A method and system for intelligently optimizing energy consumption of powder particle pneumatic conveying

Abstract

This invention relates to the field of pneumatic conveying technology for powder and granular materials, and in particular to an intelligent optimization control method and system for energy consumption in pneumatic conveying of powder and granular materials. The method includes obtaining a standardized time-series dataset; establishing a coupled energy consumption characterization model based on the time-series dataset and the physical properties of the powder and granular materials; extracting multi-scale dynamic energy consumption features from the time-series dataset and identifying key energy consumption influencing factors; constructing a set of energy consumption sensitive parameters; constructing a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries; and, based on the multi-objective energy consumption optimization function, employing a collaborative optimization strategy of an adaptive surrogate model and a multi-objective evolutionary algorithm; interpreting the optimal parameters through fuzzy decision-making and executing adaptive closed-loop optimization control; and outputting optimized operational parameter instructions. This invention effectively solves the technical problems of strong subjectivity and insufficient physical basis in the selection of sensitive parameters in traditional methods.

Description

Technical Field

[0001] This invention relates to the field of pneumatic conveying technology for powder and granular materials, and in particular to an intelligent optimization control method and system for energy consumption in pneumatic conveying of powder and granular materials. Background Technology

[0002] Pneumatic conveying technology, as a core means of long-distance transport of powder and granular materials, is widely used in industrial fields such as chemical, metallurgical, energy, and food processing. With increasingly stringent requirements for energy conservation and emission reduction, reducing the energy consumption of pneumatic conveying systems has become a focus of industry attention. In existing technologies, the operation and control of pneumatic conveying systems mainly rely on operator experience or automated control based on simple feedback regulation. Some studies have attempted to optimize conveying parameters offline by establishing empirical correlations or simplified single-phase flow models. However, these methods fail to fully consider the complex interphase coupling effects and multi-physics energy dissipation mechanisms within the gas-solid two-phase flow, resulting in a lack of sufficient physical basis for energy consumption optimization and making it difficult to achieve substantial improvements in system energy efficiency.

[0003] Specifically, existing energy consumption optimization and control technologies suffer from the following shortcomings: Traditional modeling methods often employ macroscopic empirical formulas or simplified single-phase flow assumptions, making it difficult to accurately characterize the coupled effects of particle group dynamics, interphase momentum exchange, and local turbulent dissipation on system energy consumption in gas-solid two-phase flow, resulting in insufficient accuracy in energy consumption prediction. Furthermore, existing optimization strategies are mostly limited to minimizing a single energy consumption index, neglecting the coordinated needs of multiple objectives such as transport process stability, pipeline wear suppression, and anti-clogging safety, leading to poor feasibility and robustness of optimization results in industrial settings. Finally, optimization methods based on high-fidelity numerical simulations are computationally expensive and lack a collaborative mechanism between adaptive surrogate models and intelligent evolutionary algorithms, making it difficult to meet real-time optimization requirements under complex operating conditions, resulting in a serious disconnect between theoretical optimization and actual operation. At present, there is a need for an intelligent optimization and control method and system for energy consumption in pneumatic conveying of powder and granular materials. Summary of the Invention

[0004] To address the technical problems of high unit conveying energy consumption and poor conveying process stability in existing pneumatic conveying technologies for powders and granules, this invention provides an intelligent optimization control method and system for energy consumption in pneumatic conveying of powders and granules.

[0005] In a first aspect, the present invention provides a method for intelligent optimization and control of energy consumption in pneumatic conveying of powders and granules, which adopts the following technical solution:

[0006] A method for intelligent optimization and control of energy consumption in pneumatic conveying of powders and granules includes:

[0007] A multi-source heterogeneous sensor network was constructed to collect full-condition operating data of the pneumatic conveying system and perform adaptive preprocessing to obtain a standardized time series dataset.

[0008] A coupled energy consumption characterization model is established based on time series datasets and powder and granular material properties. The coupled energy consumption characterization model is used to analyze the physical formation mechanism of transport energy consumption.

[0009] By using a coupled energy consumption characterization model, multi-scale dynamic features of energy consumption are extracted from time series datasets and key energy consumption influencing factors are identified, and a set of energy consumption sensitive parameters is constructed.

[0010] Based on the set of energy-sensitive parameters, and using adjustable operating parameters as decision variables, a multi-objective energy consumption optimization function that integrates physical constraints and operating boundaries is constructed.

[0011] Based on a multi-objective energy consumption optimization function, a collaborative optimization strategy combining an adaptive surrogate model and a multi-objective evolutionary algorithm is adopted to solve for the Pareto optimal solution set.

[0012] By utilizing the Pareto optimal solution set, the optimal parameters are interpreted through fuzzy decision-making and adaptive closed-loop optimization control is executed, outputting optimized operating parameter instructions.

[0013] Furthermore, obtaining the standardized multi-dimensional time series dataset includes:

[0014] A multi-source heterogeneous sensor network is deployed at the feeding end, along the conveying pipeline, and at the separation and collection end of the pneumatic conveying system to acquire full-condition operating data of the pneumatic conveying system.

[0015] The full-condition operation data is synchronously uploaded to the edge computing node. Within the edge computing node, the full-condition operation data is identified based on statistical test criteria, and abnormal sampled values ​​that exceed the normal fluctuation range are removed. The data sequence after removing outliers is smoothed and filtered using a weighted moving average. Standardized time series datasets are output through timestamp alignment and range normalization processing with a unified time base.

[0016] Furthermore, the establishment of a coupled energy consumption characterization model based on the time series dataset and the physical property parameters of powder particles includes:

[0017] Based on the gas-solid two-phase flow theory, a mechanism energy consumption sub-model is constructed, and the total energy consumption of the pneumatic conveying system is analyzed as the superposition of gas phase friction pressure drop energy consumption, particle acceleration energy consumption, suspension energy consumption and local resistance energy consumption.

[0018] A data-driven correction sub-model is constructed based on a standardized time series dataset. The data-driven correction sub-model adopts a gated recurrent unit and a fully connected layer in series structure. The prediction residual of the mechanism energy consumption sub-model is used as a supervision signal for iterative training to capture complex nonlinear coupling effects that the mechanism model fails to explicitly represent.

[0019] The output of the mechanistic energy consumption sub-model is residually fused with the output of the data-driven correction sub-model to obtain the coupled energy consumption characterization model.

[0020] Furthermore, the extraction of multi-scale energy consumption dynamic features from the time series dataset includes using a coupled energy consumption representation model to predict energy consumption for each operating condition in the standardized multi-dimensional time series dataset, thereby obtaining an energy consumption prediction time series.

[0021] A multi-scale decomposition is performed on the energy consumption prediction time series, and the multi-scale decomposition adopts a cascaded strategy of variational mode decomposition and wavelet packet decomposition.

[0022] The energy consumption prediction time series is adaptively decomposed into K intrinsic mode functions using variational mode decomposition. Three-level wavelet packet decomposition is performed on each intrinsic mode function component to obtain the wavelet packet coefficients corresponding to each decomposition node.

[0023] The wavelet packet energy entropy of each node is calculated based on the wavelet packet coefficients. The wavelet packet energy entropy of each node is used as a multi-scale dynamic feature to construct a multi-scale energy consumption feature matrix.

[0024] Furthermore, the identification of key energy consumption influencing factors includes calculating the nonlinear correlation strength between each operating parameter and the total energy consumption based on a multi-scale energy consumption feature matrix and using the maximum information coefficient.

[0025] The contribution of each operating parameter to energy consumption prediction is calculated, and the maximum information coefficient value is weighted and fused with the contribution of the gradient boosting decision tree to obtain the comprehensive correlation score of each operating parameter.

[0026] The operating parameters are sorted in descending order according to their comprehensive correlation scores. The operating parameters whose comprehensive correlation scores rank higher than the set threshold and meet the adjustability constraints are selected to form the energy consumption sensitive parameter set.

[0027] The formula for calculating the maximum information coefficient is as follows:

[0028] ,

[0029] in, The maximum information coefficient between the operating parameter time series X and the total energy consumption time series Y. This represents the number of mesh elements along the X-axis in a two-dimensional scattered mesh generation. This represents the number of mesh divisions along the Y-axis in a two-dimensional scattered mesh generation. Given an x×y grid, the maximum mutual information value between the operating parameter time series X and the total energy consumption time series Y is given. The upper bound of the search is dependent on the sample size. It is the smaller value between x and y.

[0030] Furthermore, the construction of a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries includes selecting adjustable operational parameters with online adjustment capabilities from a set of energy consumption-sensitive parameters and constructing a decision variable vector. ,in, For the frequency of the fan, For the frequency of the feed rotary valve, The air supply valve opening is denoted by n, and the decision variable dimension is denoted by n.

[0031] A multi-objective optimization function vector is constructed using energy consumption per unit mass of transport, root mean square of system pressure drop fluctuation, and transported mass flow rate as optimization objectives.

[0032] Construct fusion constraints, which include anti-clogging physical constraints based on minimum delivery wind speed theory, safety constraints based on pipeline pressure limit, equipment boundary constraints based on fan rated speed, and dust concentration constraints based on environmental emission requirements.

[0033] The penalty function method is used to embed the fusion constraints into the multi-objective optimization function vector to construct a standardized objective function suitable for optimization.

[0034] Furthermore, the collaborative optimization strategy of adopting an adaptive surrogate model and a multi-objective evolutionary algorithm includes generating an initial sample set in the feasible region defined by the boundaries of each component of the decision variable vector using Latin hypercube sampling, constructing an initial Kriging surrogate model based on the initial sample set, performing population evolution optimization on the Kriging surrogate model using the NSGA-II algorithm, iteratively updating the Kriging surrogate model by constructing an expected improvement point addition criterion, and outputting the final Pareto optimal solution set.

[0035] Furthermore, the construction of the expected improvement criteria includes generating a set of candidate sample points within the feasible domain of the decision variables, and calculating the expected improvement value of each candidate sample point based on the predicted mean and predicted variance of the Kriging surrogate model.

[0036] Based on the expected improvement value, each candidate sample point is sorted in descending order, and the candidate sample points that are ranked by expected improvement value and are located in the sparse region of the Pareto front or the region with high prediction uncertainty are selected as supplementary sample points.

[0037] The formula for calculating the expected improvement value is as follows:

[0038] ,

[0039] in, Let x be the expected improvement value of the candidate sample point. This represents the optimal value of the objective function corresponding to the current Pareto front. Let x be the mean of the predictions made by the Kriging surrogate model at the candidate sample point x. Let x be the standard deviation of the Kriging surrogate model's predictions at the candidate sample point x. The cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.

[0040] Furthermore, the step of interpreting the optimal parameters through fuzzy decision and executing adaptive closed-loop optimization control includes defining trapezoidal membership functions for three optimization objectives—unit energy consumption, pressure drop fluctuation, and transport flow rate—for each Pareto solution in the Pareto optimal solution set.

[0041] Based on the current operating conditions, a target weight vector is set, the weighted fuzzy comprehensive satisfaction of each Pareto solution is calculated, and the Pareto solution corresponding to the maximum value is selected as the final solution.

[0042] The final solution is interpreted into executable operating parameter instructions, which are then sent to the controller of the pneumatic conveying system to perform frequency conversion adjustment of the blower, adjustment of the opening of the feeding valve, and linkage control of the air replenishment valve.

[0043] Secondly, a smart energy consumption optimization control system for pneumatic conveying of powders and granules includes:

[0044] Multi-source data acquisition is used to collect multi-source heterogeneous operating data and powder and granular material properties of the pneumatic conveying system under all working conditions, and to perform adaptive preprocessing and standardization on the operating data to obtain a standardized multi-dimensional time series dataset.

[0045] The coupled energy consumption characterization module is used to establish a coupled energy consumption characterization model based on a standardized multi-dimensional time series dataset and powder and granular material properties, so as to analyze the physical formation mechanism of transport energy consumption.

[0046] The sensitive parameter identification module is used to extract multi-scale dynamic energy consumption features from a standardized multi-dimensional time series dataset and identify key energy consumption influencing factors by utilizing a coupled energy consumption characterization model, and to construct a set of energy consumption sensitive parameters.

[0047] The multi-objective optimization building module is used to construct a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries based on a set of energy-sensitive parameters and adjustable operating parameters as decision variables.

[0048] The collaborative optimization module is used to solve the Pareto optimal solution set based on a multi-objective energy consumption optimization function, using a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm;

[0049] The closed-loop control output module is used to utilize the Pareto optimal solution set, interpret the optimal parameters through fuzzy decision-making, and execute adaptive closed-loop optimization control, outputting the optimized operating parameter instructions.

[0050] In summary, the present invention has the following beneficial technical effects:

[0051] 1. This invention constructs a coupled energy consumption characterization model driven by mechanism and data fusion, and performs residual fusion of the gas-solid two-phase flow theoretical mechanism sub-model and the gated cyclic unit data-driven correction sub-model, thereby achieving accurate analysis of the physical formation mechanism of transmission energy consumption. This effectively overcomes the shortcomings of single mechanism models in explicitly characterizing complex nonlinear coupling effects, and significantly improves the accuracy and interpretability of energy consumption prediction.

[0052] 2. This invention extracts multi-scale dynamic energy consumption features by utilizing a coupled energy consumption characterization model and identifies key energy consumption influencing factors by employing a weighted fusion strategy of maximum information coefficient and gradient boosting decision tree. This achieves the accurate construction of a set of energy consumption sensitive parameters, providing a set of decision variables with high confidence for multi-objective optimization. It effectively solves the technical problems of strong subjectivity and insufficient physical basis in the selection of sensitive parameters in traditional methods.

[0053] 3. This invention constructs a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries, and uses a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm to solve the Pareto optimal solution set. Combined with fuzzy decision interpretation and adaptive closed-loop control mechanism, it realizes near-optimal energy consumption operation of pneumatic conveying system under all working conditions. Under the premise of ensuring conveying stability and meeting environmental emission standards, it significantly reduces the system's operating energy consumption, and has the ability to adaptively re-optimize for material property fluctuations and operating condition drift. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall process of an intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to an embodiment of the present invention.

[0055] Figure 2 This is an overall architecture diagram of an intelligent optimization control method for energy consumption in pneumatic conveying of powder and granules according to an embodiment of the present invention. Detailed Implementation

[0056] The present invention will be further described in detail below with reference to the accompanying drawings.

[0057] Example 1

[0058] Reference Figure 1 This embodiment of a method for intelligent optimization and control of energy consumption in pneumatic conveying of powder and granules includes:

[0059] S1. Construct a multi-source heterogeneous sensor network, collect full-condition operation data of the pneumatic conveying system and perform adaptive preprocessing to obtain a standardized time series dataset.

[0060] S2. A coupled energy consumption characterization model is established based on the time series dataset and the physical property parameters of powder particles. The coupled energy consumption characterization model is used to analyze the physical formation mechanism of transport energy consumption.

[0061] S3. Using a coupled energy consumption characterization model, extract multi-scale dynamic features of energy consumption from time series datasets and identify key energy consumption influencing factors to construct a set of energy consumption sensitive parameters.

[0062] S4. Based on the set of energy-sensitive parameters and using adjustable operating parameters as decision variables, construct a multi-objective energy consumption optimization function that integrates physical constraints and operating boundaries;

[0063] S5. Based on the multi-objective energy consumption optimization function, a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm is adopted to solve the Pareto optimal solution set;

[0064] S6. Using the Pareto optimal solution set, the optimal parameters are interpreted through fuzzy decision-making and adaptive closed-loop optimization control is executed, outputting the optimized operating parameter instructions.

[0065] Specifically, a method for intelligent optimization and control of energy consumption in pneumatic conveying of powders and granules includes the following steps:

[0066] like Figure 1 As shown in Figure S1, a multi-source heterogeneous sensor network is deployed at the feeding end, along the conveying pipeline, and at the separation and collection end of the pneumatic conveying system. The sensor network includes at least a pressure sensor for collecting the static and dynamic pressure of the pipeline, a temperature sensor for collecting the temperature of the gas-solid mixture, a thermal mass flow meter for collecting the mass flow rate of the gas, a power transmitter for collecting the power of the fan shaft, a capacitive tomography sensor for collecting the solid phase concentration of the pipeline, and an encoder for collecting the rotational speed of the feeding rotary valve. Each sensor collects data according to a preset sampling period. The pressure, temperature, and flow sensors collect data at a first sampling frequency, and the power and concentration sensors collect data at a second sampling frequency. The first sampling frequency is higher than the second sampling frequency. The full-condition operation data is synchronously uploaded to the edge computing node via an industrial Ethernet.

[0067] Within the edge computing node, outlier identification and removal based on statistical test criteria are first performed on the full-condition operating data. For each sensor data sequence, the Grubbs test criterion is used to calculate the statistical measure. ,in, For the data points to be tested, Let G be the sample mean of the current data sequence, and s be the sample standard deviation. The statistic G is compared with the critical value at a preset significance level. In comparison, if If the data point is found to be an abnormal sample value that exceeds the normal fluctuation range, it will be removed. For the data gaps formed after removing the outliers, a mutual information interpolation strategy based on phase space reconstruction is used to fill the gaps dynamically. That is, the phase space is reconstructed for the sequence with missing data, the mutual information value between each known point in the neighborhood of the missing point and the missing point is calculated, and the k nearest neighbor points with the largest mutual information are selected for weighted interpolation. The filling weight is determined based on the normalization of the mutual information value.

[0068] After outlier removal and missing value imputation, a weighted moving average is used to smooth the data sequence. The weighted moving average satisfies the following formula:

[0069] ,

[0070] in, Let be the filtered data at time t. The original data at time t+i Let be the filter weight for the i-th sampling point. The sliding window half-width, the filter weights Gaussian weighting function It is confirmed that, among them, This is the weight attenuation coefficient, set according to the signal noise level.

[0071] Subsequently, timestamp alignment with a unified time base is performed: linear interpolation resampling is performed on the data at the second sampling frequency using the first sampling frequency as the reference time axis, assuming that the data at the second sampling frequency are adjacent time points. and The sampled values ​​are respectively and Then at the reference time ( interpolated data satisfy:

[0072] ,

[0073] After time alignment is completed, the operating data of each channel is subjected to range normalization, which satisfies the following formula: ;

[0074] in, This is the original running data. and These represent the minimum and maximum values ​​of the data for that channel within the sliding time window, respectively. The normalized time series data has values ​​mapped to the interval [−1, 1].

[0075] Simultaneously, the physical properties of powder particles are collected through offline testing or online detection methods. These physical properties include at least the true density of the particles, median particle size, moisture content, and angle of repose. The physical properties are then used as static feature vectors and concatenated with the standardized time series dataset to obtain a standardized multi-dimensional time series dataset.

[0076] like Figure 2 As shown in Figure S2, a mechanism energy consumption sub-model is constructed based on the gas-solid two-phase flow theory to calculate the total energy consumption of the pneumatic conveying system. Analysis as voltage drop energy consumption Particle acceleration energy consumption Suspension energy consumption and local resistance energy consumption The superposition satisfies Among them, the energy consumption of pressure drop Based on the improved Barth pressure drop model, the total pressure drop in the pipeline is... The calculated total pressure drop of the pipeline satisfies the following formula:

[0077] ,

[0078] in, D is the pipe length, and D is the pipe diameter. For gas density, For the apparent velocity of the gas, For particle velocity, Particle density, The solid-gas mass ratio is given by , and m is the velocity slip correction factor. and These are the gas phase and solid phase drag coefficients, respectively; the particle acceleration energy consumption... The suspension energy consumption is determined based on the change in kinetic energy required for the particles to accelerate from the feeding end to a stable conveying speed. The local resistance energy consumption is determined based on the airflow energy required to maintain the suspension of particles in horizontal and vertical pipe sections. The determination is based on the local resistance coefficient and corresponding local velocity of the elbow, valve and separator components.

[0079] Secondly, a data-driven correction sub-model is constructed based on the standardized multi-dimensional time series dataset. The data-driven correction sub-model adopts a gated recurrent unit and a fully connected layer in series structure. The feature vector obtained by concatenating the running parameter vector and the powder and granular material property parameter vector in the standardized time series dataset is used as input. The prediction residual of the mechanism energy consumption sub-model is used as the supervision signal for iterative training. The prediction residual is the difference between the output value of the mechanism energy consumption sub-model and the measured total energy consumption of the system. The Adam optimizer is used during training. The initial learning rate is set to 0.001 and the batch size is set to 64. Iterative training is carried out until the root mean square error of the validation set converges, which is used to capture the complex nonlinear coupling effects that the mechanism model fails to explicitly represent.

[0080] Finally, the output of the mechanistic energy consumption sub-model and the output of the data-driven correction sub-model are residually fused to obtain the coupled energy consumption characterization model, whose output satisfies the following equation:

[0081] ,

[0082] in, For the mapping function of the mechanism energy consumption sub-model, The mapping function for data-driven correction of the sub-model. For the runtime parameter vector, This is a vector of powder and granular physical property parameters. This represents the energy consumption prediction value output by the coupled energy consumption characterization model.

[0083] S3. Based on the coupled energy consumption characterization model, energy consumption is predicted for each operating condition in the standardized multi-dimensional time series dataset, resulting in an energy consumption prediction time series aligned with the time axis of the standardized time series dataset. Multi-scale decomposition is then performed on the energy consumption prediction time series, employing a cascaded strategy of variational mode decomposition and wavelet packet decomposition: First, variational mode decomposition adaptively decomposes the energy consumption prediction time series into K intrinsic mode functions. This variational mode decomposition is achieved by constructing and solving a constrained variational problem, with the goal of finding the K intrinsic mode functions. and its corresponding center frequency The constraint variational problem satisfies the following conditions: The sum of the estimated bandwidths of all eigenmode functions is minimized, and the sum of all eigenmode functions equals the input signal.

[0084] ,

[0085] in, Let be the center frequency of the k-th eigenmode function. Let k be the intrinsic mode function. The input is the energy consumption prediction time series. For the Dirac function, For convolution operation, j is the imaginary unit. To obtain the partial derivative with respect to time, the above variational problem is solved iteratively using the alternating direction multiplier method, where the quadratic penalty parameter is set to 2000, the Lagrange multiplier update step size is set to 0.01, and the convergence tolerance is set to... The number of modes K is adaptively determined according to the energy conservation criterion. That is, K starts from 2 and increments. Variational mode decomposition is performed sequentially and the energy ratio of each mode component is calculated. When the energy of the Kth mode component accounts for less than 5% of the total energy of the energy consumption prediction time series, the increment stops. The current value of K is determined as the final number of decomposed modes, and K eigenmode function components distributed from high frequency to low frequency are obtained.

[0086] Subsequently, a three-level wavelet packet decomposition was performed on each intrinsic mode function component: the Daubechies fourth-order wavelet was selected as the mother wavelet, and a three-level binary tree decomposition was performed on each intrinsic mode function component. The first level decomposed the signal into low-frequency approximation coefficient nodes [2,0] and high-frequency detail coefficient nodes [2,1]. The second level further decomposed each node in the first level into nodes [3,0], [3,1], [3,2], and [3,3]. The third level further decomposed each node in the second level into nodes [4,0] to [4,7], resulting in a total of 8 terminal decomposition nodes in the third level. The wavelet packet coefficient energy of each terminal node was then calculated. ,in The coefficients of the nth wavelet packet of the i-th terminal node are used to calculate the energy proportion of each node. Where N=8 is the total number of terminal nodes; the wavelet packet energy entropy of each node is calculated based on the energy ratio, and the wavelet packet energy entropy of each node is used as a multi-scale dynamic feature to construct a multi-scale energy consumption feature matrix; the wavelet packet energy entropy satisfies the following formula:

[0087] ,

[0088] in, The wavelet packet energy entropy, Let N be the ratio of the energy of the i-th terminal node to the total energy, and N be the total number of terminal nodes.

[0089] Key energy consumption influencing factors are identified based on the multi-scale energy consumption feature matrix: the nonlinear correlation strength between each operating parameter and the total energy consumption is calculated using the maximum information coefficient, and the calculation of the maximum information coefficient satisfies the following formula:

[0090] ,

[0091] in, The maximum information coefficient between the operating parameter time series X and the total energy consumption time series Y. This represents the number of mesh elements along the X-axis in a two-dimensional scattered mesh generation. This represents the number of mesh divisions along the Y-axis in a two-dimensional scattered mesh generation. Given an x×y grid, the maximum mutual information value between the operating parameter time series X and the total energy consumption time series Y is given. The upper bound of the search is dependent on the sample size and , It is the smaller value between x and y.

[0092] Simultaneously, the contribution of each operating parameter to energy consumption prediction is calculated using a gradient boosting decision tree. The contribution is the proportion of the cumulative split gain of each operating parameter when used as a splitting feature in the gradient boosting decision tree model to the total sum of all split gains. The base learner of the gradient boosting decision tree is set to a regression tree, with a maximum tree depth of 5, a learning rate of 0.1, and 200 iterations. The maximum information coefficient value and the contribution of the gradient boosting decision tree are weighted and fused to obtain a comprehensive correlation score for each operating parameter. The weight of the maximum information coefficient value is set to 0.6, and the weight of the contribution of the gradient boosting decision tree is set to 0.4, satisfying the following formula:

[0093] ,

[0094] in, The comprehensive correlation score for the j-th running parameter. The maximum information coefficient value of the j-th running parameter. The gradient boosting decision tree contribution of the j-th operating parameter is calculated; the operating parameters are sorted in descending order according to the comprehensive correlation score, and the comprehensive correlation score screening threshold is set to 0.5. The operating parameters with comprehensive correlation scores higher than the screening threshold and satisfying the adjustability constraint are selected to form the energy consumption sensitive parameter set. The adjustability constraint means that the operating parameter has the ability to be adjusted online and in real time by the control system. The energy consumption sensitive parameter set includes at least the conveying fan frequency, solid-gas mass ratio, feeding rotary valve frequency, and air replenishment valve opening.

[0095] S4. Based on the set of energy-sensitive parameters, select adjustable operating parameters with online adjustment capabilities and construct a decision variable vector. ,in, For the frequency of the fan, For the frequency of the feed rotary valve, The air supply valve opening is denoted by n, which represents the dimension of the decision variables. The feasible region of each decision variable is determined based on the equipment's rated parameters and process safety boundaries, including the fan frequency. The adjustment range is [30, 50] Hz, and the frequency of the feed rotary valve is... The adjustment range is [10, 35] Hz, and the air supply valve opening is... The adjustment range is [20, 100]%.

[0096] Energy consumption per unit mass of transport Root mean square of system voltage drop fluctuation and conveying mass flow rate As the optimization objective, a multi-objective optimization function vector J(d) is constructed; where the energy consumption per unit mass transported is... Energy consumption prediction based on the output of the coupled energy consumption characterization model With conveying mass flow rate The ratio calculation satisfies ; Root mean square of system voltage drop fluctuation The standard deviation of the pipeline static pressure time series collected in step S1 is calculated to satisfy... ,in Let be the sampled value of the pipeline static pressure at time t. The average static pressure within the sampling period T; the mass flow rate of the conveyed product. Frequency of the feed rotary valve ratio of solid to gas mass Coupled estimation, satisfying ,in This is the volumetric efficiency coefficient of the rotary valve. Let the particle density be denoted by ; construct a multi-objective optimization function vector that satisfies:

[0097] ,

[0098] In this context, the superscript T indicates vector transpose, and the negative sign indicates that the objective of maximizing the mass flow rate is transformed into the objective of minimizing the mass flow rate.

[0099] Construct fusion constraints, which include: anti-clogging physical constraints based on the minimum delivery wind speed theory. ,in The apparent velocity of the gas is determined by the fan frequency. Calculated by converting with pipeline characteristics, The minimum conveying air velocity is determined based on the particle size and density of the powder; safety constraints are based on the pipeline's pressure bearing limit. ,in Based on the Barth pressure drop model calculation in step S2, Design pressure limit for pipelines; equipment boundary constraints based on fan rated speed. ; and dust concentration constraints based on environmental emission requirements. ,in This is the measured value of dust concentration at the separator outlet. These are the limits set by environmental emission standards.

[0100] The penalty function method is used to embed the above-mentioned fusion constraints into the multi-objective optimization function vector to construct a standardized objective function suitable for optimization; for inequality constraints Constructing penalties For equality constraints Constructing penalties The standardized objective function satisfies:

[0101] ,

[0102] in, The standardized objective function vector after embedding the penalty function. Let the weight coefficients of the penalty function be set as follows: m is the number of inequality constraints. The number of equality constraints. This is the penalty term corresponding to the i-th inequality constraint. The penalty term corresponds to the j-th equality constraint. When the decision variable d satisfies all constraints, the penalty term is zero, and the standardized objective function degenerates into the original multi-objective optimization function vector. When the decision variable violates the constraints, the penalty term generates a significant penalty value, guiding the optimization process to converge to the feasible region.

[0103] S5. Based on the multi-objective energy consumption optimization function, a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm is adopted to solve the Pareto optimal solution set. First, an initial sample set is generated within the feasible region defined by the boundaries of each component of the decision variable vector according to Latin hypercube sampling: the feasible region of each decision variable is divided into Ns layers, and a sample value is randomly selected in each layer to form Ns initial sample points. In this embodiment, Ns is set to 10(n+1), where n is the dimension of the decision variable; the initial sample set is realistically evaluated using the coupled energy consumption characterization model described in Embodiment 2 to obtain the real objective function value corresponding to each sample point, and an initial Kriging surrogate model is established, whose kernel function satisfies:

[0104] ,

[0105] in, For sample points and The kernel function values ​​between the sample points, where r is the Euclidean distance between the sample points. For length scale hyperparameters, The hyperparameter is the signal variance, which is estimated by maximizing the log-likelihood function.

[0106] Secondly, the NSGA-II algorithm with adaptive crossover and mutation operators is used to perform population evolution optimization on the Kriging surrogate model: the initial population size is... In this embodiment Set the population size to 200, randomly generate an initial population and calculate the predicted objective function value for each individual on the Kriging surrogate model; perform non-dominated sorting and crowding calculation to divide the population into different non-dominated levels; use a tournament selection strategy to select parent individuals, with crossover probabilities... With the probability of mutation Based on population diversity indicators, the system adaptively adjusts the probability of mutation when the population crowding level falls below a preset crowding threshold of 0.1. To enhance global exploration capabilities, when the population convergence exceeds a preset convergence threshold of 0.9, the crossover probability is increased to [a higher value]. To accelerate local mining, otherwise maintain the default crossover probability. With the probability of mutation The offspring population is generated by simulating binary crossover and polynomial mutation. After merging the parent and offspring populations, a non-dominated sorting is performed again, and the first generation is selected. Individuals constitute a new generation of population.

[0107] Then, based on the predicted mean of the Kriging proxy model. With the predicted standard deviation To construct the expected improvement criteria: within the feasible region of the decision variable, a set of candidate sample points is generated based on Latin hypercube sampling, with the sample size set as... Calculate the expected improvement value for each candidate sample point, wherein the expected improvement value satisfies the following formula:

[0108] ,

[0109] in, Let x be the expected improvement value of the candidate sample point. This represents the optimal value of the objective function corresponding to the current Pareto front. Let x be the mean of the predictions made by the Kriging surrogate model at the candidate sample point x. Let x be the standard deviation of the Kriging surrogate model's predictions at the candidate sample point x. The cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.

[0110] Based on the expected improvement value, the candidate sample points are sorted in descending order. Simultaneously, the crowding distance of each candidate sample point on the Pareto front is calculated. Points with the highest expected improvement values ​​and located in sparse regions or regions with high prediction uncertainty on the Pareto front are selected. The candidate sample points are used as supplementary sample points, and the number of supplementary sample points is set to n+2. The supplementary sample points are evaluated using the coupled energy consumption characterization model. The evaluation results are added to the training sample set and the Kriging proxy model is updated.

[0111] Repeat the above population evolution optimization and adaptive point addition process until the convergence criterion is met. The convergence criterion is that the rate of change of the hypervolume index of the Pareto front between two adjacent generations is less than a preset threshold. Alternatively, it can reach the maximum number of iterations, 100 generations, and output the final Pareto optimal solution set.

[0112] S6. Define the trapezoidal membership functions for the three optimization objectives: unit energy consumption, pressure drop fluctuation, and transport flow rate. For cost-related indicators such as unit mass transport energy consumption and root mean square pressure drop fluctuation (the smaller the better), their trapezoidal membership functions satisfy:

[0113] ,

[0114] For efficiency-related indicators such as mass flow rate (the larger the better), the trapezoidal membership function satisfies:

[0115] ,

[0116] in, To find the Pareto solution's objective function value for the j-th optimization objective, To optimize the historical worst value or unacceptable engineering boundary of the target, To optimize the historical best value or engineering ideal boundary of the objective, the objective function value of each Pareto solution is substituted into the corresponding trapezoidal membership function and mapped to the satisfaction interval [0,1].

[0117] Set the target weight vector based on the current operating conditions. The weight of energy consumption per unit mass transported under normal operating conditions. The weight of the root mean square of the system voltage drop fluctuation is initially set to 0.5. The weight of the mass flow rate is set to 0.3. Setting it to 0.2 will be used in special working conditions where high conveying stability is required. Increase the weights to 0.5 and adjust the remaining weights accordingly; calculate the weighted fuzzy comprehensive satisfaction of each Pareto solution, wherein the weighted fuzzy comprehensive satisfaction satisfies the following formula:

[0118] ,

[0119] in, Let be the weighted fuzzy comprehensive satisfaction of the i-th Pareto solution. The weight coefficients for the j-th optimization objective are... , Let be the trapezoidal membership function corresponding to the j-th optimization objective. Let the objective function value of the i-th Pareto solution be the value of the j-th optimization objective. Let be the decision variable vector corresponding to the i-th Pareto solution; select the Pareto solution corresponding to the maximum weighted fuzzy comprehensive satisfaction as the final solution. .

[0120] The final solution The commands are interpreted as executable operating parameter instructions and sent to the PLC controller of the pneumatic conveying system via industrial Ethernet using the Modbus TCP protocol. These instructions execute the frequency conversion adjustment of the blower, the opening adjustment of the feeding rotary valve, and the linkage control of the air replenishment valve. The blower frequency adjustment command is generated by… The D / A converter outputs a 4-20mA analog signal to the frequency converter, and the frequency command for the feeding rotary valve is... The air supply valve opening command is executed by the servo driver. Positioned via electric actuator.

[0121] An adaptive closed-loop optimization control mechanism is established: After the optimization command is executed, actual system operation feedback data is collected at fixed intervals (5 minutes in this embodiment). The feedback data includes fan shaft power, pipeline static pressure, conveyed mass flow rate, and separator outlet dust concentration. The deviation rate between the actual energy consumption and the energy consumption predicted by the coupled energy consumption characterization model is calculated, and the deviation rate satisfies:

[0122] ,

[0123] in, The actual total energy consumption calculated from the measured data will be automatically triggered when the deviation rate exceeds the preset adaptive threshold of 15%, or when a significant drift in the physical properties of the powder particles is detected (the change in particle moisture content exceeds 2% or the change in median particle size exceeds 5%), thus completing the output of the final result.

[0124] Example 2

[0125] The difference between this embodiment and Embodiment 1 is that this embodiment provides an intelligent optimization control system for energy consumption of pneumatic conveying of powder and granules, including:

[0126] Multi-source data acquisition is used to collect multi-source heterogeneous operating data and powder and granular material properties of the pneumatic conveying system under all working conditions, and to perform adaptive preprocessing and standardization on the operating data to obtain a standardized multi-dimensional time series dataset.

[0127] The coupled energy consumption characterization module is used to establish a coupled energy consumption characterization model based on a standardized multi-dimensional time series dataset and powder and granular material properties, so as to analyze the physical formation mechanism of transport energy consumption.

[0128] The sensitive parameter identification module is used to extract multi-scale dynamic energy consumption features from a standardized multi-dimensional time series dataset and identify key energy consumption influencing factors by utilizing a coupled energy consumption characterization model, and to construct a set of energy consumption sensitive parameters.

[0129] The multi-objective optimization building module is used to construct a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries based on a set of energy-sensitive parameters and adjustable operating parameters as decision variables.

[0130] The collaborative optimization module is used to solve the Pareto optimal solution set based on a multi-objective energy consumption optimization function, using a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm;

[0131] The closed-loop control output module is used to utilize the Pareto optimal solution set, interpret the optimal parameters through fuzzy decision-making, and execute adaptive closed-loop optimization control, outputting the optimized operating parameter instructions.

[0132] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent optimization and control of energy consumption in pneumatic conveying of powders and granules, characterized in that, include: A multi-source heterogeneous sensor network was constructed to collect full-condition operating data of the pneumatic conveying system and perform adaptive preprocessing to obtain a standardized time series dataset. A coupled energy consumption characterization model is established based on time series datasets and powder and granular material properties. The coupled energy consumption characterization model is used to analyze the physical formation mechanism of transport energy consumption. By using a coupled energy consumption characterization model, multi-scale dynamic features of energy consumption are extracted from time series datasets and key energy consumption influencing factors are identified, and a set of energy consumption sensitive parameters is constructed. Based on the set of energy-sensitive parameters, and using adjustable operating parameters as decision variables, a multi-objective energy consumption optimization function that integrates physical constraints and operating boundaries is constructed. Based on a multi-objective energy consumption optimization function, a collaborative optimization strategy combining an adaptive surrogate model and a multi-objective evolutionary algorithm is adopted to solve for the Pareto optimal solution set. By utilizing the Pareto optimal solution set, the optimal parameters are interpreted through fuzzy decision-making and adaptive closed-loop optimization control is executed, outputting optimized operating parameter instructions.

2. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The process of obtaining a standardized multi-dimensional time series dataset includes: A multi-source heterogeneous sensor network is deployed at the feeding end, along the conveying pipeline, and at the separation and collection end of the pneumatic conveying system to acquire full-condition operating data of the pneumatic conveying system. The full-condition operation data is synchronously uploaded to the edge computing node. Within the edge computing node, the full-condition operation data is identified based on statistical test criteria, and abnormal sampled values ​​that exceed the normal fluctuation range are removed. The data sequence after removing outliers is smoothed and filtered using a weighted moving average. Standardized time series datasets are output through timestamp alignment and range normalization processing with a unified time base.

3. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The coupled energy consumption characterization model established based on the time series dataset and the physical properties of powder particles includes: Based on the gas-solid two-phase flow theory, a mechanism energy consumption sub-model is constructed, and the total energy consumption of the pneumatic conveying system is analyzed as the superposition of gas phase friction pressure drop energy consumption, particle acceleration energy consumption, suspension energy consumption and local resistance energy consumption. A data-driven correction sub-model is constructed based on a standardized time series dataset. The data-driven correction sub-model adopts a gated recurrent unit and a fully connected layer in series structure. The prediction residual of the mechanism energy consumption sub-model is used as a supervision signal for iterative training to capture complex nonlinear coupling effects that the mechanism model fails to explicitly represent. The output of the mechanistic energy consumption sub-model is residually fused with the output of the data-driven correction sub-model to obtain the coupled energy consumption characterization model.

4. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The process of extracting multi-scale energy consumption dynamic features from time series datasets includes using a coupled energy consumption characterization model to predict energy consumption for each operating condition in a standardized multi-dimensional time series dataset, thereby obtaining an energy consumption prediction time series. A multi-scale decomposition is performed on the energy consumption prediction time series, and the multi-scale decomposition adopts a cascaded strategy of variational mode decomposition and wavelet packet decomposition. The energy consumption prediction time series is adaptively decomposed into K intrinsic mode functions using variational mode decomposition. Three-level wavelet packet decomposition is performed on each intrinsic mode function component to obtain the wavelet packet coefficients corresponding to each decomposition node. The wavelet packet energy entropy of each node is calculated based on the wavelet packet coefficients. The wavelet packet energy entropy of each node is used as a multi-scale dynamic feature to construct a multi-scale energy consumption feature matrix.

5. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 4, characterized in that, The identification of key energy consumption influencing factors includes calculating the nonlinear correlation strength between each operating parameter and total energy consumption based on a multi-scale energy consumption feature matrix and using the maximum information coefficient. The contribution of each operating parameter to energy consumption prediction is calculated, and the maximum information coefficient value is weighted and fused with the contribution of the gradient boosting decision tree to obtain the comprehensive correlation score of each operating parameter. The operating parameters are sorted in descending order according to their comprehensive correlation scores. The operating parameters whose comprehensive correlation scores rank higher than the set threshold and meet the adjustability constraints are selected to form the energy consumption sensitive parameter set. The formula for calculating the maximum information coefficient is as follows: ， in, The maximum information coefficient between the operating parameter time series X and the total energy consumption time series Y. This represents the number of mesh elements along the X-axis in a two-dimensional scattered mesh generation. This represents the number of mesh divisions along the Y-axis in a two-dimensional scattered mesh generation. Given an x×y grid, the maximum mutual information value between the operating parameter time series X and the total energy consumption time series Y is given. The upper bound of the search is dependent on the sample size. It is the smaller value between x and y.

6. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The construction of a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries includes selecting adjustable operating parameters with online adjustment capabilities from a set of energy consumption-sensitive parameters and constructing a decision variable vector. ,in, For the frequency of the fan, For the frequency of the feed rotary valve, The air supply valve opening is denoted by n, and the decision variable dimension is denoted by n. A multi-objective optimization function vector is constructed using energy consumption per unit mass of transport, root mean square of system pressure drop fluctuation, and transported mass flow rate as optimization objectives. Construct fusion constraints, which include anti-clogging physical constraints based on minimum delivery wind speed theory, safety constraints based on pipeline pressure limit, equipment boundary constraints based on fan rated speed, and dust concentration constraints based on environmental emission requirements. The penalty function method is used to embed the fusion constraints into the multi-objective optimization function vector to construct a standardized objective function suitable for optimization.

7. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The proposed collaborative optimization strategy employing an adaptive surrogate model and a multi-objective evolutionary algorithm includes generating an initial sample set within the feasible region defined by the boundaries of each component of the decision variable vector using Latin hypercube sampling, constructing an initial Kriging surrogate model based on the initial sample set, performing population evolution optimization on the Kriging surrogate model using the NSGA-II algorithm, iteratively updating the Kriging surrogate model by constructing an expected improvement point addition criterion, and outputting the final Pareto optimal solution set.

8. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 7, characterized in that, The criteria for constructing the expected improvement points include generating a set of candidate sample points within the feasible domain of the decision variables, and calculating the expected improvement value of each candidate sample point based on the predicted mean and predicted variance of the Kriging surrogate model. Based on the expected improvement value, each candidate sample point is sorted in descending order, and the candidate sample points that are ranked by expected improvement value and are located in the sparse region of the Pareto front or the region with high prediction uncertainty are selected as supplementary sample points. The formula for calculating the expected improvement value is as follows: ， in, Let x be the expected improvement value of the candidate sample point. This represents the optimal value of the objective function corresponding to the current Pareto front. Let x be the mean of the predictions made by the Kriging surrogate model at the candidate sample point x. Let x be the standard deviation of the Kriging surrogate model's predictions at the candidate sample point x. The cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.

9. The intelligent optimization control method for energy consumption of pneumatic conveying of powder and granules according to claim 1, characterized in that, The process of interpreting the optimal parameters through fuzzy decision-making and executing adaptive closed-loop optimization control includes defining trapezoidal membership functions for three optimization objectives—unit energy consumption, pressure drop fluctuation, and transport flow rate—for each Pareto solution in the Pareto optimal solution set. Based on the current operating conditions, a target weight vector is set, the weighted fuzzy comprehensive satisfaction of each Pareto solution is calculated, and the Pareto solution corresponding to the maximum value is selected as the final solution. The final solution is interpreted into executable operating parameter instructions, which are then sent to the controller of the pneumatic conveying system to perform frequency conversion adjustment of the blower, adjustment of the opening of the feeding valve, and linkage control of the air replenishment valve.

10. A smart energy consumption optimization control system for pneumatic conveying of powders and granules, executing the method described in claim 1, characterized in that, include: Multi-source data acquisition is used to collect multi-source heterogeneous operating data and powder and granular material properties of the pneumatic conveying system under all working conditions, and to perform adaptive preprocessing and standardization on the operating data to obtain a standardized multi-dimensional time series dataset. The coupled energy consumption characterization module is used to establish a coupled energy consumption characterization model based on a standardized multi-dimensional time series dataset and powder and granular material properties, so as to analyze the physical formation mechanism of transport energy consumption. The sensitive parameter identification module is used to extract multi-scale dynamic energy consumption features from a standardized multi-dimensional time series dataset and identify key energy consumption influencing factors by utilizing a coupled energy consumption characterization model, and to construct a set of energy consumption sensitive parameters. The multi-objective optimization building module is used to construct a multi-objective energy consumption optimization function that integrates physical constraints and operational boundaries based on a set of energy-sensitive parameters and adjustable operating parameters as decision variables. The collaborative optimization module is used to solve the Pareto optimal solution set based on a multi-objective energy consumption optimization function, using a collaborative optimization strategy of adaptive surrogate model and multi-objective evolutionary algorithm; The closed-loop control output module is used to utilize the Pareto optimal solution set, interpret the optimal parameters through fuzzy decision-making, and execute adaptive closed-loop optimization control, outputting the optimized operating parameter instructions.

Images