Diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering

Abstract

The invention relates to the technical field of engine control and state monitoring, and discloses a diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering, which comprises the following steps: acquiring working condition parameters and torque signals of a diesel engine; based on the collected working condition parameters, model prediction cylinder pressure is obtained; based on the collected torque signals, single-cylinder contribution is extracted, and torque reverse thrust cylinder pressure is obtained through reverse thrust; according to the real-time information entropy change of the model prediction cylinder pressure and the torque reverse thrust cylinder pressure, dynamically distributing a fusion weight, and generating a fusion observed quantity; forming an extended characteristic quantity based on the characteristic that the fusion observed quantity changes along with the crank angle; establishing a state transition equation and an observation equation; and outputting an in-cylinder state prediction and estimation result after fusion optimization. According to the method, the in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering is adopted, fusion weights of different data sources can be dynamically adjusted in real time, and the technical effect of considering steady-state accuracy and transient response speed is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engine control and state monitoring, in particular to a diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering. BACKGROUND

[0002] As a core physical quantity representing the engine combustion process, in-cylinder pressure plays a crucial role in the fields of engine performance optimization, combustion control and emission control. High-precision cylinder pressure data not only can intuitively reflect the heat release and expansion characteristics in the combustion process, but also is directly related to combustion efficiency, heat energy conversion rate and harmful emissions generation. With the increasingly stringent emission regulations and the increasing demand for energy saving and emission reduction, the dependence of engine development on real-time sensing and accurate estimation of cylinder pressure has significantly increased. Especially under complex dynamic conditions, accurate understanding of the cylinder pressure variation law is the key foundation for improving fuel economy, reducing pollution emissions and realizing closed-loop control of the combustion process.

[0003] The existing in-cylinder pressure acquisition and fusion mainly rely on three technical paths, including direct measurement method, model estimation method and indirect measurement method. The direct measurement method can real-time collect the cylinder pressure variation curve by installing high-temperature and high-pressure resistant cylinder pressure sensor in the combustion chamber, has the advantages of high precision and fast response, and is widely used in engine bench testing and a small number of special application scenarios. The model estimation method calculates the cylinder pressure evolution according to the engine operating parameters, or establishes the mapping relationship between the operating conditions and the cylinder pressure through training data-driven black box model, which is used to replace direct sensor measurement.

[0004] However, the existing cylinder pressure acquisition technology through direct measurement method has high measurement accuracy, but the cost of sensor is high, the service life is limited and the maintenance is complex. The model estimation method can provide good prediction accuracy under steady-state conditions, but under transient acceleration or deceleration or load mutation conditions, it is easily affected by non-steady-state factors in the combustion process, resulting in increased prediction error. In addition, data-driven models may have some degree of generalization problem when facing unseen conditions. The indirect measurement method has low cost, is sensitive to external noise, and the signal separation and calibration process is complex. Therefore, the present application provides a diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering to solve the problems existing in the prior art. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering, which solves the problems of insufficient cylinder pressure measurement accuracy, dynamic response lag, large prediction error under transient conditions and excessive influence of external noise.

[0006] To achieve the above object, the present application is realized by the following technical solutions: a diesel engine in-cylinder pressure multi-source fusion estimation method based on adaptive Kalman filtering, comprising the following steps: Collecting the working condition parameters and torque signals of the diesel engine as the data basis for model predicted cylinder pressure estimation and torque backstepping cylinder pressure estimation; Based on the collected working condition parameters, using zero-dimensional combustion modeling and particle swarm optimization algorithm to calibrate the combustion characteristic parameters, and combining neural network to establish the nonlinear mapping relationship between working condition parameters and in-cylinder pressure, and obtaining the model predicted cylinder pressure; Based on the collected torque signals, using the crankshaft connecting rod dynamics model to separate the gas pressure torque and inertia torque, extracting the single cylinder contribution and backstepping to obtain the torque backstepping cylinder pressure; According to the real-time information entropy change of the model predicted cylinder pressure and the torque backstepping cylinder pressure, the fusion weight is dynamically allocated, and the fusion observation is generated; Based on the characteristics of the fusion observation changing with the crankshaft angle, the first and second derivatives of the fusion observation are calculated, which correspond to the cylinder pressure change rate and the cylinder pressure change acceleration respectively, forming the extended feature quantity; Establishing the state transition equation and the observation equation, and using the adaptive Kalman filtering algorithm with variable structure switching mechanism based on the extended model to predict and estimate the state; Output the in-cylinder state prediction and estimation results after fusion optimization, which are used for engine combustion control and state monitoring.

[0007] Preferably, the collection of diesel engine working condition parameters and torque signals comprises: Collecting the real-time data of fuel injection amount, fuel injection time, intake valve opening degree, speed, and in-cylinder combustion closely related to the diesel engine; Collecting the total torque signal output by the diesel engine crankshaft as the basis for cylinder pressure backstepping; Filtering and synchronizing the collected signals to ensure time alignment and data integrity.

[0008] Preferably, the step of establishing the nonlinear mapping relationship between working condition parameters and in-cylinder pressure by combining neural network comprises: Using the Weber combustion model to describe the in-cylinder combustion rate, and using the particle swarm optimization algorithm to calibrate the combustion start angle, combustion duration and combustion shape parameters; Training a multi-layer neural network to fit the nonlinear relationship between the working condition parameters and the in-cylinder pressure curve, and outputting the model predicted cylinder pressure; Establishing the relationship between combustion heat release and pressure change, so that the change of cylinder pressure with crankshaft angle meets: ; Where, is the combustion heat release correlation coefficient; is the combustion rate index; It is an index of the combustion process; This refers to the crankshaft rotation angle; This is a reference value for the angle of combustion characteristics; This represents the heat release per unit combustion angle.

[0009] Preferably, the step of extracting the contribution of a single cylinder and reverse-engineering to obtain the torque reverse-engineering cylinder pressure includes: The total crankshaft torque is separated into gas pressure torque and reciprocating inertial torque using dynamic formulas; Extract the gas pressure and torque of a single cylinder, and deduce the cylinder pressure change of a single cylinder based on the contribution of a single cylinder. Derivation of cylinder pressure based on the torque relationship of crank-connecting rod mechanism With gas pressure and torque The calculation formula is: ; in, It is a function of the change in lever arm composed of crank radius and connecting rod length.

[0010] Preferably, the step of dynamically allocating fusion weights based on the real-time information entropy changes of cylinder pressure predicted by the model and the cylinder pressure inversely derived from torque includes: Within a sliding time window, the statistical model predicts the cylinder pressure and the torque inversely calculates the cylinder pressure data distribution. Calculate the information entropy value of each data set, and allocate the fusion weights according to the inverse ratio of the entropy value.

[0011] Preferably, the formula for calculating the fusion weight is: ; in, For the first Information entropy of the source data; For data source Weighting coefficients; For data source The predicted value; The sum of the reciprocals of the information entropy of all data sources is used to normalize the weighting coefficients.

[0012] Preferably, the step of calculating the first and second derivatives of the fused observations includes: The cylinder pressure change rate is obtained by performing a first-order difference operation on the crankshaft angle using the fused observations. The cylinder pressure change acceleration is obtained by performing a second-order difference operation on the crankshaft angle using the fused observations. The resulting extended feature quantities are used for subsequent extended state modeling, where the cylinder pressure change rate is: ; The acceleration due to cylinder pressure change is: ; in, crankshaft rotation angle The cylinder pressure at the corresponding moment; The increment of crankshaft angle; From the perspective of combustion characteristics; This represents the first-order change in pressure; This represents the second-order change in pressure.

[0013] Preferably, the step of performing state prediction and estimation based on the extended model using an adaptive Kalman filter algorithm with a variable structure switching mechanism includes: Construct a third-order extended state vector, including cylinder pressure, cylinder pressure change rate, and cylinder pressure change acceleration; An extended state transition equation and a fused observation equation are established to describe state evolution and measurement updates, respectively. An adaptive variable structure Kalman filter algorithm is adopted, which switches to a high-gain filter when combustion acceleration changes drastically and switches to a low-gain filter during the steady-state phase to improve estimation accuracy and response speed.

[0014] Preferably, the step of outputting the fused and optimized in-cylinder state prediction and estimation results includes: Extract the current cylinder pressure value based on the Kalman filter estimation results; The optimized cylinder pressure variation curve with crankshaft angle is output for the controller to use; Real-time in-cylinder pressure curves are used to monitor abnormal combustion conditions and assist in optimizing engine combustion.

[0015] It also provides a multi-source fusion estimation system for in-cylinder pressure of diesel engines based on adaptive Kalman filtering, including: The data acquisition module is used to collect diesel engine operating parameters and torque signals; The model prediction module is used to predict cylinder pressure based on zero-dimensional combustion modeling and neural networks; Torque reverse thrust module, used to reverse cylinder pressure based on crankshaft and connecting rod dynamics model; The entropy weight fusion module is used to dynamically allocate fusion weights based on the information entropy of each source data. An extended state estimation module is used for state prediction and estimation based on an extended model and a variable structure adaptive Kalman filter algorithm. The output module is used to output the fused and optimized in-cylinder pressure estimation results.

[0016] This invention provides a multi-source fusion estimation method for in-cylinder pressure in diesel engines based on adaptive Kalman filtering. It has the following beneficial effects: 1. This invention employs a multi-source fusion estimation method for in-cylinder pressure based on adaptive Kalman filtering. This method dynamically adjusts the fusion weights of different data sources in real time, achieving a balance between steady-state accuracy and transient response speed. Compared to existing technologies that rely on single-sensor cylinder pressure measurement or fixed-model estimation, this invention solves the problems of estimation lag, slow response, and large errors under complex operating conditions.

[0017] 2. This invention establishes an extended third-order state transition model and, by combining crankshaft angle difference to derive first and second-order derivative characteristics, effectively captures rapid dynamic changes during combustion, achieving the technical effect of improving the tracking accuracy of the combustion process. Compared with traditional schemes that only filter based on cylinder pressure itself, this avoids the defect of rapid accumulation of estimation errors during the rapid combustion stage.

[0018] 3. This invention introduces a dynamic weighting mechanism based on the inverse of information entropy. During the fusion model prediction and torque-based cylinder pressure calculation process, the weight allocation is automatically adjusted according to real-time data fluctuations, achieving the technical effect of suppressing noise interference and improving fusion stability. Compared to previous methods using fixed weights or manual experience weights, this overcomes the shortcomings of poor adaptability and low tolerance to extreme operating conditions.

[0019] 4. This invention incorporates an adaptive Kalman filter with a variable structure switching mechanism within the state estimation module. This filter dynamically adjusts its gain based on the intensity of combustion, achieving a comprehensive technical effect of rapid convergence during the combustion upheaval phase and smooth noise suppression during the steady-state phase. Compared to existing schemes using fixed-parameter Kalman filters, this significantly improves upon the problems of filter response lag and estimation failure under abnormal operating conditions. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 The flowchart shows the cylinder pressure multi-source fusion estimation method based on the adaptive Kalman filter algorithm. Figure 3 This is a system architecture diagram of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 --Appendix Figure 2This invention provides a multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering, comprising the following steps: S1. Collect the operating parameters and torque signals of the diesel engine as the data basis for model prediction of cylinder pressure estimation and torque back-inference of cylinder pressure estimation; S2. Based on the collected operating parameters, zero-dimensional combustion modeling and particle swarm optimization algorithm are used to calibrate combustion characteristic parameters, and a nonlinear mapping relationship between operating parameters and cylinder pressure is established by combining neural network to obtain the model-predicted cylinder pressure. S3. Based on the collected torque signal, the gas pressure torque and inertial torque are separated using the crank-connecting rod dynamics model. The contribution of a single cylinder is extracted and the torque back-reasoning cylinder pressure is obtained. S4. Based on the model-predicted cylinder pressure and the torque-inversely derived cylinder pressure, dynamically allocate fusion weights to generate fusion observations. S5. Based on the characteristics of the fusion observation as the crankshaft angle changes, calculate the first and second derivatives of the fusion observation, which correspond to the cylinder pressure change rate and cylinder pressure change acceleration, respectively, to form extended characteristic quantities; S6. Establish the state transition equation and observation equation, and use the adaptive Kalman filter algorithm with variable structure switching mechanism to predict and estimate the state based on the extended model. S7 outputs the fused and optimized in-cylinder state prediction and estimation results for engine combustion control and state monitoring.

[0023] For step S1, in this embodiment, to achieve high-precision multi-source fusion estimation of diesel engine cylinder pressure, it is first necessary to fully collect basic data closely related to in-cylinder combustion and motion states. This step provides necessary data support for subsequent combustion process modeling, neural network mapping, torque back-calculation, and fusion estimation. Generally, to ensure data integrity and synchronization, the acquisition system needs to be reasonably configured and time-series controlled to achieve temporal and spatial unification of information from different sources. Based on this, considering the impact of engine operating condition changes on combustion characteristics, it is crucial to rationally select operating condition parameters and output signals that can characterize dynamic changes in combustion.

[0024] In this embodiment, the collected data mainly includes the basic operating parameters of the diesel engine and the total crankshaft output torque signal. The operating parameters include at least the injection quantity, injection timing, intake valve opening, and engine speed. These parameters reflect the changing trends of initial combustion conditions, combustion duration, and combustion reaction rate, and therefore have a direct or indirect significant impact on in-cylinder pressure estimation.

[0025] In some embodiments, the fuel injection quantity can be calculated using a flow sensor in the fuel system or injector control commands. The injection timing can be directly read from a timing signal from the control unit. The intake valve opening is typically measured based on a position sensor installed in the intake system. The engine speed signal is acquired in real time using a crankshaft position sensor or a flywheel ring gear sensor; the accuracy of the engine speed is crucial for analyzing cylinder pressure change rate and inertial effects.

[0026] Alternatively, the total torque signal can be acquired through real-time measurement using a crankshaft torque sensor, or estimated by the engine controller based on fuel injection quantity, ignition timing, and intake parameters. Specifically, the torque signal needs to be processed by a low-pass filter to suppress high-frequency noise interference and time-corrected using a high-precision synchronizer to ensure synchronization with operating parameters.

[0027] In one possible implementation, to further improve signal quality, the torque signal can be preprocessed at the acquisition end based on sliding window filtering, while the operating parameters are optimized and estimated in real time through Kalman filtering or particle filtering to reduce the impact of measurement noise and environmental interference.

[0028] Furthermore, in this embodiment, to ensure the accuracy of cylinder pressure estimation, necessary data preprocessing is required for the collected operating parameters and torque signals. Preprocessing steps include: outlier removal, timing alignment, unit unification, and abnormal operating condition marking. For example, when a sharp fluctuation in engine speed or an interruption in fuel injection command is detected, it should be marked as an abnormal operating condition and removed or specially processed in subsequent modeling stages.

[0029] To facilitate subsequent model building, some parameters need to be normalized after acquisition. For example, the fuel injection quantity is scaled proportionally to the engine's maximum fuel injection quantity, the engine speed is normalized to the [0, 1] range according to the rated speed, and the intake valve opening is normalized to the proportion of the maximum opening. This processing helps improve the training stability and convergence speed during neural network modeling.

[0030] It should be noted that the torque signal is not only used for subsequent gas pressure torque extraction and inertial torque stripping, but also serves as an important basis for the reverse thrust cylinder pressure. In specific implementation, the total torque collected... Based on kinetic relationships, it is decomposed into gas pressure and torque. With reciprocating inertial torque The specific relationship is as follows: ; in, This represents the crankshaft rotation angle. The torque is the gas pressure generated by the expansion of the combustion gas in the cylinder; The reciprocating inertial torque generated by the mass of the piston assembly and connecting rod assembly.

[0031] Under normal circumstances, the inertial torque can be calculated using known formulas based on the engine's basic structural parameters and real-time speed signals, thereby accurately separating the gas pressure torque for use in the subsequent torque reverse cylinder pressure step.

[0032] As an improvement, auxiliary environmental parameters such as atmospheric pressure and coolant temperature can be recorded simultaneously during the data acquisition process to further refine the combustion process modeling in subsequent modeling processes, thereby improving the environmental adaptability and robustness of cylinder pressure estimation.

[0033] For step S2, in this embodiment, the collected engine operating parameters and torque signal data provide the necessary data foundation for subsequent modeling of the diesel engine in-cylinder combustion process and prediction of in-cylinder pressure. Generally, to fully characterize the dynamic behavior of in-cylinder combustion and its impact on cylinder pressure changes, it is necessary to reasonably model the key characteristics of the combustion process and further fit complex nonlinear relationships using intelligent modeling methods, thereby achieving accurate prediction of cylinder pressure changes under different operating conditions.

[0034] In this embodiment, the Weber combustion model (Wiebe function) is first used to describe the variation of the combustion rate in the diesel engine cylinder, based on the physical characteristics of the combustion process. Specifically, the Weber combustion model can effectively characterize the change in combustion heat release with crankshaft rotation angle. The evolution process, and its combustion heat release rate formula are expressed as: ; in, This represents the rate of heat release per unit crankshaft rotation. This refers to the crankshaft rotation angle; This is a dimensionless normalization coefficient for the combustion process, reflecting the intensity of combustion. The combustion rate growth index is dimensionless and characterizes the steepness of the initial growth of the combustion rate. The combustion rate decay index is dimensionless and controls the decay rate in the later stages of combustion. This is the combustion initiation angle, which is the crankshaft angle at which the combustion process begins; This represents the correlation coefficient for heat release during combustion.

[0035] Generally, the parameters of the Weber combustion model , , , Since the combustion characteristic parameters vary significantly under different operating conditions, this embodiment employs the Particle Swarm Optimization (PSO) algorithm to optimize and calibrate them, thereby improving the accuracy of combustion modeling. Particle Swarm Optimization is a swarm intelligence optimization algorithm that effectively finds the global optimum of multi-peak or complex functions by simulating the cooperative search behavior of particles in the search space, making it suitable for the multi-parameter calibration needs of combustion models.

[0036] Alternatively, when performing particle swarm optimization, the actual measured cylinder pressure data of the engine can be used as the optimization objective. The optimal parameter combination is determined by minimizing the sum of squared errors between the model's output cylinder pressure and the measured cylinder pressure. The specific optimization index can be set as the objective function: ; in, To optimize the objective function value; This refers to the crankshaft rotation angle; To measure cylinder pressure, This is the cylinder pressure calculated based on the combustion model.

[0037] In one possible implementation, to improve optimization efficiency and avoid local optima traps, mutation operations or an adaptive inertial weight strategy can be introduced into the particle swarm optimization algorithm to dynamically adjust the particle search step size.

[0038] In this embodiment, after the combustion characteristic parameters are calibrated, a multi-layer neural network is further trained to establish a nonlinear mapping relationship between operating parameters and in-cylinder pressure change curves. Generally, a feedforward neural network is used, coupled with a back propagation algorithm for training and optimization. The input layer nodes correspond to the collected operating parameters, such as fuel injection quantity, injection timing, intake valve opening, and engine speed, while the output layer nodes represent the in-cylinder pressure values ​​at different crankshaft angles.

[0039] In some embodiments, to enhance the model's fitting ability and generalization, a deep neural network (DNN) structure can be introduced, that is, multiple nonlinear transformation units are introduced in the hidden layer, and activation functions such as ReLU or Swish are selected to effectively capture the high-order nonlinear relationship between the operating conditions and cylinder pressure.

[0040] As an improvement, stochastic gradient descent (SGD) combined with a momentum term or the Adam optimizer can be used during neural network training to accelerate convergence and improve the model's fitting accuracy in different operating conditions.

[0041] Specifically, the trained neural network can quickly output the corresponding crankshaft rotation angle given real-time operating condition parameters. Model predicts cylinder pressure This provides stable and real-time basic observations for subsequent multi-source fusion estimation.

[0042] To further enhance the model's applicability, in some embodiments, sub-models can be trained separately according to different loads and speed ranges to form a working condition classification mapping system. In actual operation, the corresponding sub-network can be called according to the working condition classification to predict cylinder pressure, thereby further improving the overall prediction accuracy and robustness.

[0043] In summary, by combining Weber combustion modeling, particle swarm optimization parameter calibration, and neural network nonlinear fitting, this embodiment effectively achieves dynamic modeling of the in-cylinder combustion process of a diesel engine and efficient prediction of cylinder pressure change curves, laying a solid modeling foundation for subsequent adaptive Kalman filter fusion estimation.

[0044] For step S3, in this embodiment, after completing the cylinder pressure prediction based on combustion modeling and neural network, to further improve the accuracy of the estimation and dynamic response performance, it is also necessary to extract the gas pressure contribution based on the diesel engine crankshaft torque signal and deduce the cylinder pressure change. Generally, since the total crankshaft torque includes not only the gas pressure torque generated by the expansion of combustion gases but also the inertial torque caused by the reciprocating motion of the piston, it is necessary to extract the gas pressure torque contribution of a single cylinder through reasonable dynamic modeling and signal separation methods in order to accurately deduce the single-cylinder cylinder pressure curve.

[0045] In this embodiment, the collected crankshaft total torque signal is first processed and separated based on the fundamental principles of diesel engine dynamics. Generally, the crankshaft total torque... This can be expressed as gas pressure and torque. With reciprocating inertial torque Superposition: ; in, For crankshaft rotation The total torque measured at the location; This is the torque component generated by gas expansion; This refers to the inertial load torque generated by components such as the piston and connecting rod. Specifically, the inertial torque component can be calculated based on engine structural parameters (piston mass, connecting rod length-to-short ratio, etc.) and the rate of change of engine speed. Alternatively, an inertial torque calculation model derived from the kinematic equations of the crank-connecting rod mechanism can be used for online calculation based on engine speed and acceleration. After separating the inertial torque, the gas pressure torque is obtained. This serves as an important basis for determining the reverse thrust cylinder pressure.

[0046] In one possible implementation, to reduce the impact of rotational speed fluctuations on the accuracy of inertial torque calculation, a crankshaft position differential filter can be introduced to estimate angular acceleration in real time and correct the inertial torque calculation results accordingly.

[0047] In this embodiment, the single-cylinder gas pressure torque is obtained. Subsequently, based on the geometric torque relationship of the crank-connecting rod mechanism, the cylinder pressure was derived. With gas pressure and torque The calculation formula between them is: ; in, For the combustion gases in the cylinder at the crankshaft angle Pressure at the location; The separated gas pressure torque components; This is the lever arm length of the crank-connecting rod mechanism, and its specific expression can be derived from the crank radius. Link length and crankshaft angle The derivation yields the following general form: ; in, Where is the crank radius; The length of the link; It is a cosine function.

[0048] As an improvement, in order to further compensate for the influence of system friction torque during cylinder pressure reverse thrust, friction torque can be introduced into the total torque separation model. It also corrects the gas pressure and torque, thereby improving the accuracy of reverse propulsion.

[0049] In some embodiments, a multi-cylinder separation strategy can be introduced to distribute the total gas torque according to the combustion phase of a single cylinder, thereby obtaining an independent cylinder pressure curve for each cylinder, which is particularly suitable for cylinder pressure estimation in multi-cylinder parallel operation mode.

[0050] Under normal circumstances, the cylinder pressure pushed out in reverse The curve exhibits distinct four-stage characteristics of compression, ignition, expansion, and exhaust as the crankshaft angle rotates, closely matching the actual cylinder pressure curve. This provides an accurate and real-time reference signal for subsequent multi-source fusion and adaptive filtering.

[0051] For step S4, in this embodiment, to fully integrate the two estimation results—model-predicted cylinder pressure and torque-inverse cylinder pressure—and further improve the accuracy and robustness of diesel engine cylinder pressure estimation, it is necessary to dynamically adjust the fusion weights based on the real-time data quality of the two types of observation sources. Generally, since the reliability of different data sources varies under different operating conditions and combustion stages, it is difficult to balance steady-state accuracy and dynamic response performance using a fixed fusion weight method. Therefore, this embodiment proposes a data source weighting method based on real-time information entropy evaluation to achieve adaptive fusion of cylinder pressure observations.

[0052] In this embodiment, the cylinder pressure is first predicted by the model within a sliding time window. With torque reverse thrust cylinder pressure The data distribution characteristics are statistically analyzed in real time. Generally, by calculating the information entropy of each observation, the uncertainty and signal-to-noise ratio level of the corresponding observation can be effectively reflected. The larger the information entropy value, the stronger the data volatility and the worse the stability; the smaller the information entropy value, the more stable the data tends to be and the higher the reliability.

[0053] Specifically, the information entropy The calculation can be derived based on the standard Shannon entropy formula for discretized data sequences. Its entropy is defined as: ; in, Indicates the first Shannon information entropy of the observed sources; Indicates the first A set of values ​​for a data source or variable; For data sequences The Middle A specific data point; Indicates the first The observation source in the first The probability of each state occurring; The number of discrete states.

[0054] In one possible implementation, to reduce sampling bias in the entropy estimation process, a sliding window weighted histogram estimation method can be used, combined with kernel density estimation to further smooth the probability distribution and improve the stability of entropy calculation.

[0055] Alternatively, after the entropy value is calculated, this embodiment dynamically determines the fusion weighting coefficients based on the inverse proportion of the information entropy of each observation source. Specific fusion weights... The calculation formula is: ; in, For the first Information entropy of the source data; For data source Weighting coefficients; For data source The predicted value; The sum of the reciprocals of the information entropy of all data sources is used to normalize the weighting coefficients.

[0056] Generally, when the entropy value of a certain observation source is small, its corresponding weight... The weight of observation sources with high entropy values ​​will be automatically increased, thus giving them a larger weight in the fusion result; conversely, the weight of observation sources with high entropy values ​​will be reduced, thus minimizing their adverse impact on the final fusion result.

[0057] In some embodiments, to avoid drastic changes in weights caused by abnormal fluctuations in entropy values, upper and lower limits can be introduced for entropy values ​​or an exponential smoothing strategy can be adopted to apply hysteresis filtering to changes in fused weights, thereby improving the overall stability of the system.

[0058] In this embodiment, the fused observations The result is obtained through a weighted summation method, specifically: ; in, For the cylinder pressure observation after fusion; For data source Weighting coefficients; For the first The observation source at the crankshaft rotation angle Estimated cylinder pressure at the location.

[0059] As a further optimization method, the entropy weight calculation window size can be dynamically adjusted according to the characteristics of the combustion stage (such as compression stroke and combustion expansion stroke) during the fusion stage to adapt to the data fluctuation changes during combustion and improve real-time adaptive performance.

[0060] For step S5, in this embodiment, to further improve the dynamic response performance of the system state modeling, the cylinder pressure change rate and acceleration are introduced as extended feature quantities for the obtained fused cylinder pressure observations. Generally, the change in diesel engine cylinder pressure signal with crankshaft angle is quite complex, especially during the combustion acceleration and expansion stages. Therefore, relying solely on cylinder pressure itself is insufficient to fully characterize the dynamics of the combustion process. By introducing the first and second derivative features of cylinder pressure, the expressive power and prediction accuracy of the state model can be significantly enhanced, laying a solid foundation for subsequent Kalman filter state estimation.

[0061] In this embodiment, based on the fused cylinder pressure observation... First, perform a first-order difference operation to calculate the change with crankshaft rotation angle. The rate of change of cylinder pressure. Specifically, the rate of change of cylinder pressure. Defined as: ; in, This represents the first-order change in pressure; crankshaft rotation angle The cylinder pressure at the corresponding moment; The increment of crankshaft angle; This refers to the crankshaft rotation angle.

[0062] In general, to improve the noise resistance of differential calculations, a sliding window weighted difference method can be used to fit and differ between multiple adjacent points, thereby reducing the impact of transient noise on derivative calculations.

[0063] As an alternative, during the most intense combustion phase (e.g., the maximum heat release range), the [heat release range] can be appropriately reduced. To capture higher frequency cylinder pressure variation characteristics.

[0064] In this embodiment, after obtaining the first-order rate of change, the second derivative of the cylinder pressure with respect to the crankshaft angle is further calculated, i.e., the cylinder pressure change acceleration. The calculation formula is: ; in, crankshaft rotation angle The cylinder pressure at the corresponding moment; The increment of crankshaft angle; This refers to the crankshaft rotation angle; This represents the first-order change in pressure; This represents the second-order change in pressure.

[0065] Specifically, the acceleration due to cylinder pressure change can reflect the trend of combustion rate change, which is of great significance for identifying abnormal combustion states such as combustion phase, ignition delay, and knocking.

[0066] In one possible implementation, to improve the accuracy of cylinder pressure change acceleration, a Savitzky-Golay filter can be introduced when calculating the first derivative to perform local polynomial fitting and differentiation, and then the second derivative can be calculated based on the smoothed first derivative.

[0067] As an improvement, the length of the derivative calculation window can be adaptively adjusted for different operating conditions. For example, under low-speed, high-load conditions, the sliding window width can be appropriately increased to improve the smoothness and stability of the second derivative estimation.

[0068] It should be noted that the accuracy of derivative characteristics directly affects the state prediction performance of the subsequent Kalman filter. Therefore, in the differential and filtering design process, the sampling rate, system dynamic response characteristics and noise level should be comprehensively considered to achieve the best balance.

[0069] For step S6, in this embodiment, after extracting the fused cylinder pressure observations and their first and second derivative features, it is further necessary to establish a corresponding state-space model, and based on this, an adaptive Kalman filter algorithm is used to realize the dynamic prediction and estimation of the diesel engine cylinder pressure. Generally, traditional first-order state modeling is prone to estimation lag or oscillation problems when dealing with stages of drastic combustion changes. Therefore, this embodiment introduces extended state variables and combines them with a variable structure switching mechanism to dynamically adjust the filter gain to adapt to the nonlinear characteristic changes in different combustion stages, thereby significantly improving estimation accuracy and system response speed.

[0070] In this embodiment, a third-order extended state vector is first constructed. Specifically, the extended state vector is defined as: ; in, Indicates crankshaft rotation angle The pressure inside the cylinder at that location; This represents the first-order change in cylinder pressure; This represents the second-order change in cylinder pressure.

[0071] Generally, the state transition process can be described based on the dynamic approximation of cylinder pressure changes. Specifically, the state transition equation can be modeled as: ; in, For the corner The state vector at any given time; This is the state transition matrix, reflecting the evolutionary relationships between the state components; At crankshaft angle The system state vector at time t; Let be the process noise vector, assumed to be white noise with zero mean.

[0072] In one possible implementation, the state transition matrix It can be approximately derived from simple integrals, for example: ; in, This represents the increment of the crankshaft rotation angle; It is the square of the crankshaft angle increment.

[0073] As an alternative, in order to balance accuracy and computational complexity, the turning step size can be appropriately refined in the range of rapid combustion changes (such as the initial stage of ignition-expansion), while the step size can be widened in the steady-state stage to reduce the amount of computation.

[0074] In this embodiment, the fusion observations Its first and second derivatives constitute the observation vector. The corresponding observation equation is modeled as follows: ; in, The observation vector contains the fused cylinder pressure and its first and second derivative observations; This is the observation matrix, used to map state variables to the observation space; At crankshaft angle The system state vector at time t; To observe the noise vector, assume it to be white noise with zero mean.

[0075] Specifically, in this embodiment... Typically an identity matrix or an approximate identity matrix, it reflects the one-to-one correspondence between actual observations and extended state variables.

[0076] To address the system nonstationarity caused by drastic changes in combustion dynamics, this embodiment introduces a variable structure switching mechanism in the Kalman filter design. Generally, the filter gain is dynamically adjusted by monitoring the magnitude of the system residuals and the evolution trend of the covariance matrix in real time. Specifically, when the residuals increase rapidly and the variance expands sharply, the filter switches to a high-gain mode to improve the system's response speed to changes; when the residuals stabilize and the variance converges, the filter switches back to a low-gain mode to suppress noise amplification effects.

[0077] In some embodiments, to further enhance the adaptability of the filter, an adaptive process noise covariance adjustment strategy can be introduced to dynamically correct the process noise matrix based on the observation consistency test results. With observation noise matrix To adapt to the changes in the dynamic characteristics of the system under different combustion stages.

[0078] For step S7, in this embodiment, after completing the fused state estimation and adaptive Kalman filter optimization, in order to fully utilize the value of the diesel engine cylinder pressure estimation data, the final fused and optimized in-cylinder state needs to be output and applied to engine combustion control and real-time state monitoring. Generally, high-precision cylinder pressure information can not only be used for closed-loop control of the combustion process, but also for abnormal operating condition detection, combustion optimization, and fault prediction. Therefore, outputting the cylinder pressure estimation results and rationally organizing the data interface is an important part of the technical system of this invention.

[0079] In this embodiment, the current crankshaft angle is first extracted based on the output of the adaptive Kalman filter. The estimated cylinder pressure at that location. Generally, the estimated cylinder pressure is taken as the first term in the extended state vector, i.e.: ; in, Indicates at the corner The fusion estimation cylinder pressure value at the location; To expand the state vector The cylinder pressure component.

[0080] As an alternative, to smooth the estimated trajectory, the cylinder pressure sequence of the filtered output can be processed by moving average or low-pass filtering to remove high-frequency noise components and improve curve smoothness and system stability.

[0081] In this embodiment, the output fused and optimized cylinder pressure curve changes with the crankshaft angle, which can be used to drive the controller for combustion phase adjustment, fuel injection strategy optimization, and boost control parameter correction. Specifically, combustion phase shift detection is typically based on characteristic quantities such as the location of the maximum cylinder pressure and the angle at which the maximum pressure rise rate occurs. (Cylinder pressure maximum value) The extraction can be accomplished using the following formula: ; in, This indicates the maximum cylinder pressure value detected within the effective combustion range; Indicates at the corner The fusion estimation cylinder pressure value at the location; The effective analysis range for combustion typically covers the period from the end of compression to the initial stage of expansion.

[0082] Under normal circumstances, the maximum cylinder pressure The corresponding crankshaft angle can be used as a combustion characteristic parameter for the engine control unit (ECU) to call in real time to achieve closed-loop control or combustion anomaly diagnosis.

[0083] In one possible implementation, the system can analyze combustion stability indicators based on the cylinder pressure change rate and acceleration. For example, by detecting the extreme value distribution of cylinder pressure change acceleration, it can identify possible knocking or abnormal combustion events.

[0084] As an improvement, more comprehensive combustion indicators, such as the combustion phase index (CAI) and combustion duration (CAD) for each cycle, can be derived based on the estimated cylinder pressure curve and derivative information, providing more comprehensive data support for engine performance optimization and emission control strategies.

[0085] It should be noted that, in order to ensure the reliability of cylinder pressure data, this embodiment recommends setting up a filter consistency check module. When the system residual exceeds a preset threshold or the covariance matrix expands abnormally, the redundancy mechanism is automatically activated or the system switches to a safe mode to ensure the accuracy of the output data and the stability of the system.

[0086] The adaptive Kalman filter-based diesel engine cylinder pressure multi-source fusion estimation system described below can be referred to in conjunction with the adaptive Kalman filter-based diesel engine cylinder pressure multi-source fusion estimation method described above.

[0087] Please see the appendix Figure 2 The present invention also provides a multi-source fusion estimation system for in-cylinder pressure of diesel engines based on adaptive Kalman filtering, comprising: The data acquisition module is used to collect diesel engine operating parameters and torque signals; The model prediction module is used to predict cylinder pressure based on zero-dimensional combustion modeling and neural networks; Torque reverse thrust module, used to reverse cylinder pressure based on crankshaft and connecting rod dynamics model; The entropy weight fusion module is used to dynamically allocate fusion weights based on the information entropy of each source data. An extended state estimation module is used for state prediction and estimation based on an extended model and a variable structure adaptive Kalman filter algorithm. The output module is used to output the fused and optimized in-cylinder pressure estimation results.

[0088] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering, characterized in that, Includes the following steps: Collect the operating parameters and torque signals of the diesel engine as the data basis for model prediction of cylinder pressure estimation and torque-based cylinder pressure estimation; Based on the collected operating parameters, zero-dimensional combustion modeling and particle swarm optimization algorithm are used to calibrate combustion characteristic parameters, and a nonlinear mapping relationship between operating parameters and cylinder pressure is established by combining neural network to obtain the model-predicted cylinder pressure. Based on the collected torque signal, the gas pressure torque and inertial torque are separated using the crank-connecting rod dynamics model. The contribution of a single cylinder is extracted and the torque is used to back-calculate the cylinder pressure. Based on the real-time information entropy change of cylinder pressure predicted by the model and the reverse cylinder pressure calculated by torque, the fusion weights are dynamically allocated to generate fusion observations. Based on the characteristics of the fusion observation as a function of crankshaft angle, the first and second derivatives of the fusion observation are calculated, which correspond to the cylinder pressure change rate and cylinder pressure change acceleration, respectively, forming extended characteristic quantities. State transition equations and observation equations are established, and an adaptive Kalman filter algorithm with a variable structure switching mechanism is used for state prediction and estimation based on the extended model. The output is a fused and optimized in-cylinder state prediction and estimation result, which is used for engine combustion control and state monitoring.

2. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering as described in claim 1, characterized in that, The acquisition of diesel engine operating parameters and torque signals includes: Collect real-time data closely related to in-cylinder combustion, such as fuel injection quantity, injection timing, intake valve opening, and engine speed of the diesel engine. The total torque signal output from the diesel engine crankshaft is collected as the basis for cylinder pressure reverse thrust; The acquired signals are filtered and synchronized to ensure time alignment and data integrity.

3. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The step of establishing a nonlinear mapping relationship between operating parameters and cylinder pressure using a neural network includes: The Weber combustion model was used to describe the in-cylinder combustion rate, and the combustion initiation angle, combustion duration, and combustion shape parameters were calibrated using the particle swarm optimization algorithm. A multi-layer neural network is trained to fit the nonlinear relationship between operating parameters and cylinder pressure curves, and the output model predicts cylinder pressure. Establish the relationship between combustion heat release and pressure changes, ensuring that cylinder pressure changes with crankshaft angle in accordance with: ； in, The correlation coefficient for heat release during combustion; This is the combustion rate index; It is an index of the combustion process; This refers to the crankshaft rotation angle; This is a reference value for the crankshaft angle; It represents the combustion heat release rate per unit crankshaft rotation angle.

4. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The step of extracting the contribution of a single cylinder and reverse-engineering to obtain the torque reverse-engineering cylinder pressure includes: The total crankshaft torque is separated into gas pressure torque and reciprocating inertial torque using dynamic formulas; Extract the gas pressure and torque of a single cylinder, and deduce the cylinder pressure change of a single cylinder based on the contribution of a single cylinder. Derivation of cylinder pressure based on the torque relationship of crank-connecting rod mechanism With gas pressure and torque The calculation formula is: ； in, It is a function of the change in lever arm composed of crank radius and connecting rod length.

5. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The step of dynamically allocating fusion weights based on the real-time information entropy changes of cylinder pressure predicted by the model and the cylinder pressure inversely derived from torque includes: Within a sliding time window, the statistical model predicts the cylinder pressure and the torque inversely calculates the cylinder pressure data distribution. Calculate the information entropy value of each data set, and allocate the fusion weights according to the inverse ratio of the entropy value.

6. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 5, characterized in that, The formula for calculating the fusion weight is: ； in, For the first Information entropy of the source data; For data source Weighting coefficients; For data source The predicted value; The sum of the reciprocals of the information entropy of all data sources is used to normalize the weighting coefficients.

7. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The steps for calculating the first and second derivatives of the fused observations include: The cylinder pressure change rate is obtained by performing a first-order difference operation on the crankshaft angle using the fused observations. The cylinder pressure change acceleration is obtained by performing a second-order difference operation on the crankshaft angle using the fused observations. The resulting extended feature quantities are used for subsequent extended state modeling, where the cylinder pressure change rate is: ； The acceleration due to cylinder pressure change is: ； in, crankshaft rotation angle The cylinder pressure at the corresponding moment; The increment of crankshaft angle; This refers to the crankshaft rotation angle; This represents the first-order change in pressure; This represents the second-order change in pressure.

8. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The steps for state prediction and estimation based on the extended model using the adaptive Kalman filter algorithm with a variable structure switching mechanism include: Construct a third-order extended state vector, including cylinder pressure, cylinder pressure change rate, and cylinder pressure change acceleration; An extended state transition equation and a fused observation equation are established to describe state evolution and measurement updates, respectively. An adaptive variable structure Kalman filter algorithm is adopted, which switches to a high-gain filter when combustion acceleration changes drastically and switches to a low-gain filter during the steady-state phase to improve estimation accuracy and response speed.

9. The multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filtering according to claim 1, characterized in that, The steps for outputting the fused and optimized in-cylinder state prediction and estimation results include: Extract the current cylinder pressure value based on the Kalman filter estimation results; The optimized cylinder pressure variation curve with crankshaft angle is output for the controller to use; Real-time in-cylinder pressure curves are used to monitor abnormal combustion conditions and assist in optimizing engine combustion.

10. A multi-source fusion estimation system for in-cylinder pressure of a diesel engine based on adaptive Kalman filter, applied to the multi-source fusion estimation method for in-cylinder pressure of a diesel engine based on adaptive Kalman filter as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect diesel engine operating parameters and torque signals; The model prediction module is used to predict cylinder pressure based on zero-dimensional combustion modeling and neural networks; Torque reverse thrust module, used to reverse cylinder pressure based on crankshaft and connecting rod dynamics model; The entropy weight fusion module is used to dynamically allocate fusion weights based on the information entropy of each source data. An extended state estimation module is used for state prediction and estimation based on an extended model and a variable structure adaptive Kalman filter algorithm. The output module is used to output the fused and optimized in-cylinder pressure estimation results.

Images