An adaptive adjustment method for the motor for opening and closing ball valves based on neural network
Through the adaptive adjustment method based on neural network, the activation function is constructed using super-geometric series and Gaussian function to generate a sine wave adjustment signal, which solves the overload or undervoltage problems caused by voltage fluctuations of the electric ball valve, and improves the adjustment accuracy and stability , improves the safety and reliability of the equipment.
Patent Information
- Application Number
- CN202411151755.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-21
AI Technical Summary
The existing electric ball valves are more sensitive to voltage fluctuations during the adjustment process, which can easily lead to overload or undervoltage of the motor. The control method mainly relies on simple adjustment of feedback data, and cannot be effectively processed, affecting the normal operation of the valve.
Adaptive adjustment method based on neural network is adopted to collect motor running parameters in real time, perform data preprocessing and Fourier transform, and use super-geometric series and Gaussian function to construct activation functions to generate sine wave adjustment signals to achieve precision of the motor. adjust.
It improves the adjustment accuracy and stability of the motor, reduces energy consumption, improves the safety and reliability of the equipment, and realizes the adaptive adjustment function.
Smart Images

Figure CN118659704B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of ball valve automation control, in particular to a self-adaptive adjustment method of a ball valve starting and closing motor based on a neural network. Background Art
[0002] A ball valve is a kind of opening and closing part, which is driven by a valve stem and rotates around the axis of the ball valve. It can also be used for fluid regulation and control. The hard-sealed V-type ball valve has a strong shear force between its V-type ball core and the metal valve seat with hard alloy overlay, which is particularly suitable for media containing fibers, tiny solid particles, etc. The multi-way ball valve can not only flexibly control the confluence, diversion, and flow direction switching of the medium on the pipeline, but also close any channel to connect the other two channels. This type of valve should generally be installed horizontally in the pipeline. Ball valves are divided into: pneumatic ball valves, electric ball valves, and manual ball valves according to the driving method.
[0003] Electric ball valve is an important actuator in industrial automation control system. Electric ball valve is a valve that rotates around the axis of valve stem according to the movement of valve disc. The change of valve seat opening is proportional to the valve disc stroke. It is mainly used to cut off or connect the medium in the pipeline, and can also be used for fluid regulation and control. It is a pipeline pressure element for industrial automation process control.
[0004] The electric ball valve basically adopts simple PWM output, and performs simple PID control on the motor by adjusting the voltage. The motor is sensitive to voltage fluctuations, which may cause motor overload or undervoltage. The PWM output waveform is basically a square wave, with severe steep rises and falls, thus affecting the normal operation of the valve.
[0005] The existing intelligent ball valve controls the motor by adding data collected by sensors such as flow, pressure, and temperature, and makes precise adjustments to the medium flow, but the adjustments are limited to the feedback data and the feedback data cannot be processed. Summary of the invention
[0006] The purpose of the present invention is to provide an adaptive adjustment method for a ball valve opening and closing motor based on a neural network to solve the problems raised in the above-mentioned background technology.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] A method for adaptively adjusting a ball valve start-and-close motor based on a neural network comprises the following steps:
[0009] Step S1, real-time acquisition of motor running parameters, including but not limited to current, speed and load;
[0010] Step S2, preprocessing the collected data to make the collected data meet the input requirements of the neural network;
[0011] Step S3, performing Fourier transform on the preprocessed data, converting the time domain data into frequency domain data, and optimizing the neural network based on the frequency domain data;
[0012] Step S4, obtaining the weight parameters of the neural network by calculating the important integral of the hypergeometric series;
[0013] Step S5, performing limit check on the normal bell curve using a high power approximation method of a Gaussian function;
[0014] Step S6, generating a new activation function through fitting and interpolation algorithms, and applying it to the neural network;
[0015] Step S7, generating a sinusoidal wave adjustment signal as needed to precisely adjust the motor.
[0016] Furthermore, in step S1, parameters of the motor during operation are collected through sensors, and the sensors include but are not limited to current sensors, speed sensors, and load sensors.
[0017] Furthermore, the preprocessing in step S2 includes noise removal, normalization and interception of a time period of interest.
[0018] Furthermore, in step S3, the result of the Fourier transform is used to extract the spectral features of the motor operation, and the extracted spectral features are converted into feature vectors, and the feature vectors are input into the neural network training, and the neural network is optimized according to the training results, wherein the feature vector is a one-dimensional array for a single sample and a two-dimensional array for multiple samples.
[0019] Furthermore, the important integral calculation formula for the hypergeometric series in step S4 is:
[0020] ;
[0021] in is a hypergeometric function, x is a variable, m is an integer parameter, and m is greater than or equal to 2;
[0022] Then perform two Taylor series expansions on it, and the series expansion of the integral when x=0 is:
[0023] ;
[0024] ;
[0025] Where a is , b is , c is , z is .
[0026] Furthermore, in step S6, the neural network adopts one of a feedforward neural network, a convolutional neural network and a recurrent neural network.
[0027] Furthermore, in step S6, the process of training the neural network includes adjusting weight parameters in the network, and using gradient descent to adjust and optimize the weight parameters of the neural network so that the output of the network reaches a predetermined performance indicator.
[0028] Furthermore, the method of optimizing the weight parameters of the neural network is to generate a regulation signal according to the gradient descent adjustment rule of the weight parameters, and the generation algorithm is as follows:
[0029] Loss function L for weight parameters The gradient is: ;
[0030] Conditioning signal: ;
[0031] in is the update amount of the weight parameter, is the learning rate;
[0032] The frequency, amplitude and phase of the regulating signal are adjusted according to the actual dynamic requirements of the motor.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] 1. In the convolutional neural network of the present invention, the calculation of weight parameters is necessary, and the weight parameters are closely related to the probability distribution. Therefore, the Bell function can accurately construct different weight ratios. The present invention uses hypergeometric series, also known as hypergeometric functions, to construct bell-shaped functions and activation functions, which have strong universality and robustness. Through this equation, it can be converted into the sum of discrete data, which can be widely used in programming, such as smoothing filtering, low-pass filtering, fitting, interpolation, and even various types of neural networks;
[0035] 2. The present invention can be widely used in various types of equipment that require adaptation, so that individual equipment can also have the function of autonomous learning and autonomous training through data accumulation to achieve adaptive adjustment, such as widely used valves, frequency converters, etc., thereby effectively reducing energy consumption, improving efficiency, and enhancing the safety and reliability of equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 It is a schematic flow chart of a method for adaptively adjusting a motor for starting and closing a ball valve based on a neural network according to the present invention;
[0037] Figure 2 The bell-shaped line graph is obtained by performing limit check on the weight parameter in step S4 using the high power approximation method of the Gaussian function in the present invention;
[0038] Figure 3 It is a bell-shaped line graph of a bell-shaped function obtained by taking the derivative of the activation function when m=5 in the present invention;
[0039] Figure 4 It is a schematic diagram of converting symmetric functions and polynomials into bell-shaped functions according to the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] See also Figure 1-4 , the present invention provides the following embodiments:
[0042] A method for adaptively adjusting a ball valve start-and-close motor based on a neural network comprises the following steps:
[0043] Step S1, real-time collection of motor operating parameters, including but not limited to current, speed and load, the motor operating parameters are collected through sensors, the sensors include but are not limited to current sensors, speed sensors and load sensors, wherein the number of sensor types corresponds to the number of motor parameter types collected.
[0044] Step S2, preprocessing the collected data, the preprocessing method includes removing noise, normalizing and intercepting the time period of interest, so that the collected data meets the input requirements of the neural network.
[0045] Step S3, Fourier transform the preprocessed data to convert the time domain data into frequency domain data. The result of Fourier transform is used to extract the frequency spectrum characteristics of the motor operation to further optimize the performance of the neural network.
[0046] Step S4, by calculating the important integral of the hypergeometric series, the weight parameters of the neural network are obtained. The hypergeometric weight integral and Gaussian function approximation can ensure the accuracy and stability of the weight parameters and activation function of the neural network.
[0047] Step S5, using the high power approximation method of the Gaussian function to perform limit verification on the weight parameters in step S4 to ensure the accuracy of the adjustment.
[0048] Step S6, generating a new activation function through fitting and interpolation algorithms, and applying it to the neural network. The new activation function is dynamically adjusted according to the operating state of the motor, and the motor is adjusted in real time. The neural network can include but is not limited to a feedforward neural network, a convolutional neural network or a recurrent neural network. The process of training the neural network includes adjusting the weight parameters in the network so that the network output is as close to the expected output as possible.
[0049] Step S7, generating a sinusoidal wave adjustment signal as needed to precisely adjust the motor.
[0050] Among them, in step S3, the result of Fourier transform is used to extract the spectral features of the motor operation, and the extracted spectral features are converted into feature vectors, and the feature vectors are input into neural network training, and the neural network is optimized according to the training results, wherein the feature vector is a one-dimensional array for a single sample and a two-dimensional array for multiple samples.
[0051] In the present invention, in step S6, the process of training the neural network includes adjusting the weight parameters in the network, and using gradient descent to adjust and optimize the weight parameters of the neural network so that the output of the network reaches a predetermined performance indicator.
[0052] During the autonomous learning, training and adaptation process of the neural network, its own learning optimization results and the motor parameters of the ball valve motor controlled by the neural network are transmitted to the cloud server through communication technology, and uploaded to the terminal display through the cloud server, so as to intuitively understand the equipment status of the ball valve or other equipment when applied. The communication technologies used include 4G, 5G, TPC, IP transmission, etc., and the terminals include PC, mobile phones, display screens, etc.
[0053] The method of optimizing the weight parameters of the neural network is to generate a regulation signal according to the gradient descent adjustment rule of the weight parameters. The generation algorithm is as follows:
[0054] Loss function L for weight parameters The gradient is: ;
[0055] Conditioning signal: ;
[0056] in is the update amount of the weight parameter, is the learning rate;
[0057] The frequency, amplitude and phase of the regulation signal generation are adjusted according to the actual dynamic requirements of the motor.
[0058] In this embodiment, first, Fourier transform is performed on the collected data to convert the complex data into an understandable model. Figure 2 , Figure 3and Figure 4 shown.
[0059] Then, training the neural network requires adjusting the weight parameters in the network so that the network output is as close to the expected output as possible. This embodiment starts with an important integral involving a hypergeometric series and calculates from different angles through two series expansions:
[0060]
[0061] in is a hypergeometric function, x is a variable, m is an integer parameter, and m is greater than or equal to 2.
[0062] Then perform two Taylor series expansions on it, and the series expansion of the integral when x=0 is:
[0063] ;
[0064] ;
[0065] Where a is , b is , c is , z is .
[0066] When checking the normal bell curve in step S5, by setting m=64, a Gaussian normal distribution is performed using definite integral and indefinite integral:
[0067]
[0068] See also Figure 2 Next, we examine its limit. When the parameter m approaches infinity, it is a Gaussian normal function:
[0069]
[0070] Where x is a variable,
[0071] The above normal curve is bell-shaped, with low ends and high middle end, and is symmetrical on both sides. It is usually called a bell-shaped line.
[0072] The steps of generating the activation function in step S6 are as follows: If a random variable obeys a probability distribution with a location parameter and a scale parameter, its probability density function is:
[0073]
[0074] Among them, x is the value of the random variable, μ is the location parameter, and σ is the scale parameter;
[0075] When μ=0 and σ=1, the normal distribution becomes the standard normal distribution:
[0076]
[0077] Then calculate the error between the two. When the parameter m is set to 256, the integral difference between the two is already within 1%:
[0078] .
[0079] Finally, a new activation function is defined using the following equation and applied to the neural network:
[0080]
[0081] See also Figure 3 , when m=5, by taking the derivative of the activation function, we can get a bell-shaped function. The activation function is an important convolution kernel in approximate work such as fitting and interpolation:
[0082] .
[0083] See also Figure 4 , many orthogonal polynomials can construct more convolution kernels. These symmetric functions and polynomials can be converted into bell-shaped functions, which can be integrated to generate new activation functions and applied to neural networks:
[0084] .
[0085] In the convolutional neural network of the present invention, weight parameter calculation is necessary, and the weight parameter is closely related to the probability distribution. Therefore, the Bell function can accurately construct different weight ratios. The present invention uses the hypergeometric series, also known as the hypergeometric function, to construct the bell-shaped function and the activation function, which has strong universality and robustness. Through this equation, it can be converted into the sum of discrete data, which can be widely used in programming, such as smoothing filtering, low-pass filtering, fitting, interpolation, and even various Types of neural networks.
[0086] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0087] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for adaptively adjusting a ball valve opening and closing motor based on a neural network, characterized in that: The following steps are involved: Step S1, real-time acquisition of motor running parameters, including but not limited to current, speed and load; Step S2, preprocessing the collected data to make the collected data meet the input requirements of the neural network; Step S3, performing Fourier transform on the preprocessed data, converting the time domain data into frequency domain data, and optimizing the neural network based on the frequency domain data; Step S4, obtaining the weight parameters of the neural network by calculating the important integral of the hypergeometric series; Step S5, performing limit check on the normal bell curve using a high power approximation method of a Gaussian function; Step S6, generating a new activation function through fitting and interpolation algorithms, and applying it to the neural network; Step S7, generating a sine wave adjustment signal as needed to precisely adjust the motor; Among them, the method of optimizing the weight parameters of the neural network is to generate a regulation signal according to the gradient descent adjustment rule of the weight parameters, and the generation algorithm is as follows: The gradient of the loss function L with respect to the weight parameter w is: Conditioning signal: Where Δw is the update amount of the weight parameter, and η is the learning rate; The frequency, amplitude and phase of the regulation signal generation are adjusted according to the actual dynamic requirements of the motor; First, the acquired data is Fourier transformed; Training a neural network then requires adjusting the weight parameters in the network so that the network output is as close as possible to the expected output, starting with an important integral involving a hypergeometric series and calculated from two different perspectives through two series expansions: in is a hypergeometric function, x is a variable, m is an integer parameter, and m is greater than or equal to 2; Then perform two Taylor series expansions on it, and when x=0, the series expansion of the integral is: Where a is b is c is z is When checking the normal bell curve in step S5, by setting m=64, a Gaussian normal distribution is performed using definite integral and indefinite integral: Next, we examine its limit, which is a Gaussian normal function as the parameter m approaches infinity: Where x is a variable, The above normal curve is bell-shaped, low at both ends, high in the middle, and symmetrical; The steps of generating the activation function in step S6 are as follows: If a random variable obeys a probability distribution with a location parameter and a scale parameter, its probability density function is: Among them, x is the value of the random variable, μ is the location parameter, and σ is the scale parameter; When μ = 0 and σ = 1, the normal distribution becomes the standard normal distribution: Then calculate the error between the two. When the parameter m is set to 256, the integral difference between the two is already within 1%: Finally, a new activation function is defined using the following equation and applied to the neural network: Activation function When m=5, by taking the derivative of the activation function, we can get a bell-shaped function. The activation function is an important convolution kernel in approximate work such as fitting and interpolation: Many orthogonal polynomials can construct more convolution kernels. These symmetric functions and polynomials can be converted into bell-shaped functions, which can be integrated to generate new activation functions and applied to neural networks:
2. According to claim 1, a method for adaptively adjusting a ball valve opening and closing motor based on a neural network is characterized in that: In step S1, parameters of the motor during operation are collected through sensors, which include but are not limited to current sensors, speed sensors, and load sensors.
3. The method for adaptively adjusting a ball valve start-and-close motor based on a neural network according to claim 1, characterized in that: The pre-processing in step S2 includes noise removal, normalization and interception of a time period of interest.
4. The method for adaptively adjusting a ball valve opening and closing motor based on a neural network according to claim 1 is characterized in that: In step S3, the result of Fourier transform is used to extract the spectrum characteristics of the motor operation, and the extracted spectrum characteristics are converted into feature vectors, and the feature vectors are input into neural network training, and the neural network is optimized according to the training results, wherein the feature vector is a one-dimensional array for a single sample and a two-dimensional array for multiple samples.
