fitness values as the optimal index set MSE of the iteration by comparing the fitness degrees among different populationsmAnd simultaneously recording the corresponding optimal feature dimension set NmAnd a corresponding parameter set Pm(ii) a If it is the first iteration, then MSEmAs a global optimum fitness valueMSEbest;
Step 2-3, using MSE selected in step 2-2mReplacing global optimal fitness value MSEbestRecording the corresponding optimal feature dimension, calculation model and model parameters;
step 2-4 comparison of MSEbestWhether the iteration frequency of the model reaches the maximum value is judged;
step 2-5, if the precision requirement cannot be met and the maximum iteration number is not reached, gradually expanding the feature dimension to select the lower limit of 1/N according to the modes of [1/2,1/4, … and 1/N ], returning to the step 2-1, and starting the next iterative computation; otherwise, executing the next step operation;
step 2-6, outputting the optimal model and corresponding model parameters, giving out a final flow calculation value, and ending the loop iteration;
step 3, updating parameters
With the continuous increase of the flow monitoring data and samples, when the monitored flow data reaches a certain scale or when new characteristic samples of the flow data appear, returning to the step 1 to re-rate and update the model.
2. The flow online monitoring and calculating method according to claim 1, characterized in that:
in step 1, when the total output number of the H-ADCP cells exceeds 50, performing feature dimension reduction processing on initialized data; and (5) taking the data after the feature dimension reduction processing as input data of the step 2.
3. The flow online monitoring and calculating method according to claim 1, characterized in that:
in step 3, the new characteristic sample of the flow data is a sample with a maximum flow value, a minimum flow value or a sudden change value.
4. An on-line flow monitor, comprising:
the data acquisition processing part is used for acquiring monitored section data, H-ADCP cell flow rate data and manual comparison data, and at least carrying out abnormal value screening, model input format sorting and data normalization preprocessing on the data;
a model initialization part which is communicated with the data acquisition processing part and at least takes the number of H-ADCP cells, the cell size, the model rate periodicity, the sample number of the inspection period and the initial parameters of each sub model as DCL model parameters to carry out initialization processing;
a feature learning fusion part, which is connected with the data acquisition processing part and the model initialization part in a communication way, firstly selects N H-ADCP cells corresponding to the sample (1/2) multiplied by N dimensions, if the dimension number is less than the model number NmodelsIs then set to Nmodels(ii) a And, the PSO initial population is randomly divided into 5 parts, and each part corresponds to: the neural network model comprises a back propagation neural network model BP, an Elman neural network model ELMAN, a radial basis function neural network model RBF, a generalized regression neural network model GRNN and a support vector machine model SVM, wherein one of the 5 models is selected; are respectively marked as BPinput、ELMANinput、RBFinput、GRNNinputAnd SVMinputAnd respectively inputting the data into respective models for calculation to obtain model output value BPoutput、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutput(ii) a Then, the output value samples BP of the 5 models are respectivelyoutput、ELMANoutput、RBFoutput、GRNNoutputAnd SVMoutputFitness function BP taking root mean square error MSE of measured value samples as optimizationmes、ELMANmes、RBFmes、GRNNmesAnd SVMmesSelecting 5 MSEs with the minimum fitness values as the optimal index set MSE of the iteration by comparing the fitness degrees among different populationsmAnd simultaneously recording the corresponding optimal feature dimension set NmAnd a corresponding parameter set Pm(ii) a If it is the first iteration, then MSEmAs global optimum fitness value MSEbest(ii) a Reuse of selected MSEmReplacing global optimal fitness value MSEbestRecording the corresponding optimal feature dimension, calculation model and model parameters; to pairSpecific MSEbestWhether the iteration frequency of the model reaches the maximum value is judged; if the precision requirement is not met and the maximum iteration number is not met, press [1/2,1/4, …,1/N]Gradually enlarging the feature dimension, selecting the lower limit of 1/n, and starting the next iterative computation; otherwise, outputting the optimal model and corresponding model parameters, giving out a final flow calculation value as monitoring flow data, and finishing the loop iteration;
the parameter updating part is in communication connection with the data acquisition processing part, the model initialization part and the feature learning fusion part, and returns to the data acquisition processing part again to re-rate and update the model when the monitored flow data reaches a certain scale or when a new feature sample of the flow data appears; and
and the control part is in communication connection with the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part and controls the operation of the data acquisition processing part, the model initialization part, the feature learning fusion part and the parameter updating part.
5. An online flow monitoring system, comprising:
at least one on-line flow monitor of claim 4; and
and the monitoring terminal is in communication connection with the flow online monitor, receives the flow data sent by the flow online monitor and displays the flow data for monitoring personnel to check.
6. The online flow monitoring system according to claim 5, characterized in that:
the monitoring terminal displays the flow data monitored by each online flow monitor at a corresponding river reach in a topographic map in real time, and displays the historical flow data of the corresponding river reach according to the monitoring instructions of the monitoring personnel.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010353603.5A CN111412959B (en) | 2020-04-29 | 2020-04-29 | Flow online monitoring calculation method, monitor and monitoring system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010353603.5A CN111412959B (en) | 2020-04-29 | 2020-04-29 | Flow online monitoring calculation method, monitor and monitoring system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111412959A CN111412959A (en) | 2020-07-14 |
CN111412959B true CN111412959B (en) | 2021-07-09 |
Family
ID=71492168
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010353603.5A Expired - Fee Related CN111412959B (en) | 2020-04-29 | 2020-04-29 | Flow online monitoring calculation method, monitor and monitoring system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111412959B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112491652B (en) * | 2020-11-18 | 2023-03-24 | 国家计算机网络与信息安全管理中心 | Network flow sample processing method and device for testing |
CN113639805B (en) * | 2021-10-14 | 2022-02-18 | 成都万江港利科技股份有限公司 | Flow measurement method based on channel section flow velocity field reconstruction |
CN116124234B (en) * | 2023-02-24 | 2023-07-18 | 宁波力擎超声科技有限公司 | Ultrasonic flowmeter for gas |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE3671516D1 (en) * | 1985-09-30 | 1990-06-28 | Siemens Ag | METHOD FOR MEASURING FLOW RATES WITH ULTRASONIC VIBRATIONS. |
JP2955920B2 (en) * | 1995-12-28 | 1999-10-04 | チャンミン カンパニー リミテッド | River flow measurement device and method |
CN201828294U (en) * | 2010-08-31 | 2011-05-11 | 宇星科技发展(深圳)有限公司 | Online monitoring system for river discharge |
CN106033000A (en) * | 2015-03-18 | 2016-10-19 | 西安山脉科技发展有限公司 | Method for estimating flow by means of radar wave flow meter |
CN104897248B (en) * | 2015-06-12 | 2018-04-27 | 天津大学 | Ultrasonic flowmeter propagation time method is accurately estimated under a kind of noise background |
CN104880227B (en) * | 2015-06-12 | 2018-04-27 | 天津大学 | Ultrasonic flow rate measuring method under a kind of noise background |
CN106706215B (en) * | 2016-11-17 | 2019-02-05 | 深圳市天成智能控制科技有限公司 | A kind of therrmodynamic system valves leakage quantity monitoring method |
CN107293115B (en) * | 2017-05-09 | 2020-09-08 | 上海电科智能系统股份有限公司 | Traffic flow prediction method for microscopic simulation |
CN110906992B (en) * | 2019-11-27 | 2021-12-24 | 长江水利委员会水文局 | River flow measuring method based on horizontal ADCP measuring vertical line flow velocity distribution |
-
2020
- 2020-04-29 CN CN202010353603.5A patent/CN111412959B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN111412959A (en) | 2020-07-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111412959B (en) | Flow online monitoring calculation method, monitor and monitoring system | |
CN111366123B (en) | Part surface roughness and cutter wear prediction method based on multi-task learning | |
Ajmera et al. | Development of stage–discharge rating curve using model tree and neural networks: an application to Peachtree Creek in Atlanta | |
CN109376925B (en) | Dynamic self-adaptive optimization method for node flow of water supply pipe network | |
Wang et al. | A Review of the Extrapolation Method in Load Spectrum Compiling. | |
CN102968529B (en) | Method for quantifying computed result non-determinacy interval of water supply pipe network model | |
Ismail et al. | A hybrid model of self organizing maps and least square support vector machine for river flow forecasting | |
CN103177289B (en) | Modeling method for noise-uncertainty complicated nonlinear dynamic system | |
CN112464584B (en) | Method for estimating water level and flow of free surface flow | |
CN111353652A (en) | Wind power output short-term interval prediction method | |
CN110929359A (en) | Pipe network siltation risk prediction modeling method based on PNN neural network and SWMM technology | |
JP7382522B1 (en) | Ultra-short-term water level prediction method, system, and storage medium for reservoirs considering the influence of dynamic water storage capacity | |
Doroudi et al. | Estimation of daily suspended sediment load using a novel hybrid support vector regression model incorporated with observer-teacher-learner-based optimization method | |
CN103218664A (en) | Warning weight determination method based on wavelet neural network | |
CN109165770A (en) | Novel photovoltaic power forecasting method based on AFSA-Elman | |
CN105568732A (en) | Disc mill control method | |
CN116865343B (en) | Model-free self-adaptive control method, device and medium for distributed photovoltaic power distribution network | |
CN116596165B (en) | Wind power generation power prediction method and system | |
Anilan et al. | Application of Artificial Neural Networks and regression analysis to L-moments based regional frequency analysis in the Eastern Black Sea Basin, Turkey | |
Chen et al. | Accounting information disclosure and financial crisis beforehand warning based on the artificial neural network | |
Bi et al. | A fault diagnosis algorithm for wind turbine blades based on bp neural network | |
CN113469504A (en) | Dynamic testing method, medium and system for water traffic risk | |
CN108364136B (en) | Water resource shortage risk analysis method and system based on evidence reasoning | |
CN112819378B (en) | River ecological flow regulation and control method based on pollution condition | |
Xu et al. | Improving the real-time probabilistic channel flood forecasting by incorporating the uncertainty of inflow using the particle filter |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210709 |
|
CF01 | Termination of patent right due to non-payment of annual fee |