Air separation plant energy efficiency intelligent management system based on data analysis

Abstract

The invention relates to the technical field of air separation plant management, and discloses an air separation plant energy efficiency intelligent management system based on data analysis, which comprises a data acquisition module, an energy efficiency analysis module, an optimization strategy generation module, a real-time adjustment module and a dynamic feedback module. The data acquisition module acquires and stores data in real time in a multi-source heterogeneous manner. The energy efficiency analysis module uses a hybrid convolutional neural network and an adaptive wavelet packet decomposition algorithm to extract features. And the optimization strategy generation module formulates a strategy in combination with deep reinforcement learning and a multi-target particle swarm optimization algorithm. And the real-time adjustment module dynamically adjusts equipment parameters based on fuzzy logic control. And the dynamic feedback module forms closed-loop optimization. The system solves the problem of energy efficiency management of traditional air separation equipment, and can accurately collect and process data, deeply analyze energy efficiency, intelligently generate optimization strategies, accurately adjust parameters in real time and continuously optimize the parameters, improve the energy efficiency of the air separation equipment, reduce energy consumption, enhance the operation stability and reliability of the equipment and assist industrial sustainable development.

Description

Technical Field

[0001] The present invention relates to the technical field of air separation equipment management, and specifically to an intelligent energy efficiency management system for air separation equipment based on data analysis. Background Art

[0002] As a key piece of equipment in the industrial field, air separation equipment is widely used in many industries such as chemical engineering, steel, and electronics. It is mainly used to separate and purify gases such as oxygen, nitrogen, and argon in the air to meet the gas demand of different production processes. Against the backdrop of the global advocacy for energy conservation, emission reduction, and green development, the energy efficiency issue of air separation equipment has attracted much attention. This is not only related to the production cost control of enterprises but also of great significance to the sustainable development of the industry. However, current air separation equipment faces many challenges in energy efficiency management and urgently needs effective solutions.

[0003] During the operation of air separation equipment, various types of data are involved, such as compressor unit power data, rectification column temperature gradient data, gas purity monitoring data, cooling water flow data, etc. These data have the characteristics of multi-source heterogeneity. Traditional data acquisition methods often rely on manual regular inspection records or simple sensor acquisitions, resulting in problems such as untimely data acquisition, low accuracy, and inability to achieve synchronous acquisition at multiple time granularities. This makes the obtained data unable to comprehensively and accurately reflect the real-time operation status of the equipment and difficult to meet the needs of in-depth analysis and optimization of the equipment. At the same time, due to the lack of spatio-temporal correlation of the data, many difficulties are faced in storage and processing, and it cannot effectively support subsequent energy efficiency analysis and decision-making. Existing energy efficiency analysis methods for air separation equipment are mostly based on simple empirical formulas or traditional statistical analyses, and it is difficult to deeply explore the complex energy efficiency characteristics and correlation relationships behind the data. For example, only by monitoring the overall energy consumption and gas production of the equipment to calculate the energy efficiency, it is impossible to accurately locate the specific links and reasons for excessive energy consumption. Facing the complex operating conditions and a large amount of operating data of air separation equipment, traditional methods cannot fully utilize the value of the data and cannot analyze the equipment energy efficiency from multiple scales and dimensions, resulting in the inability to provide strong technical support for the optimized operation of the equipment.

[0004] Formulating an energy efficiency optimization strategy for air separation equipment is a complex multi-objective optimization problem that needs to comprehensively consider multiple factors such as energy consumption reduction, gas purity improvement, equipment stability, and adjustment frequency. Traditional optimization algorithms are difficult to take these objectives into account simultaneously and are inefficient in processing large-scale and high-dimensional equipment operation data. Although deep reinforcement learning and multi-objective particle swarm optimization algorithms have theoretical advantages, they are less applied in the field of air separation equipment and lack effective integration and optimization for the characteristics of air separation equipment, making it difficult to generate practical energy efficiency optimization strategies and equipment operation parameter adjustment plans.

[0005] During the operation of the air separation unit, the operating conditions can change at any time, and it is necessary to adjust the equipment operating parameters in a timely manner to maintain the best energy efficiency state. However, at present, the control methods of most air separation units are relatively rough, lacking precise and real-time adjustment means. Although fuzzy logic control has application potential, in the field of air separation units, the construction of its fuzzy rule base is often not perfect, and the membership function cannot be adjusted dynamically in real time according to the equipment operating state, resulting in poor control effects. In addition, the existing systems generally lack an effective dynamic feedback mechanism, unable to monitor the optimization effect in real time and transmit the feedback information to the system for adjustment in a timely manner, making it difficult to form a closed-loop optimization, which restricts the continuous improvement of the energy efficiency of air separation units. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent energy efficiency management system for air separation units based on data analysis to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: An intelligent energy efficiency management system for air separation units based on data analysis, the system includes:

[0008] It includes a data acquisition module, an energy efficiency analysis module, an optimization strategy generation module, a real-time adjustment module, and a dynamic feedback module;

[0009] The data acquisition module is used to collect multi-source heterogeneous data during the operation of the air separation unit. Specifically, it performs real-time data acquisition by deploying temperature sensors, pressure sensors, flow sensors, and energy consumption monitoring devices to obtain the original data set of the air separation unit, and sends the original data set of the air separation unit to the energy efficiency analysis module;

[0010] The real-time data acquisition specifically refers to synchronous acquisition based on multiple time granularities, including acquiring high-precision instantaneous data every 0.5 seconds and steady-state operation data every 30 seconds. Through the synchronous acquisition of multiple time granularities, a spatio-temporal correlated original data sequence is generated and stored in a distributed time series database;

[0011] The energy efficiency analysis module is used to extract energy efficiency features and perform correlation analysis on the original data. Specifically, through a hybrid convolutional neural network and an adaptive wavelet packet decomposition algorithm, multi-scale feature extraction is performed on the original data set of the air separation unit to obtain an energy efficiency feature matrix, and the energy efficiency feature matrix is sent to the optimization strategy generation module and the real-time adjustment module;

[0012] The optimization strategy generation module is used to generate energy efficiency optimization strategies and equipment operating parameter adjustment plans. Specifically, based on the energy efficiency feature matrix, a deep reinforcement learning combined with a multi-objective particle swarm optimization algorithm is used to generate energy efficiency optimization strategies to obtain an optimization strategy set for the air separation unit, and the optimization strategy set is sent to the real-time adjustment module;

[0013] The real-time adjustment module is used to execute the optimization strategy and dynamically adjust the device operation parameters. Specifically, by constructing a dynamic regulator based on fuzzy logic control, combining with the optimization strategy set to generate real-time control instructions, and transmitting the control instructions to the execution unit of the air separation device;

[0014] The dynamic feedback module is used to monitor the optimization effect in real time and feedback it to the system. Specifically, by constructing an incremental online learning model, dynamically updating the characteristic data of the adjusted device operation data, and feeding the updated characteristic data back to the energy efficiency analysis module to form a closed-loop optimization link.

[0015] Preferably, in the data acquisition module, the original data set of the air separation device specifically includes compressor unit power data, rectification column temperature gradient data, gas purity monitoring data, cooling water flow data, and equipment historical maintenance record data; the equipment historical maintenance record data includes equipment start-stop logs, component replacement records, and abnormal event marking data.

[0016] Preferably, in the energy efficiency analysis module, the processing steps of the hybrid convolutional neural network and the adaptive wavelet packet decomposition algorithm include:

[0017] Step S1: Decompose the original data in the time-frequency domain through the adaptive wavelet packet decomposition algorithm to generate multi-resolution sub-band signals;

[0018] Step S2: Construct three-dimensional convolution kernels for each sub-band signal, and extract spatial-frequency domain joint features through the parallel convolution layer;

[0019] Step S3: Adopt the channel attention mechanism to allocate weights to multi-channel features to generate an optimized energy efficiency feature matrix;

[0020] Step S4: Nonlinearly superimpose the features of different sub-bands through the cross-channel feature fusion algorithm to form a global energy efficiency characterization vector.

[0021] Preferably, the specific implementation steps of the deep reinforcement learning combined with the multi-objective particle swarm optimization algorithm include:

[0022] Construct a Markov decision process model, define the state space as the energy efficiency feature matrix, and the action space as the set of device parameter adjustment instructions;

[0023] Adopt a double deep Q network for policy exploration, and introduce a Pareto front screening mechanism to reduce the dimensionality of the solution set for multi-objective optimization problems;

[0024] Globally optimize the solution set through the dynamic inertia weight particle swarm algorithm to generate an optimal energy efficiency policy combination.

[0025] Preferably, the dynamic regulator with fuzzy logic control specifically includes a fuzzy rule base construction, a membership function dynamic adjustment sub-module, and a defuzzification processor;

[0026] The fuzzy rule base construction generates fuzzy control rules based on expert experience and historical optimization data, and uses a genetic algorithm to iteratively optimize the rule weights;

[0027] The membership function dynamic adjustment sub-module adjusts the width and center point of the triangular membership function in real time by online learning the change rate of the device operation state;

[0028] The defuzzification processor converts the fuzzy output into an accurate control instruction using the centroid method.

[0029] Preferably, the incremental online learning model is specifically a variational autoencoder driven by streaming data, and its processing flow includes:

[0030] Step P1: Perform sliding window sampling on the real-time input device operation data to generate dynamic data blocks;

[0031] Step P2: Perform feature reconstruction and latent space mapping on the data blocks through a variational autoencoder to generate incremental feature vectors;

[0032] Step P3: Use KL divergence to measure the difference between the new and old feature distributions, and dynamically update the encoder parameters to adapt to the change of the device state.

[0033] Preferably, in the multi-objective particle swarm optimization algorithm, the objective function is defined as:

[0034] f1 = α * energy consumption reduction rate + β * gas purity deviation

[0035] f2 = γ * device stability index - δ * adjustment frequency

[0036] Among them, α, β, γ, and δ are dynamic weight coefficients, which are adaptively adjusted according to the device operation stage.

[0037] Preferably, the construction method of the three-dimensional convolution kernel is: set separable convolution kernels in the time dimension, space dimension, and frequency domain respectively, and generate a composite convolution weight matrix through tensor product; the update of the composite convolution weight matrix adopts gradient descent combined with the Nesterov acceleration strategy.

[0038] Preferably, the sliding window sampling strategy of the dynamic data block includes:

[0039] Adaptive window size adjustment based on the detection of device operation mode switching. When a sudden change in working conditions is detected, the window is reduced to improve the response speed; during steady-state operation, the window is enlarged to enhance the robustness of feature extraction.

[0040] Preferably, the iterative optimization step of the genetic algorithm on the fuzzy rule weights includes:

[0041] Encoding phase: converting rule weights into binary gene sequences;

[0042] Selection stage: using the tournament selection mechanism to screen individuals with high fitness;

[0043] Crossover and mutation phase: Generate a new generation of population using uniform crossover operator and dynamic mutation probability strategy;

[0044] The fitness function is defined as the inverse ratio of the fitting error of the rule base to the historical optimization data and the weighted sum of the rule complexity.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The system's data acquisition module uses a variety of sensors to achieve real-time acquisition of multi-source heterogeneous data, and adopts a multi-time granularity synchronous acquisition method to collect high-precision instantaneous data every 0.5 seconds and steady-state operation data every 30 seconds, generating a time-space-related raw data sequence and storing it in a distributed time series database. Compared with traditional acquisition methods, this method not only greatly increases the acquisition frequency, but also can obtain more comprehensive and detailed operation information of the equipment, accurately reflecting the real-time status of the equipment from multiple dimensions. For example, when monitoring the power changes of the compressor unit, high-precision instantaneous data can capture the power fluctuations at the moment of startup and loading, and steady-state operation data can help analyze the energy consumption patterns during normal operation, providing a solid and reliable data foundation for subsequent energy efficiency analysis and optimization strategy formulation.

[0047] By combining a convolutional neural network with an adaptive wavelet packet decomposition algorithm, the system can perform multi-scale feature extraction on the original data, uncovering deep energy efficiency features and complex correlation relationships. The adaptive wavelet packet decomposition algorithm performs a fine decomposition of the original data in the time-frequency domain, and the multi-resolution sub-band signals can reveal the characteristics of the data at different frequencies and time scales; the three-dimensional convolutional kernels and parallel convolutional layers work together to effectively extract spatial-frequency domain joint features; the channel attention mechanism and cross-channel feature fusion algorithm further optimize and integrate the features to form a global energy efficiency representation vector. This in-depth analysis method can accurately locate the links and reasons for the excessively high energy consumption of the air separation equipment, providing strong support for the formulation of optimization strategies. Compared with traditional simple energy efficiency analysis methods, the analysis results are more accurate and comprehensive, helping to achieve more efficient energy utilization. The optimization strategy generation module uses deep reinforcement learning combined with a multi-objective particle swarm optimization algorithm to effectively solve the complex multi-objective optimization problem of air separation equipment energy efficiency optimization. The constructed Markov decision process model defines the state space and action space. The policy exploration of the double deep Q network combined with the Pareto front screening mechanism can efficiently handle multi-objective optimization problems and reduce the redundancy of the solution set; the dynamic inertia weight particle swarm algorithm performs global optimization to generate an optimal energy efficiency strategy combination that comprehensively considers the energy consumption reduction rate, gas purity deviation, equipment stability index, and adjustment frequency. Compared with traditional optimization algorithms, this method can more comprehensively balance multiple optimization objectives, and the generated optimization strategies are more in line with the actual operating requirements of the air separation equipment, helping to reduce energy consumption, improve equipment stability, and extend the service life of the equipment while ensuring the quality of gas products.

[0048] The dynamic regulator based on fuzzy logic control constructed by the real-time adjustment module realizes the real-time and precise adjustment of the device operation parameters. The fuzzy rule base is generated based on expert experience and historical optimization data, and the rule weights are iteratively optimized through the genetic algorithm to make the control rules more in line with the actual operation of the device. The membership function dynamic adjustment sub-module is corrected in real time according to the change rate of the device operation state, enhancing the adaptability of the fuzzy logic control. The defuzzification processor uses the centroid method to convert the fuzzy output into an accurate control instruction to drive the action of the air separation device execution unit. Taking the adjustment of the reflux ratio of the rectification column as an example, the system can quickly and precisely adjust the reflux ratio according to the real-time monitored data such as temperature and pressure, ensuring the efficient and stable operation of the rectification process, improving the gas separation effect, and further enhancing the energy efficiency of the device. The dynamic feedback module constructs an incremental online learning model through a variational autoencoder driven by streaming data, dynamically updates the feature data of the adjusted device operation data, and feeds the updated feature data back to the energy efficiency analysis module to form a closed-loop optimization link. This enables the system to monitor the optimization effect in real time, timely detect changes in the device operation state, and adjust the optimization strategy and operation parameters according to the new data. As time goes by and data accumulates, the system continuously optimizes its own performance, continuously improves the energy efficiency management level of the air separation device, keeps the device running efficiently under different working conditions, and adapts to complex and changeable production requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is the working principle diagram of the air separation device energy efficiency intelligent management system described in the present invention;

[0050] Figure 2 is the working flow chart of the air separation device energy efficiency analysis algorithm;

[0051] Figure 3 is the working flow chart of the air separation device energy efficiency optimization cycle. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] Please refer to Figures 1-3 , the present invention provides a technical solution: an air separation device energy efficiency intelligent management system based on data analysis. The system consists of a data acquisition module, an energy efficiency analysis module, an optimization strategy generation module, a real-time adjustment module, and a dynamic feedback module working together to realize the intelligent management and optimization of the air separation device energy efficiency.

[0054] The data acquisition module plays a crucial role in obtaining data during the operation of the air separation unit. This module deploys temperature sensors, pressure sensors, flow sensors, and energy consumption monitoring devices to collect real-time operation data of the equipment. This collection is synchronized based on multiple time granularities. High-precision instantaneous data is collected every 0.5 seconds to capture the instantaneous changes during the equipment operation; steady-state operation data is collected every 30 seconds to obtain the data characteristics under the stable operation state of the equipment. Through this multi-time granularity synchronous collection method, a spatiotemporal correlated raw data sequence is generated and stored in a distributed time-series database, forming a raw data set of the air separation unit. These data comprehensively cover the power data of the compressor unit, the temperature gradient data of the rectification column, the gas purity monitoring data, the cooling water flow data, and the equipment historical maintenance record data (including equipment start-stop logs, component replacement records, and abnormal event marking data), providing rich and accurate basic information for subsequent analysis and decision-making.

[0055] The energy efficiency analysis module receives the raw data set of the air separation unit from the data acquisition module and conducts in-depth energy efficiency feature extraction and correlation analysis on the raw data through a hybrid convolutional neural network and an adaptive wavelet packet decomposition algorithm. First, the adaptive wavelet packet decomposition algorithm is used to decompose the raw data in the time-frequency domain, decomposing the complex raw signal into multi-resolution sub-band signals to more carefully analyze the characteristics of the signal at different frequencies and time scales. Then, a three-dimensional convolutional kernel is constructed for each sub-band signal, and the spatial-frequency domain joint features are extracted through a parallel convolutional layer to fully exploit the feature information of the data in different dimensions. Next, a channel attention mechanism is used to assign weights to the multi-channel features, highlighting the features that have an important impact on energy efficiency analysis and generating an optimized energy efficiency feature matrix. Finally, through a cross-channel feature fusion algorithm, the different sub-band features are non-linearly superimposed to form a global energy efficiency characterization vector that can comprehensively represent the energy efficiency status of the air separation unit. This vector is sent to the optimization strategy generation module and the real-time adjustment module as an important basis for the decision-making of subsequent modules.

[0056] The optimization strategy generation module generates an energy efficiency optimization strategy and an equipment operation parameter adjustment plan based on the energy efficiency feature matrix provided by the energy efficiency analysis module, using deep reinforcement learning combined with a multi-objective particle swarm optimization algorithm. Specifically, a Markov decision process model is constructed, defining the energy efficiency feature matrix as the state space and the set of equipment parameter adjustment instructions as the action space. A double deep Q network is used for policy exploration, and a Pareto front screening mechanism is introduced during the exploration to reduce the dimensionality of the solution set for multi-objective optimization problems, reducing the computational complexity and screening out better solutions. Then, the dynamic inertia weight particle swarm algorithm is used to globally optimize the solution set, comprehensively considering factors such as the energy consumption reduction rate, gas purity deviation, equipment stability index, and adjustment frequency, generating an optimal energy efficiency strategy combination, that is, the optimization strategy set of the air separation unit, and sending it to the real-time adjustment module.

[0057] The real-time adjustment module is responsible for implementing the optimization strategy and dynamically adjusting the device operation parameters. This module realizes this function by constructing a dynamic regulator based on fuzzy logic control. The dynamic regulator of fuzzy logic control includes the construction of a fuzzy rule base, a sub-module for dynamically adjusting the membership function, and a defuzzification processor. The fuzzy rule base generates fuzzy control rules based on expert experience and historical optimization data, and uses a genetic algorithm to iteratively optimize the rule weights to make the rules more in line with the actual operating conditions. The sub-module for dynamically adjusting the membership function corrects the width and center point of the triangular membership function in real time by online learning the change rate of the device operating state to adapt to the change of the device operating state. The defuzzification processor uses the centroid method to convert the fuzzy output into an accurate control instruction and transmits these instructions to the execution unit of the air separation device to achieve precise adjustment of the operation parameters of the air separation device.

[0058] The dynamic feedback module updates the dynamic features of the adjusted device operation data by constructing an incremental online learning model, realizes real-time monitoring of the optimization effect, and feeds it back to the system. The variational autoencoder based on streaming data-driven is used as the incremental online learning model. First, it samples the real-time input device operation data through a sliding window to generate dynamic data blocks. Then, it reconstructs the features and maps them to the latent space of the data blocks through the variational autoencoder to generate incremental feature vectors. Finally, it uses the KL divergence to measure the difference between the new and old feature distributions, and dynamically updates the encoder parameters to adapt to the change of the device state. The updated feature data is fed back to the energy efficiency analysis module to form a closed-loop optimization link, continuously optimizing the system's management and control of the energy efficiency of the air separation device.

[0059] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0060] Embodiment 1:

[0061] During the operation of the air separation device, the accuracy and integrity of the data acquisition module play a decisive role in the performance of the entire system.

[0062] For the acquisition of the power data of the compressor unit, high-precision power sensors are installed on the power supply lines of the compressor unit. Based on the principle of electromagnetic induction, the sensors can monitor the changes in current and voltage in real time, and calculate the real-time power of the compressor unit through the internal calculation chip according to the formula (where P is power, U is voltage, I is current, is the power factor). The sensors collect high-precision instantaneous power data at intervals of 0.5 seconds to capture the instantaneous power fluctuations of the compressor unit under different working conditions such as startup, loading, and unloading. At the same time, the steady-state operating power data is collected once every 30 seconds to record the power consumption of the compressor unit in the stable operating state, providing basic data for subsequent analysis of the energy efficiency of the compressor unit.

[0063] The collection of temperature gradient data of the distillation tower depends on multiple temperature sensors installed at different heights of the distillation tower. These sensors are evenly distributed on the tower body of the distillation tower, forming an array of temperature measurement points from the top to the bottom of the tower. The sensor uses a high-precision thermistor, whose resistance value changes accurately with the change of temperature. By measuring the resistance value of the thermistor and converting the resistance value into the corresponding temperature value based on the pre-calibrated temperature-resistance relationship curve. Each temperature sensor also collects data synchronously at multiple time granularities, collecting instantaneous temperature data once every 0.5 seconds to monitor the rapid changes in temperature in the distillation tower, such as temperature fluctuations caused by feed fluctuations, reflux changes, etc.; steady-state temperature data is collected every 30 seconds to analyze the temperature distribution law of the distillation tower under stable operating conditions. Based on the collected temperature data at different heights, the temperature gradient of the distillation tower can be calculated. The temperature gradient calculation formula is: (where ΔT is the temperature difference between adjacent measuring points, and Δh is the height difference between adjacent measuring points). The temperature gradient data is of great significance for judging the distillation effect and energy utilization efficiency of the distillation tower.

[0064] The gas purity monitoring data is collected with the help of a highly sensitive gas composition analyzer. The analyzer uses advanced technologies such as chromatography, mass spectrometry or infrared absorption to accurately measure the content of various components in the gas produced by the air separation equipment. Taking oxygen purity monitoring as an example, the analyzer extracts oxygen samples produced by the air separation equipment, and undergoes a series of separation and detection processes internally to accurately calculate the purity of oxygen. Gas purity monitoring data is also collected at a frequency of 0.5 seconds for high-precision instantaneous data and 30 seconds for steady-state data, so as to promptly discover the changing trend of gas purity and provide a basis for adjusting equipment operating parameters.

[0065] The collection of cooling water flow data uses an electromagnetic flowmeter installed in the cooling water pipeline. The electromagnetic flowmeter is based on Faraday's law of electromagnetic induction. When conductive cooling water flows in a magnetic field, an induced electromotive force is generated in a direction perpendicular to the magnetic field and the direction of water flow. The magnitude of the induced electromotive force is proportional to the flow rate of the cooling water. By measuring the induced electromotive force and based on the calibration coefficient of the electromagnetic flowmeter, the flow rate of cooling water can be calculated. The calculation formula is Q = vA (where Q is the flow rate, v is the flow rate, and A is the cross-sectional area of ​​the pipe). The collected cooling water flow data is stored in a multi-time granularity synchronous collection method to provide data support for analyzing the cooling effect and energy consumption relationship of air separation equipment.

[0066] The collection of device historical maintenance record data is achieved by docking with the device maintenance management system. The equipment start-stop log records information such as the start and stop times of the air separation device each time, the operator, and whether there are any abnormalities during the start and stop processes. The component replacement record details information such as the replacement time of each component of the equipment, the model and manufacturer of the replaced component, etc. The abnormal event marking data marks abnormal situations such as faults and alarms that occur during the operation of the equipment, including the time of occurrence of the abnormality, the type of abnormality, and the handling measures, etc. These data are transmitted to the distributed time-series database through the data interface and stored together with other operation data to provide historical data reference for equipment fault diagnosis, performance evaluation, and energy efficiency optimization.

[0067] Embodiment 2:

[0068] This embodiment is used to describe the specific processing flow of the hybrid convolutional neural network and the adaptive wavelet packet decomposition algorithm of the energy efficiency analysis module. This algorithm plays a key role in mining the energy efficiency characteristics of the operation data of the air separation device.

[0069] In step S1, the adaptive wavelet packet decomposition algorithm performs time-frequency domain decomposition on the original data. Taking a certain key operation parameter in the air separation device (such as the vibration signal of the compressor unit) as an example, this signal is a complex signal that changes with time and contains different frequency components and time characteristics. The adaptive wavelet packet decomposition algorithm adaptively selects the wavelet basis function and the decomposition layer number according to the energy distribution characteristics of the signal. Assuming the original signal is x(t), it is decomposed through the wavelet packet decomposition function W j,k (t) (where j is the decomposition layer number and k is the frequency sub-band index) to obtain the multi-resolution sub-frequency band signal d j,k (t), and its calculation formula is Through this decomposition method, the original signal can be refined and analyzed at different frequencies and time scales, and more representative features can be extracted.

[0070] In step S2, a three-dimensional convolution kernel is constructed for each sub-frequency band signal. Taking one of the sub-frequency band signals d j,k (t) as an example, in the time dimension, according to the time series characteristics of the signal, a convolution kernel with an appropriate length is set. Assuming the length of the convolution kernel in the time dimension is T; in the spatial dimension, considering the spatial correlation between different components of the air separation device, the size of the convolution kernel in the spatial dimension is set. Assuming the size of the convolution kernel in the spatial dimension is S×S; in the frequency domain dimension, according to the frequency range of the sub-frequency band signal, the convolution kernel parameters in the frequency domain dimension are set. The composite convolution weight matrix C is generated through the tensor product, (where C T is the convolution kernel matrix in the time dimension, C S is the convolution kernel matrix in the spatial dimension, C Fis the convolution kernel matrix in the frequency domain dimension, is the tensor product operator). The parallel convolution layer is used to perform a convolution operation on the sub-band signal to extract the spatial-frequency domain joint features. The parallel convolution layer can simultaneously perform convolution operations on the sub-band signals of different channels, improving the feature extraction efficiency.

[0071] In step S3, a channel attention mechanism is adopted to allocate weights to the multi-channel features. Assume that the multi-channel features extracted by the parallel convolution layer are F = [f1, f2, …, f n , where n is the number of channels. First, a global average pooling operation is performed on the features of each channel to obtain the global statistical information of the channel features (where H, W, and D are the sizes of the feature map in the height, width, and depth directions respectively). Then, the global statistical information is input into a network containing a multi-layer perceptron (MLP). The structure of the MLP is MLP(g i ) = W2σ(W1g i )(where W1 and W2 are the weight matrices of the MLP, and σ is the activation function, such as the RelU function), to obtain the weight coefficient w i of each channel. Finally, the weight coefficient is multiplied by the original channel features to generate the optimized energy efficiency feature matrix F optimized = [w1f1, w2f2, …, w n f n , highlighting the feature channels important for energy efficiency analysis and suppressing irrelevant or secondary feature channels.

[0072] In step S4, different sub-band features are non-linearly superimposed through a cross-channel feature fusion algorithm. Assume that the different sub-band feature matrices optimized by the channel attention mechanism are F1, F2, …, F m (where m is the number of sub-bands). First, the feature dimension transformation is performed on each sub-band feature matrix to make them have the same dimension. Then, a non-linear fusion function, such as (where a i is the fusion coefficient obtained through training and learning, and tanh is the hyperbolic tangent activation function), is used to non-linearly superimpose different sub-band features to form the global energy efficiency representation vector G. This global energy efficiency representation vector synthesizes the feature information of different sub-bands and can more comprehensively reflect the energy efficiency status of the air separation equipment.

[0073] Example 3:

[0074] This example focuses on the specific implementation steps of the deep reinforcement learning combined with the multi-objective particle swarm optimization algorithm in claim 4. This algorithm is crucial for generating an efficient energy efficiency optimization strategy for the air separation equipment.

[0075] When constructing the Markov decision process model, the energy efficiency feature matrix output by the energy efficiency analysis module is used as the state space S. Assume the energy efficiency feature matrix is E = [e1, e2, …, e n , where e i are different energy efficiency feature indicators. The set of device parameter adjustment instructions is defined as the action space A. For example, operation instructions such as adjusting the intake air volume of the compressor unit and the reflux ratio of the distillation column. At each decision-making moment t, the system selects an action a t ∈ A according to the current state s t ∈ S. After executing this action, the system will transfer to a new state s t+1 , and obtain a reward value r t . The setting of the reward value is related to the optimization goal of the system. For example, if the energy consumption is reduced or the gas purity is increased, a positive reward is given; otherwise, a negative reward is given.

[0076] The double deep Q-network is used for policy exploration. The double deep Q-network (DDQN) solves the overestimation problem existing in the traditional deep Q-network (DQN) by introducing two neural networks, namely the evaluation network Q(s, a; θ) and the target network Q(s, a; θ - ). During the training process, first select an action from the evaluation network according to the current state s t . Then execute this action in the environment, observe the new state s t+1 and the reward r t . Next, calculate the target Q value (where γ is the discount factor, used to balance the importance of the current reward and future rewards) using the target network. By minimizing the loss function L(θ) = E[(y t - Q(s t , a t ; θ)) 2 , update the parameters θ of the evaluation network using the stochastic gradient descent algorithm. After a certain number of training steps, copy the parameters of the evaluation network to the target network, that is, θ - ← θ, to maintain the stability of the target network.

[0077] In multi-objective optimization problems, a Pareto front screening mechanism is introduced to reduce the dimensionality of the solution set. The multi-objective particle swarm optimization algorithm considers multiple objectives, such as the energy consumption reduction rate, gas purity deviation, equipment stability index, and adjustment frequency. In each iteration process, each particle in the particle swarm has a corresponding set of objective function values. The Pareto front refers to a set of non-dominated solutions in a multi-objective optimization problem, that is, among these solutions, no solution can improve a certain objective without sacrificing other objectives. By comparing the objective function values of the particles, the particles on the Pareto front are screened out. These particles are used as potential optimization solutions, reducing the number of solutions to be processed and the computational complexity.

[0078] The global optimization of the solution set is carried out by the dynamic inertia weight particle swarm algorithm. In the particle swarm algorithm, each particle represents a possible solution, and the position of the particle represents a combination of equipment operating parameters. By continuously updating the speed and position of the particles, the particle swarm gradually approaches the optimal solution and finally generates the optimal energy efficiency strategy combination, that is, the optimization strategy set of the air separation equipment.

[0079] Example 4:

[0080] This example is used to describe the fuzzy logic control dynamic regulator in the real-time adjustment module. This regulator is a key component to achieve precise dynamic adjustment of the operating parameters of the air separation equipment. Its role is to convert the fuzzy control intention into an executable precise instruction according to the strategy output by the optimization strategy generation module, so as to drive the action of the execution unit of the air separation equipment and adjust the equipment operating parameters in the direction of improving energy efficiency, ensuring the stable and efficient operation of the air separation equipment.

[0081] Construction and optimization of the fuzzy rule base: The construction of the fuzzy rule base is to deeply integrate the rich experience of experts in the field of air separation equipment and a large amount of historical optimization data. Experts have accumulated adjustment strategies for equipment operating parameters under different working conditions in long-term practice. For example, when it is observed that the temperature at the top of the rectification column continues to rise and the pressure is at a low level, it is known from experience that the reflux ratio should be appropriately increased at this time to improve the rectification effect and reduce energy consumption. The historical optimization data records the adjustment process of the operating parameters of the equipment under various working conditions in the past and the corresponding energy efficiency changes.

[0082] After sorting out these experiences and data, they are converted into specific fuzzy control rules. For example, set rules with the temperature (T), pressure (P), and reflux ratio (R) of the rectification column as parameters: If T is in the "relatively high" state and P is "relatively low", then R "increases"; if T is "moderate" and P is "moderate", then R "remains unchanged"; if T is "relatively low" and P is "relatively high", then R "decreases". Here, "relatively high", "moderate", "relatively low", "increase", "remain unchanged", and "decrease" are all fuzzy language variables, which are used to summarize different states of equipment operating parameters and corresponding adjustment directions.

[0083] To make these fuzzy rules more suitable for the actual operation of the air separation plant, the genetic algorithm is used to iteratively optimize the rule weights. First is the encoding step, where the weight of each rule is converted into a binary gene sequence.

[0084] Then comes the selection stage, adopting the tournament selection mechanism. Randomly select k individuals (k is the tournament size) from the population, compare their fitness values, and select the individual with the highest fitness to enter the next generation population. The evaluation of fitness is based on the improvement effect of system energy efficiency. The more significant the energy efficiency improvement, the higher the individual fitness.

[0085] Next are the crossover and mutation stages. During crossover, the uniform crossover operator is used to randomly select gene positions of two individuals for exchange to generate new individuals. Mutation adopts the dynamic mutation probability strategy. At the initial stage of iteration, the mutation probability is set relatively low, and as the number of iterations increases, the mutation probability is gradually increased to maintain the diversity of the population and prevent the algorithm from falling into a local optimal solution. Through continuous iteration, the rule weights are optimized so that the fuzzy rules can more accurately reflect the operating requirements of the air separation plant.

[0086] Membership function dynamic adjustment sub-module: Taking the temperature membership function as an example, it is initially set as a triangular membership function. During the operation of the air separation plant, temperature is a key parameter, and its change affects the overall performance and energy efficiency of the plant. For the fuzzy linguistic variable "high temperature", a temperature interval [a, b, c] is set, and the calculation rule of the corresponding triangular membership function μT(x) is as follows: when x ≤ a, μT(x) = 0; when a < x ≤ b, μT(x) = (x - a) / (b - a); when b < x ≤ c, μT(x) = (c - x) / (c - b); when x > c, μT(x) = 0. Here, a is a relatively high temperature threshold, b is the typical temperature value representing "high temperature", and c is a higher temperature threshold.

[0087] During the operation of the plant, the change rate of the plant operation state is obtained through online learning, and the values of a, b, and c are dynamically adjusted according to the temperature change trend. If the temperature rises rapidly and is close to the upper limit of the current membership function, it means that the plant operation state has changed and the membership function needs to be adjusted. At this time, appropriately increase the value of c and adjust the value of b to ensure that the membership function can accurately reflect the current temperature state. When adjusting, the temperature change rate and historical data are comprehensively considered. For example, when the temperature change rate exceeds the preset threshold and continues to rise for a period of time, the adjustment amplitudes of c and b can be appropriately increased according to experience; if the temperature changes relatively smoothly, the values of a, b, and c are slightly adjusted. By continuously monitoring the change rate of the plant operation state, the width and center point of the triangular membership function are continuously corrected, so that the fuzzy logic control can more accurately adapt to the actual operation state of the plant.

[0088] Defuzzification Processor: The defuzzification processor uses the center of gravity method to convert the fuzzy output into an exact control instruction. After the fuzzy inference obtains the output fuzzy set B, its membership function is μB(y), where y represents the value of the output variable.

[0089] After obtaining the exact control instruction in this way, it is transmitted to the execution unit of the air separation equipment. For example, if the fuzzy inference result indicates that the feed rate needs to be increased, the defuzzification processor will calculate the specific increased value of the feed rate, such as increasing by 5 cubic meters per hour. Then, this instruction is sent to the control unit of the feed pump, and the control unit drives the feed pump to adjust the flow rate according to the instruction, perhaps by adjusting the motor speed or valve opening of the pump, to achieve precise adjustment of the feed rate of the air separation equipment, thereby optimizing the equipment operation parameters and improving the equipment energy efficiency.

[0090] Example 5:

[0091] This example mainly elaborates on the incremental online learning model of the dynamic feedback module, which can achieve real-time monitoring of the optimization effect and dynamic update of the system.

[0092] Based on the variational autoencoder driven by streaming data as the incremental online learning model, first, sliding window sampling is performed on the real-time input device operation data. Assume that the real-time input device operation data sequence is X = [x1, x2, …, x N , where x i is the device operation data vector at each time point. Set the size of the sliding window to w, then each sampling obtains a data block X window = [x t-w+1 , x t-w+2 , …, x t , where t is the current time point. As time goes by, the sliding window moves forward continuously, generating a series of dynamic data blocks.

[0093] The variational autoencoder (VAE) consists of two parts: an encoder and a decoder. The encoder maps the input data block to the latent space. The encoder is usually implemented by a neural network, which learns the feature distribution of the input data and converts the input data into latent variables. The role of the decoder is to reconstruct the latent variables into an approximation of the original data block.

[0094] During the training process, the goal of the variational autoencoder is to maximize the evidence lower bound (ELBO). By maximizing the ELBO, the variational autoencoder can learn the latent feature distribution of the data and achieve effective reconstruction of the data.

[0095] The KL divergence is used to measure the difference between the distributions of the old and new features. If the value of the KL divergence is large, it indicates that there is a significant difference between the distributions of the old and new features, and the operating state of the device has changed significantly. At this time, it is necessary to dynamically update the encoder parameters to adapt to the change in the device state. Through the backpropagation algorithm, the parameters of the encoder and decoder are updated according to the gradient information of the ELBO, enabling the variational autoencoder to continuously adapt to the new data features.

[0096] The updated feature data is fed back to the energy efficiency analysis module. The energy efficiency analysis module re-performs energy efficiency feature extraction and correlation analysis based on the new feature data, adjusts the optimization strategy generated by the optimization strategy generation module, and the real-time adjustment module further adjusts the operating parameters of the air separation device according to the new optimization strategy. Through this closed-loop optimization link, the system can respond in real time to changes in the operating state of the device and continuously optimize the energy efficiency management of the air separation device.

[0097] For example, when the air separation device has undergone a maintenance or replaced some components, the operating state of the device may change. The dynamic feedback module can timely capture this change through the incremental online learning model, judge the difference in feature distribution through the KL divergence, update the encoder parameters, and feed the new feature data back to the energy efficiency analysis module. The energy efficiency analysis module re-evaluates the energy efficiency status of the device according to the new feature data, the optimization strategy generation module generates an optimization strategy more suitable for the current device state, and the real-time adjustment module executes the new strategy, thereby realizing the dynamic optimization of the energy efficiency of the air separation device.

[0098] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0099] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made in these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent energy efficiency management system for air separation equipment based on data analysis, characterized in that: It includes a data acquisition module, an energy efficiency analysis module, an optimization strategy generation module, a real-time adjustment module, and a dynamic feedback module; The data acquisition module is used to collect multi-source heterogeneous data during the operation of the air separation equipment. Specifically, it performs real-time data acquisition by deploying temperature sensors, pressure sensors, flow sensors, and energy consumption monitoring devices to obtain the original data set of the air separation equipment, and sends the original data set of the air separation equipment to the energy efficiency analysis module; The real-time data acquisition specifically refers to synchronous acquisition based on multiple time granularities, including acquiring high-precision instantaneous data every 0.5 seconds and steady-state operation data every 30 seconds. Through the synchronous acquisition based on multiple time granularities, a spatio-temporal correlated original data sequence is generated and stored in a distributed time series database; The energy efficiency analysis module is used to extract energy efficiency characteristics and perform correlation analysis on the original data. Specifically, through a hybrid convolutional neural network and an adaptive wavelet packet decomposition algorithm, multi-scale feature extraction is performed on the original data set of the air separation equipment to obtain an energy efficiency feature matrix, and the energy efficiency feature matrix is sent to the optimization strategy generation module and the real-time adjustment module; The optimization strategy generation module is used to generate energy efficiency optimization strategies and equipment operation parameter adjustment plans. Specifically, based on the energy efficiency feature matrix, a deep reinforcement learning combined with a multi-objective particle swarm optimization algorithm is used to generate energy efficiency optimization strategies to obtain an optimization strategy set for the air separation equipment, and the optimization strategy set is sent to the real-time adjustment module; The real-time adjustment module is used to execute the optimization strategy and dynamically adjust the equipment operation parameters. Specifically, by constructing a dynamic regulator based on fuzzy logic control, combined with the optimization strategy set, real-time control instructions are generated, and the control instructions are transmitted to the execution unit of the air separation equipment; The dynamic feedback module is used to monitor the optimization effect in real time and feedback it to the system. Specifically, by constructing an incremental online learning model, the dynamic characteristics of the adjusted equipment operation data are updated, and the updated characteristic data is feedback to the energy efficiency analysis module to form a closed-loop optimization link.

2. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 1, characterized in that In the data acquisition module, the original data set of the air separation equipment specifically includes compressor unit power data, rectification tower temperature gradient data, gas purity monitoring data, cooling water flow data, and equipment historical maintenance record data; the equipment historical maintenance record data includes equipment start-stop logs, component replacement records, and abnormal event marking data.

3. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 2, characterized in that, In the energy efficiency analysis module, the processing steps of the hybrid convolutional neural network and the adaptive wavelet packet decomposition algorithm include: Step S1: Perform time-frequency domain decomposition on the original data through the adaptive wavelet packet decomposition algorithm to generate multi-resolution sub-band signals; Step S2: Construct a three-dimensional convolutional kernel for each sub-band signal, and extract spatio-frequency domain joint features through a parallel convolutional layer; Step S3: Use a channel attention mechanism to assign weights to multi-channel features to generate an optimized energy efficiency feature matrix; Step S4: Non-linearly superimpose different sub-band features through a cross-channel feature fusion algorithm to form a global energy efficiency characterization vector.

4. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 3, wherein, The specific implementation steps of the deep reinforcement learning combined with the multi-objective particle swarm optimization algorithm are as follows: Construct a Markov decision process model, define the state space as the energy efficiency feature matrix, and the action space as the set of device parameter adjustment instructions; Use a double deep Q-network for policy exploration, and introduce a Pareto front screening mechanism to reduce the dimensionality of the solution set for the multi-objective optimization problem; Globally optimize the solution set through a dynamic inertia weight particle swarm algorithm to generate an optimal energy efficiency policy combination.

5. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 4, characterized in that, The dynamic regulator of the fuzzy logic control specifically includes a fuzzy rule base construction, a membership function dynamic adjustment sub-module, and a defuzzification processor; For the construction of the fuzzy rule base, fuzzy control rules are generated based on expert experience and historical optimization data, and a genetic algorithm is used to iteratively optimize the rule weights; The membership function dynamic adjustment sub-module real-time corrects the width and center point of the triangular membership function by online learning the change rate of the device operating state; The defuzzification processor uses the centroid method to convert the fuzzy output into an accurate control instruction.

6. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 1, wherein, The incremental online learning model is specifically a variational autoencoder driven by streaming data, and its processing flow includes: Step P1: Perform sliding window sampling on the real-time input device operation data to generate a dynamic data block; Step P2: Perform feature reconstruction and latent space mapping on the data block through a variational autoencoder to generate an incremental feature vector; Step P3: Use the KL divergence to measure the difference between the new and old feature distributions, and dynamically update the encoder parameters to adapt to the change of the device state.

7. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 4, characterized in that, In the multi-objective particle swarm optimization algorithm, the objective function is defined as: f1 = α * energy consumption reduction rate + β * gas purity deviation f2 = γ * device stability index - δ * adjustment frequency Among them, α, β, γ, and δ are dynamic weight coefficients, which are adaptively adjusted according to the device operation stage.

8. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 3, characterized in that, The construction method of the three-dimensional convolution kernel is as follows: separable convolution kernels are set in the time dimension, space dimension, and frequency domain respectively, and a composite convolution weight matrix is generated through tensor product; the update of the composite convolution weight matrix adopts gradient descent combined with the Nesterov acceleration strategy.

9. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 6, characterized in that The sliding window sampling strategy of the dynamic data block includes: Adaptive window size adjustment based on the detection of device operation mode switching. When a sudden change in working conditions is detected, the window is reduced to improve the response speed; during steady-state operation, the window is enlarged to enhance the robustness of feature extraction.

10. The intelligent energy efficiency management system for air separation equipment based on data analysis according to claim 5, characterized in that, The iterative optimization steps of the genetic algorithm for the fuzzy rule weights include: Coding stage: Convert the rule weights into a binary gene sequence; Selection stage: Use a tournament selection mechanism to screen high-fitness individuals; Crossover and mutation stage: Use a uniform crossover operator and a dynamic mutation probability strategy to generate a new generation of population; The fitness function is defined as the inverse of the weighted sum of the fitting error of the rule base to the historical optimization data and the rule complexity.

Images