High-quality biodiesel combustion performance evaluation system

Abstract

The present invention discloses a high-quality biodiesel combustion performance evaluation system, including an experimental database retrieval module, a fitting formula calculation module and a machine learning prediction module. The present invention can predict the combustion characteristics and emission characteristics of a mixed fuel after high-quality biodiesel is mixed with other fuels through the use of the three modules, and can calculate the total error through the error curve to compare the accuracy of the fitting formula calculation and the machine learning prediction to determine the actual application method. The present invention mainly solves the problem of evaluating the combustion performance when high-quality biodiesel is added and applied, so as to provide feedback to guide the biodiesel preparation process and regulate and optimize the biodiesel components. At the same time, this invention is not limited to fuel mixed with high-quality biodiesel, and the combustion performance of any fuel type involved in the system after mixing can be predicted.

Description

Technical Field

[0001] The invention relates to the field of biodiesel combustion performance, and provides a method and a system for predicting fuel combustion performance by using data software. Background Art

[0002] As early as the 1990s, biodiesel has attracted attention for its significant effect in reducing carbon dioxide emissions. However, many problems exist in the use of the first generation of biodiesel, which limits its use as a substitute for petrochemical diesel. In order to solve the problems of the first generation of biodiesel, the second generation of biodiesel, which uses oil as raw material and hydrogenation catalysis as the main production process, has become the focus of research.

[0003] Second-generation biodiesel uses waste oils, animal and plant oils, etc. as raw materials. Compared with first-generation biodiesel using oil crop seeds as raw materials, it does not need to occupy land used for growing food, its chemical structure is closer to fossil diesel, and it has advantages such as higher cetane number and lower sulfur content. Therefore, second-generation biodiesel is regarded as high-quality biodiesel. Among them, the second-generation biodiesel produced by direct hydrodeoxygenation process has a high condensation point and poor low-temperature fluidity, and can generally only be used as a high-cetane number diesel additive component. The high-quality second-generation biodiesel produced by hydrodeoxygenation isomerization process has a higher cetane number, lower condensation point and cloud point, so it has a wider range of uses.

[0004] The blending of high-quality biodiesel with traditional low-temperature petrochemical diesel can improve non-standard low-end diesel to meet national standards, while the blending with national standard diesel can make the engine achieve lower fuel consumption and emissions. In addition, the industrial preparation and market promotion of high-quality biodiesel can reduce dependence on petrochemical diesel, reduce engine exhaust emissions, and contribute to global energy conservation and emission reduction. Therefore, the research on high-quality biodiesel is important and urgent.

[0005] However, since the combustion and emission characteristics of high-quality biodiesel mixed with other fuels are unclear and the experimental cost is high, there is an urgent need to apply new technical means to predict and evaluate the combustion performance of mixed fuels with different components. Summary of the invention

[0006] To solve the above problems, the present invention provides a system for predicting the combustion performance of high-quality biodiesel based on experiments, formulas, and machine learning, which can predict and evaluate the combustion performance of mixed fuels with different components.

[0007] A high-quality biodiesel combustion performance evaluation system uses data software to establish a system including an experimental database retrieval module, a fitting formula calculation module and a machine learning prediction module, and outputs the combustion characteristics and emission characteristics of the mixed fuel to provide feedback to guide the biodiesel preparation process and regulate and optimize the biodiesel components.

[0008] In the above scheme, the experimental database retrieval module includes fuel physical and chemical properties, experimental conditions and fuel combustion performance.

[0009] In the above scheme, the physical and chemical properties of the fuel include fuel types and physical and chemical properties, wherein the fuel types include pure fuel and mixed fuel, fuel types include first-generation biodiesel, second-generation biodiesel-straight chain, second-generation biodiesel-isomerized and diesel for mixing with biodiesel, gasoline, n-dodecane, methanol, ethanol, butanol, pentanol and octanol, and the physical and chemical properties include cetane number, kinematic viscosity, density, sulfur content, condensation point and latent heat of vaporization; wherein the physical and chemical properties of pure fuel are all measured by experiments; the physical and chemical properties of mixed fuel are the data obtained by experimental measurement. For mixed fuels that have not been measured, there are two ways to obtain their physical and chemical properties:

[0010] (1) For the case where this mixing ratio has not been measured but this mixing type has been measured, it is obtained by fitting;

[0011] (2) If such a mixture has not been measured, it is calculated proportionally based on the physical and chemical properties of the pure fuel.

[0012] In the above scheme, the experimental conditions are divided into settable experimental conditions and fixed experimental conditions. The experimental devices for settable experimental conditions uniformly use high-temperature and high-pressure constant-volume combustion bombs, which can simulate the thermodynamic environment of the engine top dead center. The experimental conditions are controllable, and the ambient temperature, ambient density, ambient oxygen content, nozzle diameter and nozzle pressure can be adjusted; the experimental devices for fixed experimental conditions use an engine stand, and the fixed experimental conditions are standard thirteen conditions.

[0013] In the above scheme, fuel combustion performance includes combustion characteristics and emission characteristics. It is stipulated that all values ​​are measured using a unified definition, a unified test method and a unified experimental processing method. The combustion characteristics are obtained through constant volume combustion bomb experimental processing, including liquid phase length, ignition delay period, flame floating length, soot mass and gas phase penetration distance; the emission characteristics are obtained through engine bench tests, including brake specific fuel consumption, CO, HC, PM and NO X The specific emission of the engine bench test is based on the standard 13 working conditions, the definition and measurement of emission characteristics are fixed, and the definition of combustion characteristics is often different. Therefore, the definition, measurement method and experimental processing method of combustion characteristics are unified as follows:

[0014] The spray gas phase and liquid phase contours in the combustion characteristics are obtained by the schlieren method and the shadow method respectively, the ignition delay period is obtained by judging the schlieren image, the flame floating length is photographed by the chemical fluorescence method, and the soot mass is calculated based on the diffuse background light extinction method;

[0015] The liquid phase length is defined as the longest stable distance of the spray liquid phase in the axial direction when the fuel evaporation rate is equal to the fuel injection rate. The ignition delay period is defined as the time interval between the start of injection and the start of high-temperature combustion. The flame floatation length is defined as the average value of the shortest distance between the nozzle position and the upper and lower areas of the stable flame reaction axis, respectively. The gas phase penetration distance is defined as the distance between the nozzle vertex and the farthest point of the spray gas phase contour along the nozzle axis. The soot mass is defined by the calculation formula:

[0016] m soot =ρλAr·KL / k e (1)

[0017] Where ρ represents the soot density, λ represents the wavelength of the incident light, Ar represents the area of ​​the soot region, K represents the spatial extinction coefficient, L represents the optical length of the incident light in the soot, and k e represents the dimensionless extinction coefficient,

[0018] Among them, the liquid phase length, ignition delay period, and flame floating length are all one value in a single spray cycle. During measurement, multiple cycles are performed to eliminate unreliable data first and then take the average value. The gas phase penetration distance and carbon soot mass are a series of values ​​that change with time. During measurement, multiple cycles can also be performed to eliminate unreliable data first and then take the average value according to the time point. The final result is also a series of values.

[0019] In the above scheme, the fitting formula calculation module uses a known fitting formula to predict the combustion characteristics of the mixed fuel, and the combustion characteristics include the fitting of the liquid phase length, the ignition delay period, and the flame floating length:

[0020] Liquid phase length LL:

[0021] Ignition delay period ID:

[0022] Flame floating length LOL:

[0023] Where d is the nozzle diameter, ρ f represents the fuel density, ρ a represents the ambient density, n represents the type of mixed fuel, m i represents the mass fraction, h vap,i represents the latent heat of vaporization, T b,max Indicates the maximum boiling point, T f Indicates the initial fuel temperature, C p,i represents the specific heat of the component, C p,a represents the specific heat of the environment, T a Indicates the ambient temperature, u a represents kinematic viscosity, P injIndicates injection pressure, P a represents the ambient pressure, O2% represents the ambient oxygen content, Z st represents the stoichiometric fuel mass fraction, CN represents the cetane number of the fuel, and A, B, C, D, E, and F represent constants obtained by regression of known experimental data.

[0024] In the above scheme, the machine learning prediction module predicts the combustion performance of fuel by using machine learning. The specific steps are as follows:

[0025] S1: The experimental database used by the experimental database retrieval module is used as the data set;

[0026] S2: According to the input and output, it is divided into four models, and the data is preliminarily screened and normalized respectively, and divided into training set and test set. The prediction model is trained with the training set: Model 1 is used to predict the liquid phase length, ignition delay period and flame floatation length in the spray combustion characteristics of fuel under different experimental conditions; Model 2 is used to predict the emission characteristics of fuel under standard thirteen working conditions; Model 3 is used to predict the gas phase penetration distance of fuel; Model 4 is used to predict the carbon soot quality of fuel, among which the machine learning used in Model 1 and Model 2 is extreme learning machine, and the machine learning used in Model 3 and Model 4 is deep neural network;

[0027] S3: Use the coefficient of determination, mean square error, and mean absolute percentage error as evaluation indicators to adjust the hyperparameters in the model until the results meet the requirements.

[0028] S4: The trained model can be used to predict the combustion performance of fuel.

[0029] In the above scheme, the search results in the experimental database are used as the priority output. If no corresponding results are matched, the total error size when using the fitting formula and the machine learning method is calculated through the error curve, and the method with the smallest error is finally applied. Regarding the error curve:

[0030] Use control variables to analyze all input parameters, including various fuel components and experimental conditions, calculate the fitting value of the fitting formula or the difference between the predicted value of machine learning and the experimental value, and fit the error curve. There are m1 types of fuel in the database, m2 adjustable experimental conditions, and three outputs of liquid phase length, ignition delay period, and flame floating length. The fitting formula and machine learning error curves each have n = (m1 + m2) × 3. When a group of data that has not been experimentally measured is input, first, the values ​​corresponding to each parameter of the group of data are calculated through the fitting formula and the n error curves of machine learning to obtain n error values, and the m1 + m2 error values ​​of liquid phase length, ignition delay period, and flame floating length are summed and calculated respectively. At the same time, weights are introduced to consider the influence of each input parameter on the result, such as calculating the total error of the liquid phase length:

[0031]

[0032]

[0033] Among them, a represents the calculation of the fitting formula, b represents the prediction of machine learning, p represents the weight of each input parameter, and p i represents the weight of the i-th input parameter, e represents the error value calculated by the error curve, and e LL,ai represents the error value of the i-th parameter when calculating the liquid phase length using the fitting formula, e LL,bi Represents the error value of the i-th parameter when predicting the liquid phase length using machine learning.

[0034] In the above scheme, the experimental database retrieval module, the fitting formula calculation module or the machine learning prediction module is selected according to whether there is experimental data. When the combustion characteristics and emission characteristics of the mixed fuel are output, the experimental database retrieval module, the fitting formula calculation module or the machine learning prediction module and the method used in the system during this operation are explained at the same time. The use of the experimental database retrieval module corresponds to the matching method, the use of the fitting formula calculation module corresponds to the calculation method, and the use of the machine learning prediction module corresponds to the prediction method.

[0035] In the above scheme, there is no true value to compare with when outputting the fitting formula calculation or machine learning prediction results, and the accuracy of the two cannot be known. Therefore, all data in the experimental database are calculated using the same formula or predicted using the same model, and compared with the actual experimental results as a reference for accuracy.

[0036] The beneficial effects brought about by the above technical solution of the present invention include:

[0037] The present invention establishes a system including three modules: experimental database retrieval, fitting formula calculation, and machine learning prediction. The total error is calculated by error curve to compare the accuracy of fitting formula calculation and machine learning prediction to determine the actual application method, thereby realizing the rapid output of the combustion characteristics and emission characteristics of the mixed fuel. In addition, the system can also provide feedback on the prediction of combustion performance to guide the biodiesel preparation process and regulate and optimize the biodiesel components.

[0038] The present invention mainly solves the problem of evaluating the combustion performance when adding high-quality biodiesel, so as to provide feedback to guide the biodiesel preparation process and regulate and optimize the biodiesel components. At the same time, the present invention is not limited to fuel mixed with high-quality biodiesel, and the combustion performance of any fuel type involved in the system after mixing can be predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1A flow chart of the high-quality biodiesel combustion performance evaluation system;

[0040] Figure 2 Schematic diagram of extreme learning machine;

[0041] Figure 3 Schematic diagram of the model for predicting the spray combustion characteristics of fuel under different experimental conditions;

[0042] Figure 4 A schematic diagram of a model for predicting the emission characteristics of fuel under standard 13 operating conditions;

[0043] Figure 5 Schematic diagram of a deep neural network;

[0044] Figure 6 It is the process diagram of fitting error curve;

[0045] Figure 7 Process diagram for applying error curve;

[0046] Figure 8 This is the ambient temperature error curve of the fitting formula when calculating the liquid phase length. DETAILED DESCRIPTION

[0047] A system for high quality biodiesel combustion performance,

[0048] First, determine the types of biodiesel, including second-generation biodiesel-straight chain and second-generation biodiesel-isomeric, and also take first-generation biodiesel into account for comparison. Second, determine the common types of fuel used for blending, including diesel, gasoline, n-dodecane, methanol, ethanol, butanol, pentanol, and octanol. Then, determine the corresponding physical and chemical properties of various pure fuels, including cetane number, kinematic viscosity, density, sulfur content, condensation point, latent heat of vaporization, etc.

[0049] Mixed fuel refers to fuel obtained by adding different types of fuel in different proportions. Regarding the acquisition of its physical and chemical properties, the data obtained from experimental determination are preferred. For mixed fuels that have not been determined, there are two ways to obtain their physical and chemical properties:

[0050] (1) For the case where the mixed component has not been measured but the mixed species has been measured, it is obtained by fitting;

[0051] (2) If such a mixture has not been measured, it is calculated proportionally based on the physical and chemical properties of the pure fuel.

[0052] Collect all relevant experimental data to create an experimental database, which includes the physical and chemical properties of various fuels, the spray combustion characteristics of various fuels under different experimental conditions, and the emission characteristics of various fuels using standard thirteen working conditions. The experimental conditions include ambient temperature, ambient density, ambient oxygen content, nozzle diameter, nozzle pressure, spray combustion characteristics include liquid phase length, ignition delay period, flame floatation length, gas phase penetration distance, soot mass, and emission characteristics include BFSC (Brake Specific Fuel Consumption), CO, HC, PM, NO X Specific emissions.

[0053] This database is not only used as a basis for retrieval, but also as a basis for fitting formula regression and a data set for machine learning. The determination of the fitting formula and the training of the model can be completed after the experimental database is established.

[0054] The spray combustion characteristics under high temperature and high pressure static environment in the database are measured using unified definitions and test methods, and the experimental processing methods are also required to be consistent. The experimental device uniformly uses a high temperature and high pressure constant volume combustion bomb, which can simulate the thermodynamic environment of the engine top dead center, and the experimental conditions are highly controllable. The spray gas phase and liquid phase contours in the combustion characteristics are obtained by the schlieren method and the shadow method respectively, the ignition delay period is obtained by schlieren image judgment, the flame floating length is photographed by the chemical fluorescence method, and the soot mass is calculated based on the diffuse background light extinction method.

[0055] The liquid phase length is defined as the longest stable distance of the spray liquid phase in the axial direction when the fuel evaporation rate is equal to the fuel injection rate. The ignition delay period is defined as the time interval between the start of injection and the start of high-temperature combustion. The flame floatation length is defined as the average value of the shortest distance between the nozzle position and the upper and lower areas of the stable flame reaction axis, respectively. The gas phase penetration distance is defined as the distance between the nozzle vertex and the farthest point of the spray gas phase contour along the nozzle axis. The soot mass is defined by the calculation formula:

[0056] m soot =ρλAr·KL / k e (1)

[0057] Where ρ represents the soot density, λ represents the wavelength of the incident light, Ar represents the area of ​​the soot region, K represents the spatial extinction coefficient, L represents the optical length of the incident light in the soot, and k e represents the dimensionless extinction coefficient.

[0058] The liquid phase length, ignition delay period, and flame float length are all one value in a single spray cycle. During measurement, multiple cycles are performed to eliminate unreliable data and then take the average value. The gas phase penetration distance and soot mass are a series of values ​​that change with time. During measurement, multiple cycles can also be performed to eliminate unreliable data and then take the average value at each time point. The final result is also a series of values.

[0059] The emission characteristics in the database are also measured using unified definitions and test methods, and the experimental equipment uniformly uses engine benches and standard thirteen operating conditions.

[0060] After the database is established, the constants in the fitting formula can be regressed and the machine learning model can be trained.

[0061] The fitting formula used in this method is as follows:

[0062] Liquid phase length LL:

[0063] Ignition delay period ID:

[0064] Flame floating length LOL:

[0065] Where d is the nozzle diameter, ρ f represents the fuel density, ρ a represents the ambient density, n represents the type of mixed fuel, m i represents the mass fraction, h vap,i represents the latent heat of vaporization, T b,max Indicates the maximum boiling point, T f Indicates the initial fuel temperature, C p,i represents the specific heat of the component, C p,a represents the specific heat of the environment, T a Indicates the ambient temperature, u a represents kinematic viscosity, P inj Indicates injection pressure, P a represents the ambient pressure, O2% represents the ambient oxygen content, Z st represents the stoichiometric fuel mass fraction, CN represents the cetane number of the fuel. A, B, C, D, E, and F represent constants, whose values ​​are obtained by regression of known experimental data.

[0066] The application of machine learning in this method is divided into two types. The first type predicts a single value, such as the liquid phase length, ignition delay period, flame floatation length and emission characteristics in the spray combustion characteristics. The second type predicts a series of values ​​that change over time, such as the gas phase penetration distance and soot mass. In the present invention, different algorithms are used for these two applications. The first type uses an extreme learning machine, and the second type uses a deep neural network.

[0067] The extreme learning machine used in the first application is a neural network structure, including an input layer, a single hidden layer and an output layer, such as Figure 2 The output formula of the extreme learning machine is:

[0068]

[0069] h i (X) = g(w i X+b i ) (6)

[0070] Where Y represents the output of the extreme learning machine, Y = [y1,...,y m ] T , m represents m output values, X represents the input of the extreme learning machine, X=[x1,...,x n ] T , n represents n output values, β i represents the output weight between the i-th node in the hidden layer and the output layer, h i represents the output of the i-th node, g(·) represents the activation function, and w i represents the weight between the input layer and the i-th node of the hidden layer, b i Represents the deviation between the input layer and the i-th node in the hidden layer.

[0071] w in the model i and b i To generate values ​​randomly, the key to making the model predict well is to determine β, and its calculation formula is:

[0072] β=H′T (7)

[0073] Among them, β is the weight connecting the hidden layer and the output layer, H′ is the Moore–Penrose generalized inverse matrix of the hidden layer output, and T is the sample label.

[0074] In the application of this method, the extreme learning machine is divided into two models according to the difference in input and output. Figure 3 The schematic diagram of the model used to predict the spray combustion characteristics of fuel under different experimental conditions is shown in Figure 1. The input is the physical and chemical properties of the fuel and the experimental conditions, and the output is the liquid phase length, ignition delay period, and flame floatation length in the spray combustion characteristics. Figure 4 The schematic diagram of the model for predicting the emission characteristics of fuel under standard thirteen working conditions, whose input is the physical and chemical properties of the fuel and the output is the emission characteristics, is taken as model two.

[0075] The deep neural network used in the second application includes an input layer, multiple hidden layers, and an output layer. Figure 5As shown in the figure, it is a deep neural network consisting of an input layer, three hidden layers, and one output layer, where the number of nodes in the hidden layers are 5, 3, and 4 respectively.

[0076] The steps to build a deep neural network model are as follows:

[0077] (1) Building a neural network, including determining the number of input and output values, the number of hidden layers and the number of nodes in each hidden layer, and the activation function used;

[0078] (2) Initialize model parameters, including weights and biases of connections between layers;

[0079] (3) Forward loop: After the input value is calculated through the initial weight and bias value, the output is obtained;

[0080] (4) Calculate the loss: Calculate the error between the actual output obtained in the third step and the theoretical output through the loss function;

[0081] (5) Reverse loop: Starting from the output, calculate the partial derivative of the output with respect to the weights and biases of each layer;

[0082] (6) Update parameters: Calculate new weights and bias values ​​based on partial derivatives to replace the original values;

[0083] (7) Integration: Set multiple iterations, that is, repeat steps 3 to 6 until the error between the neural network output value and the theoretical output value is less than a given value;

[0084] (8) The neural network is used for prediction.

[0085] In this method, deep neural network is used to predict gas phase penetration distance and soot mass. Therefore, the input is the value of physical and chemical properties and experimental conditions, and the output is gas phase penetration distance and soot mass corresponding to different time points. The model used to predict gas phase penetration distance is model three, and the model used to predict soot mass is model four.

[0086] The model prediction accuracy uses the following indicators:

[0087]

[0088]

[0089]

[0090] R 2 is the determination coefficient, whose value is between 0 and 1. The closer it is to 1, the better the model prediction performance.

[0091] MSE is the mean square error, and its value is between 0 and positive infinity. The closer it is to 0, the better the model prediction performance.

[0092] MAPE is the mean absolute percentage error, and its value is between 0 and positive infinity. The closer it is to 0, the better the model prediction performance.

[0093] Where N represents the number of data points. Represents the predicted value, y i represents the true value, Represents the mean of the true values.

[0094] The comparison between the fitting formulas of liquid phase length, ignition delay period, and flame floatation length and machine learning is achieved through error calculation, such as Figure 6 and Figure 7 As shown, the specific steps are as follows:

[0095] (1) Substitute the data of all experiments in the database into the fitting formula to calculate the fitting value, which is used as the prediction value for machine learning input;

[0096] (2) Use control variables to analyze all input parameters, including various fuel components and experimental conditions, calculate the difference between the fitted value or predicted value and the experimental value, and fit the error curve, such as Figure 8 This is the ambient temperature error curve of the fitting formula when calculating the liquid phase length. There are m1 types of fuel in the database, m2 adjustable experimental conditions, and three outputs: liquid phase length, ignition delay period, and flame floating length. The fitting formula and machine learning error curves are n = (m1 + m2) × 3, respectively.

[0097] (3) The above two steps are preparatory work. When a set of data that has not been experimentally measured is input, the numerical values ​​corresponding to each parameter of the set of data are first calculated through the fitting formula and the n error curves of machine learning to obtain n error values. The m1+m2 error values ​​of the liquid phase length, ignition delay period, and flame floating length are summed and calculated respectively. At the same time, the weight is introduced to consider the influence of each input parameter on the result. Taking the liquid phase length as an example:

[0098]

[0099]

[0100] Among them, a represents the calculation of the fitting formula, b represents the prediction of machine learning, p represents the weight of each input parameter, and e represents the error value calculated by the error curve.

[0101] (4) Taking the liquid phase length as an example, compare S LL,a and S LL,b The size of , and the method with the smaller sum is taken as the method for practical application.

[0102] Finally, the three modules of the whole system are determined, namely database, fitting formula, and machine learning. The functions that can be realized by each module are matched to the given conditions. The application process is as follows Figure 1 As shown, the specific steps are as follows:

[0103] (1) Input the proportion of each fuel in the mixed fuel

[0104] (2) Determine the physical and chemical properties of the mixed fuel

[0105] ① Experimental values ​​are preferred;

[0106] ② If there is no experimental measurement, it can be obtained by fitting or proportional calculation.

[0107] (3) Choose to use a constant volume incendiary bomb or an engine test bench

[0108] ① If you choose a constant volume combustion bomb, you also need to enter the experimental conditions;

[0109] ② If you select the engine test bench, the working conditions will be fixed.

[0110] (4) In the constant volume incendiary bomb part:

[0111] ①First, match the experimental database. If there is experimental data, output the experimental data, including liquid phase length, ignition delay period, flame floating length, gas phase penetration distance, and soot mass;

[0112] ②If the experimental data cannot be matched, then:

[0113] a) Liquid phase length, ignition delay period, and flame rise length are determined based on the total error value using a fitting formula or a machine learning model.

[0114] b) Gas phase penetration distance and soot mass are predicted by machine learning three and four methods;

[0115] c) Output the results of the mixed fuel calculated by the fitting formula or predicted by machine learning under specific experimental conditions, and at the same time output the results of the mixed fuel calculated by the fitting formula or predicted by machine learning for all experimental data for reference and comparison.

[0116] (5) On the engine stand:

[0117] ①First, match the experimental database. If there is experimental data, the experimental data will be output, including BFSC, CO, HC, PM, NO X Specific emissions;

[0118] ② If the experimental data cannot be matched, the machine learning model 2 is used for prediction, and the output results include BFSC, CO, HC, PM, NO XAt the same time, the prediction results of all mixed fuels with experimental data through this model are output for reference and comparison.

[0119] The system is described below with reference to examples.

[0120] Example 1: Calculate the physical and chemical properties of a mixed fuel of second-generation biodiesel-straight chain as a test fuel and pentanol at a volume ratio of 80:20, and there is no data on the mixture of second-generation biodiesel-straight chain and pentanol in the database.

[0121] In this case, the density of the second generation biodiesel-straight chain is 786kg / m 3 The density of amyl alcohol is 815 kg / m 3 , then the density of the mixed fuel is:

[0122]

[0123] Among them, V represents the volume occupied by a certain fuel in the mixed fuel, ρ represents the density of a certain fuel in the mixed fuel, i represents the different types of fuel in the mixed fuel, and n represents the common types of fuel in the mixed fuel.

[0124] Other physical and chemical properties such as cetane number, kinematic viscosity, etc. are calculated in the same way.

[0125] Example 2: Using a constant volume combustion bomb, it is necessary to output the combustion characteristics and set two situations.

[0126] Setting 1: The experiment has measured the physical and chemical properties of the mixed fuel when the second-generation biodiesel-straight chain is used as the test fuel, mixed with methanol and n-octanol with a volume ratio of 58:25:17, and the relevant data on the corresponding combustion performance under the given experimental conditions.

[0127] Setting situation 2: The experiment did not measure the physical and chemical properties of the mixed fuel when the second-generation biodiesel-straight chain was used as the test fuel and mixed with methanol and n-octanol in a volume ratio of 70:15:15, and the relevant data on the corresponding combustion performance under the given experimental conditions.

[0128] In case 1, if reliable experimental data exists, it is directly used as the prediction result. The result includes the liquid phase length, ignition delay period, flame floating length, soot mass and gas phase penetration distance in the form of single values.

[0129] Case 2: There is no experimental data. For liquid phase length, ignition delay period, and flame floatation length, compare the total error of the fitting formula and machine learning and select the method with the smaller total error for output. For soot mass and gas phase penetration distance, directly use machine learning to predict the output. The steps are as follows:

[0130] For the liquid phase length, ignition delay period, and flame floating length, the values ​​corresponding to each parameter of this set of data are calculated through fitting formulas and n error curves of machine learning to obtain n error values, and the m1+m2 error values ​​of the liquid phase length, ignition delay period, and flame floating length are summed and calculated respectively, without considering the weight of each parameter. Calculated:

[0131] S LL,a >S LL,b ,S ID,a >S ID,b ,S LOL,a <S LOL,b (2)

[0132] Therefore, the liquid phase length and ignition delay period are calculated by fitting formulas, and the flame rise length is predicted by machine learning.

[0133] Fitting formula to determine liquid phase length and ignition delay period:

[0134]

[0135]

[0136] Substituting the required values ​​into the formula, the results including liquid phase length and ignition delay period can be easily obtained.

[0137] Then, machine learning model 1 is used to predict the flame lift length value.

[0138] For gas phase penetration distance and soot mass, the gas phase penetration distance is predicted by model three, and the soot mass is predicted by model four.

[0139] Finally, the results calculated by the fitting formula or predicted by machine learning are output, and the results predicted by the fitting formula or machine learning for all mixed fuels with experimental data are also output.

[0140] Example 3 uses an engine bench and needs to output emission characteristics. Two situations are also set.

[0141] Setting 1: The experiment has measured the physical and chemical properties and combustion performance of the mixed fuel when the second-generation biodiesel-straight chain is used as the test fuel and mixed with diesel at a volume ratio of 20:80.

[0142] Setting situation 2: The experiment did not measure the relevant data on the physical and chemical properties and combustion performance of the mixed fuel when the second-generation biodiesel-straight chain was used as the test fuel and mixed with diesel at a volume ratio of 40:60.

[0143] In case 1, if reliable experimental data exists, it is directly used as the prediction result. The result is presented as a single value of specific emissions, including BFSC, CO, HC, PM, NOX Specific emissions.

[0144] In case 2, if there is no experimental data, the machine learning model 2 is used to predict the specific emissions, and the prediction results of all mixed fuels with experimental data through this model are output at the same time.

[0145] Matters not covered in the present invention are known technologies. The above examples are only for illustrating the technical concept and features of the present invention. Their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly. They cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A high-quality biodiesel combustion performance evaluation system, characterized in that: Use data software to establish a system including an experimental database retrieval module, a fitting formula calculation module and a machine learning prediction module, and output the combustion characteristics and emission characteristics of the mixed fuel to provide feedback to guide the biodiesel preparation process and regulate and optimize the biodiesel components; The experimental database retrieval module includes fuel physical and chemical properties, experimental conditions and fuel combustion performance; The fitting formula calculation module uses the known fitting formula to predict the combustion characteristics of the mixed fuel. The combustion characteristics include the fitting of liquid phase length, ignition delay period, and flame floating length: Liquid phase length LL: Ignition delay period ID: Flame floating length LOL: Where d is the nozzle diameter, ρ f represents the fuel density, ρ a represents the ambient density, n represents the type of mixed fuel, m i represents the mass fraction, h vap,i represents the latent heat of vaporization, T b,max Indicates the maximum boiling point, T f Indicates the initial fuel temperature, C p,i represents the specific heat of the component, C p,a represents the specific heat of the environment, T a Indicates the ambient temperature, u a represents kinematic viscosity, P inj Indicates injection pressure, P a represents the ambient pressure, O2% represents the ambient oxygen content, Z st represents the stoichiometric fuel mass fraction, CN represents the cetane number of the fuel, A, B, C, D, E, and F represent constants obtained by regression of known experimental data; The machine learning prediction module predicts the combustion performance of fuel by using machine learning. The specific steps are as follows: S1: The experimental database used by the experimental database retrieval module is used as the data set; S2: According to the input and output, it is divided into four models, and the data is preliminarily screened and normalized respectively, and divided into training set and test set. The prediction model is trained with the training set: Model 1 is used to predict the liquid phase length, ignition delay period and flame floatation length in the spray combustion characteristics of fuel under different experimental conditions; Model 2 is used to predict the emission characteristics of fuel under standard thirteen working conditions; Model 3 is used to predict the gas phase penetration distance of fuel; Model 4 is used to predict the carbon soot quality of fuel, among which the machine learning used in Model 1 and Model 2 is extreme learning machine, and the machine learning used in Model 3 and Model 4 is deep neural network; S3: Use the coefficient of determination, mean square error, and mean absolute percentage error as evaluation indicators to adjust the hyperparameters in the model until the results meet the requirements. S4: Finally, the trained model is used to predict the combustion performance of fuel.

2. The high-quality biodiesel combustion performance evaluation system according to claim 1, characterized in that: The physical and chemical properties of the fuel include fuel types and physical and chemical properties, wherein the fuel types include pure fuel and mixed fuel, fuel types include first-generation biodiesel, second-generation biodiesel-straight chain, second-generation biodiesel-isomerized and diesel for mixing with biodiesel, gasoline, n-dodecane, methanol, ethanol, butanol, pentanol and octanol, and the physical and chemical properties include cetane number, kinematic viscosity, density, sulfur content, condensation point and latent heat of vaporization; wherein the physical and chemical properties of pure fuel are all measured by experiments; the physical and chemical properties of mixed fuel are the data obtained by experimental measurement. For mixed fuels that have not been measured, there are two ways to obtain their physical and chemical properties: (1) For the case where this mixing ratio has not been measured but this mixing type has been measured, it is obtained by fitting; (2) If such a mixture has not been measured, it is calculated proportionally based on the physical and chemical properties of the pure fuel.

3. The high-quality biodiesel combustion performance evaluation system according to claim 1, characterized in that: The experimental conditions are divided into settable experimental conditions and fixed experimental conditions. The experimental devices for settable experimental conditions uniformly use high-temperature and high-pressure constant-volume combustion bombs, while the experimental devices for fixed experimental conditions use engine benches. The fixed experimental conditions are standard thirteen conditions.

4. The high-quality biodiesel combustion performance evaluation system according to claim 1, characterized in that: Fuel combustion performance includes combustion characteristics and emission characteristics. It is stipulated that all values ​​are measured using a unified definition, a unified test method and a unified experimental processing method. The combustion characteristics are obtained through a constant volume combustion bomb experiment, including liquid phase length, ignition delay period, flame floatation length, soot mass and gas phase penetration distance; the emission characteristics are obtained through engine bench tests, including brake specific fuel consumption, CO, HC, PM and NO X The specific emission of the unified combustion characteristics, the definition, measurement method, and experimental treatment method are: The spray gas phase and liquid phase contours in the combustion characteristics are obtained by the schlieren method and the shadow method respectively, the ignition delay period is obtained by judging the schlieren image, the flame floating length is photographed by the chemical fluorescence method, and the soot mass is calculated based on the diffuse background light extinction method; The liquid phase length is defined as the longest stable distance of the spray liquid phase in the axial direction when the fuel evaporation rate is equal to the fuel injection rate. The ignition delay period is defined as the time interval between the start of injection and the start of high-temperature combustion. The flame floatation length is defined as the average value of the shortest distance between the nozzle position and the upper and lower areas of the stable flame reaction axis, respectively. The gas phase penetration distance is defined as the distance between the nozzle vertex and the farthest point of the spray gas phase contour along the nozzle axis. The soot mass is defined by the calculation formula: m soot =ρλAr·KL / k e (1) Where ρ represents the soot density, λ represents the wavelength of the incident light, Ar represents the area of ​​the soot region, K represents the spatial extinction coefficient, L represents the optical length of the incident light in the soot, and k e represents the dimensionless extinction coefficient, Among them, the liquid phase length, ignition delay period, and flame floating length are all one value in a single spray cycle. During the measurement, multiple cycles are performed to first eliminate unreliable data and then take the average value. The gas phase penetration distance and carbon soot mass are a series of values ​​that change with time. During the measurement, multiple cycles are performed to first eliminate unreliable data and then take the average value according to the time point. The final result is also a series of values.

5. The high-quality biodiesel combustion performance evaluation system according to claim 1, characterized in that: The search results in the experimental database are used as the priority output. If no corresponding results are matched, the total error size when using the fitting formula and the machine learning method is calculated through the error curve, and the method with the smallest error is finally applied. Regarding the error curve: Use control variables to analyze all input parameters, including various fuel components and experimental conditions, calculate the fitting value of the fitting formula or the difference between the predicted value of machine learning and the experimental value, and fit the error curve. There are m1 types of fuel in the database, m2 adjustable experimental conditions, and three outputs: liquid phase length, ignition delay period, and flame floating length. The fitting formula and machine learning error curves each have n = (m1 + m2) × 3. When a group of data that has not been experimentally measured is input, first, the values ​​corresponding to each parameter of the group of data are calculated through the fitting formula and the n error curves of machine learning to obtain n error values, and the m1 + m2 error values ​​of liquid phase length, ignition delay period, and flame floating length are summed and calculated respectively. At the same time, weights are introduced to consider the influence of each input parameter on the result, and the total error of the liquid phase length is calculated: Among them, a represents the calculation of the fitting formula, b represents the prediction of machine learning, p represents the weight of each input parameter, and p i represents the weight of the i-th input parameter, e represents the error value calculated by the error curve, and e LL,ai represents the error value of the i-th parameter when calculating the liquid phase length using the fitting formula, e LL,bi Represents the error value of the i-th parameter when predicting the liquid phase length using machine learning.

6. The high-quality biodiesel combustion performance evaluation system according to claim 1, characterized in that: Depending on whether there is experimental data, the experimental database retrieval module, the fitting formula calculation module or the machine learning prediction module is selected. When the combustion characteristics and emission characteristics of the mixed fuel are output, the experimental database retrieval module, the fitting formula calculation module or the machine learning prediction module and the method used in the system during this runtime are explained at the same time. If the experimental database retrieval module is used, it corresponds to the matching method, if the fitting formula calculation module is used, it corresponds to the calculation method, and if the machine learning prediction module is used, it corresponds to the prediction method.

7. The high-quality biodiesel combustion performance evaluation system according to claim 6, characterized in that: When outputting the fitting formula calculation or machine learning prediction results, there is no true value to compare with it, and the accuracy of the two cannot be known. Therefore, all data in the experimental database are calculated using the same formula or predicted using the same model, and compared with the actual experimental results as a reference for accuracy.

Images