Low-energy-consumption nitrogen charging grain storage system and method

Abstract

The invention discloses a low-energy-consumption nitrogen-charging grain storage system and method, and relates to the field of nitrogen-charging grain storage. The method comprises the steps that compressed air is sequentially introduced into a filter set, a drying machine and a micro-oil adsorber to be purified and dried; stripping oxygen O2 molecules and water H2O molecules in the air by using a carbon molecular sieve pressure swing adsorption tower to obtain nitrogen N2 molecular gas; inputting nitrogen N2 molecular gas into a cold drying system, and cooling to a preset temperature range; detecting whether the concentration of the output nitrogen N2 molecular gas meets a preset concentration range or not by using an oxygen analyzer; and when the detected N2 molecular gas concentration of the nitrogen is lower than the lower limit of the preset concentration range, starting a circulating pump to suck the mixed gas in the closed warehouse to the carbon molecular sieve pressure swing adsorption tower for adsorption and nitrogen production again until the N2 molecular gas concentration of the output nitrogen meets the preset concentration range. Aiming at the problem that the nitrogen concentration in nitrogen-filled grain storage is difficult to accurately control in the prior art, accurate control of the nitrogen concentration and yield is realized.

Description

Technical Field

[0001] The present application relates to the field of nitrogen-filled grain storage, and in particular to a low-energy nitrogen-filled grain storage system and method. Background Art

[0002] Nowadays, food security issues are becoming increasingly prominent, and ensuring the quality of grain storage and reducing losses have become global issues. Filling grain silos with nitrogen can effectively inhibit the growth of stored grain pests and mildew, and extend the storage period of grain. Therefore, nitrogen-filled grain storage technology has received widespread attention.

[0003] The existing nitrogen storage method mainly uses a displacement method to increase the nitrogen concentration. This method is to introduce high-purity nitrogen into the granary, and use the density difference between nitrogen and air in the granary to gradually replace and drive out the air, thereby increasing the nitrogen concentration in the granary. However, this process has a serious problem of nitrogen waste. Since nitrogen and air are not completely separated, a large amount of unsaturated nitrogen is discharged out of the granary along with the air, resulting in a large loss of nitrogen.

[0004] Nitrogen production requires a lot of electricity, and wasting nitrogen means a huge waste of energy. With the continuous growth of grain production, the demand for nitrogen is also increasing day by day, resulting in higher and higher energy consumption costs for nitrogen storage, which is not conducive to the sustainable development of agriculture. In addition, the nitrogen-rich gas discharged will have a certain impact on the atmospheric environment.

[0005] The fundamental reason for the waste of nitrogen is that the purity of the nitrogen introduced into the granary by the existing method is not high enough, and the residual oxygen cannot be completely removed during the replacement process and is discharged together with the nitrogen, so that more nitrogen needs to be introduced subsequently to achieve the required inert environment concentration, resulting in low nitrogen utilization. Summary of the invention

[0006] In view of the problem that it is difficult to accurately control the nitrogen concentration in nitrogen-filled grain storage in the prior art, the present application provides a low-energy nitrogen-filled grain storage system and method, which adopts multi-stage filtration purification and drying technology to fully remove solid impurities, moisture and oil in the air; and utilizes carbon molecular sieve pressure swing adsorption, through real-time monitoring of parameters such as the inlet and outlet pressure and temperature of the carbon molecular sieve adsorption tower, combined with the adsorbent performance model, and adopts an adaptive neural fuzzy control algorithm to optimize the adsorption operation parameters online, thereby achieving accurate control of nitrogen concentration and output.

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] One aspect of the present application provides a low-energy nitrogen-filled grain storage system, including: a compression unit, which compresses air to a preset pressure; a pretreatment unit, which is used to purify and dry the compressed air, use a filter group to remove solid impurities in the air, use a dryer to remove moisture in the air, and use a micro-oil adsorber to remove oil in the air; a nitrogen production unit, which performs pressure swing adsorption on the pretreated compressed air through a carbon molecular sieve adsorption tower, strips O2 and H2O molecules in the air, and obtains N2 molecular gas; a nitrogen re-cooling unit, which cools the generated N2 molecular gas to a preset temperature through a cold drying system and outputs it; a concentration detection unit, which uses an oxygen analyzer to detect whether the output N2 molecular gas concentration meets a preset target concentration range; a control unit, when detecting that the output nitrogen concentration is lower than the preset target concentration lower limit, generates a circulating nitrogen production instruction, starts a circulating pump to draw the mixed gas in the closed warehouse to the nitrogen production unit for pressure swing adsorption nitrogen production again, until the output nitrogen concentration reaches the preset target concentration range; a conveying unit, which conveys the nitrogen reaching the target concentration to the closed warehouse for nitrogen filling and grain storage.

[0009] Another aspect of the present application also provides a low-energy nitrogen-filled grain storage method, comprising: compressing air to a preset pressure range through an air compressor; passing the compressed air into a filter group, a dryer and a micro-oil adsorber in sequence for purification and drying; wherein the filter group removes solid impurities in the air, the dryer removes moisture in the air, and the micro-oil adsorber removes oil stains in the air to obtain pre-treated compressed air; inputting the pre-treated compressed air into a carbon molecular sieve adsorption tower, and using the carbon molecular sieve pressure swing adsorption tower to strip oxygen O2 molecules and water H2O molecules in the air to obtain nitrogen N2 molecular gas; The nitrogen N2 molecular gas is input into the cold drying system, cooled to the preset temperature range, and the cooled nitrogen N2 molecular gas is output; the oxygen analyzer is used to detect whether the concentration of the output nitrogen N2 molecular gas meets the preset concentration range; when the detected nitrogen N2 molecular gas concentration is lower than the lower limit of the preset concentration range, the circulation pump is started to suck the mixed gas in the closed warehouse to the carbon molecular sieve pressure swing adsorption tower for adsorption and nitrogen production again, until the output nitrogen N2 molecular gas concentration meets the preset concentration range; the nitrogen N2 molecular gas that meets the preset concentration range is transported to the closed warehouse for nitrogen filling and grain storage operations.

[0010] Among them, carbon molecular sieve is a highly efficient adsorbent with a uniform microporous structure, which can selectively adsorb gases of specific molecular sizes. In the nitrogen-filled grain storage scheme, the carbon molecular sieve adsorption tower uses carbon molecular sieve to preferentially adsorb oxygen O2 molecules and water H2O molecules in the air, thereby separating nitrogen N2. The carbon molecular sieve adsorption tower achieves gas separation through the principle of physical adsorption, does not require heating and regeneration, and is simple to operate. Carbon molecular sieve pressure swing adsorption (PSA) is a commonly used industrial nitrogen production method. The PSA tower achieves oxygen and nitrogen separation by changing the pressure in the adsorption tower and utilizing the difference in the adsorption capacity of carbon molecular sieves for O2 and N2 under different pressures. The cold dryer is a key component in cryogenic air separation equipment, which is used to remove moisture from compressed air. In this application, the cold drying system cools the nitrogen produced by the PSA tower to below the dew point, causing the residual water vapor in the gas to condense and precipitate.

[0011] Furthermore, the solid impurities in the air are removed by using a filter group, including: using a double-layer stainless steel sintered felt filter element to remove particles with a particle size greater than d1 in the compressed air to obtain the first filtered air; using a polyester fiber filter material to remove particles with a particle size greater than d2 in the first filtered air through an electrostatic effect to obtain the second filtered air; using a nano-PTFE fiber membrane filter material to remove particles with a particle size greater than d3 in the second filtered air to obtain air after the solid impurities are removed. Among them, the value range of d1 is: 1 to 5 microns (μm); the value range of d2 is: 0.1 to 1 micron (μm); the value range of d3 is: 0.01 to 0.1 micron (μm).

[0012] Furthermore, the micro-oil adsorber removes oil pollution in the air, including: passing the compressed air treated by the dryer into the micro-oil adsorber, the micro-oil adsorber contains a lipophilic adsorbent, and the adsorbent is used to remove oil mist and oil droplets in the compressed air; the lipophilic adsorbent includes an activated carbon adsorbent.

[0013] Preferably, a multi-stage series filtration structure is used, and different types of lipophilic adsorbents are filled inside to achieve deep removal of oil pollution through multiple filtration. The first-stage adsorption layer uses Teflon-modified polypropylene fiber felt as the first-stage adsorption layer. Its excellent hydrophobicity and lipophilicity are used to quickly aggregate the passing oil mist into larger oil droplets. The second-stage adsorption layer uses a specific surface area of ​​1000-1500m 2 / g honeycomb activated carbon is used as the second-stage adsorbent, providing a large number of microporous structures to effectively adsorb oil mist in the air. An online residual oil detector is installed at the outlet of the micro-oil adsorber to monitor the oil content of the treated compressed air in real time. When the detection value exceeds the set threshold, the adsorbent is regenerated or replaced in time.

[0014] Furthermore, the pretreated compressed air is input into the carbon molecular sieve adsorption tower, and the carbon molecular sieve pressure swing adsorption tower is used to strip oxygen O2 molecules and water H2O molecules in the air, including: the pretreated compressed air is passed into the air inlet of the carbon molecular sieve adsorption tower, and the carbon molecular adsorbent in the carbon molecular sieve adsorption tower adsorbs oxygen O2 molecules and water H2O molecules in the air to enrich nitrogen N2 molecules; pressure sensors are respectively set at the air inlet and outlet of the carbon molecular sieve adsorption tower to collect the inlet pressure and outlet pressure of the carbon molecular sieve adsorption tower; and a temperature sensor is set on the outer wall of the tower body of the carbon molecular sieve adsorption tower to collect the tower body temperature of the carbon molecular sieve adsorption tower; preferably, high-precision pressure sensors are respectively installed at the air inlet and outlet of the adsorption tower, with a range of 0 to 1MPa and an accuracy of ±0.1%, to collect the pressure changes in the tower in real time. A platinum resistance temperature sensor is installed on the outer wall of the adsorption tower, with a range of -50 to 100°C and an accuracy of ±0.2°C, to monitor the tower body temperature. According to the collected inlet pressure, outlet pressure and tower temperature, combined with the preset adsorption performance curve of the carbon molecular sieve adsorbent, the adaptive neural fuzzy inference control ANFIS algorithm is used to calculate the optimal adsorption operating point of the carbon molecular sieve adsorption tower, and dynamically adjust the air inlet valve opening of the carbon molecular sieve adsorption tower and the exhaust rate of the vacuum pump.

[0015] Among them, the optimal adsorption operating point refers to a set of operating parameter combinations that enable the adsorption tower to achieve the optimal separation performance during the carbon molecular sieve pressure swing adsorption nitrogen production process, including adsorption pressure, desorption pressure, adsorption time, desorption time, etc. At the optimal adsorption operating point, the carbon molecular sieve adsorbent has the largest adsorption of nitrogen and the smallest adsorption of oxygen and water, thereby obtaining the highest nitrogen purity and recovery rate. The optimal adsorption operating point is affected by factors such as adsorbent characteristics, tower structure, and gas composition, and needs to be solved through mathematical models and optimization algorithms.

[0016] Furthermore, the opening of the air inlet valve of the carbon molecular sieve adsorption tower and the pumping rate of the vacuum pump are dynamically adjusted, including: establishing a mathematical model of the nitrogen production process of the carbon molecular sieve adsorption tower according to the nitrogen adsorption amount of the carbon molecular sieve adsorbent, the selectivity of nitrogen and oxygen, and the adsorption and desorption rate constants; taking the inlet pressure, outlet pressure and tower temperature of the carbon molecular sieve adsorption tower as input variables of the mathematical model; using an adaptive neural fuzzy inference control ANFIS algorithm, taking the nitrogen concentration and nitrogen output of the carbon molecular sieve adsorption tower as control targets, calculating the optimal adsorption operation point of the carbon molecular sieve adsorption tower, and outputting the optimal control amount of the air inlet valve opening of the carbon molecular sieve adsorption tower and the pumping rate of the vacuum pump; adjusting the air inlet valve opening of the air inlet of the carbon molecular sieve adsorption tower and adjusting the pumping rate of the vacuum pump according to the optimal control amount of the air inlet valve opening and the pumping rate of the vacuum pump; collecting the nitrogen concentration and nitrogen flow outputted by the carbon molecular sieve adsorption tower, and using the detected nitrogen concentration and nitrogen flow to use the least square method to calculate the adsorption equilibrium constant b in the mathematical model. jand the adsorption kinetic rate constant k z Make corrections.

[0017] Specifically, in the present application, dynamically adjusting the intake valve opening and the vacuum pump exhaust rate according to the optimal adsorption operating point includes: taking the inlet pressure, outlet pressure and tower temperature of the carbon molecular sieve adsorption tower as input variables, and taking the operating parameters such as adsorption pressure, desorption pressure, adsorption time, desorption time as output variables, and establishing an ANFIS model. The ANFIS model is trained and verified through experimental data so that it can accurately predict the optimal adsorption operating point. The pressure and temperature data during the adsorption process are collected in real time by the pressure sensor installed at the inlet and outlet of the adsorption tower and the temperature sensor on the outer wall of the tower body. The collected data is filtered, denoised and normalized, and converted into the input format of the ANFIS model. The processed pressure and temperature data are input into the trained ANFIS model to calculate the optimal adsorption operating point under the current working conditions in real time. The rolling optimization strategy is adopted to repeatedly execute the ANFIS model at a certain time interval (such as 1 minute) to obtain the dynamically updated optimal operating parameters. The optimal adsorption pressure value output by the ANFIS model is converted into a control instruction for the intake valve opening, and the electric regulating valve is driven by the PID controller to achieve precise control of the adsorption pressure. Convert the optimal desorption pressure value into a control instruction for the frequency or speed of the vacuum pump, and adjust the vacuum pump's pumping rate through the frequency converter to achieve dynamic optimization of the desorption pressure. According to the optimized adsorption time and desorption time, control the switching cycle of the switching valve to achieve automatic switching of the adsorption and regeneration processes. While performing the optimization control, continuously monitor the actual operating status of the adsorption tower, including key indicators such as product nitrogen purity and pressure fluctuations. Compare the measured data with the predicted value of the ANFIS model, calculate the deviation value, and fine-tune the control parameters according to the deviation size to achieve closed-loop feedback correction. Set the judgment logic and emergency handling strategies for abnormal conditions, such as pressure or temperature exceeding the limit, adsorbent failure, etc., to ensure the safe and stable operation of the system.

[0018] Furthermore, the mathematical model expression is as follows:

[0019] Among them, q z is the dynamic adsorption amount of component z under nitrogen storage conditions, in mol / kg; is the equilibrium adsorption capacity of component z under nitrogen storage conditions, in mol / kg; k z is the adsorption kinetic rate constant of component z at the nitrogen storage temperature and pressure, in units of 1 / s; q m is the sum of the maximum adsorption capacity of carbon molecular sieve for N2 and O2 at nitrogen storage temperature, in mol / kg; b jis the Langmuir adsorption equilibrium constant of component z at the nitrogen storage temperature, in 1 / bar; p z is the partial pressure of component z under nitrogen storage conditions, in bar; p0 is the standard atmospheric pressure, with a value of 1.01325 bar; n z is the non-ideal factor of the adsorption isotherm of component z under nitrogen storage conditions, dimensionless; t is the adsorption time, unit is s.

[0020] Among them, the adsorption kinetic rate constant k z The adsorption rate of component z on carbon molecular sieve adsorbent under specific temperature and pressure conditions of nitrogen-filled grain storage is characterized. z The larger the value, the faster the adsorption process of component z on the carbon molecular sieve, and the shorter the time required to reach adsorption equilibrium. z Affected by temperature and pressure, it usually increases with increasing temperature and decreases with increasing pressure. z It is crucial to simulate and optimize the carbon molecular sieve pressure swing adsorption nitrogen production process.

[0021] Langmuir adsorption equilibrium constant b j It reflects the adsorption affinity of component z on the carbon molecular sieve adsorbent at a specific temperature of nitrogen-filled grain storage. j The larger the value, the stronger the interaction between component z and the carbon molecular sieve, and the higher the equilibrium adsorption capacity at the same pressure. j It is mainly affected by temperature and usually decreases with increasing temperature, which is consistent with the thermodynamic characteristics of the Langmuir adsorption model. j It is the prerequisite for accurately predicting the adsorption equilibrium and selectivity of carbon molecular sieves.

[0022] In actual nitrogen storage applications, the performance of carbon molecular sieve adsorbents is affected by factors such as temperature, pressure, and gas composition, resulting in its adsorption kinetic rate constant k z and the Langmuir adsorption equilibrium constant b j There is a certain deviation from the theoretical value or laboratory measurement value. During the nitrogen storage process, a series of measured adsorption data can be obtained by real-time collection of the inlet and outlet concentration, pressure, temperature and other data of the adsorption tower. Using these measured data, the least squares objective function is established, with k z and b j To optimize the variables, the modified parameter values ​​are solved through iterative calculation. The modified k z and b j It can more accurately reflect the adsorption kinetics and equilibrium characteristics of carbon molecular sieves under actual nitrogen-filled grain storage conditions, enable the mathematical model to better match the actual process, and improve the model's predictive performance and control accuracy.

[0023] Furthermore, the adaptive neural fuzzy inference control ANFIS algorithm is used to output the optimal control quantities of the carbon molecular sieve adsorption tower inlet valve opening and vacuum pump exhaust rate, including:

[0024] Set nitrogen concentration error e1, nitrogen concentration error change rate e1', nitrogen flow error e2 and nitrogen flow error change rate e2' as input variables, valve opening u1 and vacuum pump exhaust rate u2 as output variables, and use Gaussian membership function μ and Sugeno type fuzzy inference rules to establish an ANFIS controller containing a five-layer neural network.

[0025] The historical operation data of the carbon molecular sieve adsorption tower was used to optimize the mean c and variance σ of the membership function using the error back propagation algorithm. The linear function coefficients p and r of the Sugeno type fuzzy inference rule were optimized using the least squares estimation algorithm, and the fuzzy rule base for nitrogen concentration control and nitrogen flow control was established.

[0026] Specifically, the establishment of a fuzzy rule base for nitrogen concentration control includes: defining the domain and fuzzy division of input variables e1 and e1': the domain of nitrogen concentration error e1 is [-5%, 5%], which is divided into 7 fuzzy sets: NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), PB (positive large). The domain of nitrogen concentration error change rate e1' is [-1% / s, 1% / s], which is divided into 5 fuzzy sets: NB (negative large), NS (negative small), ZO (zero), PS (positive small), PB (positive large). The Gaussian membership function is used to represent the fuzzy sets A1 and B1. The least squares estimation algorithm is used to optimize the linear function coefficients of the fuzzy rules: Based on the historical operating data of the carbon molecular sieve adsorption tower, the input-output data pair is constructed: (e 1i ,e 1i ';u 1i ),i=1,2,.......,n. For each fuzzy rule, establish the least squares estimation objective function: Solve the minimum value of the objective function J and obtain the optimal coefficients p, q and t1. Construct a fuzzy rule base for nitrogen concentration control: According to the fuzzy division of e1 and e1', 35 fuzzy rules are formed. For example: If e1 is NB and e1' is NB, then: u 1i =p1e1+q1e1'+t1; if e1 is NB and e1' is NS, then: u1=p2e1+q2e1'+t2; ...; if e1 is PB and e1' is PB, then: u1=p 35 e1+q 35 e1'+t 35 ; Substitute the optimal coefficients p, q and t obtained by least squares estimation into the corresponding fuzzy rules to form a complete fuzzy rule base for nitrogen concentration control.

[0027] Specifically, a fuzzy rule base for nitrogen flow control is established, and the domain and fuzzy division of input variables e2 and e2' are defined: the domain of nitrogen flow error e2 is [-10%, 10%], which is divided into 7 fuzzy sets: NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), PB (positive large). The domain of nitrogen flow error change rate e2' is [-2% / s, 2% / s], which is divided into 5 fuzzy sets: NB (negative large), NS (negative small), ZO (zero), PS (positive small), PB (positive large). The Gaussian membership function is used to represent the fuzzy sets A2 and B2. The linear function coefficients of the fuzzy rules are optimized using the least squares estimation algorithm: According to the historical operating data of the carbon molecular sieve adsorption tower, the input-output data pair is constructed: (e 2i ,e 2i ';u 2i ),i=1,2,.......,n. For each fuzzy rule, establish the least squares estimation objective function: Solve the minimum value of the objective function J and obtain the optimal coefficients r, s and t2. Construct a fuzzy rule base for nitrogen flow control: According to the fuzzy division of e2 and e2', 35 fuzzy rules are formed, for example: if e2 is NB and e2' is NB, then: u2 = r1e2 + s1e2' + t1; if e2 is NB and e2' is NS, then: u2 = r2e2 + s2e2' + t2; ...; if e2 is PB and e2' is PB, then:

[0028] u2=r 35 e2+s 35 e2'+t 35 ; Substitute the optimal coefficients r, s and t obtained by least squares estimation into the corresponding fuzzy rules to form a complete fuzzy rule base for nitrogen flow control.

[0029] Collect the nitrogen concentration and nitrogen flow data output by the carbon molecular sieve adsorption tower, calculate the nitrogen concentration error e1, the nitrogen concentration error change rate e1', the nitrogen flow error e2 and the nitrogen flow error change rate e2' as the input layer neurons of the ANFIS controller; the ANFIS controller obtains the optimal control quantity of the intake valve opening u1 and the vacuum pump exhaust rate u2 through membership function μ calculation and fuzzy rule reasoning; specifically, activate the fuzzy rules in the reasoning layer of the ANFIS controller to calculate the membership of the input variables: according to the actual values ​​of the input variables e1, e1', e2 and e2', use the Gaussian membership function to calculate the membership of the corresponding fuzzy sets μ(e1), μ(e1'), μ(e2) and μ(e2'). The form of the Gaussian membership function is: Where c is the mean and σ is the standard deviation. Activate the corresponding fuzzy rules: Calculate the trigger strength w of each fuzzy rule according to the membership of the input variables i For the i-th fuzzy rule “Ife1 isA1_iande1'isB1_iande2 isA2_iande2'isB2_i”, then: u1=p i e1+q i e1'+t i , u2=r i e2+s i e2'+t i , its trigger strength is:

[0030] w i =μ(e1,A 1i )×μ(e1',B 1i )×μ(e2,A 2i )×μ(e2',B 2i ); trigger strength w i Indicates the contribution of this rule to the final control amount. Calculate the output of each fuzzy rule: For the i-th fuzzy rule, according to the optimized linear function coefficient p i ,q i 、r i 、s i and t i , calculate its output u 1i and u 2i :u 1i =p i e1+q i e1'+t 1i ;

[0031] u 2i =r i e2+s i e2'+t 2i Repeat the above steps to calculate the output of all activated fuzzy rules. Perform weighted summation to obtain the final control quantity: According to the trigger strength w of each fuzzy rule i and output u 1i and u 2i , calculate the final control quantities u1 and u2: The weighted sum operation considers the importance of each rule and obtains a comprehensive control decision.

[0032] Furthermore, the error back propagation algorithm is used to optimize the mean and variance of the membership function, including: initializing the mean c of the membership function ij and variance σ ij, where i is the number of the membership function, i = 1, 2, ..., m; j is the dimension of the input variable, j = 1, 2, ..., n; m is the number of membership functions, n is the dimension of the input variable; according to the mean value c of the membership function ij and variance σ ij , calculate the ANFIS controller for the kth group of input data (e 1k ,e 1k ',e 2k ,e 2k ') output value (u 1k ,u 2k ), and compared with the actual output value of the carbon molecular sieve adsorption tower under the corresponding input data The error between k1 ,ε k2 ): Among them, (e 1k ,e 1k ',e 2k ,e 2k ') is the kth group of input data in the historical operation data of the carbon molecular sieve adsorption tower, k = 1, 2, ..., N; N is the number of groups of historical operation data of the carbon molecular sieve adsorption tower; Enter data for the carbon molecular sieve adsorption tower in group k (e 1k ,e 1k ',e 2k ,e 2k ') under the actual output value; using the error (ε k1 ,ε k2 ) to the output value of the ANFIS controller (u 1k ,u 2k ) to perform error back propagation and calculate the mean value c of the membership function ij and variance σ ij The correction value Δc ij and Δσ ij : Among them, η is the learning rate, which is used to control the step size of parameter correction; according to the correction amount Δc ij and Δσ ij , for the mean value c of the membership function ij and variance σ ij To update: c ij (t+1)=c ij (t)+Δc ij ; σ ij (t+1)=σ ij (t)+Δσ ij ; Where t is the number of iterations, t = 0, 1, 2, ..., T; T is the maximum number of iterations; c ij (t) and σij (t) is the mean and variance of the membership function at the tth iteration; repeat the iteration until the error (ε k1 ,ε k2 ) is less than the preset error threshold ε, or the number of iterations t reaches the maximum number of iterations T, completing the membership function mean c ij and variance σ ij Optimization.

[0033] Furthermore, the least squares estimation algorithm is used to optimize the linear function coefficients of the Sugeno type fuzzy inference rules, including: calculating the membership degree μ of the antecedent condition of the lth rule l (e1,e1',e2,e2'), as the applicability w of the first rule l :w l =μ l (e1,e1',e2,e2'); where l is the number of the rule, l = 1,2,......,r; r is the total number of fuzzy rules; μ l (e1, e1', e2, e2') is the membership function of the antecedent condition of the first rule, which is calculated by minimum operation or product operation: in, is the membership function of the fuzzy set corresponding to the i-th input variable in the antecedent of the l-th rule, i = 1, 2, 3, 4; according to the applicability w l The linear function coefficient p of Sugeno type fuzzy inference rule lj and r l After weighting, the weighted average expression of the ANFIS controller output value is obtained:

[0034] Among them, p lij and r li is the linear function coefficient of the first rule, i = 1, 2; j = 1, 2, 3, 4; using the N sets of input and output data in the historical operation data of the carbon molecular sieve adsorption tower Construct the least squares estimation optimization objective function J: Among them, (u 1k ,u 2k ) is the ANFIS controller according to the kth group of input data (e 1k ,e 1k ',e 2k ,e 2k ') The calculated output value, k = 1, 2, ..., N; is the actual output value of the carbon molecular sieve adsorption tower under the kth group of input data; for the linear function coefficient p in the objective function J lij and r liFind the partial derivative, set it equal to 0, and establish the least squares estimation equation system: Solve the least squares estimation equations and obtain the linear function coefficient p of the Sugeno type fuzzy inference rule lij and r li The optimized value of is used to complete the parameter optimization of the fuzzy rule base.

[0035] Compared with the prior art, the advantages of this application are:

[0036] The filter group is used to filter the compressed air in multiple stages, including stainless steel sintered felt filter element, polyester fiber electrostatic filtration and PTFE nano-membrane filtration, which can effectively remove solid particles with particle sizes larger than d1, d2 and d3. At the same time, the dryer removes moisture and the micro-oil adsorber absorbs residual oil. The three-stage filtration can fully purify the compressed air, provide high-quality raw gas for the subsequent pressure swing adsorption nitrogen production, and improve the purity of nitrogen from the source.

[0037] Carbon molecular sieve adsorbent has a stronger adsorption capacity for oxygen and water molecules. By applying a variable pressure in the adsorption tower, the carbon molecular sieve adsorbent preferentially adsorbs O2 and H2O, while N2 molecules are enriched at the outlet of the adsorption tower, thereby obtaining high-purity nitrogen.

[0038] Pressure and temperature sensors are installed at the inlet and outlet of the carbon molecular sieve adsorption tower to collect the state parameters in the tower in real time. According to the carbon molecular sieve adsorption performance model, the adaptive neural fuzzy control ANFIS algorithm is used to calculate the optimal adsorption operating point online. By adjusting the opening of the air inlet valve and the vacuum pump speed, the adsorption process is always in the optimal state, ensuring high concentration, high yield and low energy consumption of nitrogen.

[0039] The self-learning optimization algorithm improves the accuracy of the ANFIS controller and enhances the nitrogen concentration control capability. The ANFIS controller uses the error back propagation algorithm and the least squares estimation algorithm, and uses the historical operation data of the carbon molecular sieve adsorption tower to perform self-learning optimization on the membership function and the fuzzy rule base. By learning the past optimal control rules, the control accuracy of the ANFIS controller is continuously improved, thereby more accurately controlling the carbon molecular sieve adsorption process, reducing nitrogen concentration fluctuations, and stabilizing at the target value.

[0040] An online nitrogen concentration detection device is installed in the grain warehouse to monitor the nitrogen purity in the warehouse in real time. When the concentration is lower than the target value, the system starts the circulation pump to pump the mixed gas in the warehouse to the carbon molecular sieve adsorption tower to re-purify it to a high concentration, and then transport it back to the warehouse, minimizing the loss of nitrogen exhaust, reducing the energy consumption of nitrogen production, and continuously maintaining a high-purity nitrogen environment in the warehouse. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0042] Figure 1 is a schematic diagram of a low-energy nitrogen-filled grain storage system according to some embodiments of the present application;

[0043] Figure 2 It is a partial schematic diagram of a low-energy nitrogen-filled grain storage system according to some embodiments of the present application;

[0044] Figure 3 It is an exemplary flow chart of a low-energy nitrogen-filled grain storage method shown in some embodiments of the present application. DETAILED DESCRIPTION

[0045] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0046] like Figure 1 As shown, this embodiment provides a low-energy nitrogen-filled grain storage system, which is installed on a remote-controlled tractor with a load capacity of 10 tons. The system includes a nitrogen grain storage and ring filling integrated machine, and its gas production capacity is 220Nm 3 / h. The parameters of the raw gas (air / circulating high-concentration nitrogen) are: pressure 0.1-0.6MPa, dew point ≤-10~-70℃, oxygen content ≤20.9%. The model of the nitrogen storage and ring filling machine is HTHR-AB220, and the output nitrogen parameters are: dew point ≤-70℃, oxygen content ≤0.1%, pressure 0.1-0.6MPa, temperature ≤25℃ (adjustable). The nitrogen filling efficiency parameters are: voltage 380V50HZ, power 66.5KW, storage capacity 2500 tons, target concentration 99%, full storage pressure 200Pa, and nitrogen filling time 35h.

[0047] Specifically, the low-energy nitrogen-filled grain storage system includes the following units: The working process of the compression unit is to compress the air to a preset pressure. The unit is mainly composed of components such as an air compressor, an air buffer tank, a high-pressure gas drainage component and auxiliary instruments. The air compressor is driven by an electric motor to suck in air under atmospheric pressure and compress it to the required pressure. The compressed air first enters the air buffer tank, which plays a role in stabilizing and buffering the pressure and reducing pressure fluctuations. The high-pressure gas drainage component is used to remove condensed water from the compressed air to prevent moisture from entering the subsequent system. Accessory instruments such as pressure gauges and thermometers are used to monitor the state parameters of the compressed air.

[0048] The function of the pretreatment unit is to purify and dry the compressed air to ensure the quality of the raw gas entering the nitrogen production unit. The unit mainly includes a filter group, a dryer, a micro-oil adsorber and corresponding auxiliary instruments. The filter group uses filter elements with different pore sizes to remove solid impurities such as particulate matter and dust in the air through inertial collision, screening and other mechanisms. The dryer uses adsorbents such as molecular sieves or silica gel to remove moisture from the air and reduce the dew point through physical adsorption. The micro-oil adsorber is filled with adsorbent materials such as activated carbon, which can absorb pollutants such as oil mist and hydrocarbons in the air. The auxiliary instruments are used to detect the dew point, oil content and other indicators of the treated air. The control system switches or regenerates the adsorber in time according to the feedback from the instrument to maintain a continuous and stable purification effect.

[0049] like Figure 2 As shown in the figure, the nitrogen production unit is a key unit for separating and producing high-purity nitrogen from compressed air using the principle of pressure swing adsorption. It is mainly composed of two carbon molecular sieve adsorption towers (Tower A and Tower B), programmable valves and auxiliary instruments. Carbon molecular sieve is a material with unique selective adsorption properties. The pore size distribution is 0.3-1.0nm, and it has significantly different adsorption capacities for nitrogen and oxygen. When compressed air is passed into the adsorption tower, under a certain pressure, the carbon molecular sieve preferentially adsorbs oxygen and water vapor molecules, while nitrogen molecules pass through the sieve bed and flow out from the top of the tower to obtain high-purity nitrogen products. When one adsorption tower is saturated, it is switched to another for adsorption, and the saturated tower is regenerated by depressurization to desorb oxygen and achieve continuous nitrogen production. The programmable valve automatically switches the adsorption-regeneration state of the two towers according to the set timing. The auxiliary instrument monitors the pressure, temperature and other parameters in the tower, and feeds the data back to the control unit to optimize the adsorption process.

[0050] The function of the nitrogen recooling unit is to cool the high-temperature nitrogen generated by the nitrogen production unit to the required temperature to meet the requirements of nitrogen filling and grain storage. The unit is mainly composed of an adjustable cold drying system and corresponding auxiliary instruments. The adjustable cold drying system usually adopts a closed refrigeration cycle, which is composed of components such as a compressor, a condenser, an expansion valve and an evaporator. The high-temperature nitrogen exchanges heat with the refrigerant in the evaporator to reduce the temperature to a preset value. The cooled nitrogen is transported to the downstream unit after the condensate is removed by the gas-liquid separator. By adjusting the opening of the expansion valve, the refrigerant flow rate can be changed, and then the outlet temperature of the nitrogen can be regulated. Auxiliary instruments such as temperature sensors and pressure gauges are used to monitor the temperature and pressure parameters of nitrogen and provide a basis for regulation and control.

[0051] The concentration detection unit is used to detect the purity of the output nitrogen in real time online to ensure that it meets the concentration requirements for nitrogen-filled grain storage. The core component of this unit is the oxygen analyzer, which uses sensors such as zirconium oxide batteries to accurately measure the oxygen concentration based on the difference in electromotive force caused by the components of the nitrogen-oxygen mixed gas, and then converts the nitrogen purity. The gas concentration automatic detection system also includes an automatic control selector and a programmable valve. The selector periodically draws a small amount of nitrogen sample from the delivery pipeline according to the set time interval or trigger signal, sends it to the oxygen analyzer for measurement, and feeds back the analysis data to the control unit. The programmable valve is used to control the entry and exit of the sample gas. At the same time, the oxygen analyzer is equipped with a calibration device, which is regularly calibrated with a standard gas of known concentration to ensure the accuracy and reliability of the measurement results.

[0052] The control unit is the brain of the entire system, coordinating the operation of each unit and optimizing the nitrogen filling process. Among them, the intelligent filling loop control system automatically determines whether cyclic nitrogen production is required based on the nitrogen purity data fed back by the concentration detection unit. When it is detected that the output nitrogen concentration is lower than the preset target value, the control unit issues a cyclic nitrogen production command, starts the axial flow fan, and sucks the mixed gas in the closed granary into the nitrogen production unit for pressure swing adsorption treatment again. At the same time, open the program-controlled valve to add the newly produced high-purity nitrogen into the granary, continuously replacing the residual oxygen until the nitrogen concentration in the silo meets the standard. The axial flow fan can be selected according to the size of the granary and the requirements of the nitrogen filling rate, and the suction process can be optimized through variable frequency speed regulation. The program-controlled valve performs precise timing and quantitative control to balance the addition of new nitrogen and the extraction of residual oxygen. The auxiliary instrument provides granary environmental parameters and assists in control.

[0053] The function of the conveying unit is to continuously, safely and reliably convey the high-purity nitrogen produced by the nitrogen production unit to the granary. It is mainly composed of components such as pipelines, valves, flow meters and nitrogen sealing devices. The pipeline material is made of stainless steel, which is resistant to high pressure and low temperature, and is well insulated to reduce heat loss during transportation. The valve is used to adjust the flow and pressure of nitrogen to ensure a stable and continuous supply of nitrogen. The flow meter monitors the instantaneous and cumulative nitrogen filling amount to provide data support for the control unit. The nitrogen sealing device is installed at the entrance of the granary. When the nitrogen is turned on, high-pressure nitrogen is used to form an air bag at the entrance to prevent outside air from entering. At the same time, it cooperates with the pressure balance valve to prevent a large amount of nitrogen from leaking out of the granary, thereby reducing unnecessary losses and improving nitrogen filling efficiency.

[0054] like Figure 3As shown, a low-energy nitrogen-filled grain storage system and method include: compressing air to a preset pressure range through an air compressor; the air compressor can be selected from screw, piston or centrifugal types, and is selected according to the system scale and pressure requirements. The parameters of the compressed air meet the air intake requirements of the subsequent pretreatment unit. The compressed air is sequentially passed into the filter group, dryer and micro-oil adsorber for purification and drying; the high-pressure air is passed into the self-cleaning primary filter unit, and a double-layer stainless steel sintered felt filter element is used to remove particles with a particle size greater than 50μm in the high-pressure air. At the same time, part of the high-pressure air flow is used to back-blow and unclog the filter element, which effectively prevents the filter element from being blocked and extends its service life. The pressure difference data and back-blow frequency data of the self-cleaning primary filter unit are compared with the preset threshold value, and the filter flow and back-blow intensity of the high-pressure air are dynamically adjusted according to the comparison results to achieve optimized control of the primary filtration and obtain the high-pressure air for the first filtration;

[0055] The high-pressure air filtered for the first time is passed into the electrostatic dust-removing intermediate filter unit, and a polyester fiber filter material with permanent static electricity is used to remove dust particles with a particle size greater than 5μm in the high-pressure air filtered for the first time through the electrostatic effect. The low-pressure pulse back-blowing cleaning technology is combined to realize the online cleaning and regeneration of the filter element. The pressure difference data and pulse back-blowing frequency data of the electrostatic dust-removing intermediate filter unit are compared with the preset threshold value. The filtration flow rate and pulse back-blowing intensity of the high-pressure air filtered for the first time are dynamically adjusted according to the comparison result to realize the optimized control of the intermediate filtration and obtain the high-pressure air filtered for the second time; the high-pressure air filtered for the second time is passed into the nanofiber sub-high efficiency filter unit, and nano-grade PTFE fiber membrane filter material is used to remove submicron particles with a particle size greater than 0.01μm in the high-pressure air filtered for the second time, and the pressure difference data of the nanofiber sub-high efficiency filter unit is collected in real time. When the pressure difference data reaches the preset replacement threshold, the automatic replacement of the filter element is controlled to always maintain the best filtration performance, and finally ultra-pure high-pressure air with a cleanliness better than level 1 in the ISO 8573-1 standard is obtained.

[0056] The ultra-pure high-pressure air is dried and then passed through a rotary adsorption dryer. The moisture in the air is adsorbed by a molecular sieve rotor with high adsorption performance. The saturated molecular sieve rotor is continuously regenerated by a regeneration heating system. The dew point data of the inlet and outlet air of the rotor are detected, and the rotation speed and regeneration temperature of the rotor are dynamically adjusted to achieve constant dew point control. The dew point of the ultra-pure high-pressure air is stabilized below -40°C to obtain dry ultra-pure high-pressure air. The high-pressure air treated by the dryer is passed into the micro-oil adsorber, which is filled with a lipophilic adsorbent to remove oil mist, oil droplets and other oil pollutants in the high-pressure air by physical adsorption. The lipophilic adsorbent is a hydrophobically modified activated carbon adsorbent, which has a strong affinity with oil molecules and an adsorption capacity of more than 100 mg / g for oil molecules. By detecting the pressure difference data of the micro-oil adsorber, when the pressure difference reaches the set replacement threshold, the control system is started to replace the micro-oil adsorber to always maintain the best adsorption performance. After being treated by the micro-oil adsorber, the residual oil concentration in the high-pressure air is less than 0.01 mg / m 3 , meeting the air cleanliness requirements of the pressure swing adsorption nitrogen production process.

[0057] The pretreated compressed air is input into the carbon molecular sieve adsorption tower, and the oxygen O2 molecules and water H2O molecules in the air are stripped by the carbon molecular sieve pressure swing adsorption tower to obtain nitrogen N2 molecular gas; including: the pretreated compressed air is passed into the air inlet of the carbon molecular sieve adsorption tower, and the oxygen O2 molecules and water H2O molecules in the air are adsorbed by the carbon molecular adsorbent in the carbon molecular sieve adsorption tower to enrich the nitrogen N2 molecules; first, the pressure sensors are installed at the air inlet and outlet of the carbon molecular sieve adsorption tower respectively, and the inlet pressure P_in and outlet pressure P_out of the adsorption tower are collected in real time. The selected pressure sensor has a measurement range of 01.6MPa, an accuracy level of 0.5, and an output signal of 420mA current signal. At the same time, a platinum resistance temperature sensor Pt100 is installed on the outer wall of the tower body of the adsorption tower to monitor the operating temperature T_tower of the adsorption tower. The measurement range of the temperature sensor is -50~200℃, and the accuracy level is A.

[0058] In order to optimize the design and operation control of the adsorption tower, a mathematical model of the nitrogen production process of the carbon molecular sieve adsorption tower was established. The model is based on the following assumptions: the carbon molecular sieve adsorbent has different adsorption selectivity for nitrogen and oxygen; the adsorption process conforms to the Langmuir adsorption isotherm and the linear drive mechanical model; the pressure gradient and temperature gradient in the adsorption tower are ignored. The control equation of the mathematical model is: Among them, q z It indicates the instantaneous adsorption amount of component z (nitrogen or oxygen) on the carbon molecular sieve under nitrogen storage conditions, in mol / kg; represents the equilibrium adsorption capacity of component z under this condition, in mol / kg; kz is the adsorption kinetic rate constant of component z under nitrogen filling temperature and pressure conditions, in units of 1 / s; p z is the partial pressure of component z in the adsorption tower, in bar; p0 is the standard atmospheric pressure, with a value of 1.01325 bar; n z is the non-ideal factor of the adsorption isotherm of component z, dimensionless; t is the adsorption time, unit is s. The equilibrium adsorption amount of component z Calculated by the following formula: In the formula, q m It indicates the sum of the maximum adsorption capacity of nitrogen and oxygen by carbon molecular sieve at nitrogen filling temperature, in mol / kg; b j represents the Langmuir adsorption equilibrium constant of component z, with the unit of 1 / bar; j represents the adsorbed component (nitrogen or oxygen).

[0059] By collecting the real-time data of the inlet and outlet pressures and the temperature of the adsorption tower, combined with the known carbon molecular sieve adsorption performance parameters (such as q m 、b j 、n j , k z By substituting the above mathematical model, the concentration distribution of nitrogen and oxygen in the adsorption tower and the change of adsorption amount over time can be solved. Based on this, the structural dimensions of the adsorption tower (such as tower height H and tower diameter D), operating parameters (such as inlet pressure P in , switching period t c etc.), and predict the nitrogen production performance and service life of the adsorption tower.

[0060] According to the collected inlet pressure, outlet pressure and tower temperature, combined with the preset adsorption performance curve of the carbon molecular sieve adsorbent, the adaptive neural fuzzy inference control ANFIS algorithm is used to calculate the optimal adsorption operation point of the carbon molecular sieve adsorption tower, and dynamically adjust the air inlet valve opening of the carbon molecular sieve adsorption tower and the vacuum pumping rate. Specifically, pressure sensors P1 and P2 are installed at the air inlet and outlet of the carbon molecular sieve adsorption tower respectively to collect the inlet pressure P of the adsorption tower in real time. in and outlet pressure P out Among them, P1 and P2 use high-precision pressure sensors with a range of 0-1.0MPa, an accuracy of ±0.1%FS, and a response time of less than 100ms.

[0061] A temperature sensor T is installed on the outer wall of the carbon molecular sieve adsorption tower to collect the tower body temperature T of the adsorption tower in real time. towerThe temperature sensor T uses a Pt100 platinum resistance temperature sensor with a temperature measurement range of -50 to 200°C, an accuracy of ±0.5°C, and a response time of less than 1s. The data P_in, P_out, and T_tower collected by the pressure sensors P1, P2, and the temperature sensor T are transmitted to the control system of the adsorption tower in real time through the RS-485 bus or 4 to 20mA analog signals. The control system uses an embedded industrial computer based on the ARMCortex-A9 architecture, equipped with 2GB RAM and 16GB Flash storage, running the Linux operating system and the self-developed adsorption tower control software.

[0062] P in , P out and T tower As the input variable of the mathematical model of the adsorption tower, combined with the preset carbon molecular sieve adsorbent for nitrogen and oxygen under different temperature (such as -2050℃) and pressure (such as 0.10.8MPa) conditions, the optimal adsorption operating pressure P of the adsorption tower is calculated in real time using the ANFIS algorithm. opt and temperature T opt Among them, the carbon molecular sieve adsorbent adopts CMS-220 model, with a specific surface area of ​​220m 2 / g, pore size distribution is 0.3 to 0.45nm, and the adsorption selectivity for nitrogen and oxygen can reach 812. The adsorption equilibrium curve q*=f(T,P) is fitted using the Sips equation, and the fitting formula is: where q m is the saturated adsorption capacity, b is the adsorption equilibrium constant, and n is the non-ideal factor, all of which are functions of temperature T.

[0063] Define the input variables of the ANFIS controller as: Nitrogen concentration error: e1 = C out -C set , indicating the actual output nitrogen concentration C of the adsorption tower out With the target concentration C set (such as 99.99%) deviation. Nitrogen concentration error change rate: Indicates the change rate of nitrogen concentration error between two adjacent sampling moments (sampling period Δt = 1s). Nitrogen flow error: e2 = F out -F set , indicating the actual nitrogen flow rate F output from the adsorption tower out With the set target flow F_set (such as 1000Nm 3 / h). Nitrogen flow error change rate: Indicates the nitrogen flow error change rate between two adjacent sampling moments. out and F outThe data are collected online in real time by a nitrogen concentration analyzer (such as a laser Raman spectrometer) and a thermal gas mass flowmeter at the outlet of the adsorption tower.

[0064] The output variables of the ANFIS controller are defined as follows: the opening of the adsorption tower air inlet valve u1: adjusted from 0 to 100% to control the air intake of the adsorption tower. The vacuum pump pumping speed u2: adjusted from 1000 to 5000 rpm to control the vacuum desorption process of the adsorption tower. Among them, the air inlet valve and vacuum pump are respectively precisely controlled by the electric regulating valve (accuracy ±1%) and the variable frequency speed control module (accuracy ±0.5%) to achieve the precise control of the opening and speed.

[0065] The membership function and fuzzy inference rule base of ANFIS controller are constructed. Each input variable e1, e1', e2, e2' of ANFIS controller is divided into 7 fuzzy subsets, which are denoted as NB (Negative Big), NM (Negative Medium), NS (Negative Small), ZO (Zero), PS (Positive Small), PM (Positive Medium), PB (Positive Big), indicating different degrees of error or error change rate.

[0066] The Gaussian function is used as the membership function of each fuzzy subset. The general form of the Gaussian function is: Where c is the mean and σ is the standard deviation, which respectively determine the center position and the degree of widening of the membership function. For example, for the 7 fuzzy subsets of e1, the following parameters can be set: NB: c = -3, σ = 0.8; NM: c = -2, σ = 0.5; NS: c = -1, σ = 0.5; ZO: c = 0, σ = 0.4; PS: c = 1, σ = 0.5; PM: c = 2, σ = 0.5; PB: c = 3, σ = 0.8; Similarly, the corresponding c and σ parameters can be set for the fuzzy subsets of e1', e2, and e2'. These parameters can be initially given based on expert experience or trial and error, and will be further optimized during the ANFIS training process.

[0067] Construct a Sugeno type fuzzy inference rule base, where each rule consists of an antecedent (IF part) and a consequent (THEN part). The antecedent consists of fuzzy subsets of the four input variables connected by AND (intersection) operations, and the consequent consists of a first-order linear function that represents the relationship between the controller outputs u1 and u2 and the input variables. Assuming that each input variable has M fuzzy subsets, the total number of rules is M 4 Take one of the rules as an example: IF e1 isNS and e1'isPS and e2 isNMande2'isZO, then u1=p 11 ×e1+p 12×e1'+p 13 ×e2+p 14 ×e2'+r1,u2=p 21 ×e1+p 22 ×e1'+p 23 ×e2+p 24 ×e2'+r2; where p 11 p 14 、p 21 p 24 Enter the rule

[0068] The linear coefficients of functions u1 and u2 with respect to the four input variables are obtained, r1 and r2 are constant terms, and their initial values ​​can be obtained by fitting the training data using the least squares method, which will be further adjusted and optimized during the ANFIS training process.

[0069] Based on the membership function and fuzzy rule base defined above, the 5-layer feedforward network structure of ANFIS is used to implement fuzzy reasoning and defuzzification operations. The functions of each layer of nodes are as follows: Layer 1: Calculate the membership of the input variable to each fuzzy subset. Layer 2: Calculate the trigger strength of each rule (i.e. the truth value of the antecedent). Layer 3: Normalize the rule trigger strength. Layer 4: Calculate the output function value of each rule. Layer 5: Sum up the output of each rule to obtain the defuzzified output results u1 and u2 of the controller.

[0070] A hybrid learning algorithm is used to train and optimize the ANFIS controller, including two processes: forward propagation and back propagation. In forward propagation, the optimal values ​​of the linear parameters p and r of each rule output function are estimated by the least squares method based on the input-output training data pairs. In back propagation, the rule output function parameters are fixed, and the nonlinear parameters c and σ of the membership function are fine-tuned by the gradient descent method to minimize the mean square error between the ANFIS output and the expected output. Through multiple rounds of iterative training, the reasoning performance of the ANFIS controller can be continuously improved, better fitting the nonlinear dynamic characteristics of the carbon molecular sieve adsorption tower, and realizing intelligent optimization control of the adsorption process.

[0071] The error back propagation algorithm is used to optimize the mean and variance of the membership function, including: initializing the mean c of the membership function ij and variance σ ij , where the number of membership functions m is 7, corresponding to 7 fuzzy subsets NB, NM, NS, ZO, PS, PM, PB; the dimension of the input variable n is 4, corresponding to 4 input variables e1, e1', e2, e2'. The initial value can be set according to the following experience: c ij The values ​​are -3, -2, -1, 0, 1, 2, and 3, respectively, indicating that the membership function is uniformly distributed on the domain [-3, 3]. ijThe uniform value is 0.6, which means that the shapes of the membership functions overlap moderately to smooth the boundaries. From the historical operation database of the carbon molecular sieve adsorption tower, N = 1000 sets of input-output data samples (e 1k ,e 1k ',e 2k ,e 2k ';u 1k *,u 2k *), where k = 1, 2, ..., 1000. Each set of samples is collected and recorded by the adsorption tower in actual production operation, which can truly reflect the dynamic characteristics of the adsorption process. According to the membership function parameter c of the current tth iteration ij (t) and σ ij (t), calculate the ANFIS controller for the kth group of input samples (e 1k ,e 1k ',e 2k ,e 2k ') output value (u 1k ,u 2k ), and the actual output value corresponding to the sample (u 1k *,u 2k *) comparison, the output error is: ε 1k =u 1k *-u 1k ; ε 2k =u 2k *-u 2k ; Use the error back propagation algorithm to solve the membership function parameter c ij and σ ij Correction amount: Among them, η is the learning rate, which is 0.01; is the gradient of the error with respect to the controller output and can be calculated directly;

[0072] The derivation of composite functions involving the ANFIS feedforward network requires the use of the chain rule to solve them layer by layer.

[0073] Use the correction amount to update the membership function parameters to prepare for the next iteration: ij (t+1)=c ij (t)+Δc ij ; σ ij (t+1)=σ ij (t)+Δσ ij ; The learning rate η has a long control effect at the beginning. If the value is too large, it may cause parameter oscillation divergence; if the value is too small, the convergence speed is slow. After completing the parameter learning of all N groups of samples, compare the output error ε k1 , ε k2The absolute value of and the preset threshold ε = 0.005, or check whether the number of iterations t reaches the maximum value T = 100: If |ε k1 |<ε and |ε k2 |<ε, or t=T, then the optimization is terminated and the optimal parameter c of the membership function is output ij (t) and σ ij (t); otherwise, set t'=t+1 to continue a new round of iteration. Through the continuous iterative optimization of the error back propagation algorithm, the parameters of the membership function can be dynamically adapted to the actual working conditions of the carbon molecular sieve adsorption tower, thereby improving the reasoning and decision-making performance of the ANFIS controller and realizing intelligent adaptive control of the adsorption process.

[0074] The linear function coefficients of the Sugeno type fuzzy inference rules are optimized by using the least squares estimation algorithm, including: determining the total number of fuzzy rules r=m according to the fuzzy rule base constructed in the above steps n =7 4 =2401, where m=7 is the number of fuzzy subsets for each input variable, and n=4 is the dimension of the input variable. For each set of input data (e 1k ,e 1k ',e 2k ,e 2k '), calculate the antecedent membership of each fuzzy rule μ1(e 1k ,e 1k ',e 2k ,e 2k '), as the applicability of the rule w lk . Take the l=1000th rule as an example: μ A1000,1 (e 1k ) means e lk The degree of belonging to the first fuzzy subset (such as NM) in the antecedent of the 1000th rule; μ A1000,2 (e 1k ') represents e 1k 'The degree of belonging to the second fuzzy subset (such as PS) in the antecedent of the 1000th rule; μ A1000,3 (e 2k ) means e 2k The degree of belonging to the third fuzzy subset (such as NS) in the antecedent of the 1000th rule; μ A1000,4 (e 2k ') represents e 2k 'The degree of belonging to the fourth fuzzy subset (such as ZO) in the antecedent of the 1000th rule; then

[0075] w 1000,k =μ A1000,1 (e 1k )×μ A1000,2 (e1k ')×μ A1000,3 (e 2k )×μ A1000,4 (e 2k '); introduce w lk As the contribution weight of rule l to the kth group of data, establish ANFIS output u 1k 、u 2k The weighted average expression of is:

[0076]

[0077] Among them, p l11 to p l14 、r l1 is the coefficient of the output function u1 of the lth rule with respect to the input variable, p l21 to p l24 、r l2 is the coefficient of the output function u2, which needs to be determined through optimization. lk It plays a weighted role in the matching degree of rule l.

[0078] Introduce performance index J to measure ANFIS output (u 1k ,u 2k ) and the expected output (u 1k *,u 2k *)’s mean square error: Among them, (u 1k ,u 2k ) is calculated by the weighted average formula and the fuzzy rule coefficient p lij 、r li Related; (u 1k ,u 2k ) is the actual output of the carbon molecular sieve adsorption tower under the kth group of input, which is the known sample data. The optimization goal is to solve a set of p lij 、r li , so that the mean square error J is minimized. For the objective function J, respectively, lij 、r li Find the partial derivative and set it equal to 0, and get the value about p lij 、r li The least squares estimating equations for are:

[0079] Since the unknown number p in the system of equations lij 、r li The number of equations is far less than 1000, which is an overdetermined system of equations. The least square method can be used to find its approximate solution. Write the system of equations in matrix form Ax = b, where:

[0080] According to the least squares method, the optimal solution x of the equation system Ax=b satisfies: x=(A T A) -1 A T b, the fuzzy rule parameter p obtained by least squares estimation lij 、r li Substituting the weighted average formula of ANFIS output into the optimized fuzzy inference system, it can be used for dynamic modeling and predictive control of carbon molecular sieve adsorption tower.

[0081] Collect the nitrogen concentration and nitrogen flow data output by the carbon molecular sieve adsorption tower, calculate the nitrogen concentration error e1, nitrogen concentration error change rate e1', nitrogen flow error e2 and nitrogen flow error change rate e2' as the input layer neurons of the ANFIS controller; install a nitrogen concentration sensor (measuring range 0100%, accuracy ±0.1%) and a nitrogen flow sensor (measuring range 05000Nm^3 / h, accuracy ±1%) at the outlet of the adsorption tower to collect the concentration of nitrogen discharged from the top of the tower in real time c N2 (t) and flow rate f N2 (t).

[0082] The nitrogen concentration setting value c is sent by the host computer N2 * = 99.99% (ultra-high purity requirement) and nitrogen flow setting value f N2 *=1000Nm 3 / h (to meet the downstream nitrogen demand). Calculate the concentration error e1(t) = c N2 *-c N2 (t), concentration error change rate Flow error e2(t) = f N2 *-f N2 (t), flow error change rate The sampling interval Δt = 1min. Input e1(t), e1'(t), e2(t), e2'(t) into the 4 input neurons of the ANFIS controller. Taking e1 as an example, set the domain [-1%, 1%], fuzzy subsets {NB, NM, NS, ZO, PS, PM, PB}, and membership function μ NB (e1)~μ PB (e1) uses Gaussian function. Perform similar fuzzification on e1', e2, e2'. Design r fuzzy control rules, such as "IF e1 isAande1'isB ande2 isCand e2'isD, THE Nu1=f1(e1,e1',e2,e2'), u2=f2(e1,e1',e2,e2')", where A, B, C, D are fuzzy subsets in the antecedent, f1, f2 are Sugeno type output functions, including the parameters to be optimized p lij and rli . Calculate the applicability of each rule Defuzzification calculation of the optimal values ​​of valve opening u1 and pumping rate u2: Among them, f l1 =p l11 ×e1+p l12 ×e1'+p l13 ×e2+p l14 ×e2'+r l1 ,

[0083] f l2 =p l21 ×e1+p l22 ×e1'+p l13 ×e2+p l24 ×e2'+r l2 .

[0084] The actuator adjusts and transmits the valve opening u1 to the electric actuator of the air inlet valve, and transmits the pumping rate u2 to the variable frequency speed control device of the vacuum pump. The opening of the air inlet valve is directly controlled by u1 (value 0-100%). The larger the opening, the greater the air intake. The pumping rate of the vacuum pump is proportional to u2 (value 0-100%). The higher the frequency, the greater the pumping rate, and the higher the vacuum degree in the tower. By adjusting the air intake and pumping rate in real time, the adsorption tower can still maintain the nitrogen concentration c under dynamic disturbance. N2 and flow rate f N2 Stable at the set value c N2 and f N2 nearby.

[0085] Mathematical model correction, the mathematical model of the carbon molecular sieve adsorption tower is as follows: i=O2、N2;q i * = b i ×p i ; Among them, q i is the adsorption amount of component i, q i * is the adsorption equilibrium amount, p i is the partial pressure of component i, b i is the equilibrium constant, k i is the kinetic rate constant. The real-time operation data of the adsorption tower is collected online. N2 (t), f N2 (t), and then substitute it into the mathematical model to obtain the theoretical value q N2 (t), q O2 (t). Compare the measured value with the theoretical value and construct the objective function where q N2 *、q O2 *Based on c N2 (t), f N2(t) The actual adsorption amount is inferred. Solved by the least squares method Get the corrected b i , k i Parameters can be used to improve the prediction accuracy of the model. The correction frequency can be adjusted according to the changes in the operating conditions of the adsorption tower (such as switching cycle, feed concentration, etc.), generally 1 to 7 days.

[0086] Nitrogen cooling and output, the nitrogen N2 produced by the nitrogen preparation unit (such as pressure swing adsorption, membrane separation, etc.) is passed into the heat exchanger of the cold drying system. The heat exchanger uses refrigerant (such as R22, R134a, etc.) as a cold source to cool the N2 gas temperature to a preset range, usually 0 to 5°C, to reduce the dew point of nitrogen. The cooled N2 gas passes through the steam-water separator of the cold drying system to further remove condensed water to obtain dry cold N2 gas with higher purity and lower moisture content. The dry cold N2 gas is transported to downstream process sections such as oxygen analyzers and closed warehouses through pipelines.

[0087] Nitrogen concentration detection, install an oxygen analyzer at the outlet of the cold drying system to detect the concentration of residual O2 in the treated N2 gas in real time. The oxygen analyzer adopts paramagnetic, electrochemical or optical principles, with a measurement range of 0 to 1000ppm, an accuracy of ±1ppm, and a response time of <30s. Set the preset concentration range of N2 gas, usually 99.99%100% (i.e. O2 concentration 0100ppm) to meet the gas purity requirements for nitrogen-filled grain storage. Compare the detection value of the oxygen analyzer with the preset range to determine whether the quality of the N2 gas is qualified.

[0088] Cyclic adsorption regeneration, continuously monitor the purity of N2 gas at the outlet of the cold drying system, and obtain real-time data through the oxygen analyzer. Set the lower limit of the preset concentration range of N2 gas to 99.99%, and the corresponding upper limit of O2 concentration to 100ppm. When the oxygen analyzer detects that the N2 concentration is lower than 99.99% (that is, the O2 concentration is higher than 100ppm), it indicates that the N2 gas produced by the cold drying system does not meet the standard and needs to be cyclically adsorbed and regenerated. Start the circulation pump to pump the mixed gas in the closed warehouse to the carbon molecular sieve pressure swing adsorption device. The composition of the mixed gas is N2 and a small amount of O2, of which the N2 concentration is usually above 99% and the O2 concentration is below 1%. According to the warehouse volume and the processing capacity of the adsorption device, set the pumping rate of the circulation pump, usually 1000~5000m^3 / h. The mixed gas passes through the pre-processor (such as filter, cold dryer, etc.) to remove impurities such as dust and moisture, and then enters the carbon molecular sieve adsorption tower.

[0089] The multi-tower pressure swing adsorption process is adopted to achieve the selective adsorption and desorption of N2 and O2 by the carbon molecular sieve through the periodic change of pressure. Usually, a 2-tower or 3-tower switching operation mode is adopted to alternately adsorb and regenerate to achieve continuous separation. Take 2 towers as an example: Adsorption stage: the mixed gas enters the adsorption tower under high pressure (0.40.8MPa), the carbon molecular sieve preferentially adsorbs O2 molecules, and the N2 molecules pass through the packing layer and are enriched and discharged from the top of the tower. The adsorption process lasts for 35 minutes until the outlet O2 concentration rises to the set value (such as 1000ppm). Regeneration stage: close the air inlet valve, open the vacuum valve, and evacuate the adsorption tower under low pressure (0.010.03MPa) to desorb the adsorbed O2 molecules from the carbon molecular sieve and discharge them from the system through the vacuum pump. The regeneration process lasts for 515 minutes until the pressure in the tower drops to the set value. Switching stage: the adsorption tower and the regeneration tower exchange roles, and adsorption and regeneration are carried out again, over and over again, to achieve continuous operation. The tower filler is made of carbon molecular sieve, with a specific surface area of ​​8001200m^2 / g, a pore size of 0.30.4nm, and a selectivity of O2 / N2 greater than 4. The amount of filler is calculated according to the intake volume and adsorption capacity, generally 1 to 3 tons / tower. An oxygen analyzer is set at the outlet of the adsorption tower to monitor the purity of the N2 gas after regeneration in real time. During the cyclic adsorption regeneration process, as the O2 impurities in the mixed gas are continuously removed, the N2 concentration at the outlet of the adsorption tower gradually increases. When the outlet N2 concentration reaches the lower limit of the preset range (99.99%), the cyclic adsorption process can be stopped and the enriched N2 gas can be directly transported to the closed warehouse. If the time of one cycle adsorption is long (such as >30min), the cycle process can be manually terminated before the oxygen analyzer monitors the qualified concentration to reduce energy consumption. The carbon molecular sieve adsorbent is activated and regenerated regularly (such as heating, vacuuming, etc.) to restore its adsorption performance and extend its service life.

[0090] Grain storage with nitrogen is to pass high-purity N2 gas output by the cold drying system into the closed warehouse to replace the grain with nitrogen and keep it fresh. The nitrogen filling process is divided into two stages: initial nitrogen filling and circulating nitrogen filling: Initial nitrogen filling: evacuate the air in the warehouse to -0.02MPa, then introduce N2 gas until the pressure in the warehouse returns to normal pressure, repeat the vacuum-N2 filling process 3 to 5 times until the O2 concentration drops below 1%. Circulating nitrogen filling: Under closed conditions, continuously introduce N2 gas, and at the same time extract an equal amount of mixed gas through the circulation pipeline to maintain the O2 concentration in the warehouse at 0.5% to 1%, and the N2 concentration at 99% to 99.5%. During the nitrogen filling process, the gas composition and pressure stability in the warehouse are controlled by adjusting the input flow rate of N2 gas and the circulating air extraction flow rate. Regularly monitor the O2 concentration, temperature and humidity in the warehouse and other parameters to ensure that the nitrogen grain storage environment meets the standard requirements.

Claims

1. A low energy consumption nitrogen-filled grain storage system, characterized in that: include: Compression unit, compresses air to a preset pressure; The pre-treatment unit is used to purify and dry the compressed air. It uses a filter group to remove solid impurities in the air, a dryer to remove moisture in the air, and a micro-oil adsorber to remove oil in the air; The nitrogen production unit uses a carbon molecular sieve adsorption tower to perform pressure swing adsorption on the pre-treated compressed air, stripping the O2 and H2O molecules in the air to obtain N2 molecular gas; The nitrogen recooling unit cools the generated N2 molecular gas to the preset temperature through the cold drying system and outputs it; The concentration detection unit uses an oxygen analyzer to detect whether the output N2 molecular gas concentration meets the preset target concentration range; The control unit, when detecting that the output nitrogen concentration is lower than the preset target concentration lower limit, generates a circulating nitrogen production instruction, starts the circulating pump to pump the mixed gas in the closed warehouse to the nitrogen production unit for pressure swing adsorption nitrogen production again, until the output nitrogen concentration reaches the preset target concentration range; The conveying unit transports nitrogen reaching the target concentration to the closed warehouse for nitrogen filling and grain storage.

2. A low-energy nitrogen-filled grain storage method, characterized in that: include: Compress the air to a preset pressure range through an air compressor; The compressed air is sequentially passed through the filter group, dryer and micro-oil adsorber for purification and drying; wherein the filter group removes solid impurities in the air, the dryer removes moisture in the air, and the micro-oil adsorber removes oil stains in the air to obtain pre-treated compressed air; The pretreated compressed air is input into a carbon molecular sieve adsorption tower, and the carbon molecular sieve pressure swing adsorption tower is used to strip oxygen O2 molecules and water H2O molecules in the air to obtain nitrogen N2 molecular gas; Inputting nitrogen N2 molecular gas into the cold drying system, cooling it to a preset temperature range, and outputting the cooled nitrogen N2 molecular gas; Use an oxygen analyzer to detect whether the concentration of the output nitrogen N2 molecular gas meets the preset concentration range; When the detected nitrogen N2 molecular gas concentration is lower than the lower limit of the preset concentration range, the circulation pump is started to pump the mixed gas in the closed warehouse to the carbon molecular sieve pressure swing adsorption tower for adsorption and nitrogen production again, until the output nitrogen N2 molecular gas concentration meets the preset concentration range; Nitrogen N2 molecular gas that meets the preset concentration range is transported to the closed warehouse for nitrogen filling and grain storage operations.

3. The low-energy nitrogen-filled grain storage method according to claim 2, characterized in that: Use filter groups to remove solid impurities from the air, including: A double-layer stainless steel sintered felt filter element is used to remove particles with a diameter greater than d1 in the compressed air to obtain the first filtered air; Polyester fiber filter material is used to remove particles with a diameter larger than d2 in the first filtered air through electrostatic effect to obtain the second filtered air; Nano-PTFE fiber membrane filter material is used to remove particles with a particle size greater than d3 in the second filtered air to obtain air with solid impurities removed.

4. The low-energy nitrogen-filled grain storage method according to claim 3, characterized in that: The micro oil absorber removes oil from the air, including: The compressed air treated by the dryer is passed into a micro-oil adsorber, which contains a lipophilic adsorbent to remove oil mist and oil droplets in the compressed air; The lipophilic adsorbent includes an activated carbon adsorbent.

5. The low-energy nitrogen-filled grain storage method according to any one of claims 2 to 4, characterized in that: The pre-treated compressed air is input into the carbon molecular sieve adsorption tower, and the carbon molecular sieve pressure swing adsorption tower is used to strip oxygen O2 molecules and water H2O molecules in the air, including: The pretreated compressed air is passed into the air inlet of the carbon molecular sieve adsorption tower, and the carbon molecular adsorbent in the carbon molecular sieve adsorption tower adsorbs oxygen O2 molecules and water H2O molecules in the air to enrich nitrogen N2 molecules; Pressure sensors are respectively arranged at the air inlet and the air outlet of the carbon molecular sieve adsorption tower to collect the inlet pressure and the outlet pressure of the carbon molecular sieve adsorption tower; and a temperature sensor is arranged on the outer wall of the tower body of the carbon molecular sieve adsorption tower to collect the tower body temperature of the carbon molecular sieve adsorption tower; According to the collected inlet pressure, outlet pressure and tower temperature, combined with the preset adsorption performance curve of the carbon molecular sieve adsorbent, the adaptive neural fuzzy inference control ANFIS algorithm is used to calculate the optimal adsorption operating point of the carbon molecular sieve adsorption tower, and dynamically adjust the air inlet valve opening of the carbon molecular sieve adsorption tower and the exhaust rate of the vacuum pump.

6. The low-energy nitrogen-filled grain storage method according to claim 5, characterized in that: Dynamically adjust the air inlet valve opening of the carbon molecular sieve adsorption tower and the vacuum pumping rate, including: According to the nitrogen adsorption capacity of carbon molecular sieve adsorbent, the selectivity of nitrogen and oxygen, and the adsorption and desorption rate constants, a mathematical model of the nitrogen production process of carbon molecular sieve adsorption tower is established; The inlet pressure, outlet pressure and tower temperature of the carbon molecular sieve adsorption tower are used as input variables of the mathematical model; Adopting adaptive neural fuzzy inference control ANFIS algorithm, taking the nitrogen concentration and nitrogen output of the carbon molecular sieve adsorption tower as the control target, the optimal adsorption operation point of the carbon molecular sieve adsorption tower is calculated, and the optimal control value of the air inlet valve opening and vacuum pump exhaust rate of the carbon molecular sieve adsorption tower is output; According to the optimal control amount of the air inlet valve opening and the vacuum pump exhaust rate, the air inlet valve opening of the carbon molecular sieve adsorption tower is adjusted, and the exhaust rate of the vacuum pump is adjusted; The nitrogen concentration and nitrogen flow rate output from the carbon molecular sieve adsorption tower are collected, and the adsorption equilibrium constant b in the mathematical model is calculated using the least squares method based on the detected nitrogen concentration and nitrogen flow rate. j and the adsorption kinetic rate constant k z Make corrections.

7. The low-energy nitrogen-filled grain storage method according to claim 6, characterized in that: The mathematical model expression is as follows: Among them, q z is the dynamic adsorption amount of component z under nitrogen storage conditions; is the equilibrium adsorption amount of component z under nitrogen storage conditions; k z is the adsorption kinetic rate constant of component z at the nitrogen storage temperature and pressure; p z is the partial pressure of component z under nitrogen storage conditions; p0 is the standard atmospheric pressure; q m is the sum of the maximum adsorption capacity of carbon molecular sieve for N2 and O2 at nitrogen storage temperature; n z is the non-ideal factor of the adsorption isotherm of component z under nitrogen storage conditions; n j is the non-ideal factor of the adsorption isotherm of component j under nitrogen storage conditions; p j is the partial pressure of component j under nitrogen storage conditions; b j is the Langmuir adsorption equilibrium constant of component j at the nitrogen storage temperature; t is the adsorption time.

8. The low-energy nitrogen-filled grain storage method according to claim 7, characterized in that: Adaptive neural fuzzy inference control ANFIS algorithm is used to output the optimal control quantity of the carbon molecular sieve adsorption tower inlet valve opening and vacuum pump exhaust rate, including: Set nitrogen concentration error e1, nitrogen concentration error change rate e1', nitrogen flow error e2 and nitrogen flow error change rate e2' as input variables, and valve opening u1 and vacuum pump exhaust rate u2 as output variables; The ANFIS controller with five-layer neural network is established by using Gaussian membership function μ and Sugeno type fuzzy inference rules. The historical operating data of the carbon molecular sieve adsorption tower is used to optimize the mean c and variance σ of the membership function by adopting the error back propagation algorithm. The least squares estimation algorithm is used to optimize the linear function coefficients p and r of the Sugeno type fuzzy inference rule, and the fuzzy rule base for nitrogen concentration control and nitrogen flow control is established; Collect the nitrogen concentration and nitrogen flow rate data output by the carbon molecular sieve adsorption tower, calculate the nitrogen concentration error e1, nitrogen concentration error change rate e1', nitrogen flow rate error e2 and nitrogen flow rate error change rate e2' as the input layer neurons of the ANFIS controller; The ANFIS controller obtains the optimal control variables of the intake valve opening u1 and the vacuum pump exhaust rate u2 through membership function μ calculation and fuzzy rule reasoning.

9. The low-energy nitrogen-filled grain storage method according to claim 8, characterized in that: The error back propagation algorithm is used to optimize the mean and variance of the membership function, including: Initialize the mean c of the membership function ij and variance σ ij , where i is the number of the membership function, i = 1, 2, ..., m; j is the dimension of the input variable, j = 1, 2, ..., n; m is the number of membership functions, and n is the dimension of the input variable; According to the mean c of the membership function ij and variance σ ij , calculate the ANFIS controller for the kth group of input data (e 1k ,e 1k ',e 2k ,e 2k ') output value (u 1k ,u 2k ), and compared with the actual output value of the carbon molecular sieve adsorption tower under the corresponding input data The error between k1 ,ε k2 ): Among them, (e 1k ,e 1k ',e 2k ,e 2k ') is the kth group of input data in the historical operation data of the carbon molecular sieve adsorption tower, k = 1, 2, ..., N; N is the number of groups of historical operation data of the carbon molecular sieve adsorption tower; Enter data for the carbon molecular sieve adsorption tower in group k (e 1k ,e 1k ',e 2k ,e 2k ') under the actual output value; Using the error (ε k1 ,ε k2 ) to the output value of the ANFIS controller (u 1k ,u 2k ) to perform error back propagation and calculate the mean value c of the membership function ij and variance σ ij The correction value Δc ij and Δσ ij : Among them, η is the learning rate, which is used to control the step size of parameter correction; According to the correction amount Δc ij and Δσ ij , for the mean value c of the membership function ij and variance σ ij To update: c ij (t+1)=c ij (t)+Δc ij ;s ij (t+1)=σ ij (t)+Δσ ij Where t is the number of iterations, t = 0, 1, 2, ..., T; T is the maximum number of iterations; c ij (t) and σ ij (t) is the mean and variance of the membership function at the tth iteration; Repeat the iteration until the error (ε k1 ,ε k2 ) is less than the preset error threshold ε, or the number of iterations t reaches the maximum number of iterations T, completing the membership function mean c ij and variance σ ij Optimization.

10. The low-energy nitrogen-filled grain storage method according to claim 9, characterized in that: The least squares estimation algorithm is used to optimize the linear function coefficients of Sugeno type fuzzy inference rules, including: Calculate the membership degree μ of the antecedent condition of the lth rule l (e1,e1',e2,e2'), as the applicability w of the first rule l :w l =μ l (e1,e1',e2,e2') Where l is the number of the rule, l = 1, 2, ..., r; r is the total number of fuzzy rules; μ l (e1, e1', e2, e2') is the membership function of the antecedent condition of the first rule, which is calculated by minimum operation or product operation: μ l (e1,e1',e2,e2')=μ Al1 (e1)∧μ Al2 (e1')∧μ Al3 (e2)∧μ Al4 (e2') Among them, μ Ali is the membership function of the fuzzy set corresponding to the i-th input variable in the antecedent of the l-th rule, i = 1, 2, 3, 4; According to the applicability w l The linear function coefficient p of Sugeno type fuzzy inference rule lj and r l After weighting, the weighted average expression of the ANFIS controller output value is obtained: Among them, p lij and r li is the linear function coefficient of the lth rule, i=1,2; j=1,2,3,4; Using the N sets of input and output data from the historical operation data of the carbon molecular sieve adsorption tower Construct the least squares estimation optimization objective function J: Among them, (u 1k ,u 2k ) is the ANFIS controller according to the kth group of input data (e 1k ,e 1k ',e 2k ,e 2k ') The calculated output value, k = 1, 2, ..., N; is the actual output value of the carbon molecular sieve adsorption tower under the kth group of input data; For the linear function coefficient p in the objective function J lij and r li Find the partial derivative, set it equal to 0, and establish the least squares estimation equation system: Solve the least squares estimation equations and obtain the linear function coefficient p of the Sugeno type fuzzy inference rule lij and r li The optimized value of is used to complete the parameter optimization of the fuzzy rule base.

Images