CN115905869A - An Ultrasonic Water Meter Fault Early Warning Method - Google Patents

Abstract

The invention relates to the technical field of water meter fault early warning, and discloses an ultrasonic water meter fault early warning method, which comprises the following steps: acquiring normal operation data and fault data of the ultrasonic water meter to form a prediction training set of an LSTM prediction model, wherein the normal operation data is used for forming a classification training set of an OCSVM classification model; preprocessing and normalizing the ultrasonic water meter data, and then forming a training set by taking the data every 6 hours as a group; constructing a prediction model according to the prediction training set, and constructing a classification model according to the classification training set; and predicting the ultrasonic water meter data of the next 6 hours through the prediction model, sending the predicted data into the classification model for fault recognition, and reporting early warning information by the ultrasonic water meter if the fault is recognized. According to the method, after the data of the ultrasonic water meter within 6 hours is predicted through the LSTM model, the fault of the ultrasonic water meter is identified through the OCSVM model, so that the fault early warning of the ultrasonic water meter within 6 hours in advance is realized, and the economic loss is reduced.

Description

An Ultrasonic Water Meter Fault Early Warning Method

technical field

The invention relates to the technical field of water meter fault early warning, in particular to an ultrasonic water meter fault early warning method.

Background technique

With the advent of the Internet of Things era, traditional mechanical water meters are gradually being replaced by smart water meters. Among them, the ultrasonic water meter is a product that has attracted much attention in the smart water meter. The ultrasonic water meter detects the time difference caused by the change of the speed of the ultrasonic sound beam when it propagates downstream and upstream in the water, analyzes and calculates the flow velocity of the water, and further calculates the water flow rate. A new type of water meter for the flow rate. Due to the very harsh working environment of water meters, in practical applications, failures such as damage to electronic components, aging of water meter transducers, and damage to flow field structures will lead to a decrease in the measurement accuracy of water meters and bring economic losses to residents or water supply companies.

The existing diagnostic methods for ultrasonic water meter faults are all real-time or lagging. When the fault is diagnosed, the fault has already damaged the safe operation of the water meter and caused losses to users or water supply companies. At the same time, there are many types of faults in ultrasonic water meters, such as the abnormal operation of the test circuit caused by the undervoltage of the battery; the degradation of the measurement performance of the water meter due to the aging of the transducer; For these faults, the conventional fault detection algorithm is very prone to false negatives or false positives.

Contents of the invention

Aiming at the deficiencies and defects existing in the prior art, the present invention provides an ultrasonic water meter failure early warning method. After predicting the data of the ultrasonic water meter within 6 hours through the LSTM prediction model, the ultrasonic water meter failure is identified through the OCSVM classification model, thereby realizing Ultrasonic water meters give 6 hours early warning of faults to reduce economic losses.

The purpose of the present invention can be achieved through the following technical solutions:

An ultrasonic water meter failure early warning method, comprising the following steps:

s1, obtain the normal operation data and fault data of the ultrasonic water meter in the recent period T to form the prediction training set of the LSTM prediction model;

Obtain the normal operation data of the ultrasonic water meter in the recent period T to form the classification training set of the OCSVM classification model;

The above data includes the propagation time difference of upstream and downstream signals, water temperature, peak-to-peak value of upstream and downstream received signals, instantaneous flow rate, and battery voltage;

s2, preprocessing the acquired data;

Normalize the preprocessed data;

For the normalized data, the data of every 6 hours is used as a group to form a prediction training set and a classification training set;

s3, build an LSTM prediction model based on the prediction training set, and use the Adam optimization algorithm to optimize the parameters of the model;

s4, extracting classification training set features;

Build an OCSVM classification model;

Use the extracted feature values to train and optimize the OCSVM classification model;

s5, predict the ultrasonic water meter data for the next 6 hours through the LSTM prediction model;

s6, send the forecast data for the next 6 hours into the OCSVM classification model for fault identification, and if a fault is identified, the ultrasonic water meter will report early warning information.

Further, the data preprocessing flow in step s2 is:

s21, use the ratio threshold method to screen the data, that is, set the threshold range of the ratio R of the upstream and downstream signal propagation time difference to the instantaneous flow rate, and discard the data outside the threshold range of R;

s22, clean the data by moving average method, the formula is as follows:

Among them, x(n) is the data before cleaning; y(m) is the data after cleaning; n is the index of ultrasonic water meter data, and the step size is 5.

Further, in the step s4, a Gaussian radial basis kernel function is used to construct an OCSVM classification model.

Beneficial technical effects of the present invention: after the data of the ultrasonic water meter within 6 hours is predicted by the LSTM prediction model, the failure of the ultrasonic water meter is identified by the OCSVM classification model. Different from the real-time and hysteresis of the existing fault diagnosis, the present invention can report the water meter fault 6 hours in advance, effectively avoiding the loss and damage caused by the water meter fault. At the same time, there are many types of ultrasonic water meter faults, and it is not feasible to classify by extracting each fault feature. As a single classification support vector machine, the OCSVM classification model judges whether the data sample is normal data of the ultrasonic water meter, and then reports it, so fault identification The rate is high.

Description of drawings

Figure 1 is a schematic diagram of the method of the present invention.

Fig. 2 is a unit structure diagram of the LSTM prediction model according to the embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

As shown in Figures 1-2, an ultrasonic water meter fault early warning method includes the following steps:

s1: Obtain the normal operation data and fault data of the ultrasonic water meter in a community in the past three years to form the prediction training set of the LSTM prediction model, and obtain the data of the normal operation of the ultrasonic water meter in the community in the past three years to form the classification training set of the OCSVM classification model , Ultrasonic water meter data include upstream and downstream signal propagation time difference, water temperature, peak-to-peak value of upstream and downstream received signals, instantaneous flow, battery voltage;

s2: Preprocess the acquired data;

Normalize the preprocessed data;

For the normalized data, every 6 hours of data is used as a group to form a prediction training set and a classification training set; specifically:

s21: Use the ratio threshold method to screen the data, and the formula for calculating the ratio R is as follows:

Screen abnormal data by setting a threshold range Rmin≤R≤Rmax for R, where Rmin is the ratio when the flow rate is Q4, Rmax is the ratio when the flow rate is Q1, and discard data outside the threshold range of R;

s22: The ultrasonic water meter data is cleaned by the moving average method, and the calculation formula is as follows:

Among them, x(n) is the data before cleaning; y(m) is the data after cleaning; n is the index of ultrasonic water meter data, and the step size is 5.

s23: Normalize the preprocessed data, and perform linear transformation on the original data, the formula is as follows:

Among them, y is the preprocessed data, is the minimum value, is the maximum value.

s24: Every 6 hours of ultrasonic water meter data is used as a set of data to form a training set; each set of data contains 4320 data points.

s3: Construct the LSTM prediction model based on the prediction training set, and use the Adam optimization algorithm to optimize the parameters of the model:

s31: Select the initial network parameters to build the LSTM prediction model. After determining the input layer, hidden layer and output layer nodes of the model, send the normalized data into the model for training. The LSTM prediction model controls the transmission state through the sigmoid activation function. For data with extremely low water consumption, the network model will selectively "forget". This greatly compresses the network training time.

s32: Combined with the prediction training set, the Adam algorithm is used to optimize the network parameters.

s33: In order to more intuitively measure the error between the model prediction results and the actual results, the commonly used mean absolute error (mae) is used as the metric. The mean absolute value error formula is shown below, which is used to express the average value of the absolute error between the predicted value and the actual value:

Among them, y is the actual ultrasonic water meter data; Output the predicted value for the model; m is the number of samples.

s4: Perform feature extraction on the classification training set, build an OCSVM classification model, and perform fault classification and identification on the training set;

s41: Assume that the ultrasonic water meter data in one of the normal states is , t is taken as 500 in the embodiment. Preprocess the normal data, and extract the following three eigenvalues S, Perc, K to construct the feature set:

1) Standard deviation S:

in, is the data mean.

2) The percentage of fluctuation range to the average value Perc:

Among them, max(L) is the maximum value; min(L) is the minimum value.

3) Data change slope K of group t data:

s42: Select the Gaussian radial basis kernel function to establish the OCSVM classification model, and set the error penalty coefficient to 0.1; use the extracted feature values to train and optimize the model, and save the optimal model.

s43: In order to better evaluate the classification effect of the model, this embodiment uses two classification indicators, the precision rate P and the recall rate R. It is defined as:

Among them, TP indicates the number of positive samples identified as positive data, FP indicates the number of positive samples identified as negative, and FN indicates the number of negative samples identified as positive.

s5: Send the data within six hours before the current moment into the LSTM prediction model to predict the ultrasonic water meter data within the next six hours, and then perform denormalization processing.

s6: Perform feature extraction on the predicted data, send it to the OCSVM classification model for fault identification, and report if it is identified as a fault.

The foregoing embodiments are descriptions of specific implementations of the present invention, rather than limitations of the present invention. Those skilled in the art may also make various transformations and changes without departing from the spirit and scope of the present invention to obtain Corresponding equivalent technical solutions, therefore all equivalent technical solutions should fall into the patent protection scope of the present invention.

Claims (3)

1. An ultrasonic water meter fault early warning method is characterized by comprising the following steps:

s1, acquiring normal operation data and fault data of the ultrasonic water meter in a period of time T to form a prediction training set of an LSTM prediction model;

acquiring normal operation data of the ultrasonic water meter in a period of time T to form a classification training set of an OCSVM classification model;

the data comprises upstream and downstream signal propagation time difference, water temperature, upstream and downstream received signal peak-to-peak values, instantaneous flow and battery voltage;

s2, preprocessing the acquired data;

normalizing the preprocessed data;

taking the data after the normalization processing as a group to form a prediction training set and a classification training set every 6 hours;

s3, constructing an LSTM prediction model according to the prediction training set, and optimizing parameters of the model by using an Adam optimization algorithm;

s4, extracting the characteristics of the classification training set;

constructing an OCSVM classification model;

training and optimizing an OCSVM classification model by using the extracted characteristic values;

s5, predicting the ultrasonic water meter data of the next 6 hours through an LSTM prediction model;

and s6, sending the prediction data of the next 6 hours into an OCSVM classification model for fault recognition, and if the fault is recognized, reporting early warning information by the ultrasonic water meter.

2. The ultrasonic water meter fault early warning method according to claim 1, wherein the data preprocessing procedure in the step s2 is as follows:

s21, screening the data by using a ratio threshold method, namely setting a threshold range of a ratio R of the propagation time difference of the upstream and downstream signals to the instantaneous flow, and discarding the data outside the threshold range of R;

s22, cleaning the data by the moving average method, the formula is as follows:

wherein x (n) is data before cleaning; y (m) is data after washing; and n is the ultrasonic water meter data index, and the step length is 5.

3. The method for warning the fault on the ultrasonic water meter according to claim 1, wherein an OCSVM classification model is constructed by using a gaussian radial basis kernel function in the step s 4.

Images