An intelligent simulation and integrated measurement and control method for non-Newtonian fluid dynamic characteristics

Abstract

The present invention discloses a non-Newtonian fluid dynamic characteristics intelligent simulation and integrated measurement and control method, comprising the following steps: sequentially reading basic data such as equipment temperature, pressure, vibration, shear rate, stress, strain, etc., using a mixed Gaussian probability density model for abnormal detection and dynamic weight local polynomial regression cleaning, and performing standardization processing through an adaptive standardization model; performing multi-scale feature extraction and correlation analysis on standardized data, and establishing a mapping relationship between rheological parameters and physical quantities; performing multivariate coupling analysis and using an improved intrinsic orthogonal decomposition method for turbulence correction; performing simulation calculations based on a non-Newtonian fluid analysis model, and evaluating model performance and optimizing model parameters through multi-dimensional error analysis. The present invention can accurately describe the dynamic characteristics of non-Newtonian fluids under complex working conditions, and improve simulation accuracy and control effects.

Description

Technical Field

[0001] The invention relates to simulation technology, in particular to an intelligent simulation and integrated measurement and control method for non-Newtonian fluid dynamic characteristics. Background Art

[0002] Non-Newtonian fluids have broad application prospects in the fields of chemical industry, petroleum, biomedicine, etc. Their unique rheological properties can meet the needs of specific process. However, compared with traditional Newtonian fluids, the flow behavior of non-Newtonian fluids exhibits significant nonlinear characteristics, and their viscosity will undergo complex changes with changes in shear rate, temperature, pressure and other factors. This nonlinear characteristic under the coupling of multiple physical quantities has brought great challenges to the industrial application of non-Newtonian fluids, especially in process processes that require precise control of fluid dynamic properties. How to accurately predict and effectively control the behavior of non-Newtonian fluids has become a key scientific issue restricting its industrial application.

[0003] At present, the research on non-Newtonian fluids mainly relies on laboratory-scale rheological tests and numerical simulations under simplified conditions. Traditional rheological testing methods mainly use rotational rheometers or capillary rheometers. Although these methods can measure the basic rheological parameters of fluids under controlled conditions, it is difficult to simulate the complex flow conditions in actual industrial processes. In terms of numerical simulation, existing research is mainly based on classical rheological constitutive models (such as Power-law model, Bingham model, Casson model, etc.) and simplified computational fluid dynamics methods. These models often show obvious limitations when dealing with multi-physical quantity coupling effects and strong nonlinear characteristics. Especially when considering the coupling effects of temperature field, pressure field and shear field, traditional models are difficult to accurately describe the dynamic response characteristics of non-Newtonian fluids.

[0004] By analyzing the existing technologies, we found that there are several key technical problems: under the coupling of multiple physical quantities, the viscosity-shear relationship of non-Newtonian fluids shows strong nonlinear and time-varying characteristics, and the existing rheological model parameter identification method is difficult to accurately describe this complex relationship; secondly, under turbulent conditions, the shear thickening or shear thinning effect of non-Newtonian fluids will significantly affect the turbulent structure, and the traditional turbulence model cannot accurately capture the impact of this rheological property on turbulent transport; thirdly, due to the lack of effective multi-scale feature extraction methods, it is difficult to identify key features that can characterize the dynamic characteristics of non-Newtonian fluids from massive sensor data; in actual processes, factors such as ambient temperature fluctuations and pressure pulsations will cause dynamic changes in rheological parameters, and existing control methods are difficult to achieve precise control of such time-varying systems. These technical problems seriously restrict the application effect of non-Newtonian fluids in industrial processes. Summary of the invention

[0005] The purpose of the invention is to provide an intelligent simulation and integrated measurement and control method for non-Newtonian fluid dynamic characteristics, in order to solve at least one technical problem in the prior art.

[0006] Technical solution, a method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics, including the following steps:

[0007] Step S1, reading basic data in sequence, and performing abnormality detection and standardization processing to obtain standardized basic data; wherein the basic data includes equipment temperature time series data, pressure time series data, shear rate time series data, stress time series data, strain time series data, vibration frequency data, ambient temperature data and pipeline parameters;

[0008] Step S2, extracting data sequence features and multi-scale time features from the standardized basic data, using a pre-built correlation data model containing rheological parameters and physical quantities to generate feature correlation data; performing multi-scale time series analysis on the feature correlation data to obtain dynamic feature data and steady-state feature data; based on the dynamic feature data and the steady-state feature data, using a nonlinear mapping method to construct a corresponding relationship between rheological parameters and physical quantities to obtain parameter mapping data;

[0009] Step S3, performing multivariable coupling analysis based on the parameter mapping data to obtain coupling characteristic data; performing turbulence effect correction through the coupling characteristic data to generate corrected characteristic data;

[0010] Step S4, performing simulation calculation of the non-Newtonian fluid analysis model based on the corrected characteristic data to obtain simulation data; comparing and analyzing the simulation data with the standardized basic data collected by the experimental device to generate comparative analysis data; establishing an evaluation index system based on the comparative analysis data to obtain model evaluation data.

[0011] The beneficial effect is that it can better fit the diverse rheological behaviors and variable viscosity characteristics of non-Newtonian fluids, effectively improve the accuracy and intelligence level of non-Newtonian fluid pipeline flow simulation, and is a new type of intelligent high-precision simulation system and method that is different from traditional Newtonian fluid simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 It is a flow chart of step S1 of the present invention.

[0014] Figure 3 It is a flow chart of step S2 of the present invention.

[0015] Figure 4 It is a flow chart of step S3 of the present invention.

[0016] Figure 5It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, according to one aspect of the present application, a method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics includes the following steps:

[0018] Step S1, reading basic data in sequence, and performing abnormality detection and standardization processing to obtain standardized basic data; wherein the basic data includes equipment temperature time series data, pressure time series data, shear rate time series data, stress time series data, strain time series data, vibration frequency data, ambient temperature data and pipeline parameters;

[0019] Step S2, extracting data sequence features and multi-scale time features from the standardized basic data, using a pre-built correlation data model containing rheological parameters and physical quantities to generate feature correlation data; performing multi-scale time series analysis on the feature correlation data to obtain dynamic feature data and steady-state feature data; based on the dynamic feature data and the steady-state feature data, using a nonlinear mapping method to construct a corresponding relationship between rheological parameters and physical quantities to obtain parameter mapping data;

[0020] Step S3, performing multivariable coupling analysis based on the parameter mapping data to obtain coupling characteristic data; performing turbulence effect correction through the coupling characteristic data to generate corrected characteristic data;

[0021] Step S4, performing simulation calculation of the non-Newtonian fluid analysis model based on the corrected characteristic data to obtain simulation data; comparing and analyzing the simulation data with the standardized basic data collected by the experimental device to generate comparative analysis data; establishing an evaluation index system based on the comparative analysis data to obtain model evaluation data.

[0022] By building a complete intelligent simulation and integrated measurement and control system, accurate description and prediction of the dynamic characteristics of non-Newtonian fluids are achieved; a closed-loop technical solution is formed from data acquisition, preprocessing, feature extraction to model construction, simulation calculation and performance evaluation; this embodiment effectively solves key problems such as complex nonlinear characteristics, multi-physical quantity coupling and turbulence effects of non-Newtonian fluid systems through multi-source data fusion, multi-scale feature extraction, deep learning mapping, turbulence correction and other methods, and improves model prediction accuracy and calculation efficiency; especially in engineering applications, it can adaptively optimize model parameters according to real-time data, realizing the intelligence and precision of non-Newtonian fluid analysis.

[0023] like Figure 2 As shown, according to one aspect of the present application, step S1 is specifically:

[0024] Step S11, collecting basic data, including: reading device temperature time series data from the device temperature sensor, reading pressure time series data from the pressure sensor, reading shear rate time series data from the shear rate sensor, reading stress time series data from the stress sensor, reading strain time series data from the strain sensor; reading vibration frequency data from the vibration sensor, reading ambient temperature data from the ambient temperature sensor; reading pipeline geometry parameters and pipeline length parameters from the preset parameter database, and reading pipeline surface roughness data from the surface roughness measuring instrument;

[0025] Step S12, obtaining basic data and processing it using a mixed Gaussian probability density model to obtain a probability density distribution function of each parameter; calculating anomaly detection thresholds of each parameter based on the probability density distribution; comparing the basic data with the anomaly detection threshold to generate an abnormal data marker sequence; using a local polynomial regression method with dynamic weights to generate cleaned basic data;

[0026] Step S13: input the cleaned basic data into an adaptive standardization model, which dynamically adjusts the standardization parameters according to the time-varying characteristics of the data to generate standardized basic data.

[0027] Among them, the mixed Gaussian probability density model is specifically P(x|θ) = ∑(k=1→K) πk·N(x|μk,Σk); P(x|θ) is the conditional probability density function of the observed data x under the given parameter θ; N(x|μk,Σk) = (2π)^(-d / 2)|Σk|^(-1 / 2)exp(-1 / 2(x-μk)^T·Σk^(-1)(x-μk)) is the kth Gaussian component; πk is the mixing coefficient and satisfies ∑πk=1; μk is the mean vector of the kth Gaussian component; Σk is the covariance matrix of the kth Gaussian component; x is a d-dimensional input vector, including parameters such as temperature, pressure, and shear rate; K is the number of Gaussian components; θ={πk,μk,Σk} is the model parameter set; the anomaly detection threshold T_i = μi ± α·σi; μi is the mean of the ith parameter; σi is the standard deviation of the ith parameter; α is the confidence coefficient, usually 2.58 (99% confidence); dynamic weight update w_i(t) = exp(-β·e_i(t)^2); e_i(t) is the residual of the ith data point at time t; β is the weight decay coefficient. ^ is a superscript symbol.

[0028] Adaptive normalization model Z(x,t) = [x(t) - μ(t)] / [σ(t) + ε]; x(t) is the original data at time t; μ(t) = λ·μ(t-1) + (1-λ)·x(t) is the dynamic mean; σ(t) = sqrt[λ·σ 2(t-1) + (1-λ)·(x(t)-μ(t)) 2 ] is the dynamic standard deviation; λ is the forgetting factor (0<λ<1); ε is the smoothing factor to prevent the denominator from being zero; time-varying weight function w(t) = exp(-γ·|dZ(t) / dt|); where γ is the weight adjustment coefficient; dZ(t) / dt is the time derivative of the standardized data; adaptive normalization function F(Z) = tanh(η·Z) when |Z|>δ, otherwise F(Z) = Z; where η is the nonlinear adjustment coefficient; δ is the linear interval threshold.

[0029] By constructing a multi-level data preprocessing system, high-quality conversion from raw sensor signals to standardized data is achieved; a mixed Gaussian probability density model is used to detect anomalies in multi-source heterogeneous data. The model can adaptively establish the statistical distribution characteristics of the data and effectively identify outliers; data cleaning is performed using a local polynomial regression method with dynamic weights, which not only maintains the temporal correlation of the data, but also can correct outliers in a targeted manner; an adaptive standardization model is used to dynamically adjust the standardization parameters according to the time-varying characteristics of the data, ensuring the comparability of data at different time scales; the reliability of subsequent analysis is improved, especially when dealing with systems with strong nonlinear characteristics such as non-Newtonian fluids, which provides a basic guarantee for the accuracy of the model; through signal conditioning, analog-to-digital conversion, and spectrum analysis, high-precision acquisition of multiple physical quantities such as temperature, pressure, and shear rate is achieved, and a complete data acquisition link is established, providing a high-quality data foundation for subsequent feature extraction and model construction.

[0030] like Figure 3 As shown, according to one aspect of the present application, step S2 is specifically:

[0031] Step S21, performing wavelet decomposition on the standardized basic data to extract feature sequences at different time scales; fusing the extracted feature sequences with the original standardized basic data, and using an adaptive feature extraction algorithm to obtain sequence feature data; based on the sequence feature data, constructing a temperature-pressure-vibration-shear rate-stress-strain multidimensional correlation model to generate feature correlation data;

[0032] Step S22, inputting the feature association data into a multi-scale time series analysis model pre-built using a recursive neural network, taking into account short-range and long-range time dependencies, extracting fast-changing features and slow-changing features respectively, and obtaining dynamic feature data characterizing transient responses and steady-state feature data characterizing steady-state behaviors;

[0033] Step S23, input the dynamic characteristic data and the steady-state characteristic data into a nonlinear mapping model pre-built using a deep neural network, establish a mapping relationship between rheological parameters and temperature, pressure, vibration and shear rate through multi-layer nonlinear transformation, and generate parameter mapping data.

[0034] Among them, the nonlinear mapping model of the deep neural network is as follows: the residual block H(x) = F(x) + x; F(x) = W2·σ(W1·x + b1) + b2 is the residual mapping; W1 and W2 are weight matrices; the adaptive gating g(x) = sigmoid(Wg·x +bg) is the gating coefficient; the feature selection y = g(x) OH(x); O is the Hadamard product; the multi-task learning objective is L = ∑iαi·Li + β·Ω(θ); Li is the loss function of the i-th task; Ω(θ) is the physical constraint term; αi, β are weight coefficients; the physical constraint Ω(θ) = ||τ - η(γ*)·γ*||² is the rheological constitutive equation constraint; τ is the shear stress; η is the apparent viscosity; γ* is the shear rate.

[0035] By constructing a multi-scale time series feature extraction and nonlinear mapping system, accurate mapping from standardized data to rheological parameters is achieved; the wavelet decomposition method is used to perform multi-scale analysis on the standardized data to extract feature sequences at different time scales; through an adaptive feature extraction algorithm, these feature sequences are fused with the original data to construct a multidimensional correlation model of temperature-pressure-vibration-shear rate-stress-strain; then a recursive neural network is used to construct a multi-scale time series analysis model, while considering short-range and long-range time dependencies, and extracting fast-changing features and slow-changing features respectively; a nonlinear mapping model is established through a deep neural network, and a mapping relationship between rheological parameters and physical quantities is established through multi-layer nonlinear transformations; through multi-level feature extraction and mapping methods, the dynamic characteristics of non-Newtonian fluids at different time scales can be effectively captured, improving the accuracy of rheological parameter prediction.

[0036] like Figure 4 According to one aspect of the present application, step S3 is specifically:

[0037] Step S31, reading and performing tensor decomposition processing on the parameter mapping data to extract the coupling mode between the variables; constructing a multivariate covariance matrix based on the extracted coupling mode, using the recursive least squares method to estimate the dynamic coupling coefficients between the variables, and generating coupling characteristic data;

[0038] Step S32, performing turbulence feature extraction processing based on the coupled characteristic data, separating the turbulent pulsation components based on the improved intrinsic orthogonal decomposition method; using the Reynolds stress correction method to compensate for the shear enhancement effect caused by turbulence, and generating corrected characteristic data.

[0039] By constructing a multivariable coupling analysis and turbulence correction system, an accurate description of the complex flow characteristics of non-Newtonian fluids is achieved; the tensor decomposition model is used to extract the coupling mode between variables, and the dynamic coupling coefficient is estimated by the recursive least squares method; the turbulent pulsation components are separated based on the improved intrinsic orthogonal decomposition method, and the shear enhancement effect caused by turbulence is compensated by the Reynolds stress correction method; this method combining coupling analysis and turbulence correction can accurately describe the nonlinear characteristics of non-Newtonian fluids under complex flow conditions, especially when considering the influence of turbulence. Through accurate turbulent pulsation separation and shear correction, the prediction accuracy of the model under high Reynolds number conditions is improved, providing a reliable theoretical basis for engineering applications.

[0040] like Figure 5 As shown, according to one aspect of the present application, step S4 is specifically:

[0041] Step S41: Based on the corrected characteristic data, a pre-built non-Newtonian fluid analysis model is used to perform non-Newtonian fluid analysis and simulation calculation, and an adaptive numerical solution method is used to generate simulation data including flow field characteristics, stress distribution and temperature field;

[0042] Step S42: Systematically compare the simulation data with the standardized basic data collected by the experimental device, use a multi-dimensional error analysis method to evaluate the model accuracy, identify key deviation points, and generate comparative analysis data; based on the comparative analysis data, optimize the model parameters until the model accuracy reaches a threshold;

[0043] Step S43, construct an evaluation index system including prediction accuracy, calculation efficiency and model stability; perform index evaluation based on comparative analysis data, conduct a comprehensive evaluation of the performance of the non-Newtonian fluid analysis model and the control parameters of the experimental device, and output model evaluation data.

[0044] By constructing a simulation calculation and model evaluation system, accurate verification and optimization of the non-Newtonian fluid analysis model are achieved; an adaptive numerical solution method is used for simulation calculation to generate simulation data containing information such as flow field characteristics, stress distribution, and temperature field; the model accuracy is evaluated through a multi-dimensional error analysis method to identify key deviation points; an evaluation index system including factors such as prediction accuracy, calculation efficiency, and model stability is established. This embodiment can not only accurately predict the flow characteristics of non-Newtonian fluids, but also continuously improve the model accuracy through systematic comparison with experimental data and parameter optimization, thereby achieving adaptive optimization and performance improvement of the model.

[0045] According to one aspect of the present application, step S11 is specifically:

[0046] Step S111, reading the original temperature signal from the device temperature sensor, performing signal conditioning and analog-to-digital conversion on the original temperature signal, and generating device temperature time series data; reading the original pressure signal from the pressure sensor, performing signal conditioning and analog-to-digital conversion on the original pressure signal, and generating pressure time series data;

[0047] Step S112, reading the original shear signal from the shear rate sensor, performing signal conditioning and analog-to-digital conversion on the original shear signal, and generating shear rate time series data; reading the original stress signal from the stress sensor, performing signal conditioning and analog-to-digital conversion on the original stress signal, and generating stress time series data;

[0048] Step S113, reading the original strain signal from the strain sensor, performing signal conditioning and analog-to-digital conversion on the original strain signal, and generating strain time series data; reading the original vibration signal from the vibration sensor, performing signal conditioning and spectrum analysis on the original vibration signal, and generating vibration frequency data;

[0049] Step S114, reading the original ambient temperature signal from the ambient temperature sensor, performing signal conditioning and analog-to-digital conversion on the original ambient temperature signal, and generating ambient temperature data;

[0050] Step S115, read the original pipeline data from the preset parameter database, convert the data format of the original pipeline data, generate pipeline geometry parameters and pipeline length parameters; read the original roughness signal from the surface roughness measuring instrument, perform signal processing and feature extraction on the original roughness signal, and generate pipeline surface roughness data.

[0051] By constructing a multi-source sensor data acquisition and signal processing system, high-precision acquisition of key physical quantities of non-Newtonian fluid systems is achieved; a hierarchical signal processing scheme is adopted to systematically acquire and process physical quantities such as temperature, pressure, shear rate, stress, and strain. Specifically, a special signal conditioning circuit is used for preprocessing of each type of sensor signal to eliminate the influence of environmental noise and interference signals; the analog signal is converted into a digital signal through a high-precision analog-to-digital converter to ensure the quantization accuracy of the signal; for vibration signals, a spectrum analysis method is used to extract frequency domain feature information; in terms of pipeline parameter acquisition, geometric parameters are read through a preset parameter database, and surface roughness measuring instruments are used to obtain surface characteristic data, achieving an accurate description of flow boundary conditions; through a systematic data acquisition method, not only the quality of the original data is improved, but also the reliability and stability of the measurement are improved through the optimized combination of signal processing technologies, laying a solid data foundation for subsequent data analysis and model construction.

[0052] According to one aspect of the present application, step S12 is specifically:

[0053] Step S121, reading basic data and grouping the data, establishing Gaussian kernel functions for temperature, pressure, vibration, shear rate, stress and strain in sequence, inputting the grouped basic data into a mixed Gaussian probability density model for parameter estimation, optimizing the model parameters by the maximum likelihood method, and obtaining the probability density distribution function of each parameter;

[0054] Step S122, performing an integral transformation on the probability density distribution function of each parameter to establish a cumulative distribution function; based on the cumulative distribution function and a preset confidence interval, using the quantile method to calculate the anomaly detection threshold of each parameter;

[0055] Step S123, comparing each parameter value in the basic data with the corresponding anomaly detection threshold; marking the data points that exceed the threshold range, and generating an abnormal data marking sequence containing abnormal data location and type information;

[0056] Step S124, construct a data weight matrix based on the abnormal data marking sequence; calculate the time weight coefficient in combination with the time correlation of the data; input the basic data, the data weight matrix and the time weight coefficient into the local polynomial regression model, correct the abnormal data through the weighted fitting method, and generate the cleaned basic data.

[0057] By constructing an anomaly detection and data cleaning system based on a mixed Gaussian probability density model, the intelligent preprocessing and quality improvement of raw data are achieved. First, different types of data are grouped, and Gaussian kernel functions are established in a targeted manner. The model parameters are optimized by the maximum likelihood method to obtain the probability density distribution function of each parameter; based on the cumulative distribution function and the preset confidence interval, the quantile method is used to calculate the anomaly detection threshold; then, through the combination of the data weight matrix and the time weight coefficient, the intelligent identification and correction of abnormal data are achieved; through the data cleaning method based on probability statistics and dynamic weights, not only can various types of abnormal data be accurately identified and processed, but also by considering the time correlation of the data, the physical continuity of the data is maintained, the reliability and representativeness of the data are improved, and high-quality data support is provided for subsequent feature extraction and model construction.

[0058] According to one aspect of the present application, step S13 is specifically:

[0059] Step S131, segmenting the cleaned basic data according to fixed time windows to generate time series segmented data; performing sliding window analysis on the time series segmented data, extracting local statistical features within each window, including mean, variance and skewness, and generating time-varying feature data;

[0060] Step S132: construct an adaptive weight function based on the time-varying feature data, calculate the data weights at different time scales, and generate time-varying weight data; input the time-varying weight data into a recursive least squares algorithm, estimate the dynamic mean and dynamic standard deviation parameters, and generate standardized parameter data;

[0061] Step S133, smoothing the standardized parameter data, using an exponentially weighted moving average method to eliminate parameter mutations, and generating smoothed parameter data; constructing an adaptive normalization function based on the smoothed parameter data, including two branches of linear transformation and nonlinear mapping, and generating normalized function data;

[0062] Step S134: input the cleaned basic data and normalized function data into the adaptive standardization model, automatically select the appropriate normalization branch according to the data distribution characteristics, realize the adaptive standardization of the data through dynamic parameter adjustment, and generate standardized basic data for subsequent analysis.

[0063] By constructing an adaptive standardization model, intelligent standardization of cleaned data is achieved; first, a sliding window analysis method is used to extract local statistical features, including time-varying features such as mean, variance and skewness; based on these features, an adaptive weight function is constructed, and the dynamic mean and standard deviation parameters are estimated by a recursive least squares algorithm; then the exponentially weighted moving average method is used to smooth the parameters to eliminate the impact of mutations; dynamic standardization of data is achieved through an adaptive normalization function. This embodiment can dynamically adjust the standardized parameters according to the time-varying characteristics of the data, effectively handle the non-stationary characteristics in non-Newtonian fluid systems, ensure the physical rationality and time consistency of the standardized results, and provide a standardized data basis for subsequent feature extraction and model construction.

[0064] According to one aspect of the present application, step S21 is specifically:

[0065] Step S211, reading and performing wavelet decomposition on the standardized basic data, performing data decomposition by a multi-scale wavelet transform method to obtain wavelet coefficient data of different scales; performing threshold processing on the wavelet coefficient data, extracting characteristic components at each scale, and generating a multi-scale characteristic sequence;

[0066] Step S212, extracting the shear rate time series data and stress time series data in the basic data and performing rheological property analysis, calculating the instantaneous shear viscosity distribution, and obtaining shear viscosity data; calculating the viscosity parameters based on the strain time series data and the stretching rate, and generating viscosity data; combining the shear viscosity data and the viscosity data, constructing a non-Newtonian index calculation model, and generating rheological index data;

[0067] The rheological properties analysis includes: instantaneous shear viscosity η(γ*,t) = τ(t) / γ*(t)·[1 + α·(dγ* / dt) / (γ*) 2 ]; where τ(t) is the shear stress; γ*(t) is the shear rate; α is the time correction coefficient; extensional viscosity ηE(ε*,t) = (τ11 - τ22) / ε*; where τ11,τ22 are the normal stress differences; ε* is the extension rate; non-Newtonian index n(t) = d(log τ) / d(log γ*); viscosity correction coefficient λ(T) = exp[-E / (R·T)]; where E is the activation energy; R is the gas constant; T is the temperature; comprehensive rheological index I = w1·n + w2·(η / η0) + w3·(ηE / η0); where η0 is the reference viscosity; w1~w3 are weight coefficients.

[0068] Step S213, feature fusion of the multi-scale feature sequence and the standardized basic data, extracting fusion features using the principal component analysis method, and generating fusion feature data; performing adaptive feature extraction processing on the fusion feature data, extracting time series features through a recursive neural network, and generating sequence feature data;

[0069] The calculation process of feature fusion is: principal component analysis Y = W^T·X; where: W is the feature vector matrix; X is the standardized data matrix; feature importance VI(i) = λi / ∑j λj; where λi is the eigenvalue; recursive feature extraction h(t) = f(Wx·x(t) + Wh·h(t-1) + b); where Wx, Wh are weight matrices; b is the bias vector; f is the activation function; feature fusion F = σ(W·[Y; h] + b); adaptive feature selection s(t) = sigmoid(v^T·tanh(W·F + b)); the final feature Z(t) = s(t) OF, O is the Hadamard product.

[0070] Step S214, read the sequence feature data, rheological index data and viscosity data and perform tensor decomposition, extract the coupling mode between the parameters, and generate coupling mode data; construct a temperature-pressure-vibration-shear rate-stress-strain-viscosity multidimensional correlation network based on the coupling mode data, learn the nonlinear mapping relationship between the parameters through a deep neural network, and generate feature correlation data.

[0071] In step S211, the wavelet decomposition and feature extraction algorithm W(a,b) = 1 / sqrt(a)·∫x(t)·ψ*((tb) / a)dt is a continuous wavelet transform; wherein: x(t) is the input signal; ψ(t) is the wavelet mother function; a>0 is the scale parameter; b is the translation parameter; ψ* is the complex conjugate of the wavelet function; multi-scale coefficient: c(j,k) = ∑nh(n-2k)c(j-1,n) is the scale coefficient; d(j,k) = ∑ng(n-2k)c(j-1,n) is the wavelet coefficient; wherein h(n) is a low-pass filter; g(n) is a high-pass filter; j is the number of decomposition levels; k is the translation parameter; characteristic energy: E(j) = ∑k|d(j,k)| 2 / N is the normalized energy of the j-th layer wavelet coefficient; N is the signal length; threshold processing: η(d) = sign(d)(|d|-λ) when |d|>λ, otherwise η(d)=0; λ is the threshold parameter; feature sequence reconstruction: x'(t) = ∑j∑k η(d(j,k))·ψj,k(t).

[0072] Multidimensional association network tensor decomposition algorithm: T ≈ G ×1U1 ×2 U2 ×3 U3 is Tucker decomposition; where: T∈R*^(I×J×K) is the original tensor; G∈R*^(P×Q×R) is the core tensor; U1∈R*^(I×P), U2∈R*^(J×Q), U3∈R*^(K×R) are factor matrices; × i is the tensor product of the i-th mode; the associated weight w(i,j,k) = |G(i,j,k)| / ||G||F; where ||G||F is the Frobenius norm; the feature association R(i,j) = ∑kw(i,j,k)·cos(U1(:,i), U2(:,j)); where cos(·,·) is the cosine similarity; the nonlinear mapping function f(x) = σ(W·[vec(G);vec(U1); vec(U2); vec(U3)]); where σ is the activation function; W is the weight matrix; vec(·) is the vectorized operation.

[0073] By constructing a multi-scale feature extraction and correlation analysis system, the systematic extraction and correlation modeling of the characteristics of the non-Newtonian fluid system are realized; first, the standardized data is decomposed by multi-scale wavelet transform to extract the characteristic components at different scales; at the same time, the instantaneous shear viscosity and extension viscosity are calculated by the rheological characteristic analysis model, and a non-Newtonian index calculation model is constructed; the multi-scale feature sequence is fused with the original data, and the fusion features are extracted by the principal component analysis method; the coupling mode between parameters is extracted by the tensor decomposition model, and a temperature-pressure-vibration-shear rate-stress-strain-viscosity multidimensional correlation network is established. This embodiment can not only fully capture the dynamic characteristics of non-Newtonian fluids at different scales, but also learn the nonlinear mapping relationship between parameters through a deep neural network, thereby realizing an accurate description of the complex behavior of the system, laying the foundation for subsequent dynamic feature extraction and parameter mapping.

[0074] According to one aspect of the present application, step S22 is specifically:

[0075] Step S221, segmenting the feature association data according to different time scales to generate short program sequence data of a short time window and long program sequence data of a long time window; performing time step normalization processing on the short program sequence data and the long program sequence data respectively to generate normalized time series data;

[0076] Step S222: input the normalized time series data into the long short-term memory network, extract the time series features through the forget gate and input gate mechanism, and obtain the memory state data; perform attention mechanism calculation on the memory state data, identify the feature weights of key time points, and generate weighted feature data;

[0077] Step S223, input the weighted feature data into the time convolution network, extract local features of different time scales through multi-layer dilated convolution, and obtain multi-scale feature data; perform time domain decomposition on the multi-scale feature data, use wavelet packet transform method to separate fast-changing components and slow-changing components, and generate decomposed feature data;

[0078] Step S224, input the fast-changing components in the decomposed feature data into the transient feature extractor, identify the jump characteristics through the adaptive threshold method, extract the system transient response characteristics, and generate dynamic feature data characterizing the transient response; input the slowly changing components in the decomposed feature data into the steady-state feature extractor, extract the system steady-state behavior characteristics through the sliding average and trend analysis method, and generate steady-state feature data characterizing the steady-state behavior.

[0079] In step S222, the long short-term memory network temporal feature extraction is specifically ft = σ(Wf·[ht-1,xt] +bf); ft is the forget gate; it = σ(Wi·[ht-1,xt] + bi), it is the input gate; ot = σ(Wo·[ht-1,xt] +bo); ot is the output gate; ct = ft O ct-1 + it O tanh(Wc·[ht-1,xt] + bc) is the unit state; ht =ot O tanh(ct) is the hidden state; where: xt is the input vector; W is the weight matrix; b is the bias vector; σ is the sigmoid function; O is the Hadamard product; attention weight αt = softmax(v^T·tanh(W1·ht + W2·s)); where s is the reference vector; weighted feature: ct = ∑t αt·ht; feature fusion F = φ(Wf·[ct; ht] + bf); where φ is a nonlinear activation function.

[0080] By constructing a multi-scale time series analysis model, the dynamic characteristics and steady-state characteristics of the non-Newtonian fluid system are accurately extracted; first, the feature-related data is segmented in time scale, and the short-range and long-range time series characteristics are processed separately; the time series characteristics are extracted through the long short-term memory network, and the attention mechanism is used to identify the key time points; then the time convolution network is used to extract multi-scale local features through multi-layer hole convolution; the fast-changing components and slow-changing components are separated by the time domain decomposition method, and the transient response characteristics and steady-state behavior characteristics are extracted respectively. This embodiment can simultaneously consider the short-range and long-range time dependencies of the system, accurately capture the dynamic behavior of non-Newtonian fluids at different time scales, and improve the accuracy and completeness of feature extraction, especially when dealing with complex flow problems with multi-time scale coupling.

[0081] According to one aspect of the present application, step S23 is specifically:

[0082] Step S231, feature fusion of the dynamic feature data and the steady-state feature data, and dimension reduction processing of the fused features by using an autoencoder to obtain compressed feature data; normalization processing of the compressed feature data, and generation of normalized feature data by a maximum and minimum value scaling method;

[0083] Step S232: input the normalized feature data into a multi-layer perceptron, extract deep features through layer-by-layer nonlinear transformation, and obtain deep feature data; perform residual connection on the deep feature data, establish a shortcut path between features at different levels, and generate residual feature data;

[0084] Step S233, inputting the residual feature data into the attention module, calculating the association weights between different physical quantities, and obtaining weight feature data; constructing an adaptive feature selector based on the weight feature data, dynamically adjusting the feature importance through a gating mechanism, and generating selected feature data;

[0085] Step S234, input the selected feature data into the mapping network, use a multi-task learning method to simultaneously predict multiple rheological parameters to obtain predicted parameter data; perform physical constraint optimization on the predicted parameter data, introduce the rheological constitutive equation as a soft constraint condition, adjust the parameter prediction value, and generate parameter mapping data with physical meaning.

[0086] By constructing a nonlinear mapping model of a deep neural network, an accurate mapping from dynamic features to rheological parameters is achieved; first, an autoencoder is used to reduce the dimension of the fused features; a multi-layer perceptron is used to perform layer-by-layer nonlinear transformation, and a residual connection is introduced to establish a shortcut path for features at different levels; then an attention module is used to calculate the association weights between different physical quantities, and the importance of features is dynamically adjusted through a gating mechanism; a multi-task learning method is used to simultaneously predict multiple rheological parameters, and a rheological constitutive equation is introduced as a soft constraint condition for physical constraint optimization. This embodiment can not only accurately establish a nonlinear correspondence between rheological parameters and physical quantities, but also ensure the physical rationality of the prediction results by introducing physical constraints, thereby improving the reliability and applicability of the model in practical engineering applications.

[0087] According to one aspect of the present application, step S31 is specifically:

[0088] Step S311, reconstruct the parameter mapping data into a three-order tensor structure, which respectively represents the time dimension, the physical quantity dimension and the parameter dimension, and generate tensor structure data; perform Tucker decomposition on the tensor structure data, extract the core tensor and the factor matrix, and obtain the decomposition feature data;

[0089] Step S312, performing singular value decomposition on the factor matrix in the decomposed characteristic data, extracting the main coupling mode, and generating mode characteristic data; projecting the mode characteristic data into the original space, reconstructing the coupling relationship between the variables, and obtaining the reconstructed characteristic data;

[0090] Step S313: construct a multivariate covariance matrix based on the reconstructed feature data, calculate the correlation coefficients between the variables, and generate correlation coefficient data; divide the correlation coefficient data into time windows, establish a sliding covariance matrix, and obtain time-varying covariance data;

[0091] Step S314: input the time-varying covariance data into a recursive least squares algorithm, adaptively estimate the coupling coefficient through a forgetting factor method, and obtain coupling coefficient data; combine the coupling coefficient data and the reconstruction feature data, construct a dynamic coupling model, and generate coupling feature data.

[0092] Among them, the multivariable coupling analysis tensor decomposition algorithm X(t) = C ×1 A(t) ×2 B(t) is a dynamic tensor decomposition; where: X(t)∈R*^(I×J×K) is the time-varying observation tensor; C∈R*^(R×R×R) is the core tensor; A(t)∈R*^(I×R), B(t)∈R*^(J×R) are time-varying factor matrices; R is the rank parameter; dynamic update: A(t) = A(t-1) + η·grad A(t); B(t) = B(t-1) + η·grad B(t); where η is the learning rate; covariance matrix: Σ(t) = λ·Σ(t-1) + (1-λ)·X(t)·X(t)^T; where λ is the forgetting factor; coupling coefficient: ρij(t) = Σij(t) / sqrt(Σii(t)·Σjj(t)); Dynamic coupling strength: S(t) = ||grad ρ(t)||F / ||ρ(t)||F; where ||·||F is the Frobenius norm.

[0093] By constructing a multivariable coupling analysis system based on tensor decomposition, an accurate description of the complex coupling relationship between parameters of a non-Newtonian fluid system is achieved; first, the parameter mapping data is reconstructed into a third-order tensor structure, and the core tensor and factor matrix are extracted by Tucker decomposition; the factor matrix is ​​subjected to singular value decomposition to extract the main coupling modes and reconstruct the coupling relationship between variables; then a multivariate covariance matrix is ​​constructed, and a time-varying covariance matrix is ​​established by a sliding window method; the coupling coefficient is adaptively estimated by a recursive least squares algorithm and a forgetting factor method. This embodiment can effectively handle the complex nonlinear coupling relationship between multiple physical quantities, especially when dealing with a non-Newtonian fluid system with strong time-varying characteristics, by adaptively estimating the dynamic coupling coefficient, the time-varying coupling characteristics between system parameters are accurately captured, and a reliable coupling feature description is provided for subsequent turbulence correction.

[0094] According to one aspect of the present application, step S32 is specifically:

[0095] Step S321, performing time-frequency analysis on the coupled characteristic data, extracting multi-scale turbulence characteristics by wavelet transform method, and obtaining turbulence scale data; performing energy spectrum analysis on the turbulence scale data, identifying turbulence energy distribution characteristics, and generating energy spectrum data;

[0096] Step S322, input the energy spectrum data into the improved intrinsic orthogonal decomposition model, extract the turbulent structure characteristics through the adaptive modal decomposition method, and obtain modal characteristic data; reconstruct the modal characteristic data, separate the turbulent pulsation components, and generate pulsation characteristic data;

[0097] Step S323, calculating the Reynolds stress tensor based on the pulsation characteristic data, adjusting the stress distribution by an anisotropic correction method, and obtaining stress correction data; inputting the stress correction data into the turbulent viscosity model, calculating the turbulent viscosity correction coefficient, and generating viscosity correction data;

[0098] Step S324: input the coupling characteristic data, the pulsation characteristic data and the viscosity correction data into the shear correction model to compensate for the shear enhancement effect caused by turbulence and generate corrected characteristic data.

[0099] Improved proper orthogonal decomposition method φ(x,t) = ∑i ai(t)·ψi(x) is POD decomposition; where: φ(x,t) is the flow field variable; ψi(x) is the spatial modal function; ai(t) is the time coefficient; modal energy Ei = λi / ∑j λj; where λi is the eigenvalue; turbulent pulsation separation u'(x,t) = φ(x,t) – φ*(x); where φ* is the time-averaged field; improved mode selection ψ'i = ψi + γ·∑j≠i (ψi^T·R·ψj)·ψj; where R is the Reynolds stress tensor; γ is the adjustment coefficient; reconstruction error E(n) = ||φ - ∑(i=1→n) ai·ψ'i|| 2 / ||φ| |2 .

[0100] By constructing a turbulence feature extraction and correction system, the turbulence effect of non-Newtonian fluids is accurately corrected; first, the time-frequency analysis is performed through the wavelet transform method, the multi-scale turbulence features are extracted and the energy spectrum analysis is performed; the turbulence structure features are extracted by the improved intrinsic orthogonal decomposition model, and the turbulence pulsation components are separated by the adaptive modal decomposition method; then the Reynolds stress tensor is calculated, and the stress distribution is adjusted by the anisotropic correction method; the correction coefficient is calculated by the turbulent viscosity model to compensate for the shear enhancement effect caused by turbulence. This embodiment can not only accurately separate and describe the turbulent pulsation characteristics, but also realize the accurate correction of the rheological characteristics of non-Newtonian fluids under turbulent conditions by considering the anisotropic effect of turbulent viscosity, and improve the prediction accuracy of the model under high Reynolds number conditions.

[0101] According to one aspect of the present application, step S41 is specifically:

[0102] Step S411, input the corrected characteristic data into the grid generator, adaptively divide the calculation domain according to the flow field characteristics, and generate grid data; perform grid quality evaluation on the grid data, improve the grid quality through node smoothing and unit optimization methods, and obtain optimized grid data;

[0103] Step S412: input the optimized grid data and the corrected characteristic data into the control equation discretizer, discretize the momentum equation and the energy equation using the finite volume method, and generate discrete equation data; construct an algebraic equation system based on the discrete equation data, and obtain the system equation data through matrix assembly;

[0104] Step S413, inputting the system equation data into the multi-grid solver, accelerating convergence through coarse grid correction and fine grid iteration method, and obtaining iterative calculation data; performing convergence judgment on the iterative calculation data, determining the final solution based on residual analysis, and generating numerical solution data;

[0105] Step S414, input the numerical solution data into the field quantity reconstruction model, calculate the flow field velocity distribution, pressure distribution and temperature distribution, and obtain the field quantity distribution data; post-process the field quantity distribution data, extract the flow field characteristics, stress distribution and temperature field information, and generate simulation data.

[0106] The numerical solution is specifically as follows: Lu = f is the discretization equation; where L is the discrete operator; multiple grid iteration: uk+1 =uk + P·(I - R·L·P)^ν·R·(f - L·uk); where P is the continuation operator; R is the restriction operator; ν is the smoothing order; grid adaptability index: η_K = h_K||[grad u]||dK; where h_K is the unit size; [grad u] is the gradient jump; adaptive time step: Δt = CFL·h_min / max(|u|); where CFL is the Courant number; h_min is the minimum grid size; residual convergence: ||r_k|| / ||r_0|| < ε; where r_k is the residual of the kth iteration.

[0107] By constructing an adaptive numerical calculation system, an efficient solution to the non-Newtonian fluid analysis model is achieved; first, the calculation domain is adaptively divided based on the flow field characteristics, and the grid quality is improved through node smoothing and unit optimization; the control equation is discretized using the finite volume method, and the system equation is constructed through matrix assembly; then a multi-grid solver is used to accelerate convergence, and the final solution is determined through residual analysis; the numerical solution is reconstructed to extract flow field characteristics, stress distribution and temperature field information. This embodiment can not only dynamically optimize the calculation grid according to the flow field characteristics, but also improve the solution efficiency through the multi-grid technology, especially when dealing with non-Newtonian fluid problems with strong nonlinear characteristics, the stability and accuracy of the numerical calculation are guaranteed through the adaptive solution strategy.

[0108] According to one aspect of the present application, step S42 is specifically:

[0109] Step S421, aligning the simulation data with the standardized basic data collected by the experimental device in the time dimension, eliminating the time deviation through the dynamic time warping algorithm, and obtaining the aligned data; performing feature space mapping on the aligned data, establishing the corresponding relationship between the simulation results and the experimental data, and generating the mapping data;

[0110] Step S422: performing error analysis on the mapping data, calculating the root mean square error and relative error of multiple physical quantities, and obtaining error statistical data; performing spatiotemporal distribution analysis on the error statistical data, identifying the main sources and distribution characteristics of the errors, and generating error distribution data;

[0111] Step S423: locate key deviation points based on the error distribution data, identify high error areas through cluster analysis, and obtain deviation feature data; perform sensitivity analysis on the deviation feature data, identify key parameters that affect model accuracy, and generate sensitivity data;

[0112] Step S424: input the sensitivity data and deviation characteristic data into a preconfigured parameter optimizer, optimize the model parameters using the gradient descent method, and obtain optimized parameter data; based on the optimized parameter data and error statistical data, generate comparative analysis data including error evaluation results and optimization suggestions.

[0113] Root mean square error RMSE = sqrt(1 / N·∑i (yi – y*i)²); relative error RE = |yi – y*i| / |yi|; where yi is the true value; y*i is the predicted value; N is the number of samples; time alignment error: DT = argmin_τ ∑t |x(t)- y(t+τ)| 2 ; where τ is the time offset; sensitivity coefficient: Si = (dy / dxi)·(xi / y); where xi is the input parameter and d is the sign of the partial derivative; parameter optimization: θk+1 = θk - α·grad L(θk); where L is the loss function; α is the learning rate; comprehensive evaluation index: Q = w1·RMSE + w2·RE + w3·DT + w4·S; where w1~w4 are weight coefficients.

[0114] By building a systematic model evaluation and optimization system, accurate verification and continuous optimization of the non-Newtonian fluid analysis model are achieved; first, the dynamic time warping algorithm is used to time align the simulation results and experimental data; the root mean square error and relative error are calculated through multi-dimensional error analysis to identify the distribution characteristics of the error; then the key deviation points are located through cluster analysis, and sensitivity analysis is performed to identify key parameters; the gradient descent method is used to optimize the model parameters, and an analysis report containing error evaluation results and optimization suggestions is generated. This embodiment can not only systematically evaluate the model performance, but also achieve continuous improvement of the model through parameter sensitivity analysis and optimization, and improve the reliability and applicability of the model in actual engineering applications.

[0115] In order to solve the problem of identifying the viscosity-shear nonlinear relationship under the coupling of multiple physical quantities, multi-source sensors are used to synchronously collect data of physical quantities such as temperature, pressure, shear rate, stress, and strain, realizing complete monitoring of the coupled physical field. Multi-scale decomposition is performed through wavelet transform, and the instantaneous shear viscosity and extensional viscosity are calculated in combination with the rheological characteristic analysis model. The coupling mode between parameters is extracted through the tensor decomposition model, and a multi-dimensional correlation network of temperature-pressure-vibration-shear rate-stress-strain-viscosity is established. It can accurately capture the nonlinear characteristics of non-Newtonian fluids under the coupling of multiple physical quantities, thus solving the problem that traditional methods are difficult to describe complex coupling relationships.

[0116] In order to address the issue of the influence of rheological properties under turbulent conditions, the wavelet transform method is used to perform time-frequency analysis and extract multi-scale turbulence characteristics; the improved intrinsic orthogonal decomposition model is used to separate the turbulent pulsation components; the correction coefficient is calculated through the turbulent viscosity model to compensate for the shear enhancement effect caused by turbulence; this systematic turbulence correction method not only considers the influence of the rheological properties of non-Newtonian fluids on the turbulent structure, but also adjusts the stress distribution through the anisotropic correction method, thereby achieving an accurate description of the turbulent transport characteristics.

[0117] To address the problem of multi-scale feature extraction, we first segment the feature-related data into time scales, extract temporal features through long short-term memory networks and attention mechanisms, and then use a temporal convolutional network to extract multi-scale local features through multi-layer dilated convolutions. This approach can simultaneously consider the short-range and long-range temporal dependencies of the system, effectively solving the problem of extracting key features from massive data.

[0118] To solve the problem of precise control of time-varying systems, an adaptive normalization model is adopted to dynamically adjust the normalization parameters according to the time-varying characteristics of the data. A nonlinear mapping model is established through a deep neural network, and the rheological constitutive equation is introduced as a soft constraint. Time alignment is performed through a dynamic time warping algorithm, and continuous improvement of the model is achieved through parameter sensitivity analysis and optimization.

[0119] In this application, other calculation processes are as follows:

[0120] The turbulent viscosity correction process is: Reynolds stress tensor τij = -ρ·u ' i.u ' j; anisotropy tensor bij = (τij / k - 2δij / 3); where k is the turbulent kinetic energy; δij is the Kronecker function; modified viscosity νt = Cμ·fμ·k 2 / ε; Cμ is the model constant; fμ is the damping function; fμ = [1 - exp(-y+ / Aμ)] 2 ·[1 + 3.45 / sqrt(Ret)]; where y+ is the dimensionless wall distance; Ret is the turbulent Reynolds number; correction factor: C = 1 + α·|S|·k / ε + β·|Ω|·k / ε; where |S| is the strain rate modulus; |Ω| is the vorticity modulus; α, β are correction coefficients.

[0121] The multi-grid solution process can be: error equation Ah·eh = rh; where Ah is the fine grid operator; coarse grid equation AH·eH = IH·rh; where IH is the restriction operator; extension correction eh = Ih·eH; where Ih is the extension operator; smooth iteration: u(k+1) = u(k) + ω·D^(-1)·(f - A·u(k)); where D is the diagonal operator; ω is the relaxation factor; V-cycle: MGV(A,u,f,l) = Sl·(A·Pl·Sl^(-1)·Rl·(f - A·u)); where l is the number of grid layers; Sl is the smoothing operator; convergence criterion: ||r(k)|| / ||r(0)|| < ε + ||A||·δ; where δ is the rounding error limit.

[0122] Parameter sensitivity analysis includes: local sensitivity Si = (dy / dxi)·(xi / y); variance decomposition V(y) = ∑iVi + ∑i∑j>i Vij + ... + V12...n; Sobol index STi = Vi / V(y); where Vi is the variance contribution of the i-th parameter; interaction effect Iij = Vij / (Vi + Vj); Morris effect μi* = 1 / r·∑k |EEi(k)|; σi =sqrt(1 / r·∑k (EEi(k) - μi) 2 );EEi is the basic effect;r is the number of samplings;comprehensive sensitivity: GSi =w1·Si + w2·STi + w3·Iij + w4·μi*;w1~w4 are weight coefficients.

[0123] According to one aspect of the present application, the data processing flow is as follows:

[0124] Step S1, establish a complete data acquisition system to monitor the key parameters of non-Newtonian fluid flowing in the pipeline in real time. Mainly collect core parameters such as temperature, pressure, shear rate, stress, strain, etc., and record environmental parameters such as vibration frequency and pipeline geometry parameters. For core parameters, a sampling frequency of 100Hz is used to ensure the capture of rapidly changing dynamic characteristics, and environmental parameters are sampled at a lower frequency of 10Hz. The collected raw data is processed by outlier detection and elimination, data normalization, and time alignment of different sensor data is ensured by timestamps, and finally a cleaned standardized data set is obtained.

[0125] The measured stress τ-strain rate du / dy relationship curve of the fluid is used to characterize the rheological behavior of the non-Newtonian fluid. The intelligent control module in the system issues instructions to the test module to adjust the test equipment. Under ideal conditions without external loads, the fluid is allowed to circulate back and forth in the closed-loop test pipeline. The online dynamic rheometer built into the module continuously changes the stirring shear rate of the fluid within the test time T. At the beginning of the test and every time t, a set of stress values ​​corresponding to the instantaneous strain rate are measured, and a total of n strain rate and stress data pairs are obtained, expressed as [(x1, y1), (x2.y2)...(xn, yn)]. Among them, y represents stress τ, and x represents strain rate du / dy. Generally, n=T / t+1. The general test time is 3~6h, and the time interval t is 10~30min.

[0126] If through observation and analysis, it is found that the stress y has an obvious and regular time lag effect when changing with the strain rate x, that is, the change of stress y always occurs later than the change of strain rate, and the hysteresis phenomenon presents a regular repetition, for example, the stress always changes after the strain rate changes by Δt, then it can be considered that the test fluid has a creep effect. For the test fluid with creep effect, the test time is 12~24h, and the time interval t is 60min.

[0127] Step S2: The strain rate and stress data of the test fluid obtained from the test are transmitted back to the intelligent controller, and a two-dimensional plane coordinate system is generated in the calculation space, and discrete data points D1, D2...Dn are plotted. The intelligent optimization algorithm is used to fit the data with high precision to obtain the stress-strain rate change curve of the test fluid. According to the optimized stress-strain rate curve of the test fluid, the rheological constitutive equation corresponding to the curve is analyzed, and archived to the database, and synchronously imported into the simulation module as the rheological model used for subsequent calculations.

[0128] Specifically, the SVM support vector machine algorithm is used for high-precision fitting. The support vector machine realizes complex nonlinear fitting by constructing a suitable high-dimensional space. When fitting the strain rate-stress of the fluid, the advantage of SVM is its strong generalization ability and robustness to small sample data. Among them, the intelligent fitting method is:

[0129] The rheological properties of non-Newtonian fluids are modeled using SVM and Neural ODE (Neural Ordinary Differential Equation) methods. The traditional rheological constitutive equation is transformed into a dynamic system based on a neural network, and the dynamic model of the fluid is obtained through training with a deep learning framework. Specifically, a neural network structure consisting of an input layer, multiple hidden layers, and an output layer is constructed. The input is the current state variable (such as stress, strain rate, etc.), and the output is the rate of change of these state variables. The Neural Ordinary Differential Equation is solved by numerical integration to obtain a prediction of the time evolution of the fluid behavior. The Adam optimizer is used during model training to minimize the error between the predicted value and the actual observed value.

[0130] The stress-strain rate relationship curve of the test fluid is obtained by fitting, and the curve is compared with the existing non-Newtonian fluid rheological curve in the database to determine the type of non-Newtonian fluid to which the test fluid belongs. The strategies for evaluating and optimizing the fitting results include: calculating the goodness of fit and using R² or mean square error (MSE) indicators to evaluate the effect of curve model fitting; analyzing the residual of the fitting results to identify specific local overfitting or underfitting data points, and further fine-tuning the fitting model and parameter values ​​until the fitting accuracy reaches the preset value. The fitting accuracy is generally controlled within ±1%.

[0131] Step S3, after determining the fluid type to which the non-Newtonian fluid belongs and the specific rheological model of the fluid used for this simulation, it is also necessary to analyze the viscosity response law of the non-Newtonian fluid under specific complex working conditions. Since the nonlinear viscosity characteristics of the non-Newtonian fluid will affect the turbulence characteristics and the selection of turbulence models during flow, it is necessary to couple the actual comprehensive impact of various different variables under complex working conditions on the fluid viscosity characteristics, quantify the functional relationship between each variable and viscosity, and thus correct the fluid flow model used in the simulation calculation, which is mainly achieved by correcting the viscosity term in the turbulence model.

[0132] Analyze and extract various conditional variables under the complex working conditions of the test fluid. Common variables include temperature, pressure, vibration, flow rate, pipe diameter, pipe roughness, etc. (each variable can be represented by letters such as A, B, C, etc.). Assuming that there are n conditional variables in this simulation, the control variable method is used to carry out single variable experiments through the test module, and the influence of these n variables on the viscosity of the test fluid is decoupled in turn (represented by mathematical functions such as F(A), F(B), F(C)).

[0133] Specifically, take temperature as an example: the letter A represents the temperature variable, and n sets of temperature variable A values ​​are preset [A1, A, 2A3...An]. The test module only adjusts the temperature without changing other variables. The viscosity of the fluid at each temperature is measured in turn by the built-in online viscometer, thereby obtaining a set of temperature-viscosity data pairs: [(A1, μ1), (A2, μ2)... (An, μn)], a total of n sets of data pairs, and a two-dimensional coordinate system is established with A on the X axis and μ on the Y axis. A nonlinear fitting algorithm is used to fit the viscosity versus temperature curve based on discrete data points, and R is controlled. 2 Reach 0.98 and above, and analyze the functional relationship between viscosity and temperature according to the curve μ A =F(A).

[0134] Step S4, decouple the influence of n variables on viscosity under complex working conditions in turn through the above method, and obtain the mathematical function expressions of n viscosities: F(A), F(B), F(C)...F(N). Use artificial intelligence algorithm to couple the influence of each variable on viscosity, generate a comprehensive change curve of the viscosity of the fluid under complex working conditions, and obtain a function expression μ= F(A, B, C...)=αF(A)+βF(B)+γF(C)+..., and use this formula to correct the viscosity term (generally a constant term) in the general simulation turbulence model.

[0135] The specific method of coupling the functional relationship of multiple viscosities using artificial intelligence algorithms is as follows:

[0136] Multivariate coupling analysis systematically analyzes the interaction effects between multiple influencing factors under complex working conditions. First, the interaction terms between variables are constructed, including first-order interactions (product terms between two variables) and higher-order interactions (combined effects of multiple variables). A gradient boosting decision tree model based on LightGBM is used to capture these nonlinear interactions, which can automatically learn complex dependencies between variables. Then, the SHAP (SHapley Additive exPlanations) value analysis method is used to quantify the contribution of each variable and its interaction terms to fluid performance, thereby identifying key influencing factors and important interaction effects.

[0137] Finally, a new function expression for the test fluid, suitable for the complex working conditions in which it is located, and reflecting the coupling effect of all influencing variables is obtained. The comprehensive function of viscosity is imported into the general turbulence model of simulation calculation to replace the existing viscosity term, thereby obtaining a modified turbulence model suitable for the test fluid and its complex working conditions, which is used for subsequent simulation calculations.

[0138] Step S5, based on the existing simulation calculation model, improve the relevant calculation parameter settings, and formally carry out dynamic simulation calculations for this specific non-Newtonian fluid and complex working conditions. Through the above steps, the type of the simulation object, that is, a certain unknown non-Newtonian fluid, and its special rheological model have been determined, and the viscosity comprehensive change function for the complex working conditions of the actual production environment of the fluid has been analyzed, and the viscosity term of the general turbulence model has been specially adaptively corrected. The rheological model of the fluid, the viscosity correction term of the turbulence model, the rheological adjustment parameters, etc. are imported into the simulation module, and other simulation settings such as grids, boundaries, and calculation parameters are improved to start calculating the dynamic characteristics of the test fluid under given working conditions (including solving the flow field, stress field, flow state simulation, etc.).

[0139] Step S6: According to the calculation results of the simulation module, the intelligent controller converts the calculation parameters (such as temperature, pressure, etc.) into control instructions and issues them to the test module. The test module self-adjusts in real time to reach the corresponding parameters and runs stably, and transmits back the measured data for dynamic fitting with the simulation results; the operating parameters of the test module are intelligently adjusted through the residual convergence algorithm until the test value approaches the calculated value and repeats stably, thereby finally verifying the accuracy of the simulation results. The running time of the test module is generally not less than 3 hours.

[0140] Step S7, the intelligent control system designs an adaptive control system to achieve precise control of the non-Newtonian fluid flow process. The controller sets different tolerance requirements for different parameters, such as 2% temperature control accuracy and 3% pressure control accuracy. The controller can dynamically adjust the control gain according to the real-time error size. When the error exceeds the set threshold, the optimal control amount is calculated by an adaptive algorithm. The calculation of the control gain adopts an adaptive method based on error exponential decay, which not only ensures the rapidity of control, but also avoids excessive control oscillations. The system also includes an online parameter identification module that can update the model parameters in real time to adapt to the dynamic changes of fluid characteristics.

[0141] The intelligent controller performs dynamic residual analysis based on the simulation calculation results and test data. If the identification residual is greater than 2%, a machine learning algorithm (such as a decision tree) is used to learn and establish the response relationship between the change in the target variable and the change in the test value. The intelligent controller is trained with historical record samples under similar working conditions in the database. The intelligent controller makes decision adjustments, including adjusting the amplitude and frequency, etc. After adaptively correcting the adjustment parameters, it issues new operating instructions to the test module, including adjusting the flow rate, adjusting the temperature, adjusting the stirring rate, etc. The test value and the calculated value are monitored in real time until the residual is within 2%.

[0142] Verification and Optimization Establish a complete performance evaluation system, and evaluate the model prediction effect and control effect through multiple indicators such as mean square error, mean absolute error, and determination coefficient. Use Bayesian optimization method to globally optimize system parameters, including model hyperparameters, controller parameters, etc. Consider actual engineering constraints during the optimization process, and search for the optimal parameter combination through multiple iterations. At the same time, establish a real-time monitoring mechanism. When the system performance indicators are abnormal, the parameter optimization process is automatically triggered to ensure that the system continues to maintain the best state.

[0143] According to one aspect of the present application, a system can be constructed by the above method, as follows:

[0144] The components include: operation terminal, simulation module, test module, intelligent control module, data-signal transmission line, database, among which:

[0145] The operation terminal is mainly used for simulation setting, test design, process control, data monitoring and system management. It has a visual human-computer interaction interface. By clicking on the interface, simulation calculations and test elements are designed and arranged before simulation. Simulation parameter setting, preprocessing, simulation dynamic management, data monitoring and analysis, and system management can be performed.

[0146] Specifically, before the simulation starts, the simulation elements are arranged and managed, including the setting of specific working conditions, calculation parameters, test variable groups, etc., as well as the packaging of data sets such as experimental related parameters and their target values, and the generation of data package files for archiving. During the simulation process, each module is dynamically controlled, data monitored, and resources allocated. Among them, the data monitoring submodule is used to monitor the real-time calculation results or experiments during the simulation process. After the simulation is completed, the results are quantitatively displayed in various forms.

[0147] The simulation module is used to perform experimental fluid flow simulation calculations and is equipped with a CFD calculation engine and components necessary for engine operation. The simulation calculation engine is responsible for allocating computing resources, managing and executing computing tasks, calling database-related fluid models, and conducting real-time data-signal communication with the intelligent decision-making control module.

[0148] Specifically, the computing engine of the simulation module is equipped with multiple computing frameworks, including: ① non-Newtonian fluid CFD computing framework, ② discrete data high-precision intelligent fitting computing framework, and ③ multi-factor intelligent coupling computing framework. The computing framework is an encapsulation of specific programming computing capabilities built on the simulation computing engine. It is at a higher level of abstraction than the computing engine and is responsible for providing the execution environment and code rules for simulation computing.

[0149] Others include operation support components, computing resource scheduling components, and data-signal interaction components.

[0150] The test module is a closed-loop non-Newtonian fluid pipeline flow test bench and online measurement and control equipment, including a test circulation pipeline, a liquid storage container, a test variable regulator, and an online non-contact measurement device.

[0151] Specifically, the test pipe circulation circuit is used to drive the test fluid to circulate in a given pipe space. Liquid storage container: used to store the test fluid and recycle it after the test. Test variable regulator: used to simulate various complex working conditions required in the test. The adjustable variables include: temperature, pressure, vibration, pipe roughness, pipe diameter, flow rate, etc. Online non-contact measurement device: used for real-time feedback of test data, integrating online rheometers, viscometers, flow meters and other non-contact sensors.

[0152] Among them, the online non-contact measurement device is generally deployed at important locations such as the inlet, characteristic sections along the test pipeline, and the outlet. The probes of the sensor cluster are arranged in a four-way distributed ring on the same measurement section. Four cameras are arranged around the test pipeline to fully cover the flow observation angle. When capturing the flow state, the camera can follow synchronously along the test pipeline and track the trajectory of the tracer particles for analysis.

[0153] Intelligent control module: Starting from the introduction of the non-Newtonian fluid to be simulated, this module drives the entire system to run in an orderly manner according to the preset workflow. It is responsible for receiving the calculation results of the simulation module and the data uploaded by the test module, and issuing control instructions to dynamically adjust the test module. It is the brain and nerve center of the entire simulation system.

[0154] Data-signal transmission line: used for real-time data-signal transmission and interaction between modules.

[0155] Database: Contains various non-Newtonian fluid rheological model data, historical calculation data and experimental data of simulation systems, etc.

[0156] The implementation process of the intelligent simulation method for non-Newtonian fluid pipeline flow includes the following steps:

[0157] 1. Build the test module. First, install the test circulation pipeline, liquid storage container and other components to ensure the effective circulation of non-Newtonian fluid in the pipeline. Among them, the test pipeline can be replaced with different shapes and materials as needed. The test pipeline is mainly used to observe the typical flow and motion parameter measurement of non-Newtonian fluids. Install the test variable regulator, involving temperature regulator, pressure regulator, vibration regulator, etc. When installing the regulator, ensure the tightness of the connection with the pipeline system. Start different regulators in turn and check the load changes of the corresponding variables. After there is no problem with the test module, connect the fluid to be tested.

[0158] 2. Build a simulation computing module, deploy a simulation computing environment, and install various components required for simulation computing, such as simulation running components, simulation interaction components, computing resource scheduling components, etc.

[0159] 3. Use the data-signal transmission line to connect the operation terminal with the corresponding data ports of the intelligent controller, test module, and simulation module, establish a connection with the database, and use test data to verify whether data transmission, data-command conversion, and command issuance are normal. After the line is connected, start the operation terminal and check whether the operation interface is normal.

[0160] 4. Perform simulation settings on the operating terminal, including working condition variables, initial calculation parameters, grid settings, etc. After the parameters are checked, start the formal simulation. The intelligent control module issues test instructions to the test module according to the preset working order of the system, starts the circulation of the fluid in the test pipeline, and measures the strain rate and stress of the fluid in the ideal state in real time through the dynamic rheometer built into the online non-contact measurement device, and transmits it back to the intelligent controller. The intelligent controller uses the built-in intelligent optimization algorithm to perform high-precision fitting to obtain the stress-strain rate curve of the fluid. The controller compares and evaluates the curve against the existing non-Newtonian fluid data in the database to determine the type of non-Newtonian fluid to which the test fluid belongs. The rheological constitutive equation corresponding to the curve is analyzed synchronously, archived to the database, and imported into the simulation module synchronously as a specific rheological model of the fluid used in subsequent calculations.

[0161] 5. The intelligent controller judges and extracts multiple variables existing in the complex working conditions that need to be simulated, such as temperature, pressure, vibration, etc., according to the established working order, and then issues instructions to the test module to conduct single variable tests on each influencing factor. The test module conducts multiple groups of control tests for each variable in turn, classifies and groups the measured test data, and transmits them back to the intelligent controller. The intelligent controller uses a nonlinear algorithm to fit the functional expression of the influence of each variable on the viscosity of the fluid based on the viscosity response data corresponding to each variable received; the built-in artificial intelligence algorithm is used to couple the influence of different variables on the viscosity of the fluid, and the fluid viscosity change function considering the comprehensive effects of multiple influencing variables for the non-Newtonian fluid under complex working conditions is analyzed. This formula is a multivariable function expression. The functional relationship of the viscosity is imported into the simulation module to modify the viscosity term of the general turbulence model, so as to obtain the turbulence model for a specific non-Newtonian fluid under specific complex working conditions.

[0162] 6. For the specific rheological model that has been determined to be suitable for the non-Newtonian fluid and the turbulence model that has been specially modified, the intelligent controller checks that all pre-simulation nodes and parameter settings of the simulation module are ready, starts the simulation calculation, and calculates the dynamic characteristics of the test fluid pipeline flow under given working conditions (including solving the flow field, stress field, flow state simulation, etc.).

[0163] 7. After the simulation module completes the calculation, the intelligent controller converts the calculated parameters (such as temperature, pressure, etc.) into control instructions and sends them to the test module. The test module starts running according to the instructions, and adjusts itself through the test variable regulator until the corresponding parameters are reached and remain stable. The test data is fed back in real time for dynamic fitting with the simulation results. The intelligent controller continuously adjusts the operating parameters of the test module through the residual convergence algorithm until the test value approaches the calculated value and repeats stably to verify the accuracy of the simulation results.

[0164] 8. The simulation controller automatically archives and packages the actual result data, operating parameters, operating records and other process data of the simulation module and test module, and saves them to the database synchronously. So far, a simulation test of a non-Newtonian fluid under a complex working condition has been completed. Later, the simulation calculation results, such as various dynamic indicators of the non-Newtonian fluid, such as flow field, stress field, flow state cloud map, etc., can be further analyzed and processed as needed. Various types of simulation results can also be further viewed in the operation terminal interface to generate a final simulation report.

[0165] The non-Newtonian fluid pipeline flow intelligent simulation system described in the present invention has both a simulation module and a test module. Through an intelligent controller, a non-Newtonian fluid intelligent simulation system with high intelligence, high pertinence and high credibility is constructed according to the built-in workflow.

[0166] In an actual scenario: a factory in an inland arid area intends to use the tail slurry after the production waste liquid is treated as a turbine power generation medium to power the production line. This tail slurry is a pollution-free high-density non-Newtonian fluid with a density higher than water, which means some polymer additives. Its flow properties are not yet known and need to be obtained through experimental simulation. The factory operator intends to test the flow properties of the tail slurry through the intelligent simulation system described in the present invention to demonstrate the feasibility. The ambient temperature of the factory area is higher than the natural temperature, and there are large-scale production equipment and vibrating machinery, which cause continuous disturbance to the ground.

[0167] Step 1: Build a non-Newtonian fluid test module. Use carbon steel of the same material as the pipelines in the plant to make the test pipeline. The pipeline circumference is 10m and the pipeline diameter is 200mm. Use a DN200 electromagnetic flowmeter and a slurry pump to form the circulation system of the operating table. Both ends of the pipeline are connected to the liquid storage tank. The liquid storage tank is 5m high and has a bottom diameter of 2m. Install a variable simulator in the test pipeline and activate the temperature module and vibration module. After the test bench is built, it is tested that the tail slurry can circulate continuously and stably in the pipeline.

[0168] Step 2. Build a non-Newtonian fluid simulation module. The Flink distributed computing engine is selected as the system's stream processing and batch processing computing engine. After processing real-time data streams and computing tasks, the intermediate data is passed to the simulation computing framework. The computing framework is connected to the open source Openfoam computational fluid program, and calls the self-programmed high-precision finite element solution code for calculation; the operation support component selects kafka to process real-time data streams and transmit asynchronous messages, and is responsible for data exchange between various computing frameworks. In terms of supporting hardware, the server architecture is configured with two 96-core high-performance servers as computing nodes, equipped with large-capacity memory and GPU to support parallel computing; Nginx is selected for task distribution to ensure high availability; Prometheus is selected to mount the Grafana monitoring solution to monitor computing resources, task execution and system operation status in real time.

[0169] Step 3: Data collection, transmission port installation, integration and connectivity between the simulation computing platform and the test operation table, including installation of various parameter online sensors, signal converters, hubs, switches, optical fibers, etc., establishment of intercommunication and feedback channels between the simulation computing platform and the test operation table, ensuring real-time data transmission and interaction between the physical layer and the functional layer, and testing whether the real-time upload of test data and the real-time issuance of control instructions are normal; enable the connection line between the computing platform and the model manager and database. Connect the computing platform to the interactive terminal of the application module, start the terminal operation interface, and check whether the application module and the overall system are running well.

[0170] Step 4: Input the test plan, fluid parameters, pipeline parameters, operating parameters, environmental parameters, etc. in the application terminal and submit them to the computing platform for processing. This test mainly tests the effect of different flow rates on the flow performance of the tail slurry. The test variable is the flow rate. The working conditions are divided into five flow rates: 1m / s, 2m / s, 3m / s, 4m / s, and 5m / s. The environmental variables are mainly temperature and vibration intensity. The test temperature is controlled to be 15°C and the vibration intensity is 0.01G. Since many physical properties of the tail slurry are unknown, they are not input. The density is initially input as 1.2×10 3kg / m³. After searching the internal database for comparison and conducting intelligent analysis, the computing platform predicts that the flow performance of the tail slurry is expected to be close to that of high-concentration brine, and gives recommended simulation schemes, calculation parameters, test schemes, fluid parameters, etc. After confirmation, complete simulation information is generated. The computing platform starts fuzzy calculations, and the test data is sent back in real time to the Kafka buffer. The Flink stream processing engine is responsible for real-time distribution of processing tasks and allocation of computing resources, which are given to the Openfoam fluid mechanics calculation framework for solution, and a preliminary rheological initial model of the tail slurry is constructed, with a K value between 400 and 800 Pa·s n The model expression is as follows: σ =K*γn; where K is the flow consistency index and n is the rheological index (n<1 indicates shear thinning, and n>1 indicates shear thickening).

[0171] Step 5: The computing platform packages the determined test plan, operation steps and test parameters into instructions and issues them to the automated test console. The console automatically executes the test instructions issued by the computing platform, regulates the tail slurry to circulate in the test pipeline at different flow rates, and uploads the rheological parameters and fluid dynamics indicators collected in real time to the computing platform.

[0172] Step 6. The Openfoam fluid mechanics calculation framework of the computing platform continuously performs dynamic fitting calculations between numerical solutions and test values, and deduces accurate solutions from fuzzy solutions. During the calculation, the average flow velocity of the measured section along the way, the stress and strain changes of the section, and the degree of shear deformation of the tail slurry along the way are combined with the internal database to fit the fluidity curve corresponding to the test value and the time series change curves of various dynamic indicators such as pressure and flow velocity.

[0173] Step 7. The calculation framework continuously compares the test value fitting curve with the calculated value time series curve in real time, distinguishes the consistency and difference, and analyzes the degree of deviation. At present, there are different fitting results in the low flow rate area below 3m / s and the high flow rate area above 3m / s. The low flow rate area fits better, and the high flow rate area deviates more. Based on the above analysis results, combined with the continuously returned test values ​​and the continuous retrieval of the historical database, the calculation platform further learns and adjusts its own fitting method to differential fitting, using 3m / s as the initial critical point and using the dichotomy approximation; at the same time, it issues control instructions to the test operation table, and executes the same working condition in two steps of 1 times viscosity μ and 2 times viscosity μ. After the calculation platform issues the control instructions, the test operation table adaptively performs the corresponding operations, adds a new test interval after each group of tests, and adjusts the stirring rate to twice the original. Through the communication and feedback between the computing platform and the test platform, as well as learning iteration, the numerical solution and the test value are fitted in real time, and finally the rheological equation of the tail slurry fluid and various model variable parameters with high confidence are analyzed. The rheological equation is in the form of a piecewise function, with a K value of approximately 550. The model expression is:

[0174] When the tail slurry velocity is less than 3.15 m / s, σ = 550*γ 0.95 ;

[0175] When the tail velocity is greater than 3.15 m / s, σ =550*γ 1.05 ;

[0176] Where K is the flow consistency index and n is the rheological index (n<1 indicates shear thinning, n>1 indicates shear thickening).

[0177] Step 8. Decouple the influencing factors of the test conditions according to the analyzed rheological and dynamic characteristics of the test fluid. The computing platform decomposes the variables into flow rate, temperature, and vibration intensity, and controls the test console to dynamically control a single variable to carry out an exhaustive test. The computing platform issues control instructions to carry out adaptive control of the flow rate, temperature, and vibration intensity in turn. The test console automatically keeps other variables unchanged. After the flow stabilizes, only the target variable is adjusted for 5 groups. After reaching a new equilibrium state, the parameter value is recorded, and the change law is analyzed by multiple groups. The computing platform performs trial calculations simultaneously, and dynamically fits and verifies with multiple groups of control data continuously obtained from the test. After a limited number of iterations, the influence of the above three variables is decoupled. Among them, the critical value of the flow rate is 3.15m / s. When the tail slurry flow rate is less than 3.15m / s, the fluid flow performance is enhanced; when the tail slurry flow rate is greater than 3.15m / s, the fluid flow performance is weakened. The temperature critical value is a range between 25°C and 40°C. When the temperature is below 25°C, the fluid fluidity is weakened; when the temperature is between 25°C and 40°C, the fluid fluidity is enhanced; when the temperature is greater than 40°C, the fluid fluidity remains almost unchanged. The vibration intensity has little effect on the flow in the pipeline, and the vibration factor can be ignored.

[0178] Step 9. Based on the obtained general solution of the tail slurry rheological model, the influence of flow rate and temperature on the tail slurry, the computing platform calculates the influence of the combination of flow rate and temperature on the flow performance of the tail slurry within a certain range. After continuous calculation and parameter iteration, the test shows that the optimal working range of the tail slurry pipeline flow is between 3.1m / s ~3.3m / s and the ambient temperature is between 32°C ~35°C. The test operation table verifies that the tail slurry can circulate effectively for a long time in this working range, and the flow state is stable.

[0179] Step 10. After the simulation is completed, the multi-dimensional data such as tail slurry flow indicators, calculation parameters, test operation parameters, etc. after the final verification are archived, and the index code is named and saved according to "tail slurry + flow rate + temperature" for subsequent retrieval. The data analysis module at the application module terminal further generates a complete tail slurry simulation result report. The simulation results show that this type of tail slurry is suitable for circulating flow in pressurized pipelines, and its rheological characteristics and flow properties are stable in carbon steel pipelines. It is a power-law fluid with variable viscosity. The main factors affecting fluidity are flow rate and temperature. The relatively good working range of flow is between 3.1m / s ~3.3m / s and the ambient temperature is between 32°C~35°C. In summary, the tail slurry can be used for hydropower generation in this plant.

[0180] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the specific details in the above embodiments, and no further description is given for some processes that can be implemented by existing methods. Within the technical concept of the present invention, a variety of equivalent transformations can be made to the technical solution of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics, characterized in that: The steps include: Step S1, reading basic data in sequence, and performing abnormality detection and standardization processing to obtain standardized basic data; wherein the basic data includes equipment temperature time series data, pressure time series data, shear rate time series data, stress time series data, strain time series data, vibration frequency data, ambient temperature data and pipeline parameters; Step S2, extracting data sequence features and multi-scale time features from the standardized basic data, using a pre-built correlation data model containing rheological parameters and physical quantities to generate feature correlation data; performing multi-scale time series analysis on the feature correlation data to obtain dynamic feature data and steady-state feature data; based on the dynamic feature data and the steady-state feature data, using a nonlinear mapping method to construct a corresponding relationship between rheological parameters and physical quantities to obtain parameter mapping data; Step S3, performing multivariable coupling analysis based on the parameter mapping data to obtain coupling characteristic data; performing turbulence effect correction through the coupling characteristic data to generate corrected characteristic data; Step S4, performing simulation calculation of the non-Newtonian fluid analysis model based on the corrected characteristic data to obtain simulation data; comparing and analyzing the simulation data with the standardized basic data collected by the experimental device to generate comparative analysis data; establishing an evaluation index system based on the comparative analysis data to obtain model evaluation data.

2. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 1 is characterized in that: The step S1 is specifically as follows: Step S11, collecting basic data, including: reading device temperature time series data from the device temperature sensor, reading pressure time series data from the pressure sensor, reading shear rate time series data from the shear rate sensor, reading stress time series data from the stress sensor, reading strain time series data from the strain sensor; reading vibration frequency data from the vibration sensor, reading ambient temperature data from the ambient temperature sensor; reading pipeline geometry parameters and pipeline length parameters from the preset parameter database, and reading pipeline surface roughness data from the surface roughness measuring instrument; Step S12, obtaining basic data and processing it using a mixed Gaussian probability density model to obtain a probability density distribution function of each parameter; calculating anomaly detection thresholds of each parameter based on the probability density distribution; comparing the basic data with the anomaly detection threshold to generate an abnormal data marker sequence; using a local polynomial regression method with dynamic weights to generate cleaned basic data; Step S13: input the cleaned basic data into an adaptive standardization model, which dynamically adjusts the standardization parameters according to the time-varying characteristics of the data to generate standardized basic data.

3. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 2 is characterized in that: The step S2 is specifically as follows: Step S21, performing wavelet decomposition on the standardized basic data to extract feature sequences at different time scales; fusing the extracted feature sequences with the original standardized basic data, and using an adaptive feature extraction algorithm to obtain sequence feature data; based on the sequence feature data, constructing a temperature-pressure-vibration-shear rate-stress-strain multidimensional correlation model to generate feature correlation data; Step S22, inputting the feature association data into a multi-scale time series analysis model pre-built using a recursive neural network, taking into account short-range and long-range time dependencies, extracting fast-changing features and slow-changing features respectively, and obtaining dynamic feature data characterizing transient responses and steady-state feature data characterizing steady-state behaviors; Step S23, input the dynamic characteristic data and the steady-state characteristic data into a nonlinear mapping model pre-built using a deep neural network, establish a mapping relationship between rheological parameters and temperature, pressure, vibration and shear rate through multi-layer nonlinear transformation, and generate parameter mapping data.

4. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 3 is characterized in that: The step S3 is specifically as follows: Step S31, reading and performing tensor decomposition processing on the parameter mapping data to extract the coupling mode between the variables; constructing a multivariate covariance matrix based on the extracted coupling mode, using the recursive least squares method to estimate the dynamic coupling coefficients between the variables, and generating coupling characteristic data; Step S32, performing turbulence feature extraction processing based on the coupled characteristic data, and separating turbulent pulsation components based on an improved intrinsic orthogonal decomposition method; The Reynolds stress correction method is used to compensate for the shear enhancement effect caused by turbulence and generate corrected characteristic data.

5. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 4 is characterized in that: The step S4 is specifically as follows: Step S41: Based on the corrected characteristic data, a pre-built non-Newtonian fluid analysis model is used to perform non-Newtonian fluid analysis and simulation calculation, and an adaptive numerical solution method is used to generate simulation data including flow field characteristics, stress distribution and temperature field; Step S42: Systematically compare the simulation data with the standardized basic data collected by the experimental device, use a multi-dimensional error analysis method to evaluate the model accuracy, identify key deviation points, and generate comparative analysis data; Based on the comparative analysis data, optimize the model parameters until the model accuracy reaches the threshold; Step S43, construct an evaluation index system including prediction accuracy, calculation efficiency and model stability; perform index evaluation based on comparative analysis data, conduct a comprehensive evaluation of the performance of the non-Newtonian fluid analysis model and the control parameters of the experimental device, and output model evaluation data.

6. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 5 is characterized in that: The step S12 is specifically as follows: Step S121, reading basic data and grouping the data, establishing Gaussian kernel functions for temperature, pressure, vibration, shear rate, stress and strain in sequence, inputting the grouped basic data into a mixed Gaussian probability density model for parameter estimation, optimizing the model parameters by the maximum likelihood method, and obtaining the probability density distribution function of each parameter; Step S122, performing an integral transformation on the probability density distribution function of each parameter to establish a cumulative distribution function; based on the cumulative distribution function and a preset confidence interval, using the quantile method to calculate the anomaly detection threshold of each parameter; Step S123, comparing each parameter value in the basic data with the corresponding anomaly detection threshold; marking the data points that exceed the threshold range, and generating an abnormal data marking sequence containing abnormal data location and type information; Step S124, constructing a data weight matrix based on the abnormal data mark sequence; calculating the time weight coefficient in combination with the time correlation of the data; The basic data, data weight matrix and time weight coefficient are input into the local polynomial regression model, and the abnormal data are corrected by the weighted fitting method to generate the cleaned basic data.

7. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 5 is characterized in that: The step S21 is specifically as follows: Step S211, reading and performing wavelet decomposition on the standardized basic data, performing data decomposition by a multi-scale wavelet transform method to obtain wavelet coefficient data of different scales; performing threshold processing on the wavelet coefficient data, extracting characteristic components at each scale, and generating a multi-scale characteristic sequence; Step S212, extracting the shear rate time series data and stress time series data in the basic data and performing rheological property analysis, calculating the instantaneous shear viscosity distribution, and obtaining shear viscosity data; calculating the viscosity parameters based on the strain time series data and the stretching rate, and generating viscosity data; Combining shear viscosity data and viscosity data, a non-Newtonian index calculation model is constructed to generate rheological index data; Step S213: perform feature fusion on the multi-scale feature sequence and the standardized basic data, extract fusion features using the principal component analysis method, and generate fusion feature data; Perform adaptive feature extraction processing on the fused feature data, extract time series features through recursive neural network, and generate sequence feature data; Step S214, reading the sequence characteristic data, rheological index data and viscosity data and performing tensor decomposition, extracting the coupling mode between the parameters, and generating coupling mode data; A multidimensional correlation network of temperature, pressure, vibration, shear rate, stress, strain and viscosity is constructed based on the coupled mode data. The nonlinear mapping relationship between parameters is learned through deep neural networks to generate feature correlation data.

8. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 5 is characterized in that: The step S31 is specifically as follows: Step S311, reconstruct the parameter mapping data into a three-order tensor structure, which respectively represents the time dimension, the physical quantity dimension and the parameter dimension, and generate tensor structure data; perform Tucker decomposition on the tensor structure data, extract the core tensor and the factor matrix, and obtain the decomposition feature data; Step S312, performing singular value decomposition on the factor matrix in the decomposed characteristic data, extracting the main coupling mode, and generating mode characteristic data; projecting the mode characteristic data into the original space, reconstructing the coupling relationship between the variables, and obtaining the reconstructed characteristic data; Step S313, constructing a multivariate covariance matrix based on the reconstructed feature data, calculating the correlation coefficients between variables, and generating correlation coefficient data; The correlation coefficient data is divided into time windows, a sliding covariance matrix is ​​established, and time-varying covariance data is obtained; Step S314, input the time-varying covariance data into a recursive least squares algorithm, adaptively estimate the coupling coefficient by a forgetting factor method, and obtain coupling coefficient data; Combining the coupling coefficient data and the reconstruction characteristic data, a dynamic coupling model is constructed to generate coupling characteristic data.

9. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 5, characterized in that: The step S41 is specifically as follows: Step S411, input the corrected characteristic data into the grid generator, adaptively divide the calculation domain according to the flow field characteristics, and generate grid data; perform grid quality evaluation on the grid data, improve the grid quality through node smoothing and unit optimization methods, and obtain optimized grid data; Step S412: inputting the optimized grid data and the corrected characteristic data into the control equation discretizer, using the finite volume method to discretize the momentum equation and the energy equation to generate discrete equation data; Construct algebraic equations based on discrete equation data, and obtain system equation data through matrix assembly; Step S413, inputting the system equation data into the multi-grid solver, accelerating convergence through coarse grid correction and fine grid iteration method, and obtaining iterative calculation data; performing convergence judgment on the iterative calculation data, determining the final solution based on residual analysis, and generating numerical solution data; Step S414, input the numerical solution data into the field quantity reconstruction model, calculate the flow field velocity distribution, pressure distribution and temperature distribution, and obtain the field quantity distribution data; post-process the field quantity distribution data, extract the flow field characteristics, stress distribution and temperature field information, and generate simulation data.

10. The method for intelligent simulation and integrated measurement and control of non-Newtonian fluid dynamic characteristics according to claim 5, characterized in that: The step S42 is specifically as follows: Step S421, aligning the simulation data with the standardized basic data collected by the experimental device in the time dimension, eliminating the time deviation through the dynamic time warping algorithm, and obtaining aligned data; Perform feature space mapping on the aligned data, establish the correspondence between the simulation results and the experimental data, and generate mapping data; Step S422: performing error analysis on the mapping data, calculating the root mean square error and relative error of multiple physical quantities, and obtaining error statistical data; performing spatiotemporal distribution analysis on the error statistical data, identifying the main sources and distribution characteristics of the errors, and generating error distribution data; Step S423: locate key deviation points based on the error distribution data, identify high error areas through cluster analysis, and obtain deviation feature data; perform sensitivity analysis on the deviation feature data, identify key parameters that affect model accuracy, and generate sensitivity data; Step S424: input the sensitivity data and deviation characteristic data into a preconfigured parameter optimizer, optimize the model parameters using the gradient descent method, and obtain optimized parameter data; based on the optimized parameter data and error statistical data, generate comparative analysis data including error evaluation results and optimization suggestions.

Images