An intelligent pressure transmitter with intelligent anti-interference optimization control and its control system

Abstract

The present invention discloses an intelligent pressure transmitter with intelligent anti-interference optimization control and a control system thereof, which relate to the field of intelligent control technology. The system comprises: a pressure transmitter acquisition module: used for using m groups of MEMS piezoelectric sensor probes to acquire pressure signals from q channels, where m is an integer greater than 1, and m=q; a pressure signal preprocessing module: used for preprocessing the pressure signal to obtain a characteristic graph, and inputting it into an interference identification module; an interference identification module: used for acquiring environmental parameters at the location of the pressure transmitter, and forming sequence data by one-to-one correspondence with the pressure signal in time, and inputting the sequence data into a pre-constructed first machine learning model to output a category probability sequence for each group of sequence data; the present invention enables the control system to control related machines based on accurate digital signals to avoid misoperation of related machines.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and more specifically, to an intelligent pressure transmitter with intelligent anti-interference optimization control and a control system thereof. Background Art

[0002] Pressure transmitters rely on core components such as pressure sensors and transmission circuits. They detect and measure pressure variables by converting pressure signals into standard electrical signal outputs. The continuous advancement of pressure sensing technology has provided a foundation for the development of pressure transmitters. In industrial production, chemical industry, energy and other fields, there is a strong demand for accurate measurement and feedback control of pressure parameters. Pressure transmitters, as key components for realizing automated control, play an important role in industrial process control.

[0003] Traditional pressure transmitters are easily affected by external interference during the measurement process, resulting in reduced accuracy and stability of the measurement results. In order to solve this problem, some anti-interference pressure transmitters have been proposed in the prior art. The Chinese patent with publication number CN117606668A discloses a high-precision intelligent pressure transmitter, which relates to the technical field of intelligent pressure transmitters and solves the technical problem that the sealing of the prior art intelligent pressure transmitters will deteriorate and leak when the gas medium is transported for a long time, and the pipe mouth connection may be cracked and damaged due to excessive air pressure or external force extrusion, thereby affecting the detection accuracy of the intelligent pressure transmitter; it includes a pipeline and a pressure transmitter body, and the pressure transmitter body includes a body, a digital display screen embedded in the side of the body, a control box arranged on the body and an air guide tube arranged at the bottom of the body; it also includes an anti-leakage mechanism; the invention can extrude and seal the connecting pipe mouth, improve the sealing of the pipe mouth connection, and effectively avoid leakage; and it can disperse the extrusion force and reduce the stress, thereby preventing the pipe mouth connection of the inner tube one and the inner tube two from being deformed due to external force extrusion, and effectively avoiding the risk of cracking and leakage due to external force at the pipe mouth connection.

[0004] However, there are still certain limitations. The pressure signal is converted into a digital signal for the control system to control the related machines based on the digital signal. Before the pressure signal is converted into a digital signal, there are environmental interference sources in the pressure signal, and the interference mixed in the pressure signal needs to be stripped away. If the control system controls the related machines based on inaccurate digital signals, it will cause the related machines to operate incorrectly, resulting in corresponding losses. Summary of the invention

[0005] In order to solve the above technical difficulties, the present invention provides an intelligent pressure transmitter control system with intelligent anti-interference optimization control, the system comprising:

[0006] Pressure transmitter acquisition module: used to collect pressure signals of q channels using m groups of MEMS piezoelectric sensor probes, where m is an integer greater than 1, m=q;

[0007] Pressure signal preprocessing module: used to preprocess the pressure signal to obtain a characteristic graph and input it into the interference recognition module;

[0008] Interference identification module: used to collect environmental parameters at the location of the pressure transmitter, and form sequence data in one-to-one correspondence with the pressure signal in time, input the sequence data to the pre-built first machine learning model to output the category probability sequence of each group of sequence data, and input the feature map to the pre-built second machine learning model to output the category label and the probability distribution corresponding to the category label;

[0009] Optimization control module: used to model interference signals according to the probability distribution corresponding to the category labels, the category labels and the category probability sequences of each group of sequence data, identify the interference signals and remove the interference signals from the pressure signals to obtain the reconstructed pressure signals, and convert the reconstructed pressure signals into digital signals. The digital signals are the final pressure values ​​output by the intelligent pressure transmitter.

[0010] Furthermore, the method of preprocessing the pressure signal to obtain a characteristic graph includes:

[0011] Use low-pass filtering to remove high-frequency noise in the pressure signal. Low-pass filtering uses a first-order low-pass filter. The calculation formula is:

[0012] y(n)=a×x(n)+(1-a)×y(n-1);

[0013] In the formula, x(n) is the sample value of the input pressure signal, which is the value taken at different time points as n increases; y(n) is the sample value of the output pressure signal; a is the filtering parameter, which takes the value (0, 1) and is used to control the cutoff frequency; y(n-1) is the sample value of the output pressure signal at the previous time point n-1; n is the time point, and the value of n starts from 0 and increases by 1 in sequence, corresponding to the order of the time points;

[0014] The time-frequency characteristics of the pressure signal are extracted using wavelet transform. The first-level wavelet transform calculation method includes:

[0015]

[0016] cD1=Σx(n)×ψ(n);

[0017] Where cA1 is the approximate component coefficient, cD1 is the detail component coefficient, φ(n) is the detection function, and ψ(n) is the detail function;

[0018] The second-level wavelet transform calculation method includes: when calculating the second-level approximate component coefficient cA2, cA1 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation; the third-level wavelet transform calculation method includes: when calculating the third-level approximate component coefficient cA3, cA2 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation, and the same is true for each level of wavelet transform detail component coefficient; the approximate component coefficient and the detail component coefficient are both time-frequency characteristics of the pressure signal;

[0019] The time-frequency characteristics use statistical methods in the small band to extract the time series characteristics of fixed characteristic parameters, including mean value, variance and entropy value; the pressure signal characteristic matrix is ​​constructed with the small band as the unit: the rows in the pressure signal characteristic matrix represent the time-frequency characteristics, and the columns represent the time series characteristics;

[0020] PCA is used to find the principal component space of the pressure signal feature matrix, and the dimension of the principal component space of the pressure signal feature matrix is ​​reduced. The reduced dimension pressure signal feature matrix is ​​embedded into a low-dimensional space using Autoencoder, and then reconstructed into a feature map.

[0021] The calculation formula for PCA dimensionality reduction is:

[0022] Y = X × U;

[0023] Wherein: X is the original pressure signal feature matrix, Y is the pressure signal feature matrix after the original pressure signal feature matrix is ​​reduced in dimension, and U is the transformation matrix of the principal component space of the pressure signal.

[0024] Furthermore, the environmental parameters include temperature, vibration and acceleration; the sequence data includes pressure sequence data, temperature sequence data, vibration sequence data and acceleration sequence data.

[0025] Furthermore, the training method of the first machine learning model includes:

[0026] A set of sequence data and the corresponding category probability sequence are combined into a set of historical training sets. G groups of historical training sets are extracted from the database as training sets to train the first machine learning model. The Adam optimizer is used, the batch size is set to 128, the initial learning rate is set to 0.001, and the sliding average method is used to update the parameter momentum and the input and output parameter matrix w. The update equation is:

[0027] vf=momentum×(vf-1)+(1-momentum)×Δwf;

[0028] wf+1=wf-η×vf;

[0029] Where f represents the sequence number of the current iteration step under the sliding average, vf represents the estimated gradient value of the current iteration step f under the sliding average, (vf-1) represents the estimated gradient value of the previous step, Δwf represents the actual gradient descent amount of this step, wf represents the parameter value of the current iteration step f, η is the learning rate that controls the step size, and wf+1 represents the updated parameter value, that is, the learning rate η is used to update wf in the reverse direction vf;

[0030] The calculation formula of loss function loos is:

[0031]

[0032] Where i is the serial number of a single historical training set, yi∈[0,1], yi is the probability distribution of the true label of the i-th historical training set, p i is the probability distribution of the i-th group of historical training sets predicted by the first machine learning model; i≤G;

[0033] When loos converges, the training is stopped and the training is completed.

[0034] Furthermore, the second machine learning model construction method includes:

[0035] The set of all feature maps is used as the input of the second machine learning model, the second machine learning model uses the predicted category label for each feature map and the probability distribution corresponding to the category label as output, uses the actual category label corresponding to each feature map and the probability distribution corresponding to the category label as the prediction target, and uses minimizing the sum of the first prediction accuracies of all predicted category labels and the probability distribution corresponding to the category label as the training target;

[0036] Among them, the calculation formula for the first prediction accuracy is: Mc e =(Zf e -Sf e ) 2 , among which Mc e is the first prediction accuracy, Zf e is the predicted category label corresponding to the e-th feature map and the probability distribution corresponding to the category label, Sf e is the probability distribution of the actual category label corresponding to the e-th feature map and the category label; e is the serial number representing the number of feature maps; the second machine learning model is trained until the sum of the first prediction accuracies converges; the second machine learning model is any one of CNN and MLP.

[0037] Furthermore, the method for modeling interference signals is a Markov process, and Markov process modeling interference signals includes:

[0038] Set each interference g, define the state space s = {s1, s2, s3}, and define the observation symbol space v = {v1, v2, v3}; calculate the state transition probability matrix D, D is a 3×3 matrix, and the probability calculation of the component element alj of the matrix D is expressed as alj = P(qti+1 = sj|qti = sl), that is, the probability that the state sl turns to sj at the ti-th moment; where P is a probability indicator, qti represents the hidden state at the ti-th moment, sj represents the j-th state in the state space s, and sl represents the l-th state in the state space s; j∈[1,3], l∈[1,3]; calculate the observation probability matrix B, B is a 3×3 matrix, and the probability calculation of the component element blj of the matrix B is expressed as blj = P(vti = vl|qti = sl), that is, the probability of observing vl in state sl; where vti represents the observation symbol at the ti-th moment, and vl is the l-th symbol in the observation symbol space v;

[0039] The initial probability distribution of the state is initialized to π, and the observation sequence is set to O = o1, o2...oT, ti∈T. The calculation formula for the observation sequence O is:

[0040] Forward calculation: αt(j) = P(o1, o2...oti, qt = sj|λ);

[0041] Backward calculation: βt(l)=P(o(ti+1),o(ti+2),o(ti+3)...oT|qt=sl,λ);

[0042] Among them, λ represents the parameters of the Markov model, λ includes the number of states in the state space s, the number of observation symbols in the observation space v, the state transition probability matrix D, the observation probability matrix B and the initial state distribution π; oti represents the ti-th observation in the observation sequence, T represents the length of the observation sequence, that is, the total number of observation points, and the length unit of the observation sequence is set to 1, so T here represents the concept of the length of the observation sequence, and also represents time; αt(j) represents the forward calculation value of the j-th state in the forward calculation, βt(l) represents the backward calculation value of the l-th state in the backward calculation, qt represents the hidden state at time t, and t represents the concept of time, that is, the universal expression symbol of time;

[0043] Calculate the hidden state probability γt(l), the hidden state probability calculation formula is γt(l) = αt(l) × βt(l) / P(O|λ); where αt(l) represents the forward calculation value of the lth state, and P(O|λ) represents the probability that the observation sequence O is generated according to the Markov model after the parameter set λ of the Markov model is given;

[0044] Calculate the adjacent state transition probability. The formula for calculating the adjacent state transition probability ξt(l,j) is:

[0045] ξt(l,j)=αt(j)×alj×bj(o(ti+1))×βt(l) / P(O|λ);

[0046] Among them, bj(o(ti+1)) is the element of B matrix, which represents the probability of occurrence of observation o(ti+1) in state sj;

[0047] Input the category probability sequence, the category label and the probability distribution corresponding to the category label into the Markov model, where the interference g is the environmental parameter, the category probability sequence is the component element in the state transition probability matrix D, and the category label and the probability distribution corresponding to the category label are the component elements in the observation probability matrix B. Iterate until λ converges to obtain the Markov model parameter λ for sensing the interference signal, that is, the modeling is completed.

[0048] Furthermore, the method of identifying the interference signal and removing the interference signal from the pressure signal to reconstruct the pressure signal includes:

[0049] A PF filter is used to eliminate interference signals to obtain a first pressure signal, a low entropy principle is used to reconstruct the first pressure signal to obtain a second pressure signal, a sampling reconstruction method is used to process the second pressure signal to obtain a third pressure signal, and a manifold recognition method is used to screen interference points in the third pressure signal to obtain a reconstructed pressure signal;

[0050] The first pressure signal is reconstructed using the low entropy principle. Let the first pressure signal be st, the reconstructed first pressure signal be st', and the suppressed first pressure signal be xt, which can be expressed as:

[0051] xt=st+st';

[0052] A set of reconstruction coefficients wi is randomly searched so that the reconstructed first pressure signal st' is expressed as:

[0053]

[0054] In the formula, is a basis function, so that the reconstructed first pressure signal st' is closest to the first pressure signal st;

[0055] Using the low entropy principle, the conditional entropy H(st|xt) of the reconstructed first pressure signal st' is defined as:

[0056] H(st|xt)=-∫p(st,xt)logp(st|xt)dstdxt;

[0057] Wherein, p(st,xt) is the joint probability density function of the first pressure signal st and the suppressed first pressure signal xt, logp(st|xt) is the conditional probability density function of the first pressure signal st and the suppressed first pressure signal xt; the pallet function E between the reconstructed first pressure signal st' and the first pressure signal st is defined as:

[0058] E = ∫(st'(x) - st(x)) 2 dx;

[0059] In the formula, x is the independent variable;

[0060] Initialize E, minimize the reconstruction coefficient wi, and take the negative minimum of the log-likelihood function, that is:

[0061] ▽wiE=0;

[0062] By taking the derivative of the first pressure signal st and φit, we get a linear equation system:

[0063] Awi = b;

[0064] Where A is the coefficient matrix obtained by differentiating the reconstruction coefficient wi, b is the coefficient vector of the linear equation system after differentiation, and is the constant term;

[0065] Using the normal equation linear equation system, we get the reconstruction coefficient wi:

[0066] wi=(A T A) -1 A T b;

[0067] Among them, A T represents the transposed matrix of A;

[0068] By minimizing the conditional entropy H(st|xt), the reconstruction coefficient wi that best matches the reconstructed first pressure signal st' and the suppressed first pressure signal xt is obtained. At this point, the reconstruction of the second pressure signal is completed.

[0069] The second pressure signal is processed by a sampling reconstruction method to obtain a third pressure signal, comprising:

[0070] The second pressure signal is uniformly sampled, a sampling point vector is extracted as a reconstruction basis, a continuous curve is reconstructed using an interpolation method according to the relationship between the sampling points, the reconstruction convention is iteratively optimized to make the reconstructed continuous curve smooth, a third pressure signal is obtained, and a third pressure signal continuous curve is output;

[0071] The reconstructed pressure signal is obtained by screening the interference points in the third pressure signal using the manifold recognition method, including:

[0072] A manifold structure of the third pressure signal is constructed, abnormal points are identified in the manifold space, and after the identified abnormal points are removed, a reconstructed pressure signal is obtained according to the manifold structure information of the third pressure signal.

[0073] Furthermore, the system further comprises:

[0074] Fluid system access module: when it is detected that the system connected to the intelligent pressure transmitter is a fluid system, the fluid data acquisition module and the fluid pressure measurement module are started;

[0075] Fluid data collection module: used to collect fluid system classification name and fluid flow rate;

[0076] Fluid pressure calculation module: used to input the fluid system classification name and fluid flow rate to the pre-built third machine learning model to output the flow rate and pressure value;

[0077] When the fluid flow rate is set as the interference signal, the pressure value obtained by subtracting the flow rate from the digital signal is used as the final pressure value output by the intelligent pressure transmitter.

[0078] When the fluid flow is set as the pressure signal source, the digital signal is used as the final pressure value.

[0079] Furthermore, the fluid system classification names include open fluid systems, pipeline transport liquid systems, constant pressure gas pipeline systems, constant flow gas pipeline systems and pump systems;

[0080] The method for constructing the third machine learning model includes:

[0081] Convert each group of fluid system classification names and fluid flow rates into the form of a first feature vector, wherein the elements in the first feature vector include an open fluid system, a pipeline transport liquid system, a constant pressure gas pipeline system, a constant flow gas pipeline system, and a pump system;

[0082] The set of all first feature vectors is used as the input of a third machine learning model, wherein the third machine learning model uses the flow pressure value predicted for each group of first feature vectors as the output, uses the actual flow pressure value corresponding to each group of first feature vectors as the prediction target, and uses minimizing the sum of the second prediction accuracies of all predicted flow pressure values ​​as the training target;

[0083] Among them, the calculation formula for the second prediction accuracy is: K d =(Z d -Q d ) 2 , where K d is the second prediction accuracy, Z d is the predicted flow pressure value corresponding to the fluid system classification name and fluid flow of the dth group, Q dis the actual flow pressure value corresponding to the dth group of fluid system classification name and fluid flow; d represents the group number of the first eigenvector; the third machine learning model is trained until the sum of the second prediction accuracies reaches convergence and the training is stopped; the third machine learning model is any one of a regression model and a decision tree.

[0084] The present invention provides an intelligent pressure transmitter with intelligent anti-interference optimization control, which is used to implement the intelligent pressure transmitter control system with intelligent anti-interference optimization control.

[0085] The technical effects and advantages of an intelligent pressure transmitter with intelligent anti-interference optimization control and its control system provided by the present invention are as follows:

[0086] Technical effect:

[0087] It can identify different types of fluid systems and calculate the theoretical pressure value according to the fluid type and flow rate; use multiple sensors to collect pressure signals, realize multi-channel collection, and improve observation accuracy; use deep learning models to identify and filter out different environmental interference signals, and extract pure pressure signals from them; use low entropy principles and sampling reconstruction methods to reconstruct pressure signals, eliminate residual interference, and output stable pressure curves;

[0088] advantage:

[0089] Combined with fluid information, the theoretical pressure value is calculated to improve detection accuracy; multi-sensor ultra-high sampling realizes multi-dimensional observation and improves observation quality; the powerful deep learning model effectively identifies different interferences and efficiently extracts effective information; the multi-link optimization processing is highly perfect, and the output is a pressure signal that meets the quality requirements of industrial applications; the use of cutting-edge algorithms realizes intelligence and automation, and can work efficiently for a long time without human intervention; the system is modularly designed, and relevant functional modules can be customized and installed according to actual needs; the pressure signal without interference after filtering and reconstruction will be converted into an accurate digital signal, and the control system will control the relevant machines based on the accurate digital signal to avoid misoperation of the relevant machines and cause corresponding losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A schematic diagram of the structure of an intelligent pressure transmitter control system with intelligent anti-interference optimization control according to Example 1;

[0091] Figure 2 This is a schematic diagram of the principle structure of an intelligent pressure transmitter control system with intelligent anti-interference optimization control according to Example 2. DETAILED DESCRIPTION

[0092] 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.

[0093] Example 1

[0094] See also Figure 1 As shown, this embodiment provides an intelligent pressure transmitter control system with intelligent anti-interference optimization control, and the system includes:

[0095] Pressure transmitter acquisition module: used to collect pressure signals from q channels using m groups of MEMS piezoelectric sensor probes, where m is an integer greater than 1, m=q.

[0096] Pressure signal preprocessing module: used to preprocess the collected pressure signal to obtain a feature map and input it into the interference recognition module.

[0097] Interference identification module: used to collect the environmental parameters of the location of the pressure transmitter, and form sequence data by one-to-one correspondence with the pressure signal in time, input the sequence data to the pre-built first machine learning model to output the category probability sequence of each group of sequence data, and input the feature map to the pre-built second machine learning model to output the category label and the probability distribution corresponding to the category label.

[0098] Optimization control module: used to model interference signals according to the probability distribution corresponding to the category labels, the category labels and the category probability sequences of each group of sequence data, identify the interference signals and remove the interference signals from the pressure signals to obtain the reconstructed pressure signals, and convert the reconstructed pressure signals into digital signals. The digital signals are the final pressure values ​​output by the intelligent pressure transmitter.

[0099] The MEMS (micro-electromechanical system) piezoelectric sensor is a piezoelectric pressure sensor, which is made of a piezoelectric film using a MEMS process and has the characteristics of small size and high sensitivity.

[0100] The method for preprocessing the pressure signal to obtain a characteristic graph comprises:

[0101] Use low-pass filtering to remove high-frequency noise in the pressure signal. Low-pass filtering uses a first-order low-pass filter. The calculation formula is:

[0102] y(n)=a×x(n)+(1-a)×y(n-1);

[0103] In the formula, x(n) is the sample value of the input pressure signal, which is the value taken at different time points as n increases; y(n) is the sample value of the output pressure signal; a is the filtering parameter, which takes the value (0, 1) and is used to control the cutoff frequency; y(n-1) is the sample value of the output pressure signal at the previous time point n-1; n is the time point, and the value of n starts from 0 and increases by 1 in sequence, corresponding to the order of the time points.

[0104] It should be noted that the formula works in the following ways: the output y(n) is a linear combination of the input x(n) and the previous output y(n-1); the filter parameter a controls the weight of the linear combination; the larger the a value, the faster the response; the smaller the a value, the slower the response; when the a value is less than 1, the filter has "inertia", and it takes a certain amount of time for the new input to affect the output; for high-frequency signals, the signal amplitude changes dramatically at adjacent sampling points, and the low-pass filter cannot respond to changes in the high-frequency signal in time due to "inertia", so the high-frequency signal component will be suppressed after passing through the filter, and the output mainly comes from the low-frequency signal, thus removing high-frequency noise and leaving low-frequency useful signals.

[0105] The time-frequency characteristics of the pressure signal are extracted using wavelet transform. The first-level wavelet transform calculation method includes:

[0106]

[0107] cD1=∑x(n)×ψ(n);

[0108] Where cA1 is the approximate component coefficient, cD1 is the detail component coefficient, φ(n) is the detection function, and ψ(n) is the detail function.

[0109] The second-level wavelet transform calculation method includes: when calculating the second-level approximate component coefficient cA2, cA1 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation; the third-level wavelet transform calculation method includes: when calculating the third-level approximate component coefficient cA3, cA2 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation, and the same is true for each level of wavelet transform detail component coefficient; the approximate component coefficient and the detail component coefficient are both time-frequency characteristics of the pressure signal.

[0110] The wavelet transform is used to naturally divide the wavelet bands. For example, the approximate component coefficient cA1 and the detail component coefficient cD1 obtained by the first-level wavelet transform can be regarded as two wavelet bands. The time-frequency characteristics use statistical methods to extract the time series characteristics of fixed feature parameters within the wavelet. The fixed feature parameters include mean, variance and entropy. The pressure signal feature matrix is ​​constructed in units of wavelet bands: the rows in the pressure signal feature matrix represent the time-frequency features or, and the columns represent the time series features.

[0111] PCA is used to find the principal component space of the pressure signal feature matrix, and the dimension of the principal component space of the pressure signal feature matrix is ​​reduced. The reduced dimension pressure signal feature matrix is ​​embedded into a low-dimensional space using Autoencoder and then reconstructed into a feature map.

[0112] The calculation formula for PCA dimensionality reduction is:

[0113] Y = X × U;

[0114] Wherein: X is the original pressure signal feature matrix, Y is the pressure signal feature matrix after the original pressure signal feature matrix is ​​reduced in dimension, and U is the transformation matrix of the principal component space of the pressure signal.

[0115] The environmental parameters at the location of the pressure transmitter are collected and corresponded one-to-one with the pressure signal in time to form sequence data, that is, the environmental parameters and pressure signals at the same time step are a group of sequence data; the sequence data includes pressure sequence data, temperature sequence data, vibration sequence data and acceleration sequence data.

[0116] The environmental parameters include temperature, vibration and acceleration. The temperature can be directly detected by a temperature sensor to obtain the temperature value. The temperature change will affect the elastic modulus of the spring, thereby affecting the pressure signal. The vibration can detect the vibration acceleration signal of the device through an acceleration sensor, and then calculate the vibration frequency and amplitude characteristics through algorithms such as FFT. Strong vibration may cause the device to generate additional pressure signal interference. The acceleration is also directly measured by an acceleration sensor in three directions of the device. Strong acceleration may generate inertial pressure and generate additional pressure signal interference.

[0117] The construction of the first machine learning model is based on the LSTM network structure. The LSTM network structure includes an input layer, two LSTM layers; an output layer, the input of the input layer is sequence data, a set of pressure sequence data and the corresponding temperature sequence data, vibration sequence data and acceleration sequence data are a sample, the sample batch of the input layer is 128, and the dimension is 4; the two LSTM layers include the first LSTM layer and the second LSTM layer, and the two LSTM layers include 4 components, including input gate, forget gate, output gate and memory unit; the activation functions of the two LSTM layers are tanh; the input dimension of the first LSTM layer is 4, and the output dimension is 128, the input dimension of the second LSTM layer is 128, and the output dimension is 64; the input dimension of the output layer is 64, the output dimension is 3, the activation function is softmax, and the loss function is cross entropy.

[0118] The training method of the first machine learning model includes:

[0119] A set of sequence data and the corresponding category probability sequence are combined into a set of historical training sets. G groups of historical training sets are extracted from the database as training sets to train the first machine learning model. The Adam optimizer is used, the batch size is set to 128, the initial learning rate is set to 0.001, and the sliding average method is used to update the parameter momentum and the input and output parameter matrix w. The update equation is:

[0120] vf=momentum×(vf-1)+(1-momentum)×Δwf;

[0121] wf+1=wf-η×vf;

[0122] Where f represents the sequence number of the current iteration step under the sliding average, vf represents the estimated gradient of the current iteration step f under the sliding average, (vf-1) represents the estimated gradient of the previous step, Δwf represents the actual gradient descent amount of this step, wf represents the parameter value of the current iteration step f, η is the learning rate that controls the step size, and wf+1 represents the updated parameter value, which is obtained by updating wf in the reverse direction of vf using the learning rate η.

[0123] It should be noted that the parameter momentum determines the weights of (vf-1) and Δwf, that is, the higher the momentum, the more vf depends on the (vf-1) of the previous step, and the stronger the smoothing effect; the lower the momentum, the more vf depends on the current gradient Δwf, and the stronger the ability to track the actual gradient changes.

[0124] The calculation formula of loss function loos is:

[0125]

[0126] Where i is the serial number of a single historical training set, yi∈[0,1], yi is the probability distribution of the true label of the i-th historical training set, p i is the probability distribution of the i-th group of historical training sets after prediction by the first machine learning model; i≤G.

[0127] Special note:

[0128] When yi=1, the term is maximum, requiring pi to be maximum, and loos is minimum when the prediction is correct;

[0129] When yi=0, the term is maximum, requiring 1-pi to be maximum, that is, when the prediction is wrong, loos is maximum;

[0130] When loos converges, the training is stopped and the training is completed.

[0131] The second machine learning model construction method includes:

[0132] The set of all feature maps is used as the input of the second machine learning model, the second machine learning model takes the predicted category label for each feature map and the probability distribution corresponding to the category label as output, takes the actual category label corresponding to each feature map and the probability distribution corresponding to the category label as the prediction target, and takes minimizing the sum of the first prediction accuracies of all predicted category labels and the probability distribution corresponding to the category label as the training target.

[0133] Among them, the calculation formula for the first prediction accuracy is: Mc e =(Zf e -Sf e ) 2 , among which Mc e is the first prediction accuracy, Zf e is the predicted category label corresponding to the e-th feature map and the probability distribution corresponding to the category label, Sf e is the probability distribution of the actual category label corresponding to the e-th feature map and the category label; e is the serial number representing the number of feature maps; the second machine learning model is trained until the sum of the first prediction accuracies converges; the second machine learning model is any one of CNN and MLP.

[0134] The category probability sequence is the probability that each environmental parameter is interference in time, and the category label includes unwanted interference signals and desired pressure signals.

[0135] The method for modeling interference signals is a Markov process, and Markov process modeling interference signals includes:

[0136] Set each interference g, define the state space s = {s1, s2, s3}, and define the observation symbol space v = {v1, v2, v3}; calculate the state transition probability matrix D, D is a 3×3 matrix, and the probability calculation of the constituent elements alj of the matrix D is expressed as alj = P(qti+1 = sj|qti = sl), that is, the probability that the state sl turns to sj at the ti-th moment; where P is a probability indicator, qti represents the hidden state at the ti-th moment, sj represents the j-th state in the state space s, and sl represents the l-th state in the state space s; j∈[1,3], l∈[1,3]; calculate the observation probability matrix B, B is a 3×3 matrix, and the probability calculation of the constituent elements blj of the matrix B is expressed as blj = P(vti = vl|qti = sl), that is, the probability of observing vl in state sl; where vti represents the observation symbol at the ti-th moment, and vl is the l-th symbol in the observation symbol space v.

[0137] The initial probability distribution of the state is initialized to π, and the observation sequence is set to O = o1, o2...oT, ti∈T. The calculation formula for the observation sequence O is:

[0138] Forward calculation: αt(j) = P(o1, o2...oti, qt = sj|λ);

[0139] Backward calculation: βt(l)=P(o(ti+1),o(ti+2),o(ti+3)...oT|qt=sl,λ);

[0140] Among them, λ represents the parameters of the Markov model, λ includes the number of states in the state space s, the number of observation symbols in the observation space v, the state transition probability matrix A, the observation probability matrix B and the initial state distribution π; oti represents the ti-th observation in the observation sequence, T represents the length of the observation sequence, that is, the total number of observation points, and the length unit of the observation sequence is set to 1, so T here represents the concept of the length of the observation sequence and also represents time; αt(j) represents the forward calculation value of the j-th state, βt(l) represents the backward calculation value of the l-th state, qt represents the hidden state at time t, and t represents the concept of time, that is, the universal expression symbol of time.

[0141] Calculate the hidden state probability γt(l), the hidden state probability calculation formula is γt(l) = αt(l) × βt(l) / P(O|λ); where αt(l) represents the forward calculation value of the l-th state, and P(O|λ) represents the probability that the observation sequence O is generated according to the Markov model after the parameter set λ of the Markov model is given.

[0142] Calculate the adjacent state transition probability. The formula for calculating the adjacent state transition probability ξt(l,j) is:

[0143] ξt(l,j)=αt(j)×alj×bj(o(ti+1))×βt(l) / P(O|λ);

[0144] Among them, bj(o(ti+1)) is the element of the B matrix, which represents the probability of occurrence of observation o(ti+1) in state sj.

[0145] Input the category probability sequence, the category label and the probability distribution corresponding to the category label into the Markov model, where the interference g is the environmental parameter, the category probability sequence is the component element in the state transition probability matrix D, and the category label and the probability distribution corresponding to the category label are the component elements in the observation probability matrix B. Iterate until λ converges to obtain the Markov model parameter λ for sensing the interference signal, that is, the modeling is completed.

[0146] It is particularly noted that the probability data used in Markov modeling comes from the output of the interference identification module; the interference signal itself is a random process, which shows randomness as it evolves over time, but its transition probability is regular and shows a certain degree of memory. The use of the Markov model to establish the state space and state transfer mechanism can describe this randomness and memory, and the probability distribution law of the state change of the interference signal at different times can be obtained through the state transfer matrix, which can be regarded as a deterministic description of the randomness and memory of the interference signal. On this basis, this deterministic description can be used to remove the random components in the interference signal through filtering; while in the Markov model, λ is initially unclear, and it is necessary to perform a reverse calculation of λ based on the output of the interference identification module. When λ converges, the Markov model modeling of the interference signal is completed.

[0147] The method for identifying interference signals and reorganizing pressure signals after removing the interference signals comprises:

[0148] A PF filter is used to eliminate the interference signal to obtain a first pressure signal, the low entropy principle is used to reconstruct the first pressure signal to obtain a second pressure signal, the sampling reconstruction method is used to process the second pressure signal to obtain a third pressure signal, and the manifold recognition method is used to screen the interference points in the third pressure signal to obtain a reconstructed pressure signal.

[0149] The first pressure signal is reconstructed using the low entropy principle. Let the first pressure signal be st, the reconstructed first pressure signal be st', and the suppressed first pressure signal be xt, which can be expressed as:

[0150] xt=st+st'

[0151] A set of reconstruction coefficients wi is randomly searched so that the reconstructed first pressure signal st' is expressed as:

[0152]

[0153] Wherein, φit is the basis function, which makes the reconstructed first pressure signal st' closest to the first pressure signal st.

[0154] Using the low entropy principle, the conditional entropy H(st|xt) of the reconstructed first pressure signal st' is defined as:

[0155] H(st|xt)=-∫p(st,xt)logp(st|xt)dstdxt;

[0156] Wherein, p(st,xt) is the joint probability density function of the first pressure signal st and the suppressed first pressure signal xt, logp(st|xt) is the conditional probability density function of the first pressure signal st and the suppressed first pressure signal xt; the pallet function E between the reconstructed first pressure signal st' and the first pressure signal st is defined as:

[0157] E = ∫(st'(x) - st(x)) 2 dx;

[0158] In the formula, x is the independent variable.

[0159] Initialize E, minimize the reconstruction coefficient wi, and take the negative minimum of the log-likelihood function, that is:

[0160] ▽wiE=0.

[0161] By taking the derivative of the first pressure signal st and φit, we get a linear equation system:

[0162] Awi = b;

[0163] Where A is the coefficient matrix obtained by differentiating the reconstruction coefficient wi, b is the coefficient vector of the linear equations after differentiation, and is the constant term.

[0164] Using the normal equation linear equation system, we get the reconstruction coefficient wi:

[0165] wi=(A T A) -1 A T b;

[0166] Among them, A T Represents the transposed matrix of A.

[0167] By minimizing the conditional entropy H(s|x), the reconstruction coefficient wi that best matches the reconstructed first pressure signal st' and the suppressed first pressure signal xt is obtained. At this point, the reconstruction of the second pressure signal is completed.

[0168] The second pressure signal is processed by a sampling reconstruction method to obtain a third pressure signal, comprising:

[0169] The second pressure signal is uniformly sampled, and the sampling point vector is extracted as a reconstruction basis. According to the relationship between the sampling points, a continuous curve is reconstructed using an interpolation method. The reconstruction convention is iteratively optimized to make the reconstructed continuous curve smooth, thereby obtaining a third pressure signal, and outputting a third pressure signal continuous curve.

[0170] The reconstructed pressure signal is obtained by screening the interference points in the third pressure signal using the manifold recognition method, including:

[0171] A manifold structure of the third pressure signal is constructed, abnormal points are identified in the manifold space, and after the identified abnormal points are removed, a reconstructed pressure signal is obtained according to the manifold structure information of the third pressure signal.

[0172] Example 2

[0173] like Figure 2 As shown, this embodiment further improves the design on the basis of Embodiment 1. The difference is that, when a smart pressure transmitter control system with intelligent anti-interference optimization control provided by this embodiment is applied to a fluid system, the fluid flow rate will be related to the pressure change amplitude. At this time, if the fluid flow rate is directly ignored and the pressure signal is transmitted, the accuracy of the final pressure value will be affected. However, when the pressure is finally measured, the fluid flow rate cannot be directly used as an interference signal. The relationship between the fluid flow rate and the pressure is an external interference factor, but it is also affected by the type of fluid system and needs to be measured separately. Based on this, a smart pressure transmitter control system with intelligent anti-interference optimization control provided by this embodiment also includes a fluid system access monitoring module, a fluid data acquisition module and a fluid pressure measurement module.

[0174] Fluid system access module: When the system connected to the intelligent pressure transmitter is monitored, the fluid data acquisition module and the fluid pressure measurement module are started.

[0175] Fluid data collection module: used to collect fluid system classification name and fluid flow rate.

[0176] The classification names of fluid systems include open fluid systems, pipeline liquid systems, constant pressure gas pipeline systems, constant flow gas pipeline systems and pump systems; the fluid flow can be collected by a meter or a flow sensor. For open fluid systems, such as water pipes, according to Bernoulli's principle, an increase in flow will lead to an increase in flow velocity, and an increase in flow velocity will lead to a decrease in static pressure; when the flow in a pipeline liquid system increases, the resistance of the fluid in the pipeline will also increase. According to fluid mechanics, an increase in resistance usually leads to an increase in pressure; the change in pressure in a constant pressure gas pipeline system depends on whether it is a constant pressure transmission. If it is a constant pressure, the pressure remains unchanged when the flow rate is increased; if it is a constant flow gas pipeline system, according to Bernoulli's principle, an increase in flow will cause a slight decrease in pressure; an increase in flow in a pump system may cause a decrease in the pump head, thereby slightly decreasing the pressure in the pipeline area.

[0177] Fluid pressure measurement module: used to input the fluid system classification name and fluid flow rate into the pre-built third machine learning model to output the flow and pressure value.

[0178] When the fluid flow rate is set as the interference signal, the pressure value obtained by subtracting the flow rate from the digital signal is used as the final pressure value output by the intelligent pressure transmitter.

[0179] When the fluid flow is set as the pressure signal source, the digital signal is used as the final pressure value.

[0180] The third method for building a machine learning model includes:

[0181] Each group of fluid system classification names and fluid flow rates are converted into the form of a first characteristic vector, wherein the elements in the first characteristic vector include an open fluid system, a pipeline transport liquid system, a constant pressure gas pipeline system, a constant flow gas pipeline system and a pump system.

[0182] The set of all first eigenvectors is used as the input of the third machine learning model. The third machine learning model takes the predicted flow pressure value for each group of first eigenvectors as output, the actual flow pressure value corresponding to each group of first eigenvectors as the prediction target, and minimizes the sum of the second prediction accuracies of all predicted flow pressure values ​​as the training target.

[0183] Among them, the calculation formula for the second prediction accuracy is: K d =(Z d -Q d ) 2 , where K d is the second prediction accuracy, Z d is the predicted flow pressure value corresponding to the fluid system classification name and fluid flow of the dth group, Q d is the actual flow pressure value corresponding to the dth group of fluid system classification name and fluid flow; d represents the group number of the first eigenvector; the third machine learning model is trained until the sum of the second prediction accuracies reaches convergence and the training is stopped; the third machine learning model is any one of a regression model and a decision tree.

[0184] Example 3

[0185] This embodiment provides an intelligent pressure transmitter with intelligent anti-interference optimization control, which implements an intelligent pressure transmitter control system with intelligent anti-interference optimization control in Embodiment 1 or Embodiment 2.

[0186] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0187] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent pressure transmitter control system with intelligent anti-interference optimization control, characterized in that the system include: Pressure transmitter acquisition module: used to collect pressure signals of q channels using m groups of MEMS piezoelectric sensor probes, where m is an integer greater than 1, m=q; Pressure signal preprocessing module: used to preprocess the pressure signal to obtain a characteristic graph and input it into the interference recognition module; The method of preprocessing the pressure signal to obtain the characteristic graph includes: Use low-pass filtering to remove high-frequency noise in the pressure signal. Low-pass filtering uses a first-order low-pass filter. The calculation formula is: y(n)=a×x(n)+(1-a)×y(n-1); In the formula, x(n) is the sample value of the input pressure signal, which is the value taken at different time points as n increases; y(n) is the sample value of the output pressure signal; a is the filtering parameter, which takes the value (0, 1) and is used to control the cutoff frequency; y(n-1) is the sample value of the output pressure signal at the previous time point n-1; n is the time point, and the value of n starts from 0 and increases by 1 in sequence, corresponding to the order of the time points; The time-frequency characteristics of the pressure signal are extracted using wavelet transform. The first-level wavelet transform calculation method includes: cD1=∑x(n)×ψ(n); Where cA1 is the approximate component coefficient, cD1 is the detail component coefficient, is the detection function, ψ(n) is the detail function; The second-level wavelet transform calculation method includes: when calculating the second-level approximate component coefficient cA2, cA1 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation; the third-level wavelet transform calculation method includes: when calculating the third-level approximate component coefficient cA3, cA2 is used as the sample value of the input pressure signal, and is substituted into the first-level wavelet transform calculation formula for calculation, and the same is true for each level of wavelet transform detail component coefficient; the approximate component coefficient and the detail component coefficient are both time-frequency characteristics of the pressure signal; The time-frequency characteristics use statistical methods in the small band to extract the time series characteristics of fixed characteristic parameters, including mean value, variance and entropy value; the pressure signal characteristic matrix is ​​constructed with the small band as the unit: the rows in the pressure signal characteristic matrix represent the time-frequency characteristics, and the columns represent the time series characteristics; PCA is used to find the principal component space of the pressure signal feature matrix, and the dimension of the principal component space of the pressure signal feature matrix is ​​reduced. The reduced dimension pressure signal feature matrix is ​​embedded into a low-dimensional space using Autoencoder, and then reconstructed into a feature map. The calculation formula for PCA dimensionality reduction is: Y = X × U; Where: X is the original pressure signal feature matrix, Y is the pressure signal feature matrix after the original pressure signal feature matrix is ​​reduced in dimension, and U is the transformation matrix of the principal component space of the pressure signal; Interference identification module: used to collect environmental parameters at the location of the pressure transmitter, and form sequence data in one-to-one correspondence with the pressure signal in time, input the sequence data to the pre-built first machine learning model to output the category probability sequence of each group of sequence data, and input the feature map to the pre-built second machine learning model to output the category label and the probability distribution corresponding to the category label; Optimization control module: used to model interference signals according to the probability distribution corresponding to the category labels, the category labels and the category probability sequence of each group of sequence data, identify the interference signals and remove the interference signals from the pressure signals to obtain the reconstructed pressure signals, and convert the reconstructed pressure signals into digital signals. The digital signals are the final pressure values ​​output by the intelligent pressure transmitter; The method for modeling interference signals is a Markov process, and Markov process modeling interference signals includes: Set each interference g, define the state space s = {s1, s2, s3}, and define the observation symbol space v = {v1, v2, v3}; calculate the state transition probability matrix D, D is a 3×3 matrix, and the probability calculation of the component element alj of the matrix D is expressed as alj = P(qti+1 = sj|qti = sl), that is, the probability that the state sl turns to sj at the ti-th moment; where P is a probability indicator, qti represents the hidden state at the ti-th moment, sj represents the j-th state in the state space s, and sl represents the l-th state in the state space s; j∈[1,3], l∈[1,3]; calculate the observation probability matrix B, B is a 3×3 matrix, and the probability calculation of the component element blj of the matrix B is expressed as blj = P(vti = vl|qti = sl), that is, the probability of observing vl in state sl; where vti represents the observation symbol at the ti-th moment, and vl is the l-th symbol in the observation symbol space v; The initial probability distribution of the state is initialized to π, and the observation sequence is set to O = o1, o2...oT, ti∈T. The calculation formula for the observation sequence O is: Forward calculation: αt(j) = P(o1, o2...oti, qt = sj|λ); Backward calculation: βt(l)=P(o(ti+1),o(ti+2),o(ti+3)...oT|qt=sl,λ); Among them, λ represents the parameters of the Markov model, λ includes the number of states in the state space s, the number of observation symbols in the observation space v, the state transition probability matrix D, the observation probability matrix B and the initial state distribution π; oti represents the ti-th observation in the observation sequence, T represents the length of the observation sequence, that is, the total number of observation points, and the length unit of the observation sequence is set to 1; αt(j) represents the forward calculation value of the j-th state, βt(l) represents the backward calculation value of the l-th state, and qt represents the hidden state at time t; Calculate the hidden state probability γt(l), the hidden state probability calculation formula is γt(l) = αt(l) × βt(l) / P(O|λ); where αt(l) represents the forward calculation value of the lth state, and P(O|λ) represents the probability that the observation sequence O is generated according to the Markov model after the parameter set λ of the Markov model is given; Calculate the adjacent state transition probability. The formula for calculating the adjacent state transition probability ξt(l,j) is: ξt(l,j)=αt(j)×alj×bj(o(ti+1))×βt(l) / P(O|λ); Among them, bj(o(ti+1)) is the element of B matrix, which represents the probability of occurrence of observation o(ti+1) in state sj; Input the category probability sequence, the category label and the probability distribution corresponding to the category label into the Markov model, where the interference g is the environmental parameter, the category probability sequence is the component element in the state transition probability matrix D, and the category label and the probability distribution corresponding to the category label are the component elements in the observation probability matrix B. Iterate until λ converges to obtain the Markov model parameter λ for sensing the interference signal, that is, the modeling is completed.

2. According to claim 1, the intelligent pressure transmitter control system with intelligent anti-interference optimization control is characterized in that: The environmental parameters include temperature, vibration and acceleration; the sequence data include pressure sequence data, temperature sequence data, vibration sequence data and acceleration sequence data.

3. According to claim 2, the intelligent pressure transmitter control system with intelligent anti-interference optimization control is characterized in that: The training method of the first machine learning model includes: A set of sequence data and the corresponding category probability sequence are combined into a set of historical training sets. G groups of historical training sets are extracted from the database as training sets to train the first machine learning model. The Adam optimizer is used, the batch size is set to 128, the initial learning rate is set to 0.001, and the sliding average method is used to update the parameter momentum and the input and output parameter matrix w. The update equation is: vf=momentum×(vf-1)+(1-momentum)×Δwf; wf+1=wf-η×vf; Where f represents the sequence number of the current iteration step under the sliding average, vf represents the estimated gradient value of the current iteration step f under the sliding average, (vf-1) represents the estimated gradient value of the previous step, Δwf represents the actual gradient descent amount of this step, wf represents the parameter value of the current iteration step f, η is the learning rate that controls the step size, and wf+1 represents the updated parameter value, that is, the learning rate η is used to update wf in the reverse direction vf; The calculation formula of loss function loos is: Where i is the serial number of a single historical training set, yi∈[0,1], yi is the probability distribution of the true label of the i-th historical training set, p i is the probability distribution of the i-th group of historical training sets predicted by the first machine learning model; i≤G; When loos converges, the training is stopped and the training is completed.

4. The intelligent pressure transmitter control system with intelligent anti-interference optimization control according to claim 1 is characterized in that: The second machine learning model construction method includes: The set of all feature maps is used as the input of the second machine learning model, the second machine learning model uses the predicted category label for each feature map and the probability distribution corresponding to the category label as output, uses the actual category label corresponding to each feature map and the probability distribution corresponding to the category label as the prediction target, and uses minimizing the sum of the first prediction accuracies of all predicted category labels and the probability distribution corresponding to the category label as the training target; Among them, the calculation formula for the first prediction accuracy is: Mc e =(Zf e -Sf e ) 2 , among which Mc e is the first prediction accuracy, Zf e is the predicted category label corresponding to the e-th feature map and the probability distribution corresponding to the category label, Sf e is the probability distribution of the actual category label corresponding to the e-th feature map and the category label; e is the serial number representing the number of feature maps; the second machine learning model is trained until the sum of the first prediction accuracies converges; the second machine learning model is any one of CNN and MLP.

5. The intelligent pressure transmitter control system with intelligent anti-interference optimization control according to claim 1 is characterized in that: The method of identifying the interference signal and removing the interference signal from the pressure signal to reconstruct the pressure signal includes: A PF filter is used to eliminate interference signals to obtain a first pressure signal, a low entropy principle is used to reconstruct the first pressure signal to obtain a second pressure signal, a sampling reconstruction method is used to process the second pressure signal to obtain a third pressure signal, and a manifold recognition method is used to screen interference points in the third pressure signal to obtain a reconstructed pressure signal; The first pressure signal is reconstructed using the low entropy principle. Let the first pressure signal be st, the reconstructed first pressure signal be st', and the suppressed first pressure signal be xt, which can be expressed as: xt=st+st'; A set of reconstruction coefficients wi is randomly searched so that the reconstructed first pressure signal st' is expressed as: In the formula, is a basis function, so that the reconstructed first pressure signal st' is closest to the first pressure signal st; Using the low entropy principle, the conditional entropy H(st|xt) of the reconstructed first pressure signal st' is defined as: H(st|xt)=-∫p(st,xt)logp(st|xt)dstdxt; Wherein, p(st,xt) is the joint probability density function of the first pressure signal st and the suppressed first pressure signal xt, logp(st|xt) is the conditional probability density function of the first pressure signal st and the suppressed first pressure signal xt; the pallet function E between the reconstructed first pressure signal st' and the first pressure signal st is defined as: E=∫(st'(x)-st(x)) 2 dx; In the formula, x is the independent variable; Initialize E, minimize the reconstruction coefficient wi, and take the negative minimum of the log-likelihood function, that is: ▽wiE=0; The first pressure signal st and Taking the derivative, we get a system of linear equations: Awi = b; Where A is the coefficient matrix obtained by differentiating the reconstruction coefficient wi, b is the coefficient vector of the linear equation system after differentiation, and is the constant term; Using the normal equation linear equation system, we get the reconstruction coefficient wi: wi=(A T A) -1 A T b; Among them, A T represents the transposed matrix of A; By minimizing the conditional entropy H(st|xt), the reconstruction coefficient wi that best matches the reconstructed first pressure signal st' and the suppressed first pressure signal xt is obtained. At this point, the reconstruction of the second pressure signal is completed. The second pressure signal is processed by a sampling reconstruction method to obtain a third pressure signal, comprising: The second pressure signal is uniformly sampled, a sampling point vector is extracted as a reconstruction basis, a continuous curve is reconstructed using an interpolation method according to the relationship between the sampling points, the reconstruction convention is iteratively optimized to make the reconstructed continuous curve smooth, a third pressure signal is obtained, and a third pressure signal continuous curve is output; The reconstructed pressure signal is obtained by screening the interference points in the third pressure signal using the manifold recognition method, including: A manifold structure of the third pressure signal is constructed, abnormal points are identified in the manifold space, and after the identified abnormal points are removed, a reconstructed pressure signal is obtained according to the manifold structure information of the third pressure signal.

6. The intelligent pressure transmitter control system with intelligent anti-interference optimization control according to claim 1 is characterized in that: The system further comprises: Fluid system access module: when it is detected that the system connected to the intelligent pressure transmitter is a fluid system, the fluid data acquisition module and the fluid pressure measurement module are started; Fluid data collection module: used to collect fluid system classification name and fluid flow rate; Fluid pressure calculation module: used to input the fluid system classification name and fluid flow rate to the pre-built third machine learning model to output the flow rate and pressure value; When the fluid flow rate is set as the interference signal, the pressure value obtained by subtracting the flow rate from the digital signal is used as the final pressure value output by the intelligent pressure transmitter. When the fluid flow is set as the pressure signal source, the digital signal is used as the final pressure value.

7. The intelligent pressure transmitter control system with intelligent anti-interference optimization control according to claim 6 is characterized in that: The fluid system classification names include open fluid systems, pipeline transport liquid systems, constant pressure gas pipeline systems, constant flow gas pipeline systems and pump systems; The method for constructing the third machine learning model includes: Convert each group of fluid system classification names and fluid flow rates into the form of a first feature vector, wherein the elements in the first feature vector include an open fluid system, a pipeline transport liquid system, a constant pressure gas pipeline system, a constant flow gas pipeline system, and a pump system; The set of all first feature vectors is used as the input of a third machine learning model, wherein the third machine learning model uses the flow pressure value predicted for each group of first feature vectors as the output, uses the actual flow pressure value corresponding to each group of first feature vectors as the prediction target, and uses minimizing the sum of the second prediction accuracies of all predicted flow pressure values ​​as the training target; Among them, the calculation formula for the second prediction accuracy is: K d =(Z d -Q d ) 2 , where K d is the second prediction accuracy, Z d is the predicted flow pressure value corresponding to the fluid system classification name and fluid flow of the dth group, Q d is the actual flow pressure value corresponding to the dth group of fluid system classification name and fluid flow; d represents the group number of the first eigenvector; the third machine learning model is trained until the sum of the second prediction accuracies reaches convergence and the training is stopped; the third machine learning model is any one of a regression model and a decision tree.

8. An intelligent pressure transmitter with intelligent anti-interference optimization control, characterized in that: An intelligent pressure transmitter control system for implementing the intelligent anti-interference optimization control described in any one of claims 1-7.

Images