Zero-gas-consumption compressed air adsorption drying intelligent control method and system

Abstract

The present application relates to compressed air drying and intelligent control technical field, specifically to a kind of zero gas consumption compressed air adsorption drying intelligent control method and system, comprising: obtaining the real-time working condition data of adsorption tower, real-time working condition data is input into pre-constructed load prediction model, based on load prediction model, real-time working condition data is extracted in time sequence characteristics, obtain the regeneration load demand curve in future preset period;Obtain the regeneration energy consumption function matched with regeneration load demand curve, and with the zero point of regeneration energy consumption function as target, the working parameters of drying machine are optimized in reverse.This application constructs load prediction model, extracts the real-time working condition data of adsorption tower in time sequence characteristics, generates the regeneration load demand curve in future preset period, so as to predict the change trend of drying agent adsorption load in advance, fundamentally avoids the problem that drying agent is excessively saturated or regenerated in time due to response lag.

Description

Technical Field

[0001] This invention relates to the field of compressed air drying and intelligent control technology, specifically to a zero-air-consumption compressed air adsorption drying intelligent control method and system. Background Technology

[0002] Compressed air adsorption drying is a core component of industrial air source treatment. Its basic principle is to selectively adsorb water vapor in compressed air using an adsorbent to obtain dry compressed air that meets process requirements. A typical dual-tower adsorption dryer adopts an alternating working mode of "one tower adsorption and one tower regeneration". The energy consumption of the regeneration process accounts for the majority of the total energy consumption of the drying system. Driven by the dual-carbon strategy, how to minimize or even eliminate the compressed air consumption of the regeneration process and achieve so-called zero-air-consumption drying has become a technical bottleneck that urgently needs to be overcome in the field of compressed air post-treatment.

[0003] Currently, the control methods for compressed air adsorption drying are mainly divided into two categories. The first category is a fixed-sequence switching control method. This method switches between adsorption and regeneration of the two towers according to a preset fixed cycle, without sensing the real-time adsorption status of the desiccant. To ensure that the outlet dew point meets the standard under the worst operating conditions, an extremely conservative switching cycle is usually adopted, resulting in excessively high regeneration frequency and serious waste of compressed air. The second category is a load adaptation control method based on outlet dew point feedback. This method dynamically adjusts the switching cycle or regeneration gas volume by monitoring the outlet dew point value. Although it can partially match load changes, it is essentially a lag response mode, that is, it passively adjusts only when the desiccant is close to saturation and the outlet dew point is about to rise, and cannot fundamentally eliminate the energy loss caused by excessive regeneration. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a zero-air-consumption compressed air adsorption drying intelligent control method, comprising: The real-time operating data of the adsorption tower is obtained, and the real-time operating data is input into the pre-constructed load prediction model. Based on the load prediction model, the time-series features of the real-time operating data are extracted to obtain the regeneration load demand curve for the future preset period. Obtain the regeneration energy consumption function that matches the regeneration load demand curve, and use the zero point of the regeneration energy consumption function as the target to perform reverse optimization on the working parameters of the dryer to generate the first working parameters; Acquire real-time exhaust data from the compressor and construct a usable energy flow spectrum of compressed air based on the real-time exhaust data; The first operating parameters and the available energy flow spectrum are input into a pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and iteratively corrects the first operating parameters based on the energy fit to obtain the second operating parameters. Based on the second operating parameter, the total flow rate of compressed air is divided into a regeneration flow path and a drying flow path, and the flow rate and heating power of the regeneration flow path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second operating parameter.

[0005] Preferably, the real-time operating data includes the current inlet temperature, inlet pressure, inlet humidity, and outlet dew point of the adsorption tower; the load prediction model includes a time-series coding network and a load decoding network; Based on the load forecasting model, time-series features are extracted from real-time operating data to obtain the regenerative load demand curve for a future preset time period, including: Real-time operating condition data is input into a time-series coding network to extract the hidden state feature vectors of the real-time operating condition data at different time scales. The latent state feature vector of the operating condition is input into the load decoding network. The load decoding network, based on the attention mechanism, predicts the regeneration load demand at each time step within the future preset period and generates the regeneration load demand curve. The regeneration load demand curve is a function curve representing the change of the theoretical minimum heat required to restore the adsorbent to the target regeneration degree over time within the future preset period.

[0006] Preferably, the temporal coding network consists of multiple parallel one-dimensional convolutional layers and a bidirectional long short-term memory network layer; Real-time operating condition data is input into a time-series coding network to extract latent state feature vectors of the real-time operating condition data at different time scales, including: By using multiple one-dimensional convolutional layers with different kernel sizes, short-term fluctuation features and long-term trend features of real-time operating data are captured in parallel to obtain multi-scale feature maps. After concatenating the multi-scale feature maps along the feature dimensions, they are input into a bidirectional long short-term memory network layer to extract temporal correlation features containing forward and backward dependencies, which are used as the latent state feature vectors of the working conditions.

[0007] Preferably, the load decoding network includes an attention weight generation unit, a load state decoding unit, and a curve fitting output unit; The latent state feature vector of the operating condition is input into the load decoding network. Based on an attention mechanism, the load decoding network predicts the regenerative load demand at each time step within a preset future period, generating a regenerative load demand curve, including: The attention weight generation unit calculates the attention weight of each time step state to the current prediction time based on the hidden state and the hidden state feature vector of the current decoding time step. The context vector of the current decoding time step is obtained by weighted summation of the hidden state feature vectors of the working condition using attention weights. The context vector is concatenated with the load prediction value of the previous decoding time step and input into the load state decoding unit to obtain the load hidden state of the current decoding time step. The hidden load state is input to the curve fitting output unit, and the regenerated load prediction value of the current decoding time step is calculated. The above steps are executed iteratively to generate the regenerative load forecast values ​​for all time steps within the preset future period. All the regenerative load forecast values ​​are then combined according to the time series to obtain the regenerative load demand curve.

[0008] Preferably, the available energy flow spectrum of compressed air is constructed based on real-time exhaust data, including: A two-dimensional energy distribution matrix is ​​constructed using exhaust temperature from real-time exhaust data as the first dimension and exhaust mass flow rate as the second dimension. The available energy conversion of the exhaust pressure corresponding to each element in the two-dimensional energy distribution matrix is ​​calculated to obtain the available heat potential of each element. Based on the available heat potential of each element, a usable energy flow spectrum is generated with temperature and flow rate as coordinates and available heat potential as the intensity.

[0009] Preferably, the dynamic matching model includes a fit calculation layer and a parameter correction layer; The first operating parameters and the available energy flow spectrum are input into a pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and iteratively corrects the first operating parameters based on the energy fit to obtain the second operating parameters, including: The target regeneration temperature and regeneration flow rate in the first working parameters are mapped to the coordinate space of the available energy flow spectrum to obtain an energy demand point; The energy fit calculation layer calculates the distance loss between the energy demand point and the high heat available potential area in the available energy flow spectrum, and outputs the energy fit based on the distance loss; When the energy fit is lower than the preset fit threshold, the parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first working parameter along the energy gradient direction of the available energy flow spectrum until the energy fit of the adjusted energy demand point reaches the preset fit threshold, and then outputs the adjusted working parameter as the second working parameter.

[0010] Preferably, the parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first operating parameters along the energy gradient direction of the available energy flow spectrum, including: Keeping the total amount of regeneration constant, and using the regeneration load demand curve as the constraint boundary, search for the target point with the smallest difference between the available heat potential value and the current energy demand point in the available energy flow spectrum. The temperature and flow rates corresponding to the target point are used as the adjusted target regeneration temperature and regeneration flow rate.

[0011] Preferably, based on the second operating parameter, the total flow rate of compressed air is divided into a regeneration flow path and a drying flow path, and the flow rate and heating power of the regeneration flow path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second operating parameter, including: Adjust the opening of the flow control valve in the regeneration flow path according to the target regeneration flow rate in the second working parameter; Adjust the heater power of the regeneration flow path according to the target regeneration temperature in the second working parameter; Obtain the real-time dew point temperature and real-time exhaust flow rate at the regenerated exhaust outlet, and calculate the deviation between the real-time dew point temperature and the corresponding point on the target dew point curve. Using the deviation value as feedback, the heater power is finely adjusted so that the regenerated exhaust dew point temperature converges to the target dew point curve.

[0012] A zero-air-consumption compressed air adsorption drying intelligent control system, applicable to the aforementioned zero-air-consumption compressed air adsorption drying intelligent control method, comprising: The load forecasting module is used to acquire real-time operating data of the adsorption tower, input the real-time operating data into the pre-built load forecasting model, extract time-series features from the real-time operating data based on the load forecasting model, and obtain the regeneration load demand curve for the future preset period. The reverse optimization module is used to obtain the regeneration energy consumption function that matches the regeneration load demand curve, and to perform reverse optimization on the working parameters of the dryer with the zero point of the regeneration energy consumption function as the target, so as to generate the first working parameters. The energy flow spectrum construction module is used to acquire real-time exhaust data of the compressor and construct an available energy flow spectrum of compressed air based on the real-time exhaust data. The dynamic matching module is used to input the first operating parameters and the available energy flow spectrum into the pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and performs iterative correction on the first operating parameters based on the energy fit to obtain the second operating parameters. The closed-loop execution module is used to divide the total flow of compressed air into a regeneration flow path and a drying flow path based on the second operating parameter, and to adjust the flow rate and heating power of the regeneration flow path in real time so that the regeneration exhaust dew point temperature of the dryer is maintained on the target dew point curve corresponding to the second operating parameter.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention constructs a load prediction model to extract time-series features from the real-time operating data of the adsorption tower, generating a regeneration load demand curve for a future preset time period. This allows for the prediction of the changing trend of the desiccant adsorption load in advance, fundamentally avoiding the problems of desiccant oversaturation or untimely regeneration caused by response lag, and improving the operational reliability of the drying system. Furthermore, it obtains a regeneration energy consumption function that matches the regeneration load demand curve, and uses the zero point of the regeneration energy consumption function as the direct target to perform reverse optimization of the dryer's operating parameters, transforming zero gas consumption into an explicit control target that can be mathematically solved, thereby optimizing the dryer's operating parameters towards the theoretical minimum energy consumption. This invention constructs an available energy flow spectrum of compressed air, abstracting the multi-dimensional parameters such as temperature, flow rate, and pressure carried by the compressor exhaust into a dynamically changing energy field. By using a dynamic matching model to calculate the energy fit between the regeneration operating parameters and this energy flow spectrum, the operating parameters are cyclically corrected. The real-time available waste heat of the upstream compressor is formally incorporated into the optimization decision-making framework for the regeneration energy supply of the dryer, realizing a leap from independent control of a single machine to coordinated optimization of system energy. Waste heat is used to replace high-quality compressed air for regeneration, reducing the overall energy consumption of the system. Attached Figure Description

[0014] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0015] In the diagram: 1. Load forecasting module; 2. Reverse optimization module; 3. Energy flow spectrum construction module; 4. Dynamic matching module; 5. Closed-loop execution module. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, please refer to Figure 1 This invention provides a technical solution: a zero-air-consumption compressed air adsorption drying intelligent control method, comprising: S1. Obtain real-time operating data of the adsorption tower, input the real-time operating data into the pre-built load prediction model, extract the time-series features of the real-time operating data based on the load prediction model, and obtain the regeneration load demand curve within the future preset time period. S2. Obtain the regeneration energy consumption function that matches the regeneration load demand curve, and use the zero point of the regeneration energy consumption function as the target to perform reverse optimization on the working parameters of the dryer to generate the first working parameters. S3. Obtain the real-time exhaust data of the compressor and construct the available energy flow spectrum of compressed air based on the real-time exhaust data; S4. Input the first operating parameters and the available energy flow spectrum into the pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum. Based on the energy fit, the first operating parameters are iteratively corrected to obtain the second operating parameters. S5. Based on the second working parameter, the total flow rate of compressed air is divided into a regeneration flow path and a drying flow path, and the flow rate and heating power of the regeneration flow path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second working parameter.

[0018] It should be noted that real-time operating data refers to the operating parameters collected in real time from temperature sensors, pressure sensors, and dew point sensors deployed on the adsorption tower. Specifically, this includes the current inlet temperature, inlet pressure, inlet humidity, and outlet dew point value of the adsorption tower. The load prediction model is a pre-trained deep learning network. Its training process involves: collecting historical operating data, using the above four operating parameters within a past time window as input samples, and using the actual heat required for adsorbent regeneration to the target dryness level in the corresponding future time period as labels, and training it through supervised learning. The temporal coding network within the model uses multiple one-dimensional convolutional kernels in parallel. Features at different time scales are extracted and then captured by a bidirectional long short-term memory network to generate a hidden state feature vector of the operating condition. The load decoding network then uses an attention mechanism to weight and decode this hidden state vector, predicting the regeneration load demand at each future time step point by point. Finally, they are combined into a regeneration load demand curve with time as the horizontal axis and the theoretical minimum regeneration heat supply as the vertical axis. For example, when the inlet humidity suddenly rises from 30%RH to 70%RH, the model will predict that the regeneration load demand curve will show a trend of first rising sharply and then falling slowly in the next 15 minutes. The heat supply corresponding to its peak is the maximum heat energy required to maintain the target regeneration degree of the adsorbent. The regeneration energy consumption function is a function that characterizes the mathematical relationship between the total energy consumption of the regeneration process and the operating parameters. Its independent variables include regeneration temperature, regeneration flow rate, and regeneration time, while the dependent variable is the energy value converted from compressed air consumed in the regeneration process. This function can be pre-established based on the thermodynamic model of the dryer and the characteristic parameters of the adsorbent, such as a desorption heat calculation model based on the adsorption isotherm. Backward optimization means that given the target value of the function as zero, it solves in reverse the problem to find the set of operating parameters that makes the energy consumption zero. In actual solutions, since absolute zero air consumption is physically constrained by factors such as ambient temperature, the "zero point" can be set as the minimum value of the energy consumption function or a feasible region below a preset threshold. For example, when the regeneration load demand curve shows that the required desorption heat in a certain period is 500 kJ and the ambient temperature is 25°C, the backward optimization algorithm may solve for: setting the regeneration temperature to 120°C, setting the regeneration flow rate to 8% of the total flow rate, and setting the regeneration time to 30 minutes. This set of parameters is the first set of operating parameters, and its corresponding regeneration energy consumption is close to the theoretical minimum. The compressor's real-time exhaust data includes exhaust temperature, exhaust mass flow rate, and exhaust pressure. This data is collected in real-time by temperature transmitters, mass flow meters, and pressure transmitters. The specific method for constructing the available energy flow spectrum is as follows: a two-dimensional grid matrix is ​​established with exhaust temperature as the first dimension and exhaust mass flow rate as the second dimension. For each grid point in the matrix, its available energy—that is, the maximum heat that the compressed air can release when passing through the heat exchanger—is calculated based on the corresponding temperature and flow rate values, combined with the exhaust pressure. This is used as the available heat potential value for that point. The available heat potential values ​​of all grid points are presented in the form of color gradients or numerical intensity, thus generating the available energy flow spectrum. For example, when the compressor is operating under rated conditions, the available heat potential value is highest in the region with an exhaust temperature of approximately 80-100℃ and a mass flow rate of approximately 500-800 kg / h, which is represented as a bright area in the spectrum, indicating that this region is the best energy source for regenerative heating. During low load or start-up / shutdown phases, the high-energy flow region shrinks and shifts towards lower temperature and lower flow rate; the spectrum reflects this dynamic change in real-time. The dynamic matching model maps the target regeneration temperature and regeneration flow rate in the first operating parameters to an energy demand point in the energy flow spectrum coordinate space, and calculates the distance loss between this point and the high energy flow region in the spectrum, thus obtaining the energy fit. When the fit is lower than a preset threshold, it indicates that the energy demand corresponding to the currently optimized operating parameters cannot be fully met by the waste heat from the compressor exhaust, and correction is required. The parameter correction layer iteratively adjusts the regeneration temperature and regeneration flow rate along the energy gradient direction of the spectrum. During the adjustment process, the regeneration load demand curve is used as a hard constraint boundary to ensure that the corrected parameters still meet the minimum heat supply required for regeneration. For example, the first operating parameters require a regeneration temperature of 130℃ and a regeneration flow rate of 10%, but the current compressor exhaust temperature is only 90℃. At this time, the energy demand point is far away from the high energy flow region in the spectrum, and the fit is only 0.3. Under the premise of keeping the total regeneration constant, the dynamic matching model lowers the regeneration temperature to 100℃ and increases the regeneration flow rate to 12%, so that the energy demand point moves into the high energy flow region, and the fit increases to 0.85. The adjusted parameters are the second operating parameters. Based on the second operating parameter, the control system adjusts the flow distribution valve of the compressed air main pipeline to the target opening, so that part of the compressed air enters the regeneration flow path as regeneration gas; at the same time, it adjusts the power of the regeneration heater so that the regeneration gas is heated to the target regeneration temperature. During execution, the dew point temperature and exhaust flow rate at the regenerated exhaust outlet are collected in real time and compared with the dew point value at the corresponding point on the target dew point curve to calculate the deviation. When the actual dew point temperature is higher than the target value, it indicates that the regeneration is insufficient, and the system fine-tunes by increasing the heater power. When the actual dew point temperature is lower than the target value, it indicates that the regeneration is sufficient and there is a tendency for overheating, and the system fine-tunes by decreasing the heater power. Through this closed-loop feedback mechanism, the regenerated exhaust dew point temperature always converges to the target dew point curve. For example, the second working parameter requires a target regeneration flow rate of 100 kg / h and a target regeneration temperature of 110°C. The corresponding target dew point curve specifies that the exhaust dew point should be -20°C when the regeneration has progressed to the 20th minute. If the actual exhaust dew point detected at the 20th minute is -15°C, the heater power is increased by 5% to gradually bring the dew point closer to -20°C, thus achieving precise control of the regeneration process.

[0019] In one optional embodiment, the real-time operating data includes the current inlet temperature, inlet pressure, inlet humidity, and outlet dew point of the adsorption tower; the load prediction model includes a time-series coding network and a load decoding network; Based on the load forecasting model, time-series features are extracted from real-time operating data to obtain the regenerative load demand curve for a future preset time period, including: Real-time operating condition data is input into a time-series coding network to extract the hidden state feature vectors of the real-time operating condition data at different time scales. The latent state feature vector of the operating condition is input into the load decoding network. The load decoding network, based on the attention mechanism, predicts the regeneration load demand at each time step within the future preset period and generates the regeneration load demand curve. The regeneration load demand curve is a function curve representing the change of the theoretical minimum heat required to restore the adsorbent to the target regeneration degree over time within the future preset period.

[0020] It should be noted that the four parameters of the real-time operating data are collected in real time by industrial sensors deployed at corresponding locations on the adsorption tower: the inlet temperature is obtained by the resistance temperature transmitter at the inlet pipe, the inlet pressure is obtained by the pressure transmitter at the inlet pipe, the inlet humidity is obtained by the humidity sensor at the inlet, and the outlet dew point value is obtained by the dew point meter at the outlet pipe. The pre-construction process of the load forecasting model is as follows: Historical data of continuous operation of the dryer under different seasons and load conditions are collected, and training samples are constructed in a sliding time window manner. Each sample takes the sequence of parameters of the above four operating conditions over the past N time steps as input and the measured value of regenerated heat supply over the next M time steps as label. The parameters of the time-series coding network and the load decoding network are jointly optimized and trained by minimizing the mean square error loss function between the predicted value and the label value. After training, the model will have the ability to predict future regenerated load demand based on the real-time operating condition sequence. The theoretical minimum heat in the regeneration load demand curve refers to the minimum heat input required to reduce the current adsorption amount to the target residual moisture content within a given time period, calculated based on the adsorption isotherm equation and desorption thermodynamic parameters of the adsorbent. This curve provides a clear energy consumption boundary constraint for subsequent reverse optimization steps. For example, when the operating conditions of an inlet temperature of 25℃, an inlet pressure of 0.7MPa, an inlet humidity of 70%RH, and an outlet dew point of -30℃ are input into the model, the load prediction model outputs the regeneration load demand curve for the next 15 minutes. The heat value gradually increases from 150kJ to 350kJ, indicating that the heat load required for regeneration in the future period is increasing.

[0021] In an alternative embodiment, the temporal coding network consists of multiple parallel one-dimensional convolutional layers and a bidirectional long short-term memory network layer; Real-time operating condition data is input into a time-series coding network to extract latent state feature vectors of the real-time operating condition data at different time scales, including: By using multiple one-dimensional convolutional layers with different kernel sizes, short-term fluctuation features and long-term trend features of real-time operating data are captured in parallel to obtain multi-scale feature maps. After concatenating the multi-scale feature maps along the feature dimensions, they are input into a bidirectional long short-term memory network layer to extract temporal correlation features containing forward and backward dependencies, which are used as the latent state feature vectors of the working conditions.

[0022] It should be noted that the kernel size of multiple one-dimensional convolutional layers in the temporal coding network is set according to the temporal response characteristics of the operating parameters: small kernels (e.g., kernel size 3) slide along the time dimension, covering only a short time window, which can keenly capture short-term changes such as sudden increases in inlet humidity or instantaneous fluctuations in outlet dew point; large kernels (e.g., kernel size 15) cover a longer time window, which can extract long-term trends such as diurnal variations or seasonal drift in environmental temperature and humidity. After multiple convolutional kernels are computed in parallel, they each output feature maps of the same length but containing information at different scales; the feature concatenation operation concatenates the feature maps of each scale along the feature channel dimension, so that the concatenated feature matrix carries information from multiple time scales simultaneously. The forward LSTM units of the bidirectional long short-term memory network process the concatenated feature sequence in the forward direction of time, while the backward LSTM units process the same sequence in the reverse direction of time. The hidden states in the two directions are concatenated or summed to obtain the hidden state feature vector of each time step. The physical meaning of this feature vector is: the compressed representation of each time point in the input operating condition sequence after multi-scale extraction and temporal dependency modeling, which can reflect the comprehensive change law of the current operating state of the adsorption tower under multiple time scales. For example, in the operating condition sequence within 5 minutes after the adsorption tower switch, the small convolutional kernel extracts the instantaneous jump feature of the outlet dew point value, and the large convolutional kernel extracts the smooth transition feature of the inlet temperature. The bidirectional LSTM further captures the correlation between the states before and after the switch, and the finally encoded hidden state feature vector of the operating condition simultaneously contains state information at two levels: "the switch has been completed" and "the new tower is stabilizing".

[0023] In an optional embodiment, the load decoding network includes an attention weight generation unit, a load state decoding unit, and a curve fitting output unit; The latent state feature vector of the operating condition is input into the load decoding network. Based on an attention mechanism, the load decoding network predicts the regenerative load demand at each time step within a preset future period, generating a regenerative load demand curve, including: The attention weight generation unit calculates the attention weight of each time step state to the current prediction time based on the hidden state and the hidden state feature vector of the current decoding time step. The context vector of the current decoding time step is obtained by weighted summation of the hidden state feature vectors of the working condition using attention weights. The context vector is concatenated with the load prediction value of the previous decoding time step and input into the load state decoding unit to obtain the load hidden state of the current decoding time step. The hidden load state is input to the curve fitting output unit, and the regenerated load prediction value of the current decoding time step is calculated. The above steps are executed iteratively to generate the regenerative load forecast values ​​for all time steps within the preset future period. All the regenerative load forecast values ​​are then combined according to the time series to obtain the regenerative load demand curve.

[0024] It should be noted that the load decoding network adopts a sequence-to-sequence decoding architecture based on an attention mechanism. Its autoregressive decoding process is as follows: Before decoding begins, the feature vector of the hidden state of the load condition is cached as the key and value matrix of the attention mechanism; decoding starts from the first future time step, with a preset start symbol or zero value as the initial input; the attention weight generation unit calculates the dot product similarity between the current decoded hidden state (query vector) and the hidden state (key vector) of each time step in the feature vector of the hidden state of the load condition, and obtains the attention weight distribution after normalization by the Softmax function. This weight distribution reflects the different contributions of each time step in the historical load condition sequence to the current prediction time; the value vector of each time step is weighted and summed using this attention weight to obtain the context vector of the current decoding time step. This vector is essentially a dynamic aggregation representation of historical key information. The load state decoding unit concatenates the context vector with the load prediction value from the previous decoding time step, and then updates the current decoded hidden state through a gated recurrent unit or a long short-term memory unit. The curve fitting output unit consists of a fully connected layer, which maps the current decoded hidden state to a scalar output, namely the regenerative load prediction value for the current time step. The above decoding process is iteratively executed until the entire future preset time period is covered. Finally, the prediction values ​​of all time steps are combined in chronological order to obtain a complete regenerative load demand curve. For example, when decoding the regenerative load demand for the 5th minute in the future, the attention mechanism may automatically assign a higher weight to the hidden states of the operating conditions from the 3rd to the 8th minute in the past, because the changes in adsorption load during this period are most predictive of the heat required for the 5th minute. The information of this period is aggregated and input together with the prediction value of the 4th minute into the decoding unit to obtain the prediction value for the 5th minute, and so on to generate a complete curve.

[0025] In an optional embodiment, constructing an available energy flow spectrum of compressed air based on real-time exhaust data includes: A two-dimensional energy distribution matrix is ​​constructed using exhaust temperature from real-time exhaust data as the first dimension and exhaust mass flow rate as the second dimension. The available energy conversion of the exhaust pressure corresponding to each element in the two-dimensional energy distribution matrix is ​​calculated to obtain the available heat potential of each element. Based on the available heat potential of each element, a usable energy flow spectrum is generated with temperature and flow rate as coordinates and available heat potential as the intensity.

[0026] It should be noted that real-time exhaust data is collected in real time by industrial sensors installed on the compressor exhaust pipeline: exhaust temperature is obtained by a thermal resistance temperature transmitter at the exhaust pipeline, exhaust mass flow rate is obtained by a mass flow meter at the exhaust pipeline, and exhaust pressure is obtained by a pressure transmitter at the exhaust pipeline; the construction of the available energy flow spectrum is based on the concept of available energy in thermodynamics. During the compression process, compressed air not only increases in pressure but also in temperature. The large amount of compression heat carried in the exhaust is a high-grade waste heat resource. However, this waste heat does not have the same utilization value under all operating conditions. The actual heat that can be used for adsorbent regeneration depends on the exhaust temperature level, mass flow rate, and pressure conditions. To quantitatively describe the availability of this dynamic change, this embodiment introduces the indicator of available heat potential. The construction process consists of three steps. The first step is to construct a two-dimensional energy distribution matrix using the exhaust temperature from real-time exhaust data as the first dimension and the exhaust mass flow rate as the second dimension. The horizontal axis of this matrix corresponds to the exhaust temperature, which typically ranges from ambient temperature to the compressor's rated exhaust temperature (e.g., 20℃ to 120℃), and is divided into grids at intervals of 1℃ or 2℃. The vertical axis corresponds to the exhaust mass flow rate, which ranges from the compressor's minimum flow rate to its rated flow rate (e.g., 100kg / h to 1000kg / h), and is divided into grids at intervals of 10kg / h or 20kg / h. Each grid point in the matrix represents a specific combination of temperature and flow rate, and its initial value is empty. The second step involves calculating the available energy conversion for each grid point in the matrix. For any given grid point, firstly, based on the corresponding temperature and flow rate, the maximum heat that the compressed air can release under ideal heat exchange conditions is calculated, i.e., the sensible heat released when the current temperature drops to the ambient temperature. Then, a pressure correction coefficient is introduced to incorporate the impact of exhaust pressure on the quality of this heat transfer—the higher the pressure, the greater the gas density, the more concentrated the heat carried per unit mass of gas, and the higher the heat exchange efficiency, thus resulting in a larger available potential value. The specific calculation formula is: Available heat potential value = mass flow rate × specific heat capacity at constant pressure × (exhaust temperature - ambient temperature) × pressure correction coefficient, where the pressure correction coefficient is a logarithmic function of the ratio of the actual exhaust pressure to the standard atmospheric pressure, characterizing the enhancing effect of pressure on the heat exchange driving force. After this calculation, each grid point in the matrix is ​​assigned a specific available heat potential value. The third step is to present the two-dimensional matrix containing the available heat potential value in the form of a color scale or contour map to generate an available energy flow spectrum. The horizontal axis of the spectrum is the exhaust temperature, and the vertical axis is the exhaust mass flow rate. The color depth or gray level of each coordinate point in the graph corresponds to the available heat potential value at that point. The bright area indicates that under the temperature and flow conditions, the exhaust has the richest extractable waste heat and is the most favorable energy source for regeneration heating; the dark area indicates that the waste heat utilization value of the exhaust is relatively low. For example, when the compressor is running stably under rated conditions, with an exhaust temperature of 95℃, an exhaust pressure of 0.8MPa, and a mass flow rate of 700kg / h, the corresponding coordinate point in the energy flow spectrum is (95℃, 700kg / h), and its usable heat potential is calculated to be 4500kJ / h, located in the bright core region of the spectrum. When the compressor is running under partial load, the exhaust temperature drops to 70℃, the mass flow rate drops to 350kg / h, and the usable heat potential at the corresponding coordinate point drops to 1200kJ / h, located in the medium brightness region of the spectrum. When the compressor has just started and is in the warm-up phase, the exhaust temperature is only 40℃, the mass flow rate is 200kg / h, and the usable heat potential at the corresponding coordinate point is only 150kJ / h, located in the dark edge region of the spectrum. The energy flow spectrum is updated in real time as the compressor's operating state changes, dynamically reflecting the distribution of the available exhaust waste heat at the current moment, providing complete energy supply information for subsequent dynamic matching steps.

[0027] In an optional embodiment, the dynamic matching model includes a fit calculation layer and a parameter correction layer; The first operating parameters and the available energy flow spectrum are input into a pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and iteratively corrects the first operating parameters based on the energy fit to obtain the second operating parameters, including: The target regeneration temperature and regeneration flow rate in the first working parameters are mapped to the coordinate space of the available energy flow spectrum to obtain an energy demand point; The energy fit calculation layer calculates the distance loss between the energy demand point and the high heat available potential area in the available energy flow spectrum, and outputs the energy fit based on the distance loss; When the energy fit is lower than the preset fit threshold, the parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first working parameter along the energy gradient direction of the available energy flow spectrum until the energy fit of the adjusted energy demand point reaches the preset fit threshold, and then outputs the adjusted working parameter as the second working parameter.

[0028] It should be noted that the dynamic matching model is a pre-built rule-algorithm hybrid model. The fit degree calculation layer is built based on the preset distance loss calculation formula and energy fit degree mapping rule, and the parameter correction layer is built based on the preset energy gradient search strategy and constraint conditions. After the model is built, it can be put into use directly without offline training. The core problem that the dynamic matching model solves is whether the optimal working parameters obtained by back-optimization can be achieved under the current compressor exhaust conditions. In actual industrial scenarios, the compressor exhaust state is dynamically changing. The target regeneration temperature and regeneration flow rate obtained by back-optimization may exceed the current exhaust waste heat supply capacity. If it is executed directly without matching verification, the actual regeneration effect will deviate significantly from the theoretical expectation. The working mechanism of the matching degree calculation layer is as follows: First, the target regeneration temperature in the first working parameter is used as the horizontal axis value and the target regeneration flow rate is used as the vertical axis value to locate a unique energy demand point in the two-dimensional coordinate space of the available energy flow spectrum. Second, the distance loss between this energy demand point and the high heat available potential value area in the available energy flow spectrum is calculated. The specific calculation method of the distance loss is as follows: search for the top K grid points with the highest heat available potential value in the energy flow spectrum (K is 3 to 10), calculate the Euclidean distance from the energy demand point to each high energy flow grid point, and take the minimum value as the distance loss; or calculate the ratio of the heat available potential value of the energy demand point to the maximum heat available potential value of the entire spectrum, and subtract this ratio from 1 as the distance loss. The energy matching degree is negatively correlated with the distance loss. The smaller the distance loss, the closer the energy conditions required by the first working parameter are to the supply conditions of the current compressor exhaust waste heat, and the higher the matching degree. The preset matching threshold is usually set to a value between 0.7 and 0.85. When the energy fit is lower than the preset fit threshold, it indicates that the first operating parameter cannot be fully supported by the current exhaust waste heat and must be corrected. The parameter correction layer is iteratively adjusted along the energy gradient direction of the energy flow spectrum. The energy gradient direction refers to the direction in which the available heat potential value in the energy flow spectrum increases the fastest, that is, the direction from the low energy flow region to the high energy flow region. During the iterative adjustment process, the regeneration load demand curve is used as a hard constraint boundary to ensure that the total heat supply corresponding to the adjusted parameters is not lower than the theoretical minimum heat required for regeneration, thereby ensuring the sufficiency of regeneration. At the same time, the total regeneration amount remains unchanged, and only the ratio adjustment between the two variables of temperature and flow rate is carried out. For example, the first working parameters output by the reverse optimization are: target regeneration temperature 130℃ and target regeneration flow rate 100kg / h. After mapping these parameters to the current available energy flow spectrum, the coordinates of the energy demand point are (130℃, 100kg / h). This point is located in the low energy flow region of the spectrum, with a usable heat potential of 800kJ / h, while the maximum usable heat potential of the spectrum is 5000kJ / h. The energy fit is only 0.16, far below the preset fit threshold of 0.75. At this time, the parameter correction layer proceeds along the energy gradient direction of the spectrum. The search is performed, and under the condition that the total regeneration volume meets the regeneration load demand curve, the target regeneration temperature is gradually lowered to 105℃ and the target regeneration flow rate is gradually increased to 140kg / h. The coordinates of the adjusted energy demand point are (105℃, 140kg / h). This point enters the high energy flow region of the spectrum, the available heat potential is 4200kJ / h, and the energy fit is improved to 0.84, reaching the preset fit threshold. The adjusted parameters are the second working parameters, which not only meet the regeneration heat demand, but also make full use of the compressor exhaust waste heat.

[0029] In an optional embodiment, the parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first operating parameters along the energy gradient direction of the available energy flow spectrum, including: Keeping the total amount of regeneration constant, and using the regeneration load demand curve as the constraint boundary, search for the target point with the smallest difference between the available heat potential value and the current energy demand point in the available energy flow spectrum. The temperature and flow rates corresponding to the target point are used as the adjusted target regeneration temperature and regeneration flow rate.

[0030] It should be noted that keeping the total regeneration amount constant means that the total regeneration heat supply obtained by multiplying the product of the target regeneration temperature and the target regeneration flow rate by the regeneration time before and after adjustment remains constant. Since the regeneration time is usually determined by the time domain span of the regeneration load demand curve and is relatively fixed, keeping the total regeneration amount constant is equivalent to keeping the product of the target regeneration temperature and the target regeneration flow rate constant. This constraint ensures that the total regeneration heat supply will not be reduced due to parameter adjustment during the correction process, thereby ensuring that the adsorbent regeneration effect is not affected. Using the regenerative load demand curve as a constraint boundary means that the adjusted regenerative heat supply must not be lower than the cumulative heat integral value of the regenerative load demand curve during that period. The regenerative load demand curve defines the minimum heat threshold required for regeneration. During the parameter correction process, this curve must be used as the lower bound constraint. Any parameter combination that causes the heat supply to be lower than this lower bound will be excluded. Under these dual constraints, the parameter correction layer searches for the target point with the smallest difference between the available heat potential value and the current energy demand point along the energy gradient direction in the energy flow spectrum. The specific search method is as follows: starting from the current energy demand point, the layer moves step-by-step along the energy gradient direction, generating a candidate point for each step. The available heat potential value of this candidate point is calculated until a candidate point that satisfies the constraints and has the highest available heat potential value is found, which is then designated as the adjustment target point. The search step size is set to 2℃ to 5℃ for temperature and 5kg / h to 10kg / h for flow rate to balance search accuracy and iteration efficiency. The iteration termination condition is when the energy fit of the candidate point reaches a preset fit threshold, or the number of iterations reaches a preset upper limit (e.g., 50 times), at which point the candidate point with the highest available heat potential value found is selected as the adjustment target point. For example, the current energy demand point is located at (130℃, 100kg / h), and its available heat potential is 800kJ / h. The parameter correction layer searches along the energy gradient direction, and calculates the available heat potential of candidate points point by point while maintaining the product of temperature and flow rate at 13000℃·kg / h. The available heat potential of candidate point (120℃, 108kg / h) is 1500kJ / h; the available heat potential of candidate point (110℃, 118kg / h) is 2800kJ / h; and the available heat potential of candidate point (105℃, 124kg / h) is 4100kJ / h. Among these, the available heat potential of candidate point (105℃, 124kg / h) has the largest difference (3300kJ / h) from the available heat potential of the current energy demand point, and it also satisfies the constraint of the regeneration load demand curve. Therefore, it is determined as the adjustment target point, and the corresponding temperature and flow rate values ​​are the adjusted target regeneration temperature and regeneration flow rate.

[0031] In an optional embodiment, based on the second operating parameter, the total flow rate of compressed air is divided into a regeneration flow path and a drying flow path, and the flow rate and heating power of the regeneration flow path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second operating parameter, including: Adjust the opening of the flow control valve in the regeneration flow path according to the target regeneration flow rate in the second working parameter; Adjust the heater power of the regeneration flow path according to the target regeneration temperature in the second working parameter; Obtain the real-time dew point temperature and real-time exhaust flow rate at the regenerated exhaust outlet, and calculate the deviation between the real-time dew point temperature and the corresponding point on the target dew point curve. Using the deviation value as feedback, the heater power is finely adjusted so that the regenerated exhaust dew point temperature converges to the target dew point curve.

[0032] It should be noted that the method of dividing the total flow of compressed air into regeneration and drying paths is as follows: a three-way proportional regulating valve or two parallel flow regulating valve groups are installed downstream of the main compressed air pipeline. The control system calculates the valve opening based on the target regeneration flow rate in the second working parameter, and diverts the corresponding proportion of compressed air to the regeneration path, while the remaining part enters the drying path. The compressed air in the regeneration path is first heated to the target regeneration temperature by the regeneration heater, and then enters the adsorption tower in the regeneration state to heat and desorb the adsorbent for regeneration. The target dew point curve is a pre-acquired function curve. Its horizontal axis represents the regeneration process time, and its vertical axis represents the target dew point temperature that the regenerated exhaust gas should reach at that time point. This curve describes the standard evolution trajectory of the exhaust gas dew point temperature over time during an ideal regeneration process—in the initial stage of regeneration, the adsorbent moisture content is high, and the exhaust gas dew point temperature is also high; as regeneration progresses, the adsorbent moisture content gradually decreases, and the exhaust gas dew point temperature should gradually decrease to the target value. The target dew point curve is obtained in the following ways: under stable dryer operation conditions, regeneration experiments are conducted with different preset combinations of regeneration temperature and regeneration flow rate. The regenerated exhaust gas dew point temperature at each time point is recorded, and the average value of multiple experiments is used to fit a standard curve; or based on the adsorbent desorption kinetic equation and the bed heat and mass transfer equation, a functional relationship between the dew point temperature and time during the regeneration process is established, and the curve parameters are determined after verification with experimental data. During execution, a dew point sensor and a flow sensor are installed at the regeneration exhaust outlet to collect the dew point temperature and exhaust flow rate of the regeneration exhaust in real time. The control system compares the real-time dew point temperature with the dew point value at the corresponding time point on the target dew point curve and calculates the deviation value. When the deviation value is positive (actual dew point temperature is higher than the target value), it indicates that the current regeneration heat is insufficient and the adsorbent desorption is incomplete. The control system then fine-tunes and increases the heater power according to the proportional-integral control algorithm. When the deviation value is negative (actual dew point temperature is lower than the target value), it indicates that the current regeneration heat is sufficient or even has a tendency to overheat. The control system then fine-tunes and reduces the heater power to avoid energy waste. The proportional coefficient of the proportional-integral control algorithm ranges from 0.5 to 2.0, and the integral time constant ranges from 30 seconds to 120 seconds. The specific parameters are tuned through step response experiments based on the dryer specifications and on-site operating conditions. This closed-loop feedback regulation ensures that the actual regeneration process always closely follows the target dew point curve, ensuring precise and controllable regeneration effect. For example, the second operating parameter requires a target regeneration flow rate of 120 kg / h and a target regeneration temperature of 110°C. The control system first adjusts the opening of the three-way proportional control valve to allow 120 kg / h of compressed air to enter the regeneration flow path, and simultaneously sets the power of the regeneration heater to the power required to heat the 120 kg / h of compressed air from the current inlet temperature to 110°C. The target dew point curve specifies that the exhaust dew point should be -25°C at the 15th minute of regeneration. If, at the 15th minute, the dew point sensor detects an actual regeneration exhaust dew point temperature of -18°C, the deviation is +7°C, indicating insufficient regeneration. The control system calculates the correction using a proportional-integral algorithm, increasing the heater power by 8% to accelerate desorption in the subsequent regeneration process, gradually bringing the exhaust dew point temperature closer to -25°C. If, at the 15th minute, the actual dew point temperature is -32°C, the deviation is -7°C, indicating overheating. The control system reduces the heater power by 5%, ensuring regeneration effectiveness while reducing unnecessary energy consumption.

[0033] Example 2, please refer to Figure 2 This invention provides a technical solution: a zero-air-consumption compressed air adsorption drying intelligent control system, which is applicable to the above-mentioned zero-air-consumption compressed air adsorption drying intelligent control method, comprising: Load forecasting module 1 is used to acquire real-time operating data of the adsorption tower, input the real-time operating data into the pre-built load forecasting model, extract time-series features from the real-time operating data based on the load forecasting model, and obtain the regeneration load demand curve for the future preset period. The reverse optimization module 2 is used to obtain the regeneration energy consumption function that matches the regeneration load demand curve, and to perform reverse optimization on the working parameters of the dryer with the zero point of the regeneration energy consumption function as the target, so as to generate the first working parameters. Energy flow spectrum construction module 3 is used to acquire real-time exhaust data of the compressor and construct an available energy flow spectrum of compressed air based on the real-time exhaust data; The dynamic matching module 4 is used to input the first operating parameters and the available energy flow spectrum into the pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and performs iterative correction on the first operating parameters based on the energy fit to obtain the second operating parameters. The closed-loop execution module 5 is used to divide the total flow of compressed air into a regeneration flow path and a drying flow path based on the second working parameters, and to adjust the flow rate and heating power of the regeneration flow path in real time so that the regeneration exhaust dew point temperature of the dryer is maintained on the target dew point curve corresponding to the second working parameters.

[0034] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A zero-air-consumption compressed air adsorption drying intelligent control method, characterized in that, include: The real-time operating data of the adsorption tower is obtained, and the real-time operating data is input into the pre-constructed load prediction model. Based on the load prediction model, the time-series features of the real-time operating data are extracted to obtain the regeneration load demand curve for the future preset period. Obtain the regeneration energy consumption function that matches the regeneration load demand curve, and use the zero point of the regeneration energy consumption function as the target to perform reverse optimization on the working parameters of the dryer to generate the first working parameters; Acquire real-time exhaust data from the compressor and construct a usable energy flow spectrum of compressed air based on the real-time exhaust data; The first operating parameters and the available energy flow spectrum are input into a pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and iteratively corrects the first operating parameters based on the energy fit to obtain the second operating parameters. Based on the second operating parameter, the total flow rate of compressed air is divided into a regeneration flow path and a drying flow path, and the flow rate and heating power of the regeneration flow path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second operating parameter.

2. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 1, characterized in that, Real-time operating data includes the current inlet temperature, inlet pressure, inlet humidity, and outlet dew point of the adsorption tower; the load prediction model includes a time-series coding network and a load decoding network. Based on the load forecasting model, time-series features are extracted from real-time operating data to obtain the regenerative load demand curve for a future preset time period, including: Real-time operating condition data is input into a time-series coding network to extract the hidden state feature vectors of the real-time operating condition data at different time scales. The latent state feature vector of the operating condition is input into the load decoding network. The load decoding network, based on the attention mechanism, predicts the regeneration load demand at each time step within the future preset period and generates the regeneration load demand curve. The regeneration load demand curve is a function curve representing the change of the theoretical minimum heat required to restore the adsorbent to the target regeneration degree over time within the future preset period.

3. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 2, characterized in that, The temporal coding network consists of multiple parallel one-dimensional convolutional layers and a bidirectional long short-term memory network layer; Real-time operating condition data is input into a time-series coding network to extract latent state feature vectors of the real-time operating condition data at different time scales, including: By using multiple one-dimensional convolutional layers with different kernel sizes, short-term fluctuation features and long-term trend features of real-time operating data are captured in parallel to obtain multi-scale feature maps. After concatenating the multi-scale feature maps along the feature dimensions, they are input into a bidirectional long short-term memory network layer to extract temporal correlation features containing forward and backward dependencies, which are used as the latent state feature vectors of the working conditions.

4. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 3, characterized in that, The load decoding network includes an attention weight generation unit, a load state decoding unit, and a curve fitting output unit; The latent state feature vector of the operating condition is input into the load decoding network. Based on an attention mechanism, the load decoding network predicts the regenerative load demand at each time step within a preset future period, generating a regenerative load demand curve, including: The attention weight generation unit calculates the attention weight of each time step state to the current prediction time based on the hidden state and the hidden state feature vector of the current decoding time step. The context vector of the current decoding time step is obtained by weighted summation of the hidden state feature vectors of the working condition using attention weights. The context vector is concatenated with the load prediction value of the previous decoding time step and input into the load state decoding unit to obtain the load hidden state of the current decoding time step. The hidden load state is input to the curve fitting output unit, and the regenerated load prediction value of the current decoding time step is calculated. The above steps are executed iteratively to generate the regenerative load forecast values ​​for all time steps within the preset future period. All the regenerative load forecast values ​​are then combined according to the time series to obtain the regenerative load demand curve.

5. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 4, characterized in that, A usable energy flow spectrum of compressed air is constructed based on real-time exhaust data, including: A two-dimensional energy distribution matrix is ​​constructed using exhaust temperature from real-time exhaust data as the first dimension and exhaust mass flow rate as the second dimension. The available energy conversion of the exhaust pressure corresponding to each element in the two-dimensional energy distribution matrix is ​​calculated to obtain the available heat potential of each element. Based on the available heat potential of each element, a usable energy flow spectrum is generated with temperature and flow rate as coordinates and available heat potential as the intensity.

6. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 5, characterized in that, The dynamic matching model includes a fit calculation layer and a parameter correction layer; The first operating parameters and the available energy flow spectrum are input into a pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and iteratively corrects the first operating parameters based on the energy fit to obtain the second operating parameters, including: The target regeneration temperature and regeneration flow rate in the first working parameters are mapped to the coordinate space of the available energy flow spectrum to obtain an energy demand point; The energy fit calculation layer calculates the distance loss between the energy demand point and the high heat available potential area in the available energy flow spectrum, and outputs the energy fit based on the distance loss; When the energy fit is lower than the preset fit threshold, the parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first working parameter along the energy gradient direction of the available energy flow spectrum until the energy fit of the adjusted energy demand point reaches the preset fit threshold, and then outputs the adjusted working parameter as the second working parameter.

7. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 6, characterized in that, The parameter correction layer iteratively adjusts the target regeneration temperature and regeneration flow rate in the first operating parameters along the energy gradient direction of the available energy flow spectrum, including: Keeping the total amount of regeneration constant, and using the regeneration load demand curve as the constraint boundary, search for the target point with the smallest difference between the available heat potential value and the current energy demand point in the available energy flow spectrum. The temperature and flow rates corresponding to the target point are used as the adjusted target regeneration temperature and regeneration flow rate.

8. The intelligent control method for zero-air-consumption compressed air adsorption drying according to claim 7, characterized in that, Based on the second operating parameter, the total compressed air flow is divided into a regeneration path and a drying path. The flow rate and heating power of the regeneration path are adjusted in real time to maintain the regeneration exhaust dew point temperature of the dryer on the target dew point curve corresponding to the second operating parameter. This includes: Adjust the opening of the flow control valve in the regeneration flow path according to the target regeneration flow rate in the second working parameter; Adjust the heater power of the regeneration flow path according to the target regeneration temperature in the second working parameter; Obtain the real-time dew point temperature and real-time exhaust flow rate at the regenerated exhaust outlet, and calculate the deviation between the real-time dew point temperature and the corresponding point on the target dew point curve. Using the deviation value as feedback, the heater power is finely adjusted so that the regenerated exhaust dew point temperature converges to the target dew point curve.

9. A zero-air-consumption compressed air adsorption drying intelligent control system, applicable to the zero-air-consumption compressed air adsorption drying intelligent control method according to any one of claims 1-8, characterized in that, include: The load forecasting module is used to acquire real-time operating data of the adsorption tower, input the real-time operating data into the pre-built load forecasting model, extract time-series features from the real-time operating data based on the load forecasting model, and obtain the regeneration load demand curve for the future preset period. The reverse optimization module is used to obtain the regeneration energy consumption function that matches the regeneration load demand curve, and to perform reverse optimization on the working parameters of the dryer with the zero point of the regeneration energy consumption function as the target, so as to generate the first working parameters. The energy flow spectrum construction module is used to acquire real-time exhaust data of the compressor and construct an available energy flow spectrum of compressed air based on the real-time exhaust data. The dynamic matching module is used to input the first operating parameters and the available energy flow spectrum into the pre-built dynamic matching model. The dynamic matching model calculates the energy fit between the first operating parameters and the available energy flow spectrum, and performs iterative correction on the first operating parameters based on the energy fit to obtain the second operating parameters. The closed-loop execution module is used to divide the total flow of compressed air into a regeneration flow path and a drying flow path based on the second operating parameter, and to adjust the flow rate and heating power of the regeneration flow path in real time so that the regeneration exhaust dew point temperature of the dryer is maintained on the target dew point curve corresponding to the second operating parameter.

Images