CN103499374B - A kind of ultrasound wave dynamic liquid level detection method based on neutral net and system - Google Patents

Abstract

The invention discloses an ultrasonic dynamic liquid level detection method based on a neural network, comprising: a single-chip microcomputer sends a control signal to an ultrasonic sensor, and simultaneously starts a timer to start timing and generate pulsed ultrasonic waves; three ultrasonic sensors convert echo signals into voltage signals , and send the voltage signal to the preamplifier circuit, and the voltage signal amplified by the preamplifier circuit is sent to the ultrasonic receiving and detecting circuit to realize the envelope detection of the high-frequency input signal; the receiving comparison circuit will pass the envelope detecting circuit The detected voltage signal is compared with the reference voltage set by the system, and a digital signal is output according to the comparison result. When the external interrupt control port receives a valid falling edge, the single-chip microcomputer generates an interrupt, records the timer value at the current moment, and converts the average value and vibration The gradient is used as the input parameter required by the neural network, and these two parameters are sent to the trained neural network mathematical model to obtain the output value of the neural network.

Description

A neural network-based ultrasonic dynamic liquid level detection method and system

technical field

The invention relates to a detection field, in particular to an ultrasonic dynamic liquid level detection method and system based on a neural network.

Background technique

The liquid will visibly slosh during transportation, and the real-time dynamic monitoring of the liquid level has always been a technical problem, which has not been well solved so far.

In the prior art, there are many methods for detecting the liquid level in the transportation process, for example: a dynamic liquid level monitoring method based on a single ultrasonic sensor. Then use the wavelet filter method to smooth the time-of-arrival curve extracted from the echo signal, combine the motion acceleration of the moving object, the ambient temperature and other parameters, use the support vector machine technology to classify and identify the motion state and mode, and get the dynamic level value. However, this method also has shortcomings. First, the ultrasonic sensor is installed above the tank, and the sound energy attenuation of the ultrasonic wave in the air is large, which is not conducive to measurement; secondly, in order to obtain accurate liquid level parameters, the monitoring time window is long. Real-time performance deteriorates.

In the prior art, there is another dynamic monitoring system for the liquid level of sonar oil storage tanks. This system adopts the principle of active sonar technology, and the computer controls the sensor to emit sonar waves from the lower end of the tank roof fixing flange along the transmission cable. Downlink, according to the different physical and chemical characteristics of oil and water, the sonar will feed back different values, and the collected data will be sent to the acquisition module, and communicate with the host computer based on the RS485 protocol to realize real-time online monitoring and analysis, and display the dynamic liquid level value. The system also has shortcomings. First, the system is complex in design and low in integration, which is not conducive to portability; second, the sonar probe is installed on the roof, and the sound energy attenuation is relatively large.

Contents of the invention

In order to overcome the problem that the accuracy of liquid level detection during liquid transportation is not high and large errors are easily generated, the present invention proposes an ultrasonic dynamic liquid level detection method and system based on a neural network.

An ultrasonic dynamic liquid level detection method based on a neural network, comprising: Step 1: the single-chip microcomputer utilizes an ultrasonic transmitting circuit to send control signals to three ultrasonic sensors, and simultaneously starts a timer to start timing and excites the three-way ultrasonic transmitting circuit to generate pulsed ultrasonic waves; 2: The ultrasonic wave is reflected after meeting the liquid surface, and the three ultrasonic sensors respectively receive the reflected echo signal, convert the echo signal into a voltage signal, and send the voltage signal to the preamplifier circuit, step 3: before After the preamplifier circuit receives the voltage signal, the voltage signal amplified by the preamplifier circuit is sent to the ultrasonic receiving and detecting circuit to realize the envelope detection of the high frequency input signal; step 4: the voltage detected by the envelope detecting circuit The signal is sent to the receiving comparison circuit, and the receiving comparison circuit compares the voltage signal detected by the envelope detection circuit with the reference voltage set by the system, outputs a digital signal according to the comparison result, and sends the output signal to The external interrupt control port of the single-chip microcomputer, the external interrupt control port is set as a jump trigger, when the external interrupt control port receives a valid falling edge, the single-chip microcomputer generates an interrupt, records the timer value at the current moment, and the timer stops timing after the single-chip microcomputer interrupts three times; Step 5: By calculating the average value and shaking gradient of the timer values recorded three times, wherein the shaking gradient is the difference between the maximum value and the minimum value during the three measurements, the average value and shaking gradient are used as the input parameters required by the neural network, Send these two parameters to the trained neural network mathematical model, and obtain the output value through the neural network, the output value is the liquid level value at the current moment.

Further, the above-mentioned ultrasonic dynamic liquid level detection method based on the neural network also includes: the error function of the mathematical model of the neural network is:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

Among them, N is the number of training samples, y(t) is the output of the BP neural network at time t, and y _d (t) is the expected output value at time t; when using the BP algorithm to train the neural network, adjust the connection weight of the network, and its adjustment expression The formula is:

$\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}$

Among them, w(t) is the connection weight of the neural network at time t, w(t+1) is the connection weight of the neural network at time (t+1), and E(t) is the mean square error of the neural network at time t , η is the learning rate.

Further, the above-mentioned ultrasonic dynamic liquid level detection method based on neural network adopts three ultrasonic sensors to form an array structure, synchronously stimulates and emits pulsed ultrasonic waves, and uses the envelope of echo signals received in a fixed-length time window to extract the instantaneous liquid surface reflection point. Latency information.

An ultrasonic dynamic liquid level detection system based on a neural network, including: a single-chip microcomputer, a conditioning circuit and three ultrasonic sensors; wherein, the single-chip microcomputer is used to load the mathematical model of the neural network, and the mathematical model based on the neural network controls and processes the ultrasonic dynamic liquid level. Bit detection; the conditioning circuit includes six parts: ultrasonic transmitting circuit, ultrasonic receiving preamplifier circuit, ultrasonic receiving signal amplifier circuit, ultrasonic receiving detection circuit, ultrasonic receiving comparison circuit and single-chip peripheral circuit. The ultrasonic transmitting circuit is used to generate pulsed ultrasonic waves. It is controlled by a single-chip microcomputer to realize three-way simultaneous transmission of signals, generating high-voltage pulses with a frequency similar to that of the ultrasonic sensor probe, and stimulating the ultrasonic sensor to transmit pulsed ultrasonic waves at the same time; the ultrasonic receiving preamplifier circuit is composed of a charge amplifier to achieve impedance matching with the ultrasonic receiving transducer The ultrasonic receiving signal amplifying circuit is composed of two-stage inverting amplifiers, which realize the amplification of the weak ultrasonic echo voltage signal; the ultrasonic receiving and detecting circuit is composed of a diode peak envelope detecting circuit, which realizes the envelope detecting of the high-frequency input voltage signal; The ultrasonic receiving comparison circuit compares the voltage signal detected by the envelope detection circuit with the reference voltage set by the system, outputs a digital signal according to the comparison result, and sends the output signal to the external interrupt control port of the microcontroller after being reversed by the inverter , the external interrupt control port is set as a jump trigger, when the external interrupt control port receives a valid falling edge, the microcontroller generates an interrupt, records the timer value at the current moment, and the timer stops timing after the microcontroller interrupts three times; the peripheral circuit of the microcontroller is composed of a crystal oscillator circuit , reset circuit, and liquid crystal display circuit, which together constitute the core control part of dynamic liquid level detection.

Further, the error function of the mathematical model of the neural network loaded by the above system is:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

Among them, N is the number of training samples, y(t) is the output of BP neural network at time t, and y _d (t) is the expected output value at time t;

When using the BP algorithm to train the neural network, adjust the connection weight of the network, and the adjustment expression is:

$\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}$

Among them, w(t) is the connection weight of the neural network at time t, w(t+1) is the connection weight of the neural network at time (t+1), and E(t) is the mean square error of the neural network at time t , η is the learning rate.

Furthermore, the above system uses three ultrasonic sensors to form an array structure, synchronously stimulates and emits pulsed ultrasonic waves, and uses the envelope of echo signals received in a fixed-length time window to extract time-delay information of instantaneous liquid surface reflection points.

Further, the single-chip microcomputer in the above-mentioned system is an AT89S52 single-chip microcomputer.

Combining the above two technical solutions, the present invention proposes an ultrasonic dynamic liquid level detection method and system based on a neural network, which can realize dynamic liquid level detection.

Description of drawings

Fig. 1 is a schematic diagram of an array of ultrasonic sensors in an embodiment of the present invention;

Fig. 2 is the principle schematic diagram of detection method in the embodiment of the present invention;

Fig. 3 is a group of ultrasonic sensor echo signals in the embodiment of the present invention;

Fig. 4 is the training error curve in the embodiment of the present invention;

Fig. 5 is the detection result of the neural network in the embodiment of the present invention.

detailed description

The technical scheme of the present invention is described in detail below in conjunction with accompanying drawing:

First, introduce the BP neural network. The BP neural network is also called the error backpropagation neural network. In the BP network, the signal is propagated forward, and the error is propagated backward. In the process of forward propagation, the input signal is processed layer by layer through the input layer and the hidden layer, and then transmitted to the output layer. If the signal cannot get the expected output at the output layer, it will be transferred to the back propagation process, and the error value will be reversed layer by layer, and the connection weights of each layer will be corrected. For a given set of training samples, the network is continuously trained with a pattern, and the process of forward propagation and error back propagation is repeated until the network output error is less than a given value. The BP algorithm changes the weights and biases along the opposite direction of the gradient. Define the error function as:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; d</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

Among them, N is the number of training samples, y(t) is the output of BP neural network at time t, and y _d (t) is the expected output value at time t.

When using the BP algorithm to train the network, adjust the connection weight of the network, and the adjustment expression is:

$\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}$

Among them, w(t) is the connection weight of the neural network at time t, w(t+1) is the connection weight of the neural network at time (t+1), and E(t) is the mean square error of the neural network at time t , η learning rate.

The BP algorithm of error backpropagation is a relatively simple and practical learning algorithm. The standard BP algorithm usually has a slow learning convergence speed and is easy to fall into a local minimum.

Introduce the ultrasonic dynamic liquid level detection method of the present invention below, as shown in Figure 1, the present invention proposes a kind of ultrasonic dynamic liquid level detection method based on neural network, comprising:

In the embodiment of the present invention, an array composed of three ultrasonic sensors is used to monitor the dynamic liquid level, and the layout of the sensor array is shown in FIG. 1 . Among them, O is the center position of the bottom of the container, A, B, and C are the probes of the three ultrasonic sensors, and a, b, and c are the liquid surface reflection points corresponding to the ultrasonic sensors A, B, and C.

The system of ultrasonic dynamic liquid level detection in the implementation of the present invention is shown in Figure 2, which includes a single-chip microcomputer, a conditioning circuit and three ultrasonic sensors A, B, and C. Wherein the single-chip microcomputer may be AT89S52 single-chip microcomputer, which is used for controlling and processing ultrasonic dynamic liquid level detection. The conditioning circuit consists of six parts: ultrasonic transmitting circuit, ultrasonic receiving preamplifier circuit, ultrasonic receiving signal amplifier circuit, ultrasonic receiving detection circuit, ultrasonic receiving comparison circuit and single-chip peripheral circuit. Specifically: the ultrasonic transmitting circuit is used to generate pulsed ultrasonic waves, It is controlled by a single-chip microcomputer to realize three-way simultaneous transmission of signals, generating high-voltage pulses with a frequency similar to that of the ultrasonic sensor probe, and stimulating the ultrasonic sensor to transmit pulsed ultrasonic waves at the same time; the ultrasonic receiving preamplifier circuit is composed of a charge amplifier to achieve impedance matching with the ultrasonic receiving transducer ;The ultrasonic receiving signal amplification circuit is composed of two-stage inverting amplifiers, which realize the amplification of weak ultrasonic echo signals; The comparison circuit compares the pulse signal detected by the envelope detection circuit with the reference voltage set by the system, the comparator will output a digital signal 1 or 0, and the output signal of the comparator is reversed by the inverter and then sent to the external interrupt of the microcontroller Control port P3.2; the peripheral circuit of the single-chip microcomputer is composed of crystal oscillator circuit, reset circuit and liquid crystal display circuit, which together constitute the core control part of dynamic liquid level detection.

The principle of the above-mentioned ultrasonic dynamic liquid level detection method based on neural network is as follows: during detection, the three ultrasonic sensors A, B, and C emit pulsed ultrasonic waves at the same time, but since the liquid level heights corresponding to the probes of the three ultrasonic sensors are different, each The echo time received by each ultrasonic sensor is also different. According to the echo signal received by each ultrasonic sensor, the time used for the sound wave signal of the three probes to go back and forth can be calculated separately, and then the time corresponding to the probe of each ultrasonic sensor can be calculated. The liquid level height, and select the average value of the liquid level height of the three probes from the liquid surface and the sloshing gradient as the characteristic parameters to train the BP neural network to obtain the mathematical model of the dynamic liquid level, and then apply the model to the real-time dynamic liquid level monitoring. Since the three probes are respectively arranged in three quadrants with the center of the bottom of the container as the origin, no matter how the liquid shakes, the liquid level measured by the three probes always has a maximum value and a minimum value, and the difference between the maximum value and the minimum value can be determined Liquid sloshing gradient.

The specific process of ultrasonic dynamic liquid level detection based on neural network is as follows:

Step 1: The single-chip microcomputer uses the ultrasonic transmitting circuit to send control signals to the three ultrasonic sensors, and at the same time starts the timer to start timing and stimulates the three-way ultrasonic transmitting circuit to generate pulsed ultrasonic waves;

Step 2: The ultrasonic waves are reflected after meeting the liquid surface, and the three ultrasonic sensors respectively receive the reflected echo signals, convert the echo signals into voltage signals, and send the voltage signals to the preamplifier circuit;

Step 3: After the pre-amplification circuit receives the voltage signal, the voltage signal amplified by the pre-amplification circuit is sent to the ultrasonic receiving and detecting circuit to realize the envelope detection of the high-frequency input signal;

Step 4: Send the voltage signal detected by the envelope detection circuit to the receiving comparison circuit, and the receiving comparison circuit compares the voltage signal detected by the envelope detection circuit with the reference voltage set by the system, and outputs a digital signal according to the comparison result, And the output signal is reversed by the inverter and then sent to the external interrupt control port of the single chip microcomputer. The external interrupt control port is set as a jump trigger. When the external interrupt control port receives a valid falling edge, the single chip microcomputer generates an interrupt and records the current time. Timer value, the timer stops counting after the microcontroller is interrupted three times;

Step 5: By calculating the average value and shaking gradient of the timer values recorded three times, wherein the shaking gradient is the difference between the maximum value and the minimum value during the three measurements, the average value and shaking gradient are used as the input parameters required by the neural network, Send these two parameters to the trained neural network mathematical model, and obtain the output value through the neural network, the output value is the liquid level value at the current moment. Then, the above-mentioned process is continuously circulated, and the system can realize real-time dynamic monitoring of the sloshing liquid. This system can also realize the monitoring of the static liquid level. The gradient of static liquid sloshing is small, close to zero, and the average value of three-point measurements is taken as the static liquid level value.

In a specific embodiment of the present invention, preferably, the ultrasonic sensors all adopt a transceiver-integrated type, and the probe frequency is 1 MHz. Under laboratory conditions, shake the water in the iron tank by hand to simulate the sloshing of the liquid during transportation. Figure 3 shows the different echo signals of the three ultrasonic probes measured in the experiment when the sloshing liquid is measured, where (a), (b), (c) are the echo signals measured by sensors A, B, and C respectively. During implementation, 100 sets of data measured from the liquid level from 2cm to 50cm are used as training samples, and ten sets of data are drawn within this range as test samples.

The BP neural network in the embodiment of the present invention has two input parameters and one output parameter, then the input layer of the BP network is set as 2 neurons, the hidden layer is set as one layer, and the output layer is 1 neuron. During implementation, the mean square error of the neural network is set to 1×10 ^-5 , the learning rate is 0.05, and the maximum number of training steps is 500. It can be seen from experiments that when the number of neurons in the hidden layer is 8, the number of training times is the least. Therefore, a network of 2×8×1 is selected for training. The transfer function of the hidden layer of the network is tansig, and the output layer is purelin. In order to make the network converge quickly, the learning algorithm of Levenberg-Marquardt is adopted. It can be seen from Figure 4 that after the network has been trained for 90 times, the mean square error of the network meets the design requirements, and the network stops training.

Input 10 groups of test samples into the trained BP neural network, and the test results are shown in Figure 5. It can be seen from Figure 5 that the overall relative error of the test sample is small, the accuracy of the network output is high, and the output result is stable. Although the network training takes a certain amount of time, after the network training is completed, the time required for practical application is extremely short, and the system can completely monitor the dynamic liquid level.

Claims (7)

1. the ultrasound wave dynamic liquid level detection method based on neutral net, it is characterised in that including:

Step 1: single-chip microcomputer utilizes ultrasonic emitting circuit to send control signal to three ultrasonic sensors, starts intervalometer simultaneously and starts timing and excitation three tunnel ultrasonic emitting circuit generation pulse ultrasonic wave;

Step 2: ultrasound wave meets liquid level back reflection, after three ultrasonic sensors are respectively received the echo-signal after reflection, and echo-signal is converted into voltage signal, and this voltage signal is sent to pre-amplification circuit;

Step 3: after pre-amplification circuit have received voltage signal, the voltage signal after pre-amplification circuit amplifies is sent to ultrasound wave and receives detecting circuit, it is achieved the envelope detection to high-frequency input signal;

Step 4: the voltage signal after envelope detection circuit detection is sent to reception comparison circuit, receive comparison circuit to be compared by the reference voltage that the voltage signal after envelope detection circuit detection and system are arranged, according to comparative result output digit signals, and by export the inverted device of signal reversely after give the external interrupt control mouth of single-chip microcomputer, external interrupt control mouth is set to saltus step to be triggered, after external interrupt control mouth receives effective trailing edge, single-chip microcomputer produces to interrupt, the timer value of record current time, after singlechip interruption three times, intervalometer stops timing;

Step 5: by calculating the meansigma methods of the timer value of three records and rocking gradient, wherein rock the difference of liquid level maxima and minima when gradient is record for three times, using meansigma methods with rock gradient as the input parameter needed for neutral net, the two parameter is sent to the neutral net mathematical model trained, and obtaining the output valve through neutral net, output valve is the level value of current time.

2. method as claimed in claim 1, it is characterised in that the error function of the mathematical model of neutral net is:

Wherein, N is number of training, and y (t) exports for t BP neutral net, y_dT () is t desired output; During by BP Algorithm for Training neutral net, adjusting the connection weights of network, its adjustment expression formula is:

Wherein, the connection weights of the neutral net that w (t) is t, w (t+1) is the connection weights of t+1 moment neutral net, and E (t) is the mean square error of t neutral net, and η is learning rate.

3. method as claimed in claim 1, it is characterised in that adopt three ultrasonic sensor composition array structures, synchronization motivationtheory emission pulse ultrasonic, utilizes the Delay of the window interior echo signal envelope extraction instantaneous liquid level pip received during fixed length.

4. the ultrasound wave dynamic fluid flow position detecting system based on neutral net, it is characterised in that including: single-chip microcomputer, modulate circuit and three ultrasonic sensors;

Wherein, single-chip microcomputer is for loading the mathematical model of neutral net, and the mathematical model based on neutral net controls and processes the detection of ultrasound wave dynamic liquid level; Modulate circuit includes ultrasonic emitting circuit, ultrasound wave receives pre-amplification circuit, ultrasound wave receives signal amplification circuit, ultrasound wave receives detecting circuit, ultrasound wave receives comparison circuit and SCM peripheral circuit six part composition, ultrasonic emitting circuit is used for producing pulse ultrasonic wave, realized three roads by Single-chip Controlling and launch signal simultaneously, produce the high-voltage pulse close with ultrasonic sensor probe frequency, excitation sonac emission pulse ultrasonic simultaneously; Ultrasound wave receives pre-amplification circuit and is made up of charge amplifier, it is achieved with the impedance matching of ultrasonic reception transducer; Ultrasound wave receives signal amplification circuit and is made up of two-stage inverting amplifier, it is achieved the amplification to Weak Ultrasonic echo voltage signal; Ultrasound wave receives detecting circuit and is made up of diode peak envelope detection circuit, it is achieved the envelope detection to high-frequency input voltage signal; Ultrasound wave receives comparison circuit and is compared by the reference voltage that the voltage signal after envelope detection circuit detection and system are arranged, according to comparative result output digit signals, and by export the inverted device of signal reversely after give the external interrupt control mouth of single-chip microcomputer, external interrupt control mouth is set to saltus step to be triggered, after external interrupt control mouth receives effective trailing edge, single-chip microcomputer produces to interrupt, the timer value of record current time, and after singlechip interruption three times, intervalometer stops timing; SCM peripheral circuit is made up of crystal oscillating circuit, reset circuit, liquid crystal display circuit, collectively forms the core control portions of dynamic liquid level detection.

5. system as claimed in claim 4, it is characterised in that the error function of the mathematical model of neutral net is:

Wherein, N is number of training, and y (t) exports for t BP neutral net, y_dT () is t desired output;

During by BP Algorithm for Training neutral net, adjusting the connection weights of network, its adjustment expression formula is:

Wherein, the connection weights of the neutral net that w (t) is t, w (t+1) is the connection weights of t+1 moment neutral net, and E (t) is the mean square error of t neutral net, and η is learning rate.

6. system as claimed in claim 4, it is characterised in that adopt three ultrasonic sensor composition array structures, synchronization motivationtheory emission pulse ultrasonic, utilizes the Delay of the window interior echo signal envelope extraction instantaneous liquid level pip received during fixed length.

7. system as claimed in claim 4, it is characterised in that single-chip microcomputer is AT89S52 single-chip microcomputer.