Diesel engine thermal compression ratio intelligent calculation method based on machine learning

Abstract

This invention discloses an intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning, comprising the following steps: S1, acquiring diesel engine thermal operation data and generating a state data sequence; S2, performing continuous-time state encoding on the state data sequence to construct a state representation sequence; S3, performing feature contribution analysis on the state representation sequence to generate a correlated feature sequence; S4, performing state correlation propagation and compression ratio optimization calculation on the correlated feature sequence to obtain a predicted compression ratio sequence; S5, constructing a theoretical pressure curve and an in-cylinder pressure curve based on the predicted compression ratio sequence, and calculating the curve deviation sequence; S6, performing feedback updates based on the curve deviation sequence and outputting the calculation results. This invention, combined with dual-state flow liquid neural networks, possesses advantages such as strong adaptability to thermal engine states, high accuracy in compression ratio calculation, high stability in complex operating condition analysis, and strong continuous dynamic optimization capability.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a machine learning-based intelligent calculation method for the thermal compression ratio of a diesel engine. Background Technology

[0002] Compression ratio is a crucial operating parameter for diesel engines, directly affecting their combustion efficiency, power output, fuel economy, and emissions performance. The geometric compression ratio is typically defined as the ratio between the total cylinder volume when the piston is at bottom dead center and the combustion chamber volume when the piston is at top dead center. In traditional technologies, diesel engine compression ratios are mostly calculated based on cold-state structural parameters, obtaining the corresponding geometric compression ratio by measuring the combustion chamber volume and cylinder structural parameters.

[0003] However, for high-performance diesel engines, under hot-engine conditions, the high in-cylinder combustion pressure leads to significant cylinder clearance leakage, piston thermal expansion, and combustion chamber thermal deformation. This results in a substantial difference between the actual combustion chamber volume under operating conditions and the cold-state volume, causing the actual hot-engine compression ratio to deviate significantly from the theoretical geometric compression ratio. If the cold-state geometric compression ratio is continued to be used in diesel engine thermodynamic analysis, combustion state analysis, and performance prediction, it can easily lead to increased pressure prediction errors, decreased accuracy of explosion pressure analysis, and distortion of operating state analysis results, making it difficult to meet the precise analysis requirements of high-performance diesel engines under complex operating conditions.

[0004] In existing technologies, some solutions calculate the compression ratio of diesel engines using pressure curve fitting or machine learning methods, predicting the compression ratio by establishing a mapping relationship between operating parameters and the compression ratio. However, existing technologies lack effective modeling capabilities for the continuous-time state evolution relationship between different time steps, the state change relationship across multiple time scales, and the pressure propagation deviation relationship under diesel engine operating conditions. Furthermore, they lack a closed-loop optimization mechanism to update model parameters based on pressure curve deviation results, leading to low accuracy and stability in compression ratio calculations under complex thermal engine conditions.

[0005] Therefore, how to provide a machine learning-based intelligent calculation method for the thermal compression ratio of diesel engines is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent calculation method for the compression ratio of a diesel engine based on machine learning. This invention performs multi-scale continuous-time state encoding operations on the state data sequence through a dual-state flow liquid neural network, and combines the temporal SHAP algorithm, particle swarm optimization algorithm, and improved Hausdorff algorithm to perform state analysis on the state evolution relationship, pressure propagation relationship, and compression state consistency relationship between different time steps during the operation of the diesel engine. This enables intelligent optimization calculation and feedback update of the diesel engine compression ratio, and has the advantages of strong adaptability to the engine state, high accuracy of compression ratio calculation, high stability in complex operating condition analysis, and strong continuous dynamic optimization capability.

[0007] A machine learning-based intelligent calculation method for the thermal compression ratio of a diesel engine, according to an embodiment of the present invention, includes the following steps: S1. Acquire and preprocess the operating data of the diesel engine under hot engine operation conditions to generate a state data sequence; S2. Using a dual-state flow liquid neural network, perform continuous-time state encoding on the state data sequence, extract state change features based on the state evolution relationship between the encoded states at different time steps, and construct a state representation sequence. S3. Based on the state representation sequence, the temporal SHAP algorithm is used to perform feature contribution calculation. Key state features are selected according to the feature contribution corresponding to each state change feature to form an associated feature sequence. S4. In the dual-state flow liquid neural network, state association propagation operation is performed on the associated feature sequence, and the particle swarm optimization algorithm is used to perform compression ratio optimization calculation based on the propagation result to obtain the predicted compression ratio sequence. S5. Based on the predicted compression ratio sequence and the state data sequence, construct the theoretical pressure curve and the in-cylinder pressure curve, and perform trend smoothing. Use the improved Hausdorff algorithm to perform compression state consistency analysis on the smoothing results and calculate the curve deviation sequence. S6. Based on the curve deviation sequence, perform feedback update operation on the parameters of the dual-state flow liquid neural network, and re-execute the compression ratio optimization calculation to obtain the actual heat engine compression ratio.

[0008] Optionally, the operating data includes cylinder pressure data, crankshaft angle data, and speed data collected by intelligent sensors when the diesel engine is in a hot-engine operating state. The preprocessing includes time synchronization, abnormal data removal, and normalization processing. Compared with traditional liquid neural networks, the dual-state flow liquid neural network introduces a dual-state evolution structure based on short-period state flow and long-period state flow. Through the dual-state evolution structure, multi-scale continuous-time state encoding and state evolution calculation operations are performed on the state data sequence. The state evolution relationship represents the continuous-time state association relationship formed between the encoded states corresponding to different time steps, based on the state fluctuation relationship and the state evolution relationship.

[0009] Optionally, S2 specifically includes: S21. Input the state data sequence into the dual-state flow liquid neural network, perform continuous-time state mapping operation on the state data sequence, and generate the initial encoded state corresponding to different time steps. S22. In the dual-state evolution structure, high-frequency instantaneous response calculation is performed on the initial encoded state through a short-cycle state flow, and short-cycle state features are extracted based on the state fluctuation relationship between adjacent time steps to generate a short-cycle state sequence. S23. In the dual-state evolution structure, long-cycle state evolution calculation operations are performed on the initial encoded state through a long-cycle state flow. Long-cycle state features are extracted based on the state evolution relationship between different time steps to generate a long-cycle state sequence. S24. Perform multi-scale state fusion operation on the short-period state sequence and the long-period state sequence to form a state representation sequence.

[0010] Optionally, the high-frequency instantaneous response calculation and long-period state evolution calculation specifically include: In a short-cycle state flow, a continuous state difference calculation operation is performed on the initial encoded states of adjacent time steps to calculate the magnitude and direction of state change between different time steps. Calculate the difference in the amplitude of state change between adjacent time steps, and perform a direction consistency judgment operation on the direction of state change. Define the same direction of change as 1 and the opposite direction of change as 0 to obtain the direction consistency result. A weighted correlation calculation operation is performed based on the result of consistency between the amplitude difference of state change and direction to obtain the short-period state characteristics corresponding to each time step; In a long-cycle state flow, a cross-time-step state association calculation operation is performed on the initial encoded state at different time steps to calculate the state difference value between the corresponding encoded states at different time steps. Then, a time decay calculation operation is performed on the state difference value in combination with the time span to obtain the corresponding state association value. Calculate long-period state characteristics based on the state correlation values ​​corresponding to different time steps; The short-period state features and long-period state features are arranged and combined respectively to generate the corresponding short-period state sequences and long-period state sequences.

[0011] Optionally, S3 specifically includes: S31. Using the temporal SHAP algorithm, perform feature contribution calculation operations on the feature changes of each state in the state representation sequence, specifically including: According to the preset contribution calculation rules, the contribution value calculation operation is performed on each state change feature in the state representation sequence to obtain the single-step contribution value of different state change features at each time step. Based on the single-step contribution value of the same state change characteristics at different time steps, perform time-related statistical operations to obtain the corresponding cumulative contribution value; Based on the cumulative contribution value, generate a sequence of contribution values ​​corresponding to the characteristics of each state change; S32. Based on the sequence of contribution values, calculate the magnitude and direction of contribution change of different state change characteristics at each time step; S33. Perform contribution fluctuation statistics operation on the contribution change amplitude at different time steps, and calculate the state contribution value corresponding to each state change characteristic in combination with the contribution change direction. S34. Extract state change features whose state contribution value is greater than the preset contribution threshold as key state features and generate associated feature sequences.

[0012] Optionally, S33 specifically includes: S331. According to the preset time window size, the contribution change amplitude and contribution change direction of each time step are divided into sliding windows to generate multiple contribution fluctuation intervals. S332. In each contribution fluctuation interval, calculate the difference between the contribution change amplitudes of adjacent time steps to obtain the corresponding contribution fluctuation value. S333. Based on the direction of contribution change, count the number of times the direction is consistent and the number of times the direction switches between adjacent time steps in each contribution fluctuation interval to obtain the corresponding direction change results. S334. Adaptive Kalman filtering algorithm is used to perform state prediction calculation operation on each contribution fluctuation value, generate predicted contribution state value, and construct state observation value based on the direction change result. S335. Calculate the state residual for the corresponding time step based on the difference between the predicted contribution state value and the observed state value, and perform a filter gain correction operation on the predicted contribution state value based on the state residual to obtain the state contribution value corresponding to each state change feature. S336. Based on the contribution values ​​of each state, form the corresponding state contribution sequence.

[0013] Optionally, S4 specifically includes: S41. In a dual-state flow liquid neural network, perform a state association propagation operation on the associated feature sequence and calculate the state propagation value corresponding to the key state features between different time steps. S42. Calculate the propagation change value and propagation direction between the state propagation values ​​of each adjacent time step, and perform state propagation stability analysis to generate a state stable sequence. S43. Based on the stable state sequence, perform a propagation weight allocation operation on each state propagation value, and perform a compression ratio parameter mapping operation on the associated feature sequence according to the propagation weight to generate a compression ratio parameter sequence. S44. Using the particle swarm optimization algorithm, multiple compression ratio search particles are generated according to the preset number of particles, and the compression ratio parameter initialization operation is performed on each compression ratio search particle according to the compression ratio parameter sequence to generate the particle parameter sequence. S45. Based on the particle parameter sequence, perform fitness calculation operations on each search particle with a compression ratio, and filter target particles according to fitness to generate a target particle sequence. S46. Based on the target particle sequence, update the compression ratio parameters of each compression ratio search particle, and iteratively execute the fitness calculation and compression ratio parameter update operations until the fitness reaches the preset convergence threshold. S47. Use the compression ratio parameter at the end of the iteration as the predicted compression ratio to form a predicted compression ratio sequence.

[0014] Optionally, S45 specifically includes: S451. Based on the particle parameter sequence, perform compression state deduction operation on the compression ratio parameters corresponding to each compression ratio search particle to generate theoretical pressure sequences corresponding to different time steps. S452. Based on the theoretical pressure sequence and the cylinder pressure data, calculate the pressure difference, pressure change direction difference, and pressure evolution trend difference for each time step to generate a pressure deviation sequence. S453. Based on the pressure deviation sequence, perform a time-series deviation propagation calculation operation on the pressure difference corresponding to different time steps, calculate the deviation propagation value between each time step, and generate a deviation propagation sequence. S454. Using the HDBSCAN algorithm, perform pressure instability density clustering on the deviation propagation sequence, specifically including: A state feature vector is constructed based on the deviation propagation value, pressure difference, and pressure evolution trend difference at different time steps; The Euclidean distance between each state feature vector is calculated as the state distance value. The core neighborhood range of each time step is determined based on the state distance value. The local density value of the corresponding time step is calculated based on the state distance value in each core neighborhood range. Hierarchical clustering and connection operations are performed on time steps where local density values ​​continuously increase, generating multiple pressure instability clusters; Calculate the changing trend of deviation propagation value at different time steps in each pressure instability cluster, and take the cluster with continuously increasing deviation propagation value as the pressure instability interval to generate a pressure instability sequence; S455. Based on the pressure instability sequence, perform state-coupled fluctuation analysis on the difference between the pressure change direction and the pressure evolution trend at different time steps, calculate the state instability value of each pressure instability interval, and generate the state instability sequence. S456. Based on the state instability sequence and the state stability sequence, calculate the pressure stability value, state stability value and deviation propagation suppression value corresponding to each compression ratio search particle, and generate a multi-dimensional state evaluation sequence. S457. Based on the multidimensional state evaluation sequence, perform multi-objective score calculation operation on each compression ratio search particle to obtain the score value of each compression ratio search particle. S458. Extract the compression ratio search particles whose score values ​​meet the preset screening conditions as target particles to form a target particle sequence.

[0015] Optionally, S5 specifically includes: S51. Based on the predicted compression ratio sequence, perform theoretical pressure mapping calculation operation on the predicted compression ratio corresponding to different time steps to generate theoretical pressure values ​​for each time step and form a theoretical pressure curve. S52. Based on the cylinder pressure data in the state data sequence, perform pressure curve reconstruction operation in chronological order to generate the cylinder pressure curve. S53. Perform Savitzky-Golay trend smoothing on the theoretical pressure curve and the in-cylinder pressure curve respectively to generate the corresponding smoothed pressure sequence. S54. Based on each smoothed pressure sequence, perform a local pressure change interval division operation on the theoretical pressure curve and the cylinder pressure curve to generate multiple pressure change intervals. S55. Using the improved Hausdorff algorithm, a compression state consistency analysis is performed on the theoretical pressure curve and the in-cylinder pressure curve in each pressure variation range, specifically including: Based on the time window corresponding to each pressure change range, the theoretical pressure value of each time step in the theoretical pressure curve is used as the reference point. Multiple neighboring pressure points are searched within the time window corresponding to the cylinder pressure curve to generate the corresponding candidate matching point set. Calculate the corresponding local curve distance based on the pressure distance between each neighboring pressure point and the corresponding reference point in the candidate matching point set; The pressure change direction of each adjacent pressure point and the corresponding reference point is statistically analyzed, and a direction consistency judgment operation is performed. Adjacent pressure points with the same pressure change direction are marked as valid matching points. Based on the local curve distance, perform distance weight allocation operation on each valid matching point according to the preset weight allocation rule, and calculate the corresponding weighted curve distance value; Based on the distance values ​​of each weighted curve, calculate the Hausdorff distance value corresponding to each pressure change interval; S56. Based on the Hausdorff distance values ​​of each interval, calculate the distance change value and direction of distance change between adjacent time steps, and perform curve deviation correlation analysis operation in combination with the pressure change values ​​between corresponding time steps to generate a curve deviation sequence.

[0016] Optionally, S6 specifically includes: S61. Based on the curve deviation sequence, calculate the curve deviation change value and the curve deviation change direction corresponding to different time steps, and generate the deviation change sequence. S62. Perform feedback update operation on the parameters in the dual-state flow liquid neural network according to the deviation change sequence; S63. Based on the updated dual-state flow liquid neural network, re-execute the compression ratio optimization calculation operation to generate an updated compression ratio sequence; S64. Reconstruct the theoretical pressure curve based on the updated compression ratio sequence, and calculate the corresponding updated curve deviation sequence; S65. Determine whether the updated curve deviation sequence meets the preset deviation convergence condition: If the conditions are not met, then iteratively execute parameter update, compression ratio optimization calculation, and curve deviation calculation operations; If the conditions are met, the corresponding updated compression ratio sequence will be used as the intelligent calculation result of the compression ratio.

[0017] The beneficial effects of this invention are: First, this invention performs multi-scale continuous-time state encoding on the state data of a diesel engine under thermal operation by using a dual-state flow liquid neural network. This enables the simultaneous characterization of short-period state fluctuations and long-period state evolution relationships, thereby improving the accuracy and stability of compression ratio calculation under complex thermal engine conditions.

[0018] Secondly, this invention combines the temporal SHAP algorithm, particle swarm optimization algorithm and improved Hausdorff algorithm to perform state analysis and compression ratio optimization calculation on state change characteristics, compression ratio parameters and pressure curve consistency relationship. This can more accurately reflect the actual compression ratio change under the hot engine operation state of the diesel engine and improve the dynamic matching capability of the compression ratio.

[0019] Finally, based on the deviation between the theoretical pressure curve and the in-cylinder pressure curve, this invention performs a feedback update operation on the dual-state flow liquid neural network to construct a dynamic optimization closed loop for the compression ratio. This reduces the impact of pressure fluctuations on the compression ratio calculation results under complex operating conditions and improves the continuous dynamic optimization capability and operating condition adaptability in the intelligent calculation process of the diesel engine's thermal compression ratio. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a machine learning-based intelligent calculation method for the thermal compression ratio of a diesel engine, as proposed in this invention. Figure 2 This is a flowchart of the compression ratio optimization calculation process for a machine learning-based intelligent calculation method for diesel engine heat engine compression ratio proposed in this invention. Figure 3 This is a flowchart illustrating the compression ratio calculation feedback optimization process of a machine learning-based intelligent calculation method for diesel engine thermal compression ratio proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A machine learning-based intelligent calculation method for the thermal compression ratio of a diesel engine includes the following steps: S1. Acquire and preprocess the operating data of the diesel engine under hot engine operation conditions to generate a state data sequence; S2. Using a dual-state flow liquid neural network, perform continuous-time state encoding on the state data sequence, extract state change features based on the state evolution relationship between the encoded states at different time steps, and construct a state representation sequence. S3. Based on the state representation sequence, the temporal SHAP algorithm is used to perform feature contribution calculation. Key state features are selected according to the feature contribution corresponding to each state change feature to form an associated feature sequence. S4. In the dual-state flow liquid neural network, state association propagation operation is performed on the associated feature sequence, and the particle swarm optimization algorithm is used to perform compression ratio optimization calculation based on the propagation result to obtain the predicted compression ratio sequence. S5. Based on the predicted compression ratio sequence and the state data sequence, construct the theoretical pressure curve and the in-cylinder pressure curve, and perform trend smoothing. Use the improved Hausdorff algorithm to perform compression state consistency analysis on the smoothing results and calculate the curve deviation sequence. S6. Based on the curve deviation sequence, perform feedback update operation on the parameters of the dual-state flow liquid neural network, and re-execute the compression ratio optimization calculation to obtain the actual heat engine compression ratio.

[0023] In this embodiment, the operating data includes cylinder pressure data, crankshaft angle data and speed data collected by intelligent sensors when the diesel engine is in a hot-engine operating state. The preprocessing includes time synchronization, abnormal data removal and normalization. Compared with traditional liquid neural networks, dual-state flow liquid neural networks introduce a dual-state evolution structure based on short-period state flow and long-period state flow. Through the dual-state evolution structure, multi-scale continuous-time state encoding and state evolution calculation operations are performed on the state data sequence. The state evolution relationship represents the continuous-time state association relationship formed between the encoded states corresponding to different time steps, based on the state fluctuation relationship and the state evolution relationship.

[0024] In this embodiment, S2 specifically includes: S21. Input the state data sequence into the dual-state flow liquid neural network, perform continuous time state mapping operation on the state data sequence, and perform state association encoding processing according to the crankshaft angle change order based on the cylinder pressure data, crankshaft angle data and speed data corresponding to each time step in the state data sequence to generate the initial encoded state corresponding to different time steps. S22. In the dual-state evolution structure, high-frequency instantaneous response calculation is performed on the initial encoded state through a short-cycle state flow, and short-cycle state features are extracted based on the state fluctuation relationship between adjacent time steps to generate a short-cycle state sequence. S23. In the dual-state evolution structure, long-cycle state evolution calculation operations are performed on the initial encoded state through a long-cycle state flow. Long-cycle state features are extracted based on the state evolution relationship between different time steps to generate a long-cycle state sequence. S24. Perform a multi-scale state fusion operation on the short-period state sequence and the long-period state sequence to obtain the state change characteristics at each time step and form a state representation sequence.

[0025] In this embodiment, the high-frequency instantaneous response calculation and long-period state evolution calculation specifically include: In the short-cycle state flow, continuous state difference calculation is performed on the initial encoded state of adjacent time steps. Based on the state difference between the in-cylinder pressure encoded value and the speed encoded value corresponding to adjacent time steps, the state change amplitude and state change direction between different time steps are calculated. The crankshaft angle interval between adjacent time steps is set to 1°CA, forming a state fluctuation sequence. Based on the state fluctuation sequence, the difference in the amplitude of state change between adjacent time steps is calculated, and a direction consistency judgment operation is performed on the direction of state change. The direction of change of cylinder pressure is consistent with the direction of change of speed, which is defined as 1. The direction of change of cylinder pressure is opposite to the direction of change of speed, which is defined as 0. The direction consistency result is obtained. A weighted correlation calculation operation is performed based on the difference in amplitude of state change and the result of direction consistency. The difference in amplitude of state change is assigned an amplitude correlation weight of 0.7, and the result of direction consistency is assigned a direction correlation weight of 0.3, so as to obtain the short-period state characteristics corresponding to each time step. In a long-period state flow, a cross-time-step state association calculation operation is performed on the initial encoded states at different time steps. The Euclidean distance between the corresponding encoded states at different time steps is calculated as the state difference value. A time decay calculation operation is performed on the state difference value according to the time span between different time steps. The time span is calculated according to 20 consecutive time steps, and the time decay coefficient is set to 0.85 to obtain the corresponding state association value. A state evolution sequence is generated based on the state association value. Calculate long-period state characteristics based on the state correlation values ​​corresponding to different time steps in the state evolution sequence; The short-period state features and long-period state features are arranged and combined respectively to generate the corresponding short-period state sequences and long-period state sequences.

[0026] In this embodiment, S3 specifically includes: S31. Using the temporal SHAP algorithm, perform feature contribution calculation operations on the feature changes of each state in the state representation sequence, specifically including: According to the preset contribution calculation rules, the contribution value calculation operation is performed on each state change feature in the state representation sequence to obtain the single-step contribution value of different state change features at each time step. The time steps corresponding to the single-step contribution value are divided according to the 1°CA crankshaft rotation angle interval. Based on the single-step contribution values ​​of the same state change characteristics at different time steps, perform time correlation statistics operation, and perform cumulative summation on the single-step contribution values ​​within a range of 20 consecutive time steps to obtain the corresponding cumulative contribution value; Based on the cumulative contribution value, generate a sequence of contribution values ​​corresponding to the characteristics of each state change; S32. Based on each contribution value sequence, calculate the contribution change amplitude and contribution change direction of different state change characteristics at each time step. Define the increase of the cumulative contribution value between adjacent time steps as a positive contribution change and the decrease of the cumulative contribution value between adjacent time steps as a negative contribution change. S33. Perform contribution fluctuation statistics operation on the contribution change amplitude at different time steps, count the number of fluctuations and fluctuation range of contribution change amplitude within a continuous time window, and calculate the state contribution value corresponding to each state change characteristic in combination with the contribution change direction. The time window length is set to 15 consecutive time steps, the contribution change amplitude is assigned a fluctuation weight of 0.6, the contribution change direction is assigned a direction weight of 0.4, and a state contribution sequence is generated. S34. Sort the state contribution sequence, extract the state change features with a state contribution value greater than 0.75 as key state features, and generate the associated feature sequence. The key state features include cylinder pressure change features, compression state change features and combustion state change features.

[0027] In this embodiment, S33 specifically includes: S331. According to the preset time window size, the contribution change amplitude and contribution change direction of each time step are divided into sliding windows, wherein the time window size is set to 15 consecutive time steps, the window sliding step size is set to 5 time steps, and multiple contribution fluctuation intervals are generated. S332. In each contribution fluctuation interval, calculate the difference between the contribution change amplitudes of adjacent time steps to obtain the corresponding contribution fluctuation value. S333. Based on the direction of contribution change, count the number of times the direction is consistent and the number of times the direction switches between adjacent time steps in each contribution fluctuation interval to obtain the corresponding direction change results. S334. Adaptive Kalman filtering algorithm is used to perform state prediction calculation operation on each contribution fluctuation value. The predicted contribution state value of the current time step is predicted based on the contribution fluctuation value corresponding to the previous time step, and the state observation value is constructed based on the direction change result. S335. Calculate the state residual for the corresponding time step based on the difference between the predicted contribution state value and the observed state value, and perform a filter gain correction operation on the predicted contribution state value based on the state residual. Increase the observation weight corresponding to the observed state value when the state residual increases, and increase the prediction weight corresponding to the predicted contribution state value when the state residual decreases, to obtain the state contribution value corresponding to each state change feature. S336. Based on the contribution values ​​of each state, form the corresponding state contribution sequence.

[0028] In this embodiment, S4 specifically includes: S41. In a dual-state flow liquid neural network, a state association propagation operation is performed on the associated feature sequence. Based on the state association relationship between the in-cylinder pressure change features, compression state change features and combustion state change features corresponding to different time steps, the state propagation value corresponding to the key state features between different time steps is calculated. The state propagation value corresponds to the state influence value between different time steps. S42. Calculate the propagation change value and propagation direction between the state propagation values ​​of each adjacent time step, and perform state propagation stability analysis. Define the increase of the state propagation value between adjacent time steps as the forward propagation state, and define the decrease of the state propagation value between adjacent time steps as the reverse propagation state. Generate a state stable sequence based on the consistency between the number of fluctuations of the propagation change value and the direction of propagation change between consecutive time steps. S43. Based on the stable state sequence, perform a propagation weight allocation operation on each state propagation value. The time step with a higher degree of state stability corresponds to a larger propagation weight, and the time step with a lower degree of state stability corresponds to a smaller propagation weight. Then, perform a compression ratio parameter mapping operation on the associated feature sequence according to the propagation weight to generate a compression ratio parameter sequence. S44. Using the particle swarm optimization algorithm, 50 compression ratio search particles are generated. Based on the compression ratio parameter sequence, the compression ratio parameter initialization operation is performed on each compression ratio search particle, mapping each compression ratio parameter to the initial position parameter and initial velocity parameter of the corresponding compression ratio search particle, and generating a particle parameter sequence. S45. Based on the particle parameter sequence, perform fitness calculation operations on each search particle with a compression ratio, and filter target particles according to fitness to generate a target particle sequence. S46. Based on the target particle sequence, update the compression ratio parameters of each compression ratio search particle, and iteratively perform fitness calculation and compression ratio parameter update operations until the fitness reaches 0.95. S47. Use the compression ratio parameter at the end of the iteration as the predicted compression ratio to form a predicted compression ratio sequence.

[0029] In this embodiment, S45 specifically includes: S451. Based on the particle parameter sequence, perform compression state deduction operation on the compression ratio parameter corresponding to each compression ratio search particle. Calculate the theoretical pressure value corresponding to different time steps based on the compression ratio parameter, cylinder pressure data and crankshaft angle data corresponding to each time step. The theoretical pressure value is continuously calculated according to the crankshaft angle range from 60°CA before the compression top dead center to the injection time, generating a theoretical pressure sequence corresponding to different time steps. S452. Based on the theoretical pressure sequence and the in-cylinder pressure data, calculate the pressure difference, pressure change direction difference, and pressure evolution trend difference for each time step to generate a pressure deviation sequence. The pressure difference represents the difference between the theoretical pressure value and the actual in-cylinder pressure value. The pressure change direction difference represents the difference between the theoretical pressure change direction and the actual pressure change direction between adjacent time steps. The pressure evolution trend difference represents the degree of deviation between the change trend of the theoretical pressure curve and the change trend of the in-cylinder pressure curve. S453. Based on the pressure deviation sequence, perform a time-series deviation propagation calculation operation on the pressure difference corresponding to different time steps. Calculate the deviation propagation value based on the change result of the pressure difference between adjacent time steps. A continuous increase in the deviation propagation value indicates the pressure deviation diffusion state, and a continuous decrease in the deviation propagation value indicates the pressure deviation convergence state. Generate the deviation propagation sequence. S454. Using the HDBSCAN algorithm, perform pressure instability density clustering on the deviation propagation sequence, specifically including: A state feature vector is constructed based on the deviation propagation value, pressure difference, and pressure evolution trend difference at different time steps; The Euclidean distance between each state feature vector is calculated as the state distance value. The core neighborhood range of each time step is determined based on the state distance value. The local density value of the corresponding time step is calculated based on the state distance value in each core neighborhood range. Hierarchical clustering and connection operations are performed on time steps where local density values ​​continuously increase, generating multiple pressure instability clusters; Calculate the changing trend of deviation propagation value at different time steps in each pressure instability cluster, and take the cluster where the deviation propagation value continues to increase for more than 5 consecutive time steps as the pressure instability interval to generate a pressure instability sequence. S455. Based on the pressure instability sequence, perform state coupling fluctuation analysis on the difference between the pressure change direction and the pressure evolution trend at different time steps. Calculate the state instability value of each pressure instability interval based on the degree of synchronous fluctuation between the pressure change direction difference and the pressure evolution trend difference. The higher the degree of synchronous fluctuation, the larger the corresponding state instability value, and generate a state instability sequence. S456. Based on the state instability sequence and the state stability sequence, calculate the pressure stability value, state stability value and deviation propagation suppression value corresponding to each compression ratio search particle, and generate a multi-dimensional state evaluation sequence. Among them, the pressure stability value represents the overall stability between the theoretical pressure sequence and the cylinder pressure data, the state stability value represents the stability of state propagation changes, and the deviation propagation suppression value represents the degree of suppression of deviation propagation values. S457. Based on the multidimensional state evaluation sequence, perform multi-objective score calculation operation on each compression ratio search particle, assign a stability weight of 0.5 to the pressure stability value, assign a state weight of 0.3 to the state stability value, and assign a suppression weight of 0.2 to the deviation propagation suppression value to obtain the score value of each compression ratio search particle. S458. Extract the top 10% of the compression ratio search particles with the highest scores as target particles to form a target particle sequence.

[0030] In this embodiment, S5 specifically includes: S51. Based on the predicted compression ratio sequence, perform theoretical pressure mapping calculation for the predicted compression ratio corresponding to different time steps. Perform continuous mapping calculation according to the crankshaft angle range from 60°CA before the compression top dead center to the injection time to generate the theoretical pressure value for each time step and form a theoretical pressure curve. S52. Based on the cylinder pressure data in the state data sequence, perform pressure curve reconstruction operation in time order, and continuously arrange the cylinder pressure values ​​corresponding to different time steps in the order of crankshaft angle change to generate cylinder pressure curve, wherein the crankshaft angle sampling interval is set to 1°CA. S53. Set the Savitzky-Golay smoothing window length to 11 and the fitting polynomial order to 3. Perform Savitzky-Golay trend smoothing on the local pressure fluctuations in the theoretical pressure curve and the cylinder pressure curve respectively to generate the corresponding smoothed pressure sequence. S54. Based on each smooth pressure sequence, perform local pressure change interval division operation on the theoretical pressure curve and the cylinder pressure curve. Based on the changes in the pressure change amplitude and pressure change direction between adjacent time steps, divide the time steps with consistent pressure change trends into the same pressure change interval, and generate multiple pressure change intervals. S55. Using the improved Hausdorff algorithm, a compression state consistency analysis is performed on the theoretical pressure curve and the in-cylinder pressure curve in each pressure variation range, specifically including: Based on the time window corresponding to each pressure change range, the theoretical pressure value of each time step in the theoretical pressure curve is used as the reference point. Multiple neighboring pressure points are searched within the time window corresponding to the cylinder pressure curve. The search range of neighboring pressure points is set to 5 consecutive time steps before and after the current time step, and a corresponding set of candidate matching points is generated. Based on the pressure distance values ​​between each neighboring pressure point and the corresponding benchmark point in the candidate matching point set, the corresponding local curve distance is calculated, where the pressure distance value represents the absolute difference between the theoretical pressure value and the neighboring pressure value. The pressure change direction of each adjacent pressure point and the corresponding reference point is statistically analyzed, and a direction consistency judgment operation is performed. Adjacent pressure points with the same pressure change direction are marked as valid matching points. Based on the local curve distance, perform distance weight allocation operation on each valid matching point according to the preset weight allocation rule. Valid matching points with smaller distances correspond to larger distance weights, and valid matching points with larger distances correspond to smaller distance weights. Calculate the corresponding weighted curve distance value. Based on the distance values ​​of each weighted curve, calculate the interval Hausdorff distance value corresponding to each pressure change interval. The interval Hausdorff distance value represents the overall curve deviation between the theoretical pressure curve and the in-cylinder pressure curve. S56. Based on the Hausdorff distance values ​​of each interval, calculate the distance change value and direction of distance change between adjacent time steps, and perform curve deviation correlation analysis operation in combination with the pressure change values ​​between corresponding time steps to generate a curve deviation sequence.

[0031] In this embodiment, S6 specifically includes: S61. Based on the curve deviation sequence, calculate the curve deviation change value and the curve deviation change direction corresponding to different time steps, and generate the deviation change sequence. S62. Perform feedback update operation on the parameters in the dual-state flow liquid neural network according to the deviation change sequence; S63. Based on the updated dual-state flow liquid neural network, re-execute the compression ratio optimization calculation operation to generate an updated compression ratio sequence; S64. Reconstruct the theoretical pressure curve based on the updated compression ratio sequence, and calculate the corresponding updated curve deviation sequence; S65. Determine whether the update curve deviation sequence satisfies the condition that the Hausdorff distance value of the update interval corresponding to 10 consecutive time steps is less than 0.05: If the conditions are not met, then iteratively execute parameter update, compression ratio optimization calculation, and curve deviation calculation operations; If the conditions are met, the corresponding updated compression ratio sequence will be used as the intelligent calculation result of the compression ratio.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to a scenario of intelligent compression ratio calculation for a high-performance diesel engine under hot-engine operation. During continuous hot-engine operation, the engine speed fluctuated between 1800 r / min and 3200 r / min, and the peak in-cylinder pressure varied between 165 bar and 248 bar. Because the diesel engine operates under high temperature and high pressure for an extended period, combustion chamber deformation, in-cylinder leakage, and pressure propagation fluctuations are significantly amplified. This results in a significant deviation between the actual hot-engine compression ratio and the cold-state geometric compression ratio. Traditional compression ratio calculation methods based on fixed thermodynamic formulas are prone to problems such as large pressure curve matching errors, low dynamic compression ratio tracking capability, and insufficient adaptability to complex operating conditions.

[0033] In this embodiment, intelligent sensors installed in the diesel engine cylinder block continuously collect in-cylinder pressure data, crankshaft angle data, and engine speed data. The collected operating data undergoes time synchronization, abnormal data removal, and normalization processing to form a state data sequence. Subsequently, the state data sequence is input into a dual-state-flow liquid neural network. Short-cycle state flow extracts instantaneous pressure and combustion fluctuation characteristics between adjacent time steps, while long-cycle state flow extracts thermodynamic state evolution and pressure propagation change characteristics between different time steps. A multi-scale state fusion operation is then performed on the short-cycle and long-cycle state sequences to form a state representation sequence.

[0034] After the state representation sequence is constructed, the temporal SHAP algorithm is used to perform state contribution analysis on different state change features, and key state features with high state contribution values ​​are selected. Subsequently, state association propagation operation is performed on the key state features in the dual-state flow liquid neural network, and particle swarm optimization algorithm is used to perform dynamic optimization calculation on the compression ratio parameter. In the compression ratio optimization process, a total of 120 sets of compression ratio search particles are generated, the number of particle iterations is set to 80, and the fitness convergence threshold is set to 0.003. The theoretical pressure curve is constructed based on the predicted compression ratio sequence, and the actual in-cylinder pressure curve is constructed by combining it with the in-cylinder pressure data. Then, Savitzky-Golay trend smoothing is used to eliminate the influence of local pressure fluctuations. Then, an improved Hausdorff algorithm is used to analyze the consistency relationship of compression state between the theoretical pressure curve and the actual in-cylinder pressure curve, generating the corresponding curve deviation sequence. The dual-state flow liquid neural network is subjected to feedback update operation based on the curve deviation sequence, and the compression ratio optimization calculation is re-executed until the curve deviation meets the preset convergence condition, and finally the actual hot engine compression ratio result of the diesel engine under hot engine operation is obtained.

[0035] Table 1. Performance analysis results of different compression ratio calculation methods under complex thermal engine conditions.

[0036] As can be seen from Table 1, after adopting the method of the present invention, the average compression ratio error is reduced to 1.84%, and the maximum compression ratio error is reduced to 2.63%, which are 6.89% and 10.01% lower than the traditional thermodynamic calculation method, respectively. This shows that the present invention can more accurately reflect the actual compression ratio change under the hot engine operation state of the diesel engine.

[0037] Meanwhile, the pressure curve matching deviation between the theoretical pressure curve and the actual cylinder pressure curve was reduced to 2.16 bar, and the pressure peak deviation was reduced to 1.95 bar, indicating that the present invention has a higher pressure curve consistency analysis capability under complex thermal engine conditions.

[0038] The state fluctuation error was reduced to 1.95%, and the state propagation stability value was increased to 0.94, indicating that the dual-state flow liquid neural network can more effectively characterize the continuous state evolution relationship between different time steps. The number of compression ratio convergences was reduced to 5, and the average convergence deviation was reduced to 0.48%, indicating that the present invention improves the dynamic optimization efficiency of the compression ratio through the feedback update mechanism.

[0039] In addition, the curve consistency accuracy reached 96.85%, the dynamic operating condition adaptability reached 95.73%, the pressure instability identification accuracy reached 94.62%, and the compression ratio dynamic tracking accuracy reached 96.18%, further demonstrating that the present invention can effectively reduce the impact of pressure propagation fluctuations on the compression ratio calculation results under complex thermal engine conditions, and improve the stability and continuous dynamic optimization capability in the intelligent calculation process of diesel engine thermal engine compression ratio.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A machine learning-based intelligent calculation method for the thermal compression ratio of a diesel engine, characterized in that, Includes the following steps: S1. Acquire and preprocess the operating data of the diesel engine under hot engine operation conditions to generate a state data sequence; S2. Using a dual-state flow liquid neural network, perform continuous-time state encoding on the state data sequence, extract state change features based on the state evolution relationship between the encoded states at different time steps, and construct a state representation sequence. S3. Based on the state representation sequence, the temporal SHAP algorithm is used to perform feature contribution calculation. Key state features are selected according to the feature contribution corresponding to each state change feature to form an associated feature sequence. S4. In the dual-state flow liquid neural network, state association propagation operation is performed on the associated feature sequence, and the particle swarm optimization algorithm is used to perform compression ratio optimization calculation based on the propagation result to obtain the predicted compression ratio sequence. S5. Based on the predicted compression ratio sequence and the state data sequence, construct the theoretical pressure curve and the in-cylinder pressure curve, and perform trend smoothing. Use the improved Hausdorff algorithm to perform compression state consistency analysis on the smoothing results and calculate the curve deviation sequence. S6. Based on the curve deviation sequence, perform feedback update operation on the parameters of the dual-state flow liquid neural network, and re-execute the compression ratio optimization calculation to obtain the actual heat engine compression ratio.

2. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, The operating data includes cylinder pressure data, crankshaft angle data, and speed data collected by intelligent sensors when the diesel engine is in hot operation. The preprocessing includes time synchronization, abnormal data removal, and normalization. Compared with traditional liquid neural networks, the dual-state flow liquid neural network introduces a dual-state evolution structure based on short-period state flow and long-period state flow. Through the dual-state evolution structure, multi-scale continuous-time state encoding and state evolution calculation operations are performed on the state data sequence. The state evolution relationship represents the continuous-time state association relationship formed between the encoded states corresponding to different time steps, based on the state fluctuation relationship and the state evolution relationship.

3. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, S2 specifically includes: S21. Input the state data sequence into the dual-state flow liquid neural network, perform continuous-time state mapping operation on the state data sequence, and generate the initial encoded state corresponding to different time steps. S22. In the dual-state evolution structure, high-frequency instantaneous response calculation is performed on the initial encoded state through a short-cycle state flow, and short-cycle state features are extracted based on the state fluctuation relationship between adjacent time steps to generate a short-cycle state sequence. S23. In the dual-state evolution structure, long-cycle state evolution calculation operations are performed on the initial encoded state through a long-cycle state flow. Long-cycle state features are extracted based on the state evolution relationship between different time steps to generate a long-cycle state sequence. S24. Perform a multi-scale state fusion operation on the short-period state sequence and the long-period state sequence to form a state representation sequence.

4. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 3, characterized in that, The high-frequency instantaneous response calculation and long-period state evolution calculation specifically include: In a short-cycle state flow, a continuous state difference calculation operation is performed on the initial encoded states of adjacent time steps to calculate the magnitude and direction of state change between different time steps. Calculate the difference in the magnitude of state changes between adjacent time steps, and perform a direction consistency determination operation on the direction of state changes. Define the same direction of change as 1 and the opposite direction of change as 0 to obtain the direction consistency result. A weighted correlation calculation is performed based on the result of consistency between the amplitude difference of state changes and the direction to obtain the short-period state characteristics corresponding to each time step; In a long-cycle state flow, a cross-time-step state association calculation operation is performed on the initial encoded state at different time steps to calculate the state difference value between the corresponding encoded states at different time steps. Then, a time decay calculation operation is performed on the state difference value in combination with the time span to obtain the corresponding state association value. Calculate long-period state characteristics based on the state correlation values ​​corresponding to different time steps; The short-period state features and long-period state features are arranged and combined respectively to generate the corresponding short-period state sequences and long-period state sequences.

5. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, S3 specifically includes: S31. Using the temporal SHAP algorithm, perform feature contribution calculation operations on the feature changes of each state in the state representation sequence, specifically including: According to the preset contribution calculation rules, the contribution value calculation operation is performed on each state change feature in the state representation sequence to obtain the single-step contribution value of different state change features at each time step. Based on the single-step contribution value of the same state change characteristics at different time steps, perform time-related statistical operations to obtain the corresponding cumulative contribution value; Based on the cumulative contribution value, generate a sequence of contribution values ​​corresponding to the characteristics of each state change; S32. Based on each contribution value sequence, calculate the contribution change amplitude and contribution change direction of different state change characteristics at each time step; S33. Perform contribution fluctuation statistics operation on the contribution change amplitude at different time steps, and calculate the state contribution value corresponding to each state change characteristic in combination with the contribution change direction. S34. Extract state change features whose state contribution value is greater than the preset contribution threshold as key state features and generate associated feature sequences.

6. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 5, characterized in that, Specifically, S33 includes: S331. According to the preset time window size, the contribution change amplitude and contribution change direction of each time step are divided into sliding windows to generate multiple contribution fluctuation intervals. S332. In each contribution fluctuation interval, calculate the difference between the contribution change amplitudes of adjacent time steps to obtain the corresponding contribution fluctuation value. S333. Based on the direction of contribution change, count the number of times the direction is consistent and the number of times the direction switches between adjacent time steps in each contribution fluctuation interval to obtain the corresponding direction change results. S334. Adaptive Kalman filtering algorithm is used to perform state prediction calculation operation on each contribution fluctuation value, generate predicted contribution state value, and construct state observation value based on the direction change result. S335. Calculate the state residual for the corresponding time step based on the difference between the predicted contribution state value and the observed state value, and perform a filter gain correction operation on the predicted contribution state value based on the state residual to obtain the state contribution value corresponding to each state change feature. S336. Based on the contribution values ​​of each state, form the corresponding state contribution sequence.

7. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, S4 specifically includes: S41. In a dual-state flow liquid neural network, perform a state association propagation operation on the associated feature sequence and calculate the state propagation value corresponding to the key state features between different time steps. S42. Calculate the propagation change value and propagation direction between the state propagation values ​​of each adjacent time step, and perform state propagation stability analysis to generate a state stable sequence. S43. Based on the stable state sequence, perform a propagation weight allocation operation on each state propagation value, and perform a compression ratio parameter mapping operation on the associated feature sequence according to the propagation weight to generate a compression ratio parameter sequence. S44. Using the particle swarm optimization algorithm, multiple compression ratio search particles are generated according to the preset number of particles, and the compression ratio parameter initialization operation is performed on each compression ratio search particle according to the compression ratio parameter sequence to generate the particle parameter sequence. S45. Based on the particle parameter sequence, perform fitness calculation operations on each search particle with a compression ratio, and filter target particles according to fitness to generate a target particle sequence. S46. Based on the target particle sequence, update the compression ratio parameters of each compression ratio search particle, and iteratively execute the fitness calculation and compression ratio parameter update operations until the fitness reaches the preset convergence threshold. S47. Use the compression ratio parameter at the end of the iteration as the predicted compression ratio to form a predicted compression ratio sequence.

8. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 7, characterized in that, Specifically, S45 includes: S451. Based on the particle parameter sequence, perform compression state deduction operation on the compression ratio parameters corresponding to each compression ratio search particle to generate theoretical pressure sequences corresponding to different time steps. S452. Based on the theoretical pressure sequence and the cylinder pressure data, calculate the pressure difference, pressure change direction difference, and pressure evolution trend difference for each time step to generate a pressure deviation sequence. S453. Based on the pressure deviation sequence, perform a time-series deviation propagation calculation operation on the pressure difference corresponding to different time steps, calculate the deviation propagation value between each time step, and generate a deviation propagation sequence. S454. Using the HDBSCAN algorithm, perform pressure instability density clustering on the deviation propagation sequence, specifically including: A state feature vector is constructed based on the deviation propagation value, pressure difference, and pressure evolution trend difference at different time steps; The Euclidean distance between each state feature vector is calculated as the state distance value. The core neighborhood range of each time step is determined based on the state distance value. The local density value of the corresponding time step is calculated based on the state distance value in each core neighborhood range. Hierarchical clustering and connection operations are performed on time steps where local density values ​​continuously increase, generating multiple pressure instability clusters; Calculate the changing trend of deviation propagation value at different time steps in each pressure instability cluster, and take the cluster with continuously increasing deviation propagation value as the pressure instability interval to generate a pressure instability sequence; S455. Based on the pressure instability sequence, perform state-coupled fluctuation analysis on the difference between the pressure change direction and the pressure evolution trend at different time steps, calculate the state instability value of each pressure instability interval, and generate the state instability sequence. S456. Based on the state instability sequence and the state stability sequence, calculate the pressure stability value, state stability value and deviation propagation suppression value corresponding to each compression ratio search particle, and generate a multi-dimensional state evaluation sequence. S457. Based on the multidimensional state evaluation sequence, perform multi-objective score calculation operation on each compression ratio search particle to obtain the score value of each compression ratio search particle. S458. Extract the compression ratio search particles whose score values ​​meet the preset screening conditions as target particles to form a target particle sequence.

9. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, S5 specifically includes: S51. Based on the predicted compression ratio sequence, perform theoretical pressure mapping calculation operation on the predicted compression ratio corresponding to different time steps to generate theoretical pressure values ​​for each time step and form a theoretical pressure curve. S52. Based on the cylinder pressure data in the state data sequence, perform pressure curve reconstruction operation in chronological order to generate the cylinder pressure curve. S53. Perform Savitzky-Golay trend smoothing on the theoretical pressure curve and the in-cylinder pressure curve respectively to generate the corresponding smoothed pressure sequence. S54. Based on each smoothed pressure sequence, perform a local pressure change interval division operation on the theoretical pressure curve and the cylinder pressure curve to generate multiple pressure change intervals. S55. Using the improved Hausdorff algorithm, a compression state consistency analysis is performed on the theoretical pressure curve and the in-cylinder pressure curve in each pressure variation range, specifically including: Based on the time window corresponding to each pressure change range, the theoretical pressure value of each time step in the theoretical pressure curve is used as the reference point. Multiple neighboring pressure points are searched within the time window corresponding to the cylinder pressure curve to generate the corresponding candidate matching point set. Calculate the corresponding local curve distance based on the pressure distance between each neighboring pressure point and the corresponding reference point in the candidate matching point set; The pressure change direction of each adjacent pressure point and the corresponding reference point is statistically analyzed, and a direction consistency judgment operation is performed. Adjacent pressure points with the same pressure change direction are marked as valid matching points. Based on the local curve distance, perform distance weight allocation operation on each valid matching point according to the preset weight allocation rule, and calculate the corresponding weighted curve distance value; Based on the distance values ​​of each weighted curve, calculate the Hausdorff distance value corresponding to each pressure change interval; S56. Based on the Hausdorff distance values ​​of each interval, calculate the distance change value and direction of distance change between adjacent time steps, and perform curve deviation correlation analysis operation in combination with the pressure change values ​​between corresponding time steps to generate a curve deviation sequence.

10. The intelligent calculation method for the thermal compression ratio of a diesel engine based on machine learning according to claim 1, characterized in that, S6 specifically includes: S61. Based on the curve deviation sequence, calculate the curve deviation change value and the curve deviation change direction corresponding to different time steps, and generate the deviation change sequence. S62. Perform feedback update operation on the parameters in the dual-state flow liquid neural network according to the deviation change sequence; S63. Based on the updated dual-state flow liquid neural network, re-execute the compression ratio optimization calculation operation to generate an updated compression ratio sequence; S64. Reconstruct the theoretical pressure curve based on the updated compression ratio sequence, and calculate the corresponding updated curve deviation sequence; S65. Determine whether the updated curve deviation sequence meets the preset deviation convergence condition: If the conditions are not met, then iteratively execute parameter update, compression ratio optimization calculation, and curve deviation calculation operations; If the conditions are met, the corresponding updated compression ratio sequence will be used as the intelligent calculation result of the compression ratio.

Images