A water level measurement method for water collection wells based on twin networks

Abstract

A twin network-based method for measuring water level in a water collection well includes an audio generation module, an audio acquisition module, and an audio processing module. A cylinder with the audio generation and acquisition modules mounted on top is placed in the water collection well. The audio generation module generates a fixed-frequency audio signal, which is reflected from the water surface at the bottom and superimposed into a new audio signal that is collected by the audio acquisition module. The twin network model within the audio processing module identifies the collected audio signal and determines the current water level in the water collection well, thereby enabling control and early warning of the water level. This method addresses the issues of zero drift and inconvenient installation and maintenance associated with submersible pressure sensors, which can occur over long periods of time.

Description

Technical Field

[0001] The invention relates to a water level measurement method for a water collection well based on a twin network. Background Art

[0002] At present, the water level control of the water collection well is that the PLC controller collects the 4-20mA current value output by the pressure sensor in real time, converts it into the water level engineering value, and starts and stops the pump according to the set value, thereby controlling the water level of the water collection well.

[0003] The drawbacks are: 1. Submersible pressure sensors can inevitably experience zero drift during use due to blockage of the pressure tapping hole, corrosion of the pressure measuring element, and other factors, leading to measurement errors. 2. Submersible pressure sensors are typically tied to a wire rope with a weight hanging from the bottom, making replacement both time-consuming and labor-intensive. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a water level measurement method for a water collection well based on a twin network, which is used to solve the problems of zero drift and inconvenience in installation and maintenance of immersion pressure sensors during long-term use.

[0005] In order to solve the above problems, the technical solution of the present invention is:

[0006] A water level measurement method for a water collection well based on a twin network includes an audio generation module, an audio acquisition module and an audio processing module. A cylinder with the audio generation module and the audio acquisition module installed on the top is placed in the water collection well. The audio generation module generates an audio signal of a fixed frequency, which is reflected by the bottom water surface and superimposed into a new audio signal, which is collected by the audio acquisition module. The collected audio signal is identified by the twin network model in the audio processing module to determine the current water level of the water collection well, thereby realizing control and early warning of the water level of the water collection well.

[0007] The beneficial effects of the present invention are: the use of a non-contact water level measurement method effectively reduces errors caused by various environmental interference factors, improves the accuracy of water level measurement, and reduces the risk of factory flooding. Submersible water level gauges are complex to replace and require personnel to climb into the water collection well, which poses certain safety risks. The present invention can effectively reduce the risks of falling from height, drowning, and inhaling toxic gases. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present invention will be further described below with reference to the accompanying drawings:

[0009] Figure 1 This is the twin network architecture diagram of the present invention.

[0010] Figure 2 This is a flow chart of water level measurement of water collection well based on twin network of the present invention.

[0011] Figure 3 This is the structural diagram of the water level measurement system of the water collection well based on the twin network of the present invention.

[0012] Figure 4 It is a schematic structural diagram of the present invention when it is implemented.

[0013] In the figure: water collection well 1, cable 2, audio generation module 3, audio collection module 4, cylinder 5. DETAILED DESCRIPTION

[0014] like Figures 1 to 4 As shown, a twin network-based method for measuring water level in a water collection well includes an audio generation module, an audio acquisition module, and an audio processing module. A cylinder with the audio generation and acquisition modules mounted on top is placed in the water collection well. The audio generation module generates a fixed-frequency audio signal, which is reflected by the water surface at the bottom and superimposed into a new audio signal that is collected by the audio acquisition module. The cylinder consists of two layers, an inner and outer wall, with a vacuum between them. A through-hole is provided at the top for cable passage and ventilation.

[0015] Fluctuations in the water level in the collection well cause changes in the height of the air column inside the cylinder, which in turn causes changes in the frequency of the collected audio signal. The twin network model within the audio processing module recognizes the collected audio signal and determines the current water level in the collection well, enabling control and early warning of the water level.

[0016] Deep neural network models often require large amounts of training data to train the model, and when the data is insufficient, they will fail. However, by using two networks with the same architecture, the Siamese network finds similarities between two input values ​​to learn, using fewer sample points to train the model and complete related tasks.

[0017] Step 1: Data acquisition and processing. Install the audio generation module and acquisition module inside a cylinder. The cylinder adopts a double-layer structure with a vacuum in the middle to effectively avoid signal interference. Use a fixed guide rail that can move up and down to place the cylinder in the water collection well. The acquisition module collects background environmental noise, determines its frequency range, and selects frequency signals outside the background environmental noise frequency range as the audio signal of the audio generation module. Move the cylinder up and down to change the height of the air column to simulate the change of the water level in the water collection well. Collect two audio signals every 0.1m.

[0018] Step 2: Create sample pair data. The audio signal is digitized, pre-filtered, pre-emphasized, endpoint detected, framed, and windowed to make the signal features more distinct and remove redundant data. The audio signal is converted from the time domain to the frequency domain using the Fast Fourier Transform (FFT). The audio signal is converted into a short-term power spectrum within the nonlinear Mel frequency range using the linear cosine transform of the Mel-frequency cepstral coefficients (MFCCs). The short-term power spectrum features of two audio signals at the same water level are combined into a positive sample pair and labeled as 1. The short-term power spectrum features of two audio signals at different water levels are combined into a negative sample pair and labeled as 0. The sample pair data is divided into 75% for model training and 25% for accuracy testing.

[0019] Step 3: Build a twin network. Create a neural network A with three dense layers and a dropout layer in the middle: And neural network B: , and they have the same weight w , this neural network is only used for feature vector extraction. When the model has many parameters and a small number of samples, the model is prone to overfitting, that is, the prediction accuracy is very high on the training set, but the accuracy drops significantly on the test set. The dropout layer randomly shuts down some neurons during model training, which can effectively avoid overfitting and keep the accuracy of the training set and test set consistent. Using Euclidean distance: ; As an energy function, it is used to determine the similarity of two inputs. X 1 and X 2 similar, The value is smaller, otherwise The value is large, such as Figure 1 shown.

[0020] Step 4: Model training. The positive and negative sample pairs generated in step 2 are used as the input of the twin network. X 1. X 2. Perform model training, select 10 iterations, and use the RMSProp algorithm for optimization. Define the contrast loss function ; Y Is the label of whether the two inputs are similar, 1 if they are similar and 0 if they are not similar; m is the lower limit of the Euclidean distance of dissimilar samples, that is, the Euclidean distance> m No loss will be incurred. Observe whether the contrast loss function converges to the expected value. If so, complete the training. If not, increase the number of iterations to 12 until convergence and complete the training.

[0021] Step 5: Model accuracy test. Input the accuracy test data generated in step 2 into the trained model in step 4 to test whether the model accuracy meets the requirements. If not, repeat step 4, adjust the model parameters, add more training data, and retrain the model until it meets the requirements.

[0022] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A water level measurement method for a water collection well based on a twin network, characterized by: The system includes an audio generation module, an audio acquisition module, and an audio processing module. A cylinder with the audio generation module and the audio acquisition module installed on top is placed in a water collection well. The audio generation module generates an audio signal of a fixed frequency, which is reflected by the water surface at the bottom and superimposed into a new audio signal that is collected by the audio acquisition module. The collected audio signal is recognized by the twin network model in the audio processing module to determine the current water level in the water collection well, thereby achieving control and early warning of the water level in the water collection well. The twin network uses two networks with the same architecture. The training method of the twin network model includes the following steps: Step 1: data collection and processing; collecting background environmental noise, determining its frequency range, and selecting frequency signals outside the background environmental noise frequency range as audio signals of the audio generation module; Step 2: Create sample pair data; process the audio signal and remove redundant data; use fast Fourier transform to convert the audio signal from the time domain to the frequency domain; convert the audio signal into a short-term power spectrum within the nonlinear Mel frequency range through linear cosine transform of Mel-frequency cepstral coefficients (MFCCs); combine the short-term power spectrum features of two audio signals at the same water level into a positive sample pair and set the label to 1; combine the short-term power spectrum features of two audio signals at different water levels into a negative sample pair and set the label to 0; Step 3: Build a twin network; create a neural network A with multiple dense layers and a dropout layer in the middle: And neural network B: , and they have the same weight w ,The neural network is used for feature vector extraction; Step 4: Model training; use the positive and negative sample pairs generated in step 2 as the input of the twin network X 1. X 2. Perform model training, select 10 iterations, and use RMSProp algorithm for optimization; define the contrast loss function Where is the Euclidean distance Y Is the label of whether the two inputs are similar, 1 if they are similar and 0 if they are not similar; m is the lower limit of the Euclidean distance of dissimilar samples, that is, the Euclidean distance> m No loss will be caused; observe whether the contrast loss function converges to the expected value. If it converges, the training is completed. If it does not converge, increase the number of iterations to 12 until it converges and the training is completed. Step 5: Model accuracy test: input the accuracy test data generated in step 2 into the model trained in step 4 to test whether the model accuracy meets the requirements. If not, repeat step 4, adjust the model parameters, increase the training data, and retrain the model until it meets the requirements.

2. The method for measuring water level in a water collection well based on a twin network according to claim 1, characterized in that: Step 3 also includes using Euclidean distance when the model is overfitting: As an energy function, it is used to determine the similarity of two inputs. X 1 and X 2 similar, The value is small, otherwise The value is larger.

3. The method for measuring water level in a water collection well based on a twin network according to claim 1, characterized in that: In step 2, the sample data is divided into 75% for model training and 25% for accuracy testing.

Images