Air compressor system optimization control method based on multiple sensors and modeling

Abstract

The invention provides an air compressor system optimization control method based on multiple sensors and modeling. The method comprises the steps of sensor installation, wherein an environment temperature and humidity sensor, an atmospheric pressure sensor, an air supply flow sensor and the like are installed in an air compressor system; data acquisition: acquiring operation data of the air compression system; model training: carrying out data cleaning by utilizing the collected data, and then training each equipment model in the air compression system; performing optimization calculation: according to the topological structure of the air compression system, connecting all equipment models of the air compression system together, and calculating an optimal control parameter by using a genetic algorithm by taking the lowest energy consumption and the highest dew point temperature as targets; and control parameter issuing: the optimized parameters are issued to equipment for execution. Operation parameters of the air compressor can be optimized, and energy consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of optimization control of air compressor systems, and in particular to an optimization control method for air compressor systems based on multiple sensors and modeling. Background Art

[0002] The air compressor system is an important power source equipment in industrial production, and its operating efficiency directly affects energy consumption and production costs. Traditional air compressor system control methods usually rely on experience or fixed rules, which are difficult to adapt to complex and changing working environments. At the same time, they cannot adapt to the operating efficiency characteristics of air compressors of different brands and models, resulting in energy waste and low equipment efficiency. In addition, the operating status of equipment such as dryers and cooling towers in the air compressor system will also affect the performance of the overall system. Therefore, how to achieve efficient operation of the air compressor system through data acquisition, modeling and optimization control has become a technical problem that needs to be solved urgently.

[0003] Patent application document CN116877414A discloses an air compressor temperature control optimization test device and measurement method based on orthogonal test of response surface model, including a variable frequency oil pump, a flow meter and a tested temperature control valve connected in series in the main oil circuit. The oil circuit is divided into two branches after passing through the tested temperature control valve, one is a non-cooling branch, and the other is a cooling branch. The two branches are mixed before entering the air compressor. The tested temperature control valve is used to control the flow of the two branches, wherein the cooling branch includes a variable frequency fan for cooling, a quick-release cooler, a flow meter, and a flow regulating valve. However, this patent cannot completely solve the existing technical problems, nor can it meet the needs of the present invention. Summary of the invention

[0004] In view of the defects in the prior art, the object of the present invention is to provide an air compressor system optimization control method based on multi-sensor and modeling.

[0005] The air compressor system optimization control method based on multi-sensor and modeling provided by the present invention includes: Step 1: Install environmental temperature and humidity sensors, atmospheric pressure sensors, air supply flow sensors, cooling water flow sensors, cooling water temperature sensors, air compressor power sensors, dryer humidity sensors and cooling tower power sensors in the air compressor system to collect corresponding data in real time; Step 2: Use the quartile method combined with the sliding window method to clean the collected data, remove outliers and smooth the noise data; Step 3: Based on the cleaned data, train the air compressor power model, dryer humidity model, cooling pump frequency model and cooling tower power model respectively; Step 4: According to the topological structure of the air compressor system, the models of each device are coupled into an overall system model, and the total energy consumption and dew point temperature of the system are output; Step 5: Taking the minimum energy consumption and the maximum dew point temperature as the optimization objectives, a genetic algorithm is used to perform multi-objective optimization on the system model to generate the optimal control parameters; Step 6: Send the optimized control parameters to the air compressor, dryer, cooling pump and cooling tower for execution.

[0006] Preferably, in step 2: The quartile method specifically includes: setting 25% of the data in the data set less than or equal to Q1 as the first quartile, setting 75% of the data in the data set less than or equal to Q1 as the third quartile, calculating the interquartile range IQR of the data set, setting the lower limit to Q1-1.5×IQR, the upper limit to Q3+1.5×IQR, and eliminating outliers outside the range; The sliding window method specifically includes: using a sliding window with a window size of 10 and a sliding step size of 1, calculating the mean of the data in the window, and eliminating data points that deviate from the mean by more than 2%.

[0007] Preferably, in step 3: The air compressor power model adopts the ridge regression method. The input includes the ambient temperature, ambient humidity, each stage outlet pressure of the compressor and the cooling water inlet temperature. The output is the instantaneous power of the air compressor. The objective function is:

[0008] in, For the The actual value of the instantaneous power of the air compressor of the sample, For the The sample Features, is the regularization parameter, n is the number of samples, p is the number of features, is the initial regression coefficient, For the The regression coefficient of each feature; The training process is: divide the data set into a training set and a test set, and use cross-validation to select the best Value, fit the ridge regression model to the training set, and solve for the regression coefficient , calculate the mean absolute percentage error MAPE of the predicted value on the test set, the expression is:

[0009] in, For the The predicted value of instantaneous power of air compressor for samples; If MAPE<0.03, the model training is complete; otherwise, data needs to be collected again or features need to be adjusted; The power model of the cooling tower adopts an LSTM neural network. The inputs include ambient temperature, cooling water flow rate, and inlet and outlet water temperatures, and the output is the power of the cooling tower. The accuracy requirement for model training is MAPE < 0.05.

[0010] Preferably, in the step 5: The population size of the genetic algorithm is 400, and the maximum number of iterations is 500. The process of a single iteration is as follows: calculate the fitness value for each individual, eliminate the individuals that do not meet the preset conditions, and then randomly select two parent individuals to exchange part of the genes at the preset parameter positions; The fitness function aims at the lowest total power of the system. The constraint conditions include that the fluctuation range of the air supply flow rate is ±10% of the current value, and the dew point temperature meets the process requirements.

[0011] Preferably, in the step 6: The issued parameters include the start / stop instruction of the air compressor, the pressure set value, the regeneration heating instruction of the dryer, the frequency set value of the cooling pump, and the frequency set value of the cooling tower fan; The parameters are executed by a PLC / DDC controller and communicate using industrial protocols such as BACNet, ModBus, or OPC.

[0012] Compared with the prior art, the present invention has the following beneficial effects: (1) By installing a variety of sensors and communicating with equipment, including but not limited to compressed air flow rate, ambient temperature and humidity, ambient wet bulb temperature, atmospheric pressure, air supply pressure, air supply temperature, cooling water supply and return water temperature and flow rate, compressor power consumption, pump power consumption, and cooling tower power consumption, etc., the present invention collects and monitors all data affecting the operating energy efficiency of the air compressor system, realizing the full-state perception of the air compressor system; (2) By using the quartile method and the sliding window method, the present invention removes the data that cannot truly reflect the equipment performance, and then trains the performance models of equipment such as air compressors, dryers, cooling pumps, and cooling towers through machine learning modeling. For example, for an air compressor, the inputs of the model are ambient temperature, ambient humidity, ambient pressure, cooling water inlet temperature, cooling water outlet temperature, air temperature at each stage of the compressor inlet and outlet, air pressure at the outlet of each stage of the compressor, air flow rate at the outlet of the air compressor, opening degree of the compressor inlet valve, and opening degree of the compressor outlet valve, and the output of the model is the power consumption of the air compressor. This method realizes the quantitative parameter relationship between control parameters and equipment energy consumption, changing the state where the energy efficiency of equipment is unknown under the original group control state; (3) By constructing a system model, the present invention couples device models together, thereby realizing a method that can simulate the actual energy efficiency for any set of control and environmental parameters. Then, with the goal of minimizing energy consumption, the air supply volume as a constraint, the control parameters as variables, and the environmental parameters as fixed variables, the genetic algorithm is called in real time to optimize the current control parameters of the air compressor and send the control parameters to the air compressor system for execution, so that the air compressor system is always in the best energy efficiency state and energy consumption is reduced. Brief Description of the Drawings

[0013] By reading the following detailed description of the non-restrictive embodiments with reference to the accompanying drawings, other features, objectives, and advantages of the present invention will become more apparent: Figure 1 Schematic diagram of the optimization of the air compressor system of the present invention; Figure 2 Schematic diagram of the construction of the system model of the present invention. Detailed Embodiment

[0014] The present invention will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can still be made. These all fall within the protection scope of the present invention.

[0015] Embodiment The present invention provides an optimized control method for an air compressor system based on multi-sensor data acquisition and modeling, including the following steps: Sensor installation step: In the air compressor system, necessary sensors are added to collect data and train the model. Data acquisition step: The operating data of the air compressor system is collected. Model training step: Using the collected data, after data cleaning, the models of the air compressor, dryer, water pump, and cooling tower are trained. Optimization algorithm step: According to the topological structure of the air compressor system, the models of the air compressor, dryer, water pump, and cooling tower are connected together, and with the goal of minimizing energy consumption and maximizing the dew point temperature, the genetic algorithm is used to calculate the optimal control parameters. Control parameter sending step: The control parameters calculated in the previous step are sent to the air compressor, dryer, water pump, and cooling tower for execution.

[0016] In the sensor installation step, the following sensors and actuators are installed in the air compressor system: Atmospheric pressure sensor: Used to monitor the ambient atmospheric pressure; Ambient temperature and humidity sensor: Used to monitor the ambient temperature and humidity; Compressed air flow sensor at the outlet of the air compressor: Used to monitor the compressed air flow at the outlet of the air compressor. When installed, the requirements for the straight pipe section lengths upstream and downstream should be met; Dryer inlet regulating valve: Used to regulate the compressed air flow at the dryer inlet (not required if there is only one dryer); Cooling water flow sensor: Used to monitor the cooling water flow. When installed, the requirements for the upstream and downstream straight pipe section lengths should be met; Cooling water inlet and outlet temperature sensors: Used to monitor the inlet / outlet temperature of the cooling water for the air compressor; Cooling water inlet and outlet pressure sensors: Used to monitor the inlet / outlet temperature of the cooling water for the air compressor.

[0017] The above sensors and actuators are sensors that are not configured in a general air compressor system, so they need to be added. The data acquisition range required is larger than that of the sensors listed above. The final data acquisition content is subject to the variables listed in the data acquisition steps. Sensors need to be added for variables that are not collected. The above sensors and actuators all need to be connected to the PLC / DDC controller to ensure stable data control and reliability.

[0018] In the data acquisition step, the following data is collected in real-time through the newly installed sensors and the original air compressor system: Environmental parameters: Ambient temperature (Tdb), ambient humidity (RH), ambient pressure (P); Cooling water parameters: Cooling water inlet temperature (TcdIn), cooling water outlet temperature (TcdOut), cooling water flow (Fcd), cooling water inlet and outlet pressure differential (PcdIn and PcdOut), cooling pump frequency (cdp_Hz), cooling pump power (cdp_E); Air compressor parameters: Inlet and outlet air temperature of each stage of the compressor (TairIni and TairOuti), outlet pressure of each stage of the compressor (PairOuti), air compressor outlet flow (Fair), air compressor instantaneous power (ca_E), compressor inlet valve opening (ca_ValIn), compressor outlet valve opening (ca_ValOut) (for air compressors without regulating valves, valve parameters do not need to be collected), pressure loss rate (dPcoef); Dryer parameters: Dryer inlet flow (dry_F), dryer inlet air humidity (dry_HIn), dryer outlet air humidity (dry_HOut), regeneration heating temperature (dry_Treg), regeneration heating electric power (dry_E), regeneration heating duration (dry_Timereg), adsorption / regeneration state (dry_State); Cooling tower parameters: Cooling water inlet and outlet temperature (TcdIn and TcdOut), cooling water flow (Fcd), cooling tower fan frequency (ct_Hz), cooling tower fan power (ct_E).

[0019] The collection of the above variables is required. For variables that do not exist in the original air compressor system, sensors need to be added to supplement them. The above data collection needs to be collected into the PLC / DDC controller and then collected into the server through industrial protocols such as BACNet, ModBus, OPC, etc., and stored in the server at a minute-level frequency. If the storage frequency is less than 1 minute, it is easy to cause deviations in the indicators of the next model training, resulting in a reduction in the reliability of the model.

[0020] In the model training step, data cleaning is carried out first.

[0021] Data cleaning uses the quartile method and the sliding window method to clean the collected data and exclude abnormal data: Quartile method: By calculating the interquartile range (IQR) of the data, identify and remove outliers outside the reasonable range; A possible implementation of the quartile method is: Quartiles: After sorting a set of data from smallest to largest, it is divided into four equal parts, and the value corresponding to each quantile is called a quartile. Specifically, it includes: First quartile (Q1): 25% of the data in the dataset is less than or equal to Q1; Second quartile (Q2, i.e., the median): 50% of the data in the dataset is less than or equal to Q2; Third quartile (Q3): 75% of the data in the dataset is less than or equal to Q3.

[0022] Interquartile range (IQR): IQR is the difference between Q3 and Q1, i.e., IQR = Q3 - Q1. IQR reflects the distribution range of the middle 50% of the data.

[0023] Lower Bound: LB = Q1 - k × IQR Upper Bound: UB = Q3 + k × IQR, where k is a constant, usually taking a value of 1.5.

[0024] For each data point di in the dataset, if it meets one of the following conditions, it is considered an outlier: di < LB or di > UB. Remove the outliers from the dataset.

[0025] Sliding window method: Smooth the data through a sliding window to eliminate the noise caused by short-term fluctuations.

[0026] A possible implementation of the sliding window method is: (1) Determine the window size and sliding step: According to the data characteristics and application requirements, determine the window size w and the sliding step s. After multiple tests, the window size w takes a value of 10, and the sliding step s takes a value of 1.

[0027] (2) Initialize the window position. Place the starting position of the window at the first point of the data.

[0028] (3) Process the data within the window. Apply a window function to the data within the window, such as calculating the mean, median, or standard deviation. After testing, using the mean calculation is the best method. Then compare each data within the window with the average value. Data with a deviation exceeding x% is considered an outlier and marked for deletion. The best value of x after testing is 2.

[0029] (4) Slide the window. Slide the window s data points to the right and repeat step (3) until the window covers the entire dataset.

[0030] (5) Delete the samples marked as outliers, and the remaining samples are used for model training.

[0031] Train the model based on the data jointly cleaned by the two methods.

[0032] In the model training step, the second step is to perform model training: Air compressor model Model input: Ambient temperature, ambient humidity, ambient pressure, cooling water inlet temperature, cooling water outlet temperature, air temperature at the inlet and outlet of each stage of the compressor, outlet pressure of each stage of the compressor, air flow at the outlet of the air compressor, opening degree of the inlet valve of the compressor, opening degree of the outlet valve of the compressor; Model output: Instantaneous power of the air compressor.

[0033] Model training method: Divide the model input into two groups. The first group is ambient temperature, ambient humidity, and ambient pressure, and the second group is the others. Perform quadratic polynomial feature engineering expansion on the two groups separately, and the parameters of the two groups cannot perform cross-feature engineering. Through testing, the input variables do not need to be normalized or changed, and they can be directly brought into the training in their original state to make the model reach the required accuracy. Then use the ridge regression method for model training. After testing, compared with the least squares method, the ridge regression method can better adapt to the working conditions with a relatively small sample ratio. The model training results are evaluated using MAPE (Mean Absolute Percentage Error). MAPE < 0.03 represents the completion of model training, otherwise, data needs to be collected again and the model needs to be retrained until MAPE < 0.03.

[0034] A possible implementation of the mean absolute percentage error is:

[0035] Among them, MAPE represents the mean absolute percentage error; actual(t) represents the actual value of the t-th sample; forecast(t) represents the predicted value of the t-th sample; n represents the number of samples, and this formula can be applied to the calculation of the MAPE index of all models in this article.

[0036] Dryer model Model input: dryer inlet flow rate, dryer inlet cumulative flow rate (accumulated from the most recent regeneration to the current), dryer inlet air humidity, regeneration heating temperature, regeneration heating electric power, regeneration heating duration; Model output: dryer outlet air humidity, total power consumption of regeneration heating.

[0037] Model training method: Perform quadratic polynomial feature expansion on the model input variables (excluding the dryer inlet cumulative flow rate), and use the least squares method for model training. After testing other regression model training methods such as ridge regression and elastic net, it is found that they cannot significantly improve the model robustness. Therefore, the most classic least squares method can be applied here. The model training results are evaluated using MAPE (mean absolute percentage error). MAPE < 0.1 indicates that the model training is completed; otherwise, data needs to be collected again and the model needs to be retrained until MAPE < 0.1.

[0038] Water pump model Model input: cooling water flow rate; Model output: cooling pump frequency, cooling pump power.

[0039] Model training method: Use the cooling water flow rate as the input variable, and train the cooling pump frequency model by linear regression using the least squares method. Then, change the cooling pump frequency to cooling pump frequency^3, cooling pump frequency^2, and cooling pump frequency, and use them together with the cooling water flow rate and the cooling water pressure difference as the model input, and use the least squares method for model training. After testing other regression model training methods such as ridge regression and elastic net, it is found that they cannot significantly improve the model robustness. Therefore, the most classic least squares method can be applied here. The model training results are evaluated using MAPE (mean absolute percentage error). MAPE < 0.02 indicates that the model training is completed; otherwise, data needs to be collected again and the model needs to be retrained until MAPE < 0.02.

[0040] Cooling tower model Model input: ambient temperature, ambient humidity, cooling water inlet and outlet temperatures, cooling water flow rate; Model output: cooling tower power.

[0041] Model training method: Normalize the training data, and then use the LSTM model form for training. The LSTM model should include at least 1 LSTM layer and 2 fully connected layers with no less than 64 neurons. The activation function should be tanh. After testing, non-neural network models such as the least squares method, ridge regression, and support vector regression cannot be effectively applied to the training of the cooling tower model. Adding more than 3 fully connected layers is of little help in improving the model accuracy. The model training results are evaluated using MAPE (Mean Absolute Percentage Error). MAPE < 0.05 indicates that the model training is completed; otherwise, data needs to be collected again and the model retrained until MAPE < 0.05.

[0042] In the optimization algorithm steps, in the first step, according to the topological structure of the air compressor system, connect the inputs and outputs of the air compressor, dryer, water pump, and cooling tower models together to form a system model. Specifically: The total inputs of the system model are ambient temperature / humidity / pressure, cooling water inlet / outlet temperature, compressed air flow rate, air compressor start / stop status, main pipeline pressure, pipeline pressure loss rate, etc. The input variables of equipment models not listed in the above parameters are constants, and those that are not the main process connection parameters are temporarily omitted. Then connect the inputs and outputs of the models together in the following order: Using the main pipeline pressure, pipeline pressure loss rate, and compressed air flow rate, calculate the compressor outlet pressure using the formula: Compressor outlet pressure = Main pipeline pressure + Pipeline pressure loss rate * Compressed air flow rate * Compressed air flow rate; Substitute the ambient temperature / pressure, cooling water inlet / outlet temperature, compressed air flow rate, and compressor outlet pressure into the air compressor model to calculate the humidity of the compressed air at the compressor outlet and the air compressor power, and sum up the air compressor powers; Substitute the compressed air flow rate and cooling water inlet temperature into the dryer model to calculate the total regeneration heating power and the humidity of the compressed air at the dryer outlet; Calculate the cooling water flow rate based on the total air compressor power and the cooling water inlet / outlet temperature using the formula: Total air compressor power = (Cooling water outlet temperature - Cooling water inlet temperature) * Cooling water flow rate * 1.163; Substitute the cooling water flow rate into the cooling pump model to calculate the cooling pump power and frequency; Substitute the cooling water inlet / outlet temperature, cooling water flow rate, and ambient temperature / relative humidity into the cooling tower model to calculate the cooling tower power; Sum up the powers of the air compressor, dryer, cooling pump, and cooling tower as the output of the system model, and at the same time, take the dew point temperature at the dryer outlet as the output of the system model; The construction of the system model is completed above. The expression of the system model is: Total system power = sum(air compressor power) + sum(dryer power) + sum(cooling pump power) + sum(cooling tower power). The specific structure can also be referred toFigure 2 。

[0043] In the optimization algorithm steps, the second step is to use the genetic algorithm to optimize the system model: The model genetic algorithm (Genetic Algorithm, GA) is a global optimization algorithm based on natural selection and genetic mechanisms, which is widely used in solving complex problems. Its core idea is to search for the optimal solution by simulating the biological evolution process (such as selection, crossover, mutation). The following are the specific steps of the genetic algorithm: Step 1: Initialize the population Randomly generate the initial population with a population size of N; after testing, setting the population size to 400 can balance the calculation efficiency and the accuracy of the optimization algorithm. In this case, the variables included in the population are the variable inputs of the system model, specifically the environmental temperature / humidity / pressure, the inlet / outlet water temperature of the cooling water, the compressed air flow rate, the start / stop state of the air compressor, the main pipeline pressure, the pipeline pressure loss rate, etc. In this case, the model input parameters carried by each population individual are the sum of three groups of model inputs. Among these three groups of parameters, the environmental parameters and the unit combination are the same values. For example, the environmental temperature, relative humidity, environmental pressure, and unit opening state are the same values in the three groups of parameters. For other parameters, such as the air supply flow rate, they are three different values and can be unequal in the three groups.

[0044] Each individual is represented by a chromosome, and the chromosome can be a binary string, a real number vector, or other coding forms. After testing, the input and output of the system model are both real continuous variables, so the real number vector coding form needs to be used.

[0045] Step 2: Calculate the fitness Calculate the fitness value for each individual, and the fitness value is determined by the objective function; in this case, the three groups of parameters in the population individuals are respectively substituted into the system model for calculation, and the sum of the variable powers output by the system model calculated three times is used as the fitness value. With the lowest power as the goal, and at the same time, the dew point temperature of the output variable of the system model each time is used as a constraint. Individuals with a dew point temperature higher than the compressed air process requirement value in the calculation will be marked as eliminated individuals and removed when updating the population. At the same time, the compressed air flow rate is also used as a constraint. The current total air supply flow rate, the current total air supply flow rate * 0.9, and the current total air supply flow rate * 1.1 are used as the constraints for the system model calculation three times respectively. Individuals below the requirements will be eliminated. The coefficients 0.9 and 1.1 for calculating the air supply volume usually do not need to be changed.

[0046] Step 3: Selection Select excellent individuals according to the fitness value to enter the next generation; Common selection methods include roulette wheel selection, tournament selection, etc. After testing, there is no significant difference among various methods, and each classical method can be selected.

[0047] Step 4: Crossover Randomly select two individuals from the selected individuals as parents; Generate two new individuals through the crossover operation; Common crossover methods include single-point crossover, multi-point crossover, uniform crossover, etc.

[0048] Step 5: Mutation Perform mutation operations on the newly generated individuals to increase the diversity of the population; Common mutation methods include bit flipping, random perturbation, etc.

[0049] Step 6: Update the population Replace some or all of the individuals in the current population with the newly generated individuals; After testing, it is the best choice to retain 50% of the current individuals; Retain the optimal individual (elitist strategy).

[0050] Step 7: Judge the termination condition If the termination condition is met (such as reaching the maximum number of iterations or the fitness value converges), the algorithm ends; In this case, the maximum number of iterations is set to 500 times.

[0051] Otherwise, return to Step 2.

[0052] After the genetic algorithm calculation is completed, the optimal control parameters can be used for the next control parameter distribution. At the same time, since the unit combination fully considers the fluctuations within ±10% of the current gas supply flow, it can effectively reduce the fluctuations of the optimized unit combination caused by the characteristics of its own mutation steps, and reduce the number of start-stop operations in the actual control process.

[0053] In the control parameter distribution step, the control parameters calculated in the previous step, including the start-stop instruction of the air compressor, the pressure setting value of the air compressor, the regeneration instruction of the dryer, the start-stop instruction and frequency setting value of the cooling pump, the start-stop instruction and frequency setting value of the cooling tower, are distributed to the PLC / DDC controller, and the controller executes the start-stop actions and parameter setting actions of the air compressor, dryer, water pump and cooling tower.

[0054] Those skilled in the art know that, in addition to implementing the systems, devices, and their respective modules provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the systems, devices, and their respective modules provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc., to achieve the same program. Therefore, the systems, devices, and their respective modules provided by the present invention can be considered as a kind of hardware components, and the modules included therein for implementing various programs can also be regarded as the structures within the hardware components; the modules for implementing various functions can also be regarded as either software programs for implementing methods or structures within hardware components.

[0055] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An air compressor system optimization control method based on multi-sensor and modeling, characterized in that: include: Step 1: Install environmental temperature and humidity sensors, atmospheric pressure sensors, air supply flow sensors, cooling water flow sensors, cooling water temperature sensors, air compressor power sensors, dryer humidity sensors and cooling tower power sensors in the air compressor system to collect corresponding data in real time; Step 2: Use the quartile method combined with the sliding window method to clean the collected data, remove outliers and smooth the noise data; Step 3: Based on the cleaned data, train the air compressor power model, dryer humidity model, cooling pump frequency model and cooling tower power model respectively; Step 4: According to the topological structure of the air compressor system, the models of each device are coupled into an overall system model, and the total energy consumption and dew point temperature of the system are output; Step 5: Taking the minimum energy consumption and the maximum dew point temperature as the optimization objectives, a genetic algorithm is used to perform multi-objective optimization on the system model to generate the optimal control parameters; Step 6: Send the optimized control parameters to the air compressor, dryer, cooling pump and cooling tower for execution.

2. The air compressor system optimization control method based on multi-sensor and modeling according to claim 1 is characterized in that: In step 2: The quartile method specifically includes: setting 25% of the data in the data set less than or equal to Q1 as the first quartile, setting 75% of the data in the data set less than or equal to Q1 as the third quartile, calculating the interquartile range IQR of the data set, setting the lower limit to Q1-1.5×IQR, the upper limit to Q3+1.5×IQR, and eliminating outliers outside the range; The sliding window method specifically includes: using a sliding window with a window size of 10 and a sliding step size of 1, calculating the mean of the data in the window, and eliminating data points that deviate from the mean by more than 2%.

3. The air compressor system optimization control method based on multi-sensor and modeling according to claim 1 is characterized in that: In step 3: The air compressor power model adopts the ridge regression method. The input includes the ambient temperature, ambient humidity, each stage outlet pressure of the compressor and the cooling water inlet temperature. The output is the instantaneous power of the air compressor. The objective function is: in, For the The actual value of the instantaneous power of the air compressor of the sample, For the The sample Features, is the regularization parameter, n is the number of samples, p is the number of features, is the initial regression coefficient, For the The regression coefficient of each feature; The training process is: divide the data set into a training set and a test set, and use cross-validation to select the best Value, fit the ridge regression model to the training set, and solve for the regression coefficient , calculate the mean absolute percentage error MAPE of the predicted value on the test set, the expression is: in, For the The predicted value of instantaneous power of air compressor for samples; If MAPE<0.03, the model training is complete; otherwise, data needs to be collected again or features need to be adjusted; The cooling tower power model adopts an LSTM neural network, the input includes ambient temperature, cooling water flow rate and inlet and outlet water temperature, the output is cooling tower power, and the model training accuracy requires MAPE<0.

05.

4. The air compressor system optimization control method based on multi-sensor and modeling according to claim 1 is characterized in that: In step 5: The population size of the genetic algorithm is 400, the maximum number of iterations is 500, and the single iteration process is: calculate the fitness value of each individual, eliminate individuals that do not meet the preset conditions, and then randomly select two parent individuals to exchange some genes at the preset parameter position; The fitness function aims to minimize the total system power, and the constraints include that the fluctuation range of the gas supply flow rate is ±10% of the current value and the dew point temperature meets the process requirements.

5. The air compressor system optimization control method based on multi-sensor and modeling according to claim 1 is characterized in that: In step 6: The parameters sent include air compressor start and stop instructions, pressure setting value, dryer regeneration heating instruction, cooling pump frequency setting value and cooling tower fan frequency setting value; The parameters are executed by a PLC / DDC controller and communicate using BACNet, ModBus or OPC industrial protocols.

Images