Engine oil consumption prediction and optimization method fused with digital twinborn technology
Through digital twin technology, combined with mechanism model and dual-domain data calibration module, the accuracy and efficiency problems of engine fuel consumption prediction are solved, and high-precision fuel consumption prediction and optimization are achieved.
Patent Information
- Application Number
- CN202411978470.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-05-30
AI Technical Summary
The prior art is difficult to accurately predict engine fuel consumption, and traditional methods require a large amount of data and computing resources, and it is difficult to describe the complex in-cylinder flow and combustion processes of the engine system.
Using digital twin technology, the engine digital twin model is designed, combining mechanism model with dual-domain data calibration module to calibrate and predict fuel consumption data. The model is calibrated in the mechanism domain and timing domain to improve prediction accuracy, and optimize PID parameters through the fuel consumption control optimization algorithm to reduce fuel consumption.
High-precision prediction and optimization of engine fuel consumption is achieved, fuel consumption is reduced, and the energy efficiency of the engine is improved.
Smart Images
Figure CN120068587A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to digital twin and deep learning technologies, and particularly to an engine fuel consumption prediction and optimization method integrating digital twin technology. Background Art
[0002] Under the objective factors of strict international emission standards and rising fuel costs, new energy technologies are in a period of rapid development. There are still some problems to be solved urgently in their energy storage, use, endurance, power performance, stability and reliability. In the next few decades, engines will continue to play an important role in the fields of construction machinery and military equipment until the arrival of mature new energy power devices. During this period, engines need to continuously update technologies to meet the requirements of energy conservation, emission reduction, low carbon and environmental protection under the current background.
[0003] The fuel efficiency of an engine is one of the key factors for engine performance and sustainability. The fuel consumption rate of an engine is an important indicator to measure the fuel efficiency of the engine, and it is very sensitive to changes in engine operating conditions. Therefore, accurately predicting the fuel consumption of an engine is an important challenge for optimizing engine performance and achieving energy efficiency. Traditional engine fuel consumption estimation methods include physics-based models, data-based models, and statistics-based models. However, these single methods often require a large amount of data such as prior knowledge and feature engineering, and consume a large amount of computing resources. Moreover, current computer simulation technologies are difficult to accurately describe the complex in-cylinder flow and combustion processes of the engine system through mathematical formulas, and it is difficult to establish a high-precision mapping relationship from control parameters to fuel consumption and emission performance for coupled multi-parameter inputs.
[0004] Currently in the field of computer data regression analysis. Well-trained artificial neural networks can reasonably predict different data and show relatively superior performance. However, general artificial neural networks such as BP neural networks and RNN neural networks usually require a long training time and a large amount of historical data, and problems such as gradient explosion or gradient disappearance are likely to occur during the training process. Summary of the Invention
[0005] Objective of the Invention: Aiming at the problems existing in the prior art, the present invention provides a method for predicting and optimizing engine fuel consumption by integrating digital twin technology. In the present invention, the fuel consumption data simulated by the mechanism model is calibrated twice in the mechanism domain and the time series domain through the dual-domain data calibration module, solving the problems of inaccurate simulation data and large noise in real data. At the same time, the engine mechanism model and the dual-domain data calibration module are taken as a whole and called the digital twin model. This model can accurately predict the fuel consumption at the next moment based on the state parameters at the current moment and the historical data before the current moment. Subsequently, a fuel consumption control optimization algorithm integrating the engine digital twin model is designed. Taking the above-built digital twin model as the objective function and minimizing the fuel consumption at the future moment as the optimization objective, the PID parameters are optimized, and the PID parameters of the three control objects of injection timing, injection quantity, and injection angle are comprehensively optimized to achieve the purpose of fuel consumption optimization.
[0006] The technical solution of the present invention is as follows:
[0007] A method for predicting and optimizing engine fuel consumption by integrating digital twin technology according to the present invention includes:
[0008] (1) Design an engine digital twin model, including a mechanism model corresponding to the physical entity and a dual-domain data calibration module. This model can accurately predict the fuel consumption at the next moment based on the state parameters at the current moment and the historical data before the current moment for subsequent fuel consumption optimization;
[0009] (2) Design an engine modular mechanism model, including parts such as a fuel injection pump sub-module, an intake pipe sub-module, an exhaust pipe sub-module, and a governor sub-module. Set the corresponding boundary conditions for the intake pipe sub-module, exhaust pipe sub-module, cylinder sub-module, etc., and realize the mechanism simulation and deduction of the relevant parameters of the engine through the energy conservation equation, mass conservation equation, ideal gas state equation, etc.;
[0010] (3) Input the initial load, initial speed, target load, target speed, PID, etc. at time t as state parameters into the mechanism model, set the corresponding simulation time, and obtain the simulation data at the next moment, i.e., time t + 1, including fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure, etc.; Collect the big data during the operation of the engine to obtain the real operation performance parameter data corresponding to the time from t - k to t + 1, including fuel consumption, speed, load, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure, etc.;
[0011] (4) Design a dual-domain data calibration module, which includes a mechanism domain calibration module and a timing domain calibration module; in the mechanism domain calibration module, design a mechanism domain deep learning calibration model, using the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure, etc. at time t+1 in the real data as labels, calibrate the corresponding simulation data at time t+1 obtained from the mechanism model, improve the understanding of mechanism knowledge by the mechanism domain deep learning model to optimize the fuel consumption index prediction performance; in view of the fact that the mechanism domain calibration process lacks the support of a large amount of timing domain knowledge and is prone to model overfitting, design a timing domain calibration module, introduce timing information to further calibrate the fuel consumption obtained after mechanism domain calibration in the timing domain; design a timing domain deep learning calibration model, fuse the fuel consumption after mechanism domain calibration, the target load, target speed, PID in the state parameters, and the performance parameters from time t-k to time t in the real data as the input of the timing domain calibration module; use the real fuel consumption at time t+1 as the corresponding label; combine the mechanism domain features and timing domain feature knowledge, mine the mechanism features and timing features related to engine fuel consumption, suppress overfitting, and reduce the influence of noise interference, so as to achieve high-precision calibration and prediction of the fuel consumption data at time t+1.
[0012] (5) Design an optimization algorithm for fuel consumption control that integrates the engine digital twin model. Use the digital twin model constructed above as the objective function, with the lowest fuel consumption at future times as the optimization goal, and comprehensively optimize the PID parameters of the three control objects of injection timing, injection quantity, and injection angle; based on the actual PID parameter matrix set during the corresponding process of the physical engine, design a random variable initialization optimization algorithm and limit the optimization space to improve the optimization efficiency of the algorithm and avoid falling into local optima.
[0013] (6) Obtain the comprehensive optimal PID parameter matrix of the injection timing, injection quantity, and injection angle control objects during the change of different load speeds of the engine through step (5), and then set the optimized PID parameter matrix during the corresponding process of the physical engine to achieve fuel consumption optimization during the actual operation process of the engine.
[0014] Furthermore, in step (1), the steps of designing an engine digital twin model, including a mechanism model corresponding to the physical entity and a dual-domain data calibration module, are as follows:
[0015] (1-1) Design a mechanism model, model the essential laws of the physical system, comprehensively describe the key factors affecting fuel consumption, and provide physical constraints and preliminary prediction capabilities for the digital twin system.
[0016] (1-2) Design a dual-domain data calibration module, which calibrates the simulation data by combining the characteristics of the mechanism domain and the timing domain to further improve the prediction accuracy.
[0017] Further, the steps for designing a modular mechanism model of the engine in step (2) are as follows:
[0018] (2-1) Build modules such as the fuel injection pump sub-module, intake pipe sub-module, exhaust pipe sub-module, governor sub-module, etc. Set corresponding boundary conditions for each module, and the sub-modules are interconnected through the transfer of energy and mass. All sub-modules satisfy the zero-dimensional assumption;
[0019] (2-2) The system satisfies assumptions such as the state inside the cylinder is uniform, the working medium is an ideal gas, and its specific heat, specific internal energy, and specific enthalpy are only related to the gas temperature and composition;
[0020] (2-3) Describe the state of the working medium in the cylinder by regarding the cylinder as a thermodynamic system. The boundary of the system consists of the piston top, cylinder head, and cylinder liner wall;
[0021] (2-4) Use the energy conservation equation, mass conservation equation, and ideal gas state equation to link the entire working process through the in-cylinder pressure, in-cylinder temperature, and working medium mass. Take the four-stroke as a cycle and solve the differential equation segment by segment to simulate the relevant parameters of the engine.
[0022] The energy conservation equation is as follows:
[0023]
[0024] Mass conservation equation:
[0025]
[0026] Ideal gas state equation:
[0027] PV = mRT
[0028] Take the four-stroke as a cycle and solve the differential equation segment by segment, and realize the simulation of the parameters by solving the equations of each parameter.
[0029] Further, the steps for obtaining the simulation data at time t + 1 and the real operation performance parameter data from time t - k to time t + 1 in step (3) are as follows:
[0030] (3-1) Input the load, speed, target load, target speed, and PID at the current time t into the mechanism model and set the corresponding simulation time to obtain the simulation output data at time t + 1, including fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, and intake pressure;
[0031] (3-2) Collect big data during the engine operation process, and obtain the real operating performance parameter data from the t-k to t+1 moment in the time series data of the corresponding state parameters, including fuel consumption, rotational speed, load, in-cylinder temperature, in-cylinder pressure, intake air temperature, intake air pressure, etc.;
[0032] (3-3) Standardize the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake air temperature, intake air pressure in the simulation output data and the rotational speed, load, in-cylinder temperature, in-cylinder pressure, intake air temperature, intake air pressure in the real data so that the data is in the same scale. The standardization formula is as follows:
[0033]
[0034] In the formula, x' is the normalized eigenvalue; x i is the actual value of the feature before normalization; x max is the maximum value of the feature data; x min is the minimum value of the feature data.
[0035] Furthermore, in step (4), design a dual-domain data calibration module. The steps of this module including a mechanism domain calibration module and a time series domain calibration module are as follows:
[0036] (4-1) Divide the preprocessed simulation data and real data into a training set, a validation set, and a test set according to 8:1:1, and use ten-fold cross-validation in the test set;
[0037] (4-1) In the mechanism domain calibration module, design a mechanism domain deep learning calibration model based on a neural network. The input of this model is the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake air temperature, intake air pressure at the t+1 moment of the simulation output data, and the predicted output of the model is the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake air temperature, intake air pressure at the t+1 moment of the corresponding real data. The forward propagation formula of the neural network is as follows:
[0038]
[0039] In the formula, w i is the weight parameter, b i is the bias term, and f is the Sigmoid activation function used to introduce non-linearity. The formula of the Sigmoid function is as follows:
[0040]
[0041] (4-2) In the time-domain calibration module, a time-domain deep learning calibration model is designed. First, a feature extraction module is built based on LSTM. The input of this module is the performance parameters from t-k to t in the real data. The time-series features of the real data are extracted through this module. Then, the extracted time-series features, the fuel consumption calibrated in the mechanism domain, the target load, the target speed, and the PID in the state parameters are feature-fused. Finally, the fused features are input into a regressor to further calibrate the fuel consumption data.
[0042] (4-4) Use the dataset divided in (4-1) to train the dual-domain data calibration module. The training process is to first train the mechanism domain calibration module and then train the time-domain calibration module. The mean square error is used as the objective function. The error E is used to measure the error between the predicted value and the real value. By taking the partial derivatives of the objective function error E with respect to the weight w and the bias term b, the parameter update of the neural network is realized. The calculation formula of the mean square error loss function is as follows:
[0043]
[0044] In the formula, N is the number of samples, y is the real value, and a is the predicted value. The calculation formula of the predicted value a is as follows:
[0045] a = f(z) = f(w * x + b)
[0046] In the formula, x is the input, w is the weight, b is the bias term, and f is the activation function.
[0047] Furthermore, in step (5), a fuel consumption control optimization algorithm integrating the engine digital twin model is designed as follows:
[0048] (5-1) The optimization algorithm uses the particle swarm optimization algorithm. In the particle swarm optimization algorithm, the number of individuals in the population is set to 20, the initialized population dimension is set to 9, the initialized population iteration times are set to 100, the learning factor 1 is set to 2, the learning factor 2 is set to 2, and the inertia weight is set to 1.5;
[0049] In each population iteration algorithm, the velocity of each particle is calculated, and then the position of each particle is updated according to the velocity. The velocity update formula is as follows:
[0050]
[0051] In the formula, w is the inertia weight, which controls the exploration ability of the particle; c 1 , c 2 is the learning factor, which measures the influence degree of the particle by its own experience and the group experience respectively; r 1 , r 2 is a random number, and the range is [0, 1]; p iis the historical optimal position of particle i; g is the global optimal position.
[0052] Among them, the inertia weight w controls the ability of the particle to search the range. Usually, a linear decreasing strategy is adopted, and the decreasing formula is as follows:
[0053]
[0054] In the formula, w max and w min are the maximum and minimum inertia weights respectively.
[0055] The position update formula is as follows:
[0056]
[0057] In the formula, is the position of particle i at time t + 1, is the velocity of particle i at time t + 1;
[0058] (5-2) Taking the engine digital twin model constructed above as the objective function of the particle swarm optimization algorithm, the predicted fuel consumption of the model is optimized. Taking one set of PID parameters as an example, the objective function formula of the particle swarm optimization algorithm is as follows:
[0059]
[0060] In the formula, P = (K p , K i , K d ) represents the three parameters of the PID controller (proportional gain, integral gain, derivative gain), f(P) is the digital twin model, is the PID parameter combination after our optimization;
[0061] (5-3) In order to improve the optimization efficiency of the algorithm and avoid falling into the local optimal solution, based on the PID parameter matrix manually tuned in the corresponding process of the physical engine, the initial position vectors of some particles are obtained by introducing random variables, and the corresponding solution space is set for the particles;
[0062] Taking one set of PID parameters as an example, the initialization of the particle position vector is as follows:
[0063]
[0064] where i = 1, 2, 3, 4, 5, are the PID manually tuned respectively, and σ p , σ i , σ d correspond to the perturbation coefficients of the three parameters respectively.
[0065] The initialization of the position vectors of the remaining particles is as follows:
[0066]
[0067]
[0068] where j = 6, 7, 8, 9, …, 20, (K p,min , K p,max ), (K i,min , K i,max ), (K d,min , K d,max ) respectively represent the solution spaces of k p , k i , k d .
[0069] Beneficial effects: Compared with the prior art, the significant advantages of the present invention are as follows: The present invention provides a method for predicting and optimizing engine fuel consumption by integrating digital twin technology. First, a dual-domain data calibration module is proposed, which calibrates the simulation data twice in the mechanism domain and the time series domain respectively, solving the problem of low accuracy of current simulation data. Subsequently, the engine mechanism model and the dual-domain data calibration module are taken as a whole and called the digital twin model, which can accurately predict the fuel consumption at the next moment based on the state parameters at the current moment and the historical data before the current moment. Subsequently, an optimization algorithm for fuel consumption control integrating the engine digital twin model is designed. Taking the above-constructed digital twin model as the objective function and minimizing the fuel consumption at the future moment as the optimization goal, the PID parameters are optimized, and the PID parameters of the three control objects of injection timing, injection quantity, and injection angle are comprehensively optimized to achieve the purpose of fuel consumption optimization. Brief Description of the Drawings
[0070] Figure 1 is the flowchart of the present invention;
[0071] Figure 2 is the working process diagram in the cylinder;
[0072] Figure 3 is the calibration effect diagram;
[0073] Figure 4 is the digital twin model structure diagram;
[0074] Figure 5 is the fuel consumption optimization flowchart Detailed Embodiments
[0075] To more clearly understand the object, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0076] As Figure 1 shown, the present invention discloses an engine fuel consumption prediction and optimization method integrating digital twin technology, including:
[0077] (1) Design an engine modular mechanism model, including building a compressor sub-module, an intercooler sub-module, an intake pipe sub-module, a cylinder sub-module, an exhaust pipe sub-module, a turbine sub-module, a supercharger rotor sub-module, a governor sub-module, and a fuel injection pump sub-module. Corresponding boundary conditions are set for each module. The sub-modules are interconnected through the transfer of energy and mass, and the system satisfies the following basic assumptions: the state inside the cylinder is uniform, the working medium is an ideal gas, its specific heat, specific internal energy, and specific enthalpy are only related to the gas temperature and composition, the gas flowing into or out of the cylinder is a quasi-steady flow, the flow kinetic energy at the inlet and outlet of the working medium is ignored, and the process of releasing chemical energy by fuel combustion is regarded as a thermodynamic process in which the outside heats the working medium in the system according to a known apparent heat release law.
[0078] The working process inside a diesel engine cylinder is complex. To describe the state of the working medium inside the cylinder, the cylinder is regarded as a thermodynamic system, and the boundary of the system is composed of the piston top, the cylinder head, and the cylinder liner wall. The working process inside the cylinder is as Figure 2 shown.
[0079] Using the energy conservation equation, mass conservation equation, and ideal gas state equation through the in-cylinder pressure (P), in-cylinder temperature (T), and working medium mass (M), the entire working process is connected, and the four-stroke cycle is used to solve the differential equation in segments;
[0080] The energy conservation equation is as follows:
[0081]
[0082] The mass conservation equation:
[0083]
[0084] The ideal gas state equation:
[0085] PV = mRT
[0086] Generally, the specific internal energy u and mass m inside the system change simultaneously, so there is:
[0087]
[0088] From \(u = u(T,\lambda)\), by total differentiation, we have:
[0089]
[0090] From the thermodynamic formula: Finally, the temperature \(T\) with respect to the crankshaft angle gives the differential equation:
[0091]
[0092] According to the characteristics of each stage, solve the differential equation in stages. Usually, select the actual compression start point, the intake valve closing moment as the calculation start point, and calculate step by step until the intake valve closing moment of the next cycle.
[0093] (2) Input the load, speed at time \(t\), as well as the target load, target speed, PID, etc. into the mechanism model and set the corresponding simulation time, and obtain the simulation output data at time \(t + 1\), including fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure, etc. Obtain the real data corresponding to the time from \(t - k\) to \(t + 1\) from the engine historical big data in the time series data of the same starting load, starting speed, target load, target speed, and PID, that is, performance parameters, including fuel consumption, speed, load, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure, etc., where \(k = 50\).
[0094] Standardize the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure in the simulation output data and the speed, load, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure in the real data so that the data is on the same scale. The standardization formula is as follows:
[0095]
[0096] In the formula, \(x'\) is the normalized eigenvalue; \(x\) i is the actual value of the feature before normalization; \(x\) max is the maximum value of the feature data; \(x\) min is the minimum value of the feature data.
[0097] (3) Divide the preprocessed simulation data and real data into training set, validation set, and test set according to 8:1:1, and use ten-fold cross-validation in the test set. Build a mechanism domain calibration module based on the neural network. The input of this module is the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure at time \(t + 1\) of the simulation output data, and the predicted output of the model is the fuel consumption, in-cylinder temperature, in-cylinder pressure, intake temperature, intake pressure at time \(t + 1\) of the corresponding real data. The forward propagation formula of the neural network is as follows:
[0098] y = f(b i +∑w i x i )
[0099] where w i is the weight parameter, b i is the bias term, and f is the Sigmoid activation function used to introduce non-linearity. The formula for the Sigmoid function is as follows:
[0100]
[0101] Build a time-domain data calibration module. First, build a feature extraction module based on LSTM. The input of this module is the performance parameters from t - k to t in the real data. The LSTM (Long Short-Term Memory) neural network extracts the temporal features of the data. The neurons in the LSTM will extract and filter the data in the model, thus improving the effect of the model. The internal calculation process of its neurons is as follows. First, for each time step t, the input gate determines how much new information should be added to the cell state. The output i t of the input gate is calculated by the following formula:
[0102] i t = σ(w i [h t-1 , x t ) + b i
[0103] At the same time, use the tanh activation function to add a candidate memory unit:
[0104] c' t = tanh(w c [h t-1 , x t ) + b c )
[0105] where σ is the Sigmoid activation function, w i , w c are the weights of i t , c' t respectively; b i , b c are the bias terms of i t , c' t respectively, x t is the input of the input sequence at time step t, and h t-1 is the hidden state at the previous time step t - 1.
[0106] The forget gate determines which information in the new state should be forgotten or retained. The output f t of the forget gate is calculated by the following formula:
[0107] f t = σ(w f [h t-1 , x t + b f )
[0108] Next, use the output i of the input gate t and the output f of the forget gate t to update the cell state c t :
[0109] c t = f t c t-1 + i t c' t
[0110] The output gate calculates the output information o based on h t-1 , x t , and the tanh activation function combines with the output gate information o t to obtain the current hidden layer state t
[0111]
[0111] o t = σ(w o [h t-1 , x t + b o )
[0112] h t = o t * tanh(c t )
[0113] In the formula, w o is the output gate weight; b o is the output gate bias term
[0114] The hidden layer state ht is transmitted to the next time step, and at the same time, the cell state Ct is also passed to the next step to maintain long-term memory. Thus, the neuron of the LSTM neural network completes a calculation process according to the input
[0115] Then, fuse the time-series features extracted by the feature extractor, the fuel consumption calibrated by the mechanism domain, the target load, target speed, and PID in the state parameters. Finally, input the fused features into the regressor to further calibrate the fuel consumption data in the time series domain
[0116] Use the previously partitioned dataset to train the dual-domain data calibration module. The training process is to first train the mechanism domain calibration module and then train the time series domain calibration module. The mean square error is used as the objective function, and the error E is used to measure the error between the predicted value and the true value. By taking the partial derivatives of the objective function error E with respect to the weight w and the bias term b, the parameter update of the neural network is realized. The calculation formula of the mean square error loss function is as follows:
[0117]
[0118] In the formula, N is the number of samples, y is the true value, and a is the predicted value. The calculation formula of the predicted value a is as follows:
[0119] a = f(z) = f(w * x + b)
[0120] In the formula, x is the input, w is the weight, b is the bias term, and f is the activation function.
[0121] Through training, the dual-domain calibration module can achieve high-precision calibration of fuel consumption data. The calibration results are as Figure 3 shown. The engine mechanism model and the dual-domain calibration module as a whole are called the digital twin model. The results of the digital twin model are as Figure 4 shown.
[0122] (4) Obtain the current speed and load, set the target speed, target load, and PID, and input them into the engine digital twin model to predict the next fuel consumption data. Take this engine digital twin model as the objective function, with the lowest fuel consumption at the future moment as the optimization goal, and comprehensively optimize the PID parameter matrix of the three control objects of injection timing, injection quantity, and injection angle. The optimization flowchart is as Figure 5 shown. In the particle swarm optimization algorithm, the number of population individuals is set to 20, the initialization population dimension is set to 9, the initialization population iteration times are set to 100, the learning factor 1 is set to 2, the learning factor 2 is set to 2, and the inertia weight is set to 1.5.
[0123] In each population iteration of the particle swarm optimization algorithm, the velocity of each particle is calculated, and then the position of each particle is updated according to the velocity. The velocity update formula is as follows:
[0124]
[0125] In the formula, w is the inertia weight, which controls the exploration ability of the particle; c 1 , c 2 are the learning factors, which respectively measure the influence degrees of the particle by its own experience and the group experience; r 1 , r 2 are random numbers, with the range of [0, 1]; p iis the historical optimal position of particle i; g is the global optimal position.
[0126] Among them, the inertia weight e controls the ability of the particle to search the range. Usually, a linear decreasing strategy is adopted, and the decreasing formula is as follows:
[0127]
[0128] In the formula, w max and w min are the maximum and minimum inertia weights respectively.
[0129] The position update formula is as follows:
[0130]
[0131] In the formula, is the position of particle i at time t + 1, is the velocity of particle i at time t + 1.
[0132] In order to improve the optimization efficiency of the algorithm and avoid falling into the local optimal solution, based on the PID parameter matrix manually tuned in the corresponding process of the solid engine, the initial position vectors of some particles are obtained by introducing random variables, and the corresponding solution spaces are set for the particles.
[0133] Taking a set of PID parameters as an example, the initialization of the particle position vector is as follows:
[0134]
[0135] where i = 1, 2, 3, 4, 5, are the PID manually tuned respectively, and σ p , σ i , σ d correspond to the random perturbations of the three parameters respectively.
[0136] For the initialization of the position vectors of the remaining particles, it is as follows:
[0137]
[0138]
[0139] where j = 5, 6, 7,..., 20, (K p,min , K p,max ), (K i,min , K i,max ), (K d,min , K d,max represent the solution spaces of k p , k i , k d respectively.
[0140] The objective function formula of the particle swarm optimization algorithm is as follows:
[0141]
[0142] In the formula, P = (K p , K i , K d ) represents the three parameters (proportional gain, integral gain, derivative gain) of the PID controller, f(P) is the digital twin model, is the PID parameter combination after our optimization.
[0143] (6) Through the engine digital twin model and the particle swarm optimization algorithm, obtain the comprehensive optimal PID parameter matrix of the injection timing, injection quantity, and injection angle control objects during the change of different load speeds of the engine. Subsequently, set the optimized PID parameter matrix during the corresponding process of the physical engine to achieve fuel consumption optimization.
[0144] PID adjusts the control output according to the system error, so that the system reaches the predetermined target value. Its calculation formula is as follows:
[0145]
[0146] In the formula, K p is the proportional gain, e(t) is the error, K i is the integral gain, K d is the derivative gain, u t is the output.
[0147] Matters not covered in this invention are well-known techniques.
[0148] The above embodiments are only used to illustrate the technical concept and characteristics of the present invention. The purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and it 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 covered within the protection scope of the present invention.
Claims
1. A method for predicting and optimizing engine fuel consumption by integrating digital twin technology, characterized in that: The following steps are involved: (1) Design an engine digital twin model, including a mechanism model corresponding to the physical entity and a dual-domain data calibration module. The model can accurately predict the fuel consumption at the next moment based on the state parameters at the current moment and the historical data before the current moment for subsequent fuel consumption optimization; (2) Design a modular mechanism model of the engine, including a fuel injection pump submodule, an intake pipe submodule, an exhaust pipe submodule, a speed regulator submodule, etc., set corresponding boundary conditions for the intake pipe submodule, the exhaust pipe submodule, the cylinder submodule, etc., and realize the mechanism simulation deduction of engine-related parameters through energy conservation equations, mass conservation equations, and ideal gas state equations; (3) Input the initial load, initial speed, target load, target speed, PID, etc. at time t as state parameters into the mechanism model, set the corresponding simulation time, and obtain the simulation data at the next time, i.e., time t+1, including fuel consumption, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc.; collect the big data of the engine operation process to obtain the real operation performance parameter data corresponding to time tk to time t+1, including fuel consumption, speed, load, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc.; (4) Design a dual-domain data calibration module, which includes a mechanism domain calibration module and a timing domain calibration module; In the mechanism domain calibration module, a mechanism domain deep learning calibration model is designed. The fuel consumption, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc. at time t+1 in the real data are used as labels to calibrate the corresponding simulation data at time t+1 obtained from the mechanism model, so as to improve the mechanism domain deep learning calibration model's understanding of mechanism knowledge and optimize the prediction performance of fuel consumption indicators. In view of the lack of massive time series domain knowledge support in the mechanism domain calibration process, which may lead to model overfitting, a time series domain calibration module is designed to introduce time series information to the fuel consumption obtained after the mechanism domain calibration. Further calibration in the sequence domain; design a deep learning calibration model in the time series domain, and fuse the fuel consumption calibrated in the mechanism domain, the target load, target speed, PID in the state parameters, and the performance parameters from time tk to time t in the real data as the input of the time series domain calibration module; use the real fuel consumption at time t+1 as the corresponding label; combine the mechanism domain features with the time series domain feature knowledge to mine the engine fuel consumption related mechanism features and time series features, suppress overfitting, and reduce the impact of noise interference, so as to achieve high-precision calibration and prediction of the fuel consumption data at time t+1; (5) Design a fuel consumption control optimization algorithm that integrates the digital twin model of the engine. The digital twin model of the engine constructed above is used as the objective function, and the lowest fuel consumption at the future moment is used as the optimization goal. The PID parameters of the three control objects, injection timing, injection amount, and injection angle, are comprehensively optimized. Based on the actual PID parameter matrix adjusted in the corresponding process of the physical engine, a random variable initialization optimization algorithm is designed and the optimization space is limited to improve the algorithm optimization efficiency and avoid falling into the local optimum. (6) Through step (5), the comprehensive optimal PID parameter matrix of the injection timing, injection amount, and injection angle control objects of the engine under different load and speed changes is obtained, and then the optimized PID parameter matrix is set in the corresponding process of the physical engine to achieve fuel consumption optimization during the actual operation of the engine.
2. The engine fuel consumption prediction and optimization method integrating digital twin technology according to claim 1 is characterized in that: In the step (1), an engine digital twin model is designed, including a mechanism model corresponding to the physical entity and a dual-domain data calibration module, specifically in the following manner: (1-1) Design a mechanism model that comprehensively describes the key factors affecting fuel consumption by modeling the essential laws of the physical system, and provides physical constraints and preliminary prediction capabilities for the digital twin system. (1-2) Design a dual-domain data calibration module, which combines the characteristics of the mechanism domain and the timing domain to calibrate the simulation data and improve the calibration accuracy.
3. The engine fuel consumption prediction and optimization method integrating digital twin technology according to claim 1 is characterized in that: In the step (2), a modular mechanism model of an engine is designed, specifically in the following manner: (2-1) Build the fuel injection pump submodule, intake pipe submodule, exhaust pipe submodule, speed regulator submodule and other modules. Set corresponding boundary conditions for each module and the submodules are interconnected through the transfer of energy and mass. All submodules satisfy the zero-dimensional assumption. (2-2) The system satisfies the following assumptions: the state in the cylinder is uniform, the working fluid is an ideal gas, and its specific heat, specific internal energy and specific enthalpy are only related to the gas temperature and composition; (2-3) The working medium state in the cylinder is described by considering the cylinder as a thermodynamic system. The boundary of the system is composed of the piston top, cylinder head and cylinder liner wall. (2-4) The entire working process is linked together through the in-cylinder pressure, in-cylinder temperature, and mass using the energy conservation equation, mass conservation equation, and ideal gas state equation. The four-stroke cycle is taken as a cycle and the differential equations are solved piecewise to simulate the engine's related parameters.
4. The engine fuel consumption prediction and optimization method integrating digital twin technology according to claim 1 is characterized in that: In step (3), the simulation data at time t+1 and the actual operating performance parameter data corresponding to time tk to time t+1 are obtained in the following manner: (3-1) Input the load, speed, target load, target speed, PID, etc. at time t in the mechanism model and set the corresponding simulation time to obtain the simulation output data at time t+1, including fuel consumption, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc.; (3-2) Collect big data of the engine operation process and obtain the actual operating performance parameter data from time tk to time t+1 from the time series data of the corresponding state parameters, including fuel consumption, speed, load, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc.
5. The engine fuel consumption prediction and optimization method integrating digital twin technology according to claim 1 is characterized in that: In the step (4), a dual-domain data calibration module is designed, which includes a mechanism domain calibration module and a timing domain calibration module, and specifically adopts the following method: (4-1) In the mechanism domain calibration module, a mechanism domain deep learning calibration model is designed. The fuel consumption, cylinder temperature, cylinder pressure, intake temperature, intake pressure, etc. at time t+1 in the real data are used as labels to calibrate the corresponding simulation data at time t+1 obtained from the mechanism model, thereby improving the mechanism domain deep learning model's understanding of mechanism knowledge and optimizing the prediction performance of fuel consumption indicators. (4-2) Considering that the mechanism domain calibration process lacks massive time domain knowledge support and is prone to model overfitting, a time domain calibration module is designed to introduce time series information to further calibrate the fuel consumption obtained after the mechanism domain calibration in the time domain; a time domain deep learning calibration model is designed to fuse the fuel consumption after the mechanism domain calibration, the target load, target speed, PID in the state parameters, and the performance parameters from tk to t in the real data as the input of the time domain calibration module; the real fuel consumption at t+1 is used as the corresponding label; (4-3) Divide the simulation data and real data obtained in the previous step into training set, validation set, and test set according to the ratio of 8:1:
1. Use ten-fold cross validation in the test set to design the training mechanism domain deep learning calibration model and the time series domain deep learning calibration model.
6. The engine fuel consumption prediction and optimization method integrating digital twin technology according to claim 1 is characterized in that: In the step (5), a fuel consumption control optimization algorithm integrating the engine digital twin model is designed, specifically in the following manner: The optimization algorithm described in (5-1) uses a particle swarm optimization algorithm, in which the number of individuals in the population is set to 20, the initial population latitude is set to 9, the number of initial population iterations is set to 100, the learning factor 1 is set to 2, the learning factor 2 is set to 2, and the inertia weight is set to 1.5; (5-2) The digital twin model of the engine constructed above is used as the objective function of the particle swarm optimization algorithm, and the lowest fuel consumption in the future is used as the optimization goal. The PID parameters of the three control objects, injection timing, injection amount, and injection angle, are comprehensively optimized. Based on the actual PID parameter matrix adjusted in the corresponding process of the physical engine, random variables are designed to initialize particles and set the corresponding solution space to improve the algorithm's optimization efficiency and avoid falling into local optimality.