Method for predicting performance map of turbocharger compressor for marine
By using similarity indices to generate adaptive weights and apply joint constraints in the turbocharger compressor performance map prediction, the problem of low prediction accuracy across speeds is solved, achieving higher prediction accuracy and reliability, and ensuring the stability of the pressure ratio curve and efficiency value.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN ENG UNIV
- Filing Date
- 2026-04-20
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies suffer from low prediction accuracy and morphological mismatch issues in the cross-speed prediction of turbocharger compressor performance maps. Especially under sparse data or extrapolation conditions, they cannot effectively guarantee the stability of the pressure ratio curve and the rationality of the efficiency value.
By acquiring sparse performance feature point data of the compressor at multiple speeds, grouping by speed and constructing a subset of speed line data, generating adaptive weights using similarity indexes, combining multi-dimensional morphological similarity for weighted prediction, and applying joint constraints, an objective function is constructed to generate a compressor performance map.
It significantly improves the accuracy and reliability of cross-speed prediction, avoids non-physical mismatch problems such as peak rebound of pressure ratio curve, efficiency value overshoot and boundary jump, and enhances the physical rationality and engineering credibility of prediction results.
Smart Images
Figure CN122241922A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for predicting the performance of marine turbocharger compressors using a performance map, belonging to the field of marine power. Background Technology
[0002] Performance maps are used to describe the correspondence between performance parameters such as (corrected) mass flow rate, pressure ratio, and isentropic efficiency of compressors in power equipment such as turbochargers, gas turbines, and aero engines at different speeds. They are a key basis for compressor component selection, performance evaluation, control strategy design, and overall machine performance simulation.
[0003] In existing technologies, compressor performance maps are typically presented as isodynamic lines, surge lines, choking lines, and isoefficiency lines. Industry and various engineering manuals generally use the surge line as the left boundary of the performance map, i.e., the boundary of the unstable flow region, and the choking line as the right boundary. The operating point performance between each speed line is usually estimated using interpolation.
[0004] At the software implementation level, maps are often organized into an M-by-N matrix, where M represents the number of constant speed lines and N represents the number of operating points on each speed line. This is accompanied by a constant speed vector of length M and an auxiliary coordinate β of length N. This type of data structure, composed of matrices and index vectors, can be directly used for table lookups and interpolation, and is a common parameterized input form in control and simulation models.
[0005] Relatedly, the introduction of corrected mass flow rate and corrected rotational speed is to convert performance data obtained under different inlet pressure and temperature conditions to a unified reference state, so that a single map can characterize compressor performance under multiple inlet conditions.
[0006] The typical level of existing technology and its main shortcomings are as follows.
[0007] First, compressor maps obtained through experiments or calculations typically only cover the speed range from idle to full power, and the number of operating points obtained is relatively limited. However, under extreme operating conditions such as engine start-up, data from lower speed ranges, or even conditions with a pressure ratio less than 1.0, are often required; therefore, the original map must be refined, reconstructed, or expanded.
[0008] Secondly, to meet the engineering requirements of matrix table input, each speed curve typically needs to be resampled to the same number of points. Traditional methods mostly employ single interpolation to fill in the discrete points, such as linear interpolation, spline interpolation, or ordinary cubic splines. Some methods use simple linear interpolation or two-dimensional scattered point interpolation in the speed direction. Engineering practice has shown that to construct a map table that can be directly used in compressor models, the data must be organized into a regular matrix, and when extracting variables such as efficiency, interpolation is needed to ensure consistency with the points on each speed curve. This processing inevitably introduces errors, which can usually only be mitigated by adjusting the point selection strategy.
[0009] Furthermore, when operating points are sparse or located in boundary regions, relying solely on mathematical interpolation or extrapolation can easily lead to non-physical risks and engineering usability risks. Typical problems include at least the following categories.
[0010] First, the pressure ratio curve rebounds or overshoots after the peak, which makes the interpolation results unstable or unreliable.
[0011] Secondly, the efficiency value exceeds the limit, for example, exceeding 100% or exceeding the reasonable engineering limit.
[0012] Third, the surge boundary and the blockage boundary are discontinuous in the cross-speed direction, and may even jump.
[0013] Fourth, the cross-rotation direction is not smooth enough, which causes numerical jitter in the lookup table results during dynamic simulation.
[0014] Fifth, the lack of clear risk indicators in the extrapolation area makes it impossible for subsequent callers to judge the reliability of the results.
[0015] It should be noted that even with conformal interpolation, the coupling problem between the aforementioned engineering constraints cannot be automatically resolved. Taking PCHIP as an example, the advantage of this method is that it better preserves the shape of the original data, avoids obvious overshoot, and tries to follow monotonicity, thus exhibiting less oscillation when processing non-smooth data. However, it is essentially still an interpolation tool on a single curve, and cannot naturally guarantee the boundary continuity across speed curves, nor can it simultaneously guarantee controllable extrapolation risks and the coordinated satisfaction of multiple engineering constraints. Summary of the Invention
[0016] To address the problems of low prediction accuracy and morphological mismatch in traditional cross-speed prediction methods under sparse data or extrapolation conditions, this invention provides a marine turbocharger compressor performance map prediction method.
[0017] The present invention provides a method for predicting the performance of a marine turbocharger compressor by mapping, comprising:
[0018] S1. Obtain a sparse performance feature point dataset of the compressor at multiple speeds, where each feature point data includes at least the speed, mass flow rate, pressure ratio and isentropic efficiency.
[0019] S2. Group the data point set according to rotational speed to obtain multiple subsets of rotational speed line data;
[0020] S3. Determine the target rotational speed to be predicted. Select two rotational speed data subsets that are close to the target rotational speed from multiple rotational speed data subsets. Construct feature point data of the target rotational speed through linear interpolation. Based on the feature point data of the target rotational speed, calculate the similarity index of each known rotational speed data subset relative to the target rotational speed. The similarity index includes at least the rotational speed distance difference, the flow rate difference corresponding to the pressure ratio peak position, the pressure ratio peak amplitude difference, the flow rate difference corresponding to the surge endpoint, the flow rate difference corresponding to the blockage endpoint, the local slope difference of the rotational speed line, and the local curvature difference of the rotational speed line.
[0021] S4. Based on the similarity index, generate adaptive weights, perform weighted calculations on all known speed line candidate curves for pressure ratio and isentropic efficiency, and predict the target speed for pressure ratio and isentropic efficiency.
[0022] S5. Perform joint constraint solution on the pressure ratio prediction curve and isentropic efficiency prediction curve of the target speed to construct the objective function, and apply constraint conditions at the same time to form the target speed line prediction result that satisfies all constraint conditions.
[0023] S6. Based on the obtained target speed curve prediction results, generate a compressor performance map.
[0024] Preferably, S3 includes:
[0025] S31. By using linear interpolation of adjacent speed lines, estimate the peak pressure ratio position, peak pressure ratio amplitude, surge endpoint, and blockage endpoint at the target speed, respectively.
[0026] ;
[0027] The characteristic point data for the target rotational speed includes the peak pressure ratio location, peak pressure ratio amplitude, surge endpoint, or blockage endpoint.
[0028] , This refers to the feature point data adjacent to the rotational speed line;
[0029] For the target speed, , The speed is the speed of the adjacent speed line;
[0030] S32. Each subset of speed curve data is compared item by item with the feature points of the target speed to construct a subset of speed curve data. Differences related to target speed , , respectively representing the speed distance difference, the flow rate difference corresponding to the pressure ratio peak position, the pressure ratio peak amplitude difference, the flow rate difference corresponding to the surge end, the flow rate difference corresponding to the blockage end, the local slope difference of the speed line, and the local curvature difference of the speed line;
[0031] S33, Based on the subset of speed line data Differences related to target speed Obtain the similarity index of each known speed curve data subset relative to the target speed. :
[0032] ;
[0033] These are the weighting coefficients for the corresponding difference terms.
[0034] Preferably, in step S4, adaptive weights are generated based on the similarity index. :
[0035] ;
[0036] Represents the sensitivity coefficient. Indicates the number of subsets of speed curve data;
[0037] Candidate pressure ratio curves for all known speed lines With isentropic efficiency candidate curve By performing weighted calculations, the pressure ratio prediction curve for the target rotational speed is obtained. With isentropic efficiency prediction curve :
[0038] ;
[0039] .
[0040] Preferably, the method for obtaining the pressure ratio curve and isentropic efficiency curve of the adjacent speed line is as follows:
[0041] Within each subset of speed lines, the data points are sorted by mass flow rate from smallest to largest. Interpolation is performed on the pressure ratio sequence and isentropic efficiency sequence of each subset of speed lines to obtain the corresponding pressure ratio curve and isentropic efficiency curve.
[0042] The pressure ratio peak value is determined based on each pressure ratio curve, and a post-peak monotonically non-increasing constraint is applied to the high flow rate region after the pressure ratio peak value to obtain a candidate pressure ratio curve that is monotonically non-increasing after the peak value.
[0043] An efficiency limit constraint is applied to each of the isentropic efficiency curves to obtain isentropic efficiency candidate curves that satisfy the efficiency limit.
[0044] As a preferred option, piecewise cubic Hermite conformal interpolation is performed on the pressure ratio sequence and isentropic efficiency sequence of each subset of rotational speed lines.
[0045] As a preferred option, the constraints include post-peak monotonicity constraint, efficiency limit constraint, surge boundary continuity constraint, blockage boundary continuity constraint, speed-line smoothness constraint, and speed-crossing smoothness constraint.
[0046] As a preferred option, the objective function is:
[0047]
[0048] in, For the total loss, To fit the weighting coefficients for efficiency, The robust interpolation pressure ratio curve is obtained from robust conventional interpolation (such as PCHIP). The robust interpolation isentropic efficiency curve is obtained by robust conventional interpolation (such as PCHIP);
[0049] This is the loss due to the smoothing constraint within the speed range line;
[0050] Loss due to smoothness constraints across rotational speeds;
[0051] The loss is due to the continuity constraint of the surge boundary;
[0052] The loss is due to the blockage of boundary continuity constraints;
[0053] , , , Represents the pre-trained constraint weight coefficients;
[0054] Post-peak monotonically non-increasing constraint is The efficiency limit constraint is ;
[0055] The position index of the pressure ratio amplitude in the subset of speed curve data sorted by mass flow rate. The pressure ratio amplitude, It is the position index where the pressure ratio amplitude reaches its maximum value; This represents the isentropic efficiency within a subset of the rotational speed data. This represents the maximum value of the isentropic efficiency.
[0056] Preferably, the performance map includes at least a pressure ratio table, an efficiency table, and confidence indices corresponding to the pressure ratio table and efficiency table. The confidence indices include: prediction uncertainty based on the difference between adjacent speed lines, back-substitution error of the original measurement points, and consistency of indices for multiple interpolation or fitting strategies.
[0057] When the confidence index falls below a set threshold, the system automatically switches to a conservative reconstruction strategy, which includes strengthening the constraint strength, limiting the extrapolation interval, or changing the weighted prediction model.
[0058] Preferably, the compressor performance map is a matrix table composed of a speed vector and an auxiliary coordinate parameter vector, wherein the auxiliary coordinate parameter has a value range of [0,1] and is used to represent the position index of each speed line.
[0059] The beneficial effects of this invention are that, by constructing an adaptive weighted prediction mechanism based on multi-dimensional morphological similarity, it effectively solves the problems of low prediction accuracy and easy morphological mismatch in traditional cross-speed prediction methods under sparse data or extrapolation conditions. Specifically, this method first uses the characteristic parameters of adjacent speed lines (such as pressure ratio peak position, peak amplitude, surge endpoint, and blockage endpoint flow rate) to construct virtual feature points of the target speed through linear interpolation. Then, it integrates multiple morphological indicators such as speed distance, peak difference, boundary difference, and local slope and curvature difference of each known speed line into a similarity difference with the target speed, and generates adaptive weights through exponential mapping. This weight can automatically identify and enhance the contribution of known speed lines with high morphological consistency with the target speed in the weighted prediction, while suppressing the influence of speed lines with large morphological differences. As a result, the initial curve obtained by weighted prediction is closer to the real physical change law in terms of peak position, boundary direction, and overall shape. Compared with traditional linear interpolation methods that rely solely on speed distance, this scheme significantly reduces prediction errors caused by large differences in speed curve morphology. Especially in high-risk conditions such as low-speed extrapolation and sparse data interpolation, it can effectively avoid non-physical morphological mismatch problems such as pressure ratio curve peak rebound, efficiency value exceeding limits, and boundary jumps, thus greatly improving the accuracy and reliability of cross-speed prediction. Attached Figure Description
[0060] Figure 1 Compressor performance map resampled from the original sparse data;
[0061] Figure 2 Candidate curves are generated using conformal interpolation for single rotational speeds;
[0062] Figure 3 For peak pressure ratio identification and post-peak monotonic non-increasing constraint (compare before and after constraint);
[0063] Figure 4Extraction of surge endpoints and blockage endpoints, and processing of boundary curve continuity;
[0064] Figure 5 Adaptive weighted prediction based on similarity criteria for target rotational speed (including similarity index and weight generation);
[0065] Figure 6 This is a framework for solving joint constraints (objective function + constraint set);
[0066] Figure 7 The matrix table organization method for the output map (corrected rotational speed vector + auxiliary coordinate β);
[0067] Figure 8 For confidence index output and strategy switching;
[0068] Figure 9 This refers to the call chain of the map in the simulation or control system. Detailed Implementation
[0069] The technical solutions of 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.
[0070] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0071] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.
[0072] The marine turbocharger compressor performance map prediction method of this embodiment includes:
[0073] Step 1: Obtain a sparse performance feature point dataset of the compressor at multiple speeds, wherein each feature point data includes at least the speed, mass flow rate, pressure ratio, and isentropic efficiency; group the data point set according to the speed to obtain multiple speed line data subsets;
[0074] Compressor tests or numerical calculations typically only yield discrete performance points at finite speeds, and the number of points on each speed line is small (sparse). For subsequent interpolation and prediction, the raw data needs to be grouped and sorted by speed to provide ordered independent variables for conformal interpolation. Specifically:
[0075] Acquire compressor performance data points at multiple speeds *n*. Each data point includes: speed, mass flow rate, pressure ratio, and isentropic efficiency. The speed values divide the data points into multiple subsets, each subset corresponding to an isentropic speed line. Within each subset, sort the mass flow rate in ascending order to obtain an increasing flow rate sequence and its corresponding pressure ratio and efficiency sequences.
[0076] A structured subset of rotational speed line data is obtained, laying an ordered data foundation for conformal interpolation and feature extraction.
[0077] Step 2, Single-speed linear conformal interpolation and constraint correction:
[0078] The sparse raw data points will produce large errors if used directly for weighted prediction. Therefore, we first perform densification interpolation on each known speed line to obtain continuous and shape-preserving candidate curves. However, even shape-preserving interpolation (such as PCHIP) cannot completely guarantee that the monotonicity and efficiency after the peak will not exceed the limit, so additional hard constraints need to be applied.
[0079] For each speed curve, the pressure ratio and efficiency sequences are subjected to piecewise cubic Hermite conformal interpolation (PCHIP). PCHIP connects every two adjacent data points with a cubic polynomial, and its derivative is locally estimated to ensure that the interpolation curve retains the monotonicity and extremum characteristics of the original data, avoiding overshoot and oscillations common in ordinary cubic splines. The PCHIP interpolation function is then applied to the pressure ratio and efficiency respectively to obtain the corresponding pressure ratio curve and isentropic efficiency curve.
[0080] At a constant compressor speed, the pressure ratio initially rises to a peak value with increasing flow rate, then decreases monotonically and does not rise again. Sparse data or interpolation may produce spurious bounces after the peak. Specifically:
[0081] Applying a post-peak monotonically increasing constraint to the pressure ratio curve This ensures that the pressure ratio does not increase with increasing mass flow rate after the peak point, resulting in a candidate pressure ratio curve that monotonically does not increase after the peak.
[0082] Apply efficiency boundary constraints to each of the aforementioned efficiency candidate curves to ensure that the efficiency meets the following conditions. Thus, candidate isentropic efficiency curves that satisfy the efficiency limit are obtained;
[0083] This implementation method effectively suppresses pressure ratio rebound and overshoot after the peak value by combining conformal interpolation with a post-peak monotonic non-increasing constraint. Even when the original data is relatively sparse or located in the boundary region, it can maintain the stability of the post-peak shape, thereby improving the physical rationality and engineering reliability of the pressure ratio curve.
[0084] For the efficiency curve, upper and lower limit constraints are applied during the reconstruction process to ensure that the efficiency remains within a preset reasonable range, thus avoiding non-physical results that exceed the physical upper limit or are significantly distorted. This constraint is a hard constraint for engineering usability, used to ensure that the output performance diagram can be directly used for subsequent thermal calculations, control lookup tables, and whole-machine simulation.
[0085] Step 3: Determine the target rotational speed to be predicted. Select two subsets of rotational speed data that are close to the target rotational speed from multiple subsets of rotational speed data. Construct feature point data of the target rotational speed through linear interpolation. Based on the feature point data of the target rotational speed, calculate the similarity index of each known subset of rotational speed data relative to the target rotational speed. This similarity index includes at least the difference in rotational speed distance, the difference in flow rate corresponding to the pressure ratio peak position, the difference in pressure ratio peak amplitude, the difference in flow rate corresponding to the surge endpoint, the difference in flow rate corresponding to the blockage endpoint, the difference in local slope of the rotational speed line, and the difference in local curvature of the rotational speed line. Generate adaptive weights based on the similarity indexes, and perform weighted calculation on the pressure ratio candidate curves and isentropic efficiency candidate curves of all known rotational speed lines to predict the pressure ratio prediction curve and isentropic efficiency prediction curve of the target rotational speed.
[0086] Since there is no measured curve for the target rotational speed (to be predicted), its morphological difference from known rotational speed lines cannot be directly calculated. Therefore, virtual features of the target rotational speed are first constructed using linear interpolation based on the characteristics of adjacent rotational speed lines (peak position, peak amplitude, surge point, blockage point). Then, the features of each known rotational speed line are compared item by item with these virtual features to obtain a comprehensive difference, which is then converted into an adaptive weight. This weight reflects the degree of similarity in physical morphology between each known rotational speed line and the target rotational speed. Specifically, it includes:
[0087] Step 31: Estimate the peak pressure ratio position, peak pressure ratio amplitude, surge endpoint, and blockage endpoint at the target speed by linear interpolation of adjacent speed lines.
[0088] ;
[0089] The characteristic point data for the target rotational speed includes the peak pressure ratio location, peak pressure ratio amplitude, surge endpoint, or blockage endpoint.
[0090] , This refers to the feature point data adjacent to the rotational speed line;
[0091] For the target speed, , The speed is the speed of the adjacent speed line;
[0092] Peak pressure ratio position at the target rotational speed obtained by interpolation Peak amplitude of pressure ratio Surge endpoint and blocked endpoints ;
[0093] Step 32: Compare each subset of speed curve data with the feature points of the target speed item by item to construct the speed curve data subset. Differences related to target speed , , respectively representing the speed distance difference, the flow rate difference corresponding to the pressure ratio peak position, the pressure ratio peak amplitude difference, the flow rate difference corresponding to the surge end, the flow rate difference corresponding to the blockage end, the local slope difference of the speed line, and the local curvature difference of the speed line;
[0094] rotational speed Distance difference:
[0095]
[0096] Peak pressure ratio position Corresponding flow difference:
[0097]
[0098] Peak amplitude of pressure ratio Difference:
[0099]
[0100] surge endpoint Corresponding flow difference:
[0101]
[0102] Blocked endpoints Corresponding flow difference:
[0103]
[0104] Local slope difference of the speed line:
[0105]
[0106] Local curvature difference of rotational speed line:
[0107]
[0108] Reference curve It can be obtained by weighted average or interpolation of adjacent rotational speeds;
[0109] In terms of boundary processing, the boundary points between the low-flow and high-flow ends of each speed curve are extracted to form surge boundaries and blockage boundaries, respectively. Furthermore, the continuity and stability of the boundary curves across speed directions are constrained or smoothed to avoid significant jumps between adjacent speeds, thereby improving the usability of the boundaries in control protection and operating range division.
[0110] Step 33: Based on the subset of speed curve data Differences related to target speed Obtain the similarity index of each known speed curve data subset relative to the target speed. :
[0111] ;
[0112] These are the weighting coefficients for the corresponding difference terms.
[0113] The similarity index in this embodiment not only considers the difference in rotational speed, but also comprehensively considers morphological information such as differences in peak position, peak size, surge endpoint and blockage endpoint, as well as differences in local slope and curvature.
[0114] In step 34, adaptive weights are generated based on the similarity index. :
[0115] ;
[0116] Represents the sensitivity coefficient. Indicates the number of subsets of speed curve data;
[0117] Step 35: For all candidate pressure ratio curves with known rotational speeds... With isentropic efficiency candidate curve By performing weighted calculations, the pressure ratio prediction curve for the target rotational speed is obtained. With isentropic efficiency prediction curve :
[0118] ;
[0119] .
[0120] Compared to linear interpolation based solely on rotational speed distance, this method incorporates multi-dimensional morphological information such as peak value, boundary, slope, and curvature, enabling the weights to adaptively favor known rotational speed lines that are more consistent with the target morphology, thus significantly improving extrapolation and interpolation accuracy.
[0121] Step 4: Perform joint constraint solution on the pressure ratio prediction curve and isentropic efficiency prediction curve of the target speed to construct the objective function, and apply constraint conditions at the same time to form the target speed line prediction result that satisfies all constraint conditions;
[0122] The objective function is:
[0123]
[0124] in, For the total loss, To fit the weighting coefficients for efficiency, The robust interpolation pressure ratio curve is obtained from robust conventional interpolation (such as PCHIP). The robust interpolation isentropic efficiency curve is obtained by robust conventional interpolation (such as PCHIP);
[0125] The loss is due to the smoothing constraint within the speed range line. This helps to suppress drastic local fluctuations. The loss is due to the smoothness constraint across rotational speeds. The cross-speed smoothness constraint makes the pressure ratio curves of adjacent speed lines second-order smooth at the same flow rate.
[0126] The loss is due to the continuity constraint at the surge boundary. ; This represents the surge point flow rate of the i-th rotational speed line;
[0127] To mitigate the loss caused by blocking boundary continuity constraints, ; This represents the flow rate at the blockage point of the i-th rotational speed line;
[0128] , , , Represents the pre-trained constraint weight coefficients;
[0129] The following conditions must be enforced during the optimization process:
[0130] Post-peak monotonically non-increasing constraint is The efficiency limit constraint is ;
[0131] The position index of the pressure ratio amplitude in the subset of speed curve data sorted by mass flow rate. The pressure ratio amplitude, It is the position index where the pressure ratio amplitude reaches its maximum value; This represents the isentropic efficiency within a subset of the rotational speed data. This represents the maximum value of the isentropic efficiency.
[0132] This implementation method unifies the solution framework for constraints such as post-peak monotonicity constraints, efficiency limit constraints, and boundary continuity constraints, and jointly solves the reconstruction results. This joint solution framework can simultaneously consider the fitting requirements of the original data, the smoothing requirements within a single speed line, the smoothing requirements between different speed lines, and the boundary consistency requirements, and restricts non-physical changes in pressure ratio and efficiency through hard constraints. This avoids the local distortion and constraint conflicts caused by interpolation followed by patching in traditional methods.
[0133] This implementation improves the continuity, stability, and interpretability of boundary curves by extracting surge and blockage boundaries and constraining or smoothing their continuity across speed ranges. Since surge and blockage boundaries directly relate to control and protection logic and the definition of the available operating range, this improvement directly enhances the engineering application value of performance diagrams.
[0134] The above optimization problem can be solved using sequential quadratic programming (SQP) or constrained gradient descent. The initial value for each iteration is a weighted prediction curve, and after several iterations, the optimal curve satisfying all constraints is obtained. By solving the problem jointly, multiple physical and engineering constraints are satisfied at once, avoiding constraint conflicts and local distortions inherent in traditional interpolation-then-repair methods. The resulting target rotational speed curve is physically reasonable, numerically stable, and continuous at its boundaries.
[0135] This step involves constructing a multi-feature difference model to transform the differences between speed curves into differences related to the target speed, and obtaining adaptive weights through exponential mapping to achieve weighted prediction of the target speed curve.
[0136] Step 5: Generate a compressor performance map based on the obtained target speed curve prediction results.
[0137] In sparse data or extrapolated regions, predictions may be unreliable. To provide risk warnings to project users, it is necessary to quantify the confidence level of each prediction point and automatically switch to a more conservative reconstruction strategy when the confidence level is too low.
[0138] To improve engineering reliability under extrapolation conditions, this embodiment further sets up a confidence assessment and strategy switching mechanism. For each reconstructed speed line or each table point, a corresponding confidence index or error index can be output. The performance map of this embodiment includes at least a pressure ratio table, an efficiency table, and confidence indices corresponding to the pressure ratio table and efficiency table. The confidence indices include: prediction uncertainty based on the difference between adjacent speed lines, back-substitution error of the original measurement point, and consistency of indices for multiple interpolation or fitting strategies.
[0139] Uncertainty in predicting the difference between adjacent speed lines: Change the selection combination of adjacent speed lines, make multiple predictions, and calculate the standard deviation of the prediction results.
[0140] Back-substitution error of the original measurement points: The relative root mean square error between the predicted value and the measured value is calculated at the original data point location.
[0141] Consistency of indicators for multiple interpolation or fitting strategies: The mean absolute difference between the results is calculated by independently reconstructing the data using three interpolation methods: PCHIP, cubic spline, and radial basis function.
[0142] When the confidence index falls below a set threshold, the system automatically switches to a conservative reconstruction strategy. This strategy includes strengthening constraints, limiting the extrapolation interval, or changing the weighted prediction model. This embeds the extrapolation risk control mechanism into the entire reconstruction process.
[0143] Control systems and simulation software typically require performance maps to be input in matrix form, with each speed line having the same number of points, and auxiliary coordinates for interpolation lookup. In this embodiment, the compressor performance map is a matrix composed of a speed vector and an auxiliary coordinate parameter vector. The auxiliary coordinate parameters have values ranging from [0,1] and are used to represent the position index of each speed line.
[0144] The predicted pressure ratio and efficiency data are organized into a matrix table structure commonly used in engineering. This matrix table uses the corrected rotational speed vector and auxiliary coordinate vector as indexes, and outputs corresponding confidence tables or error tables. This allows the results to be directly used as parameterized inputs for control and simulation models, reducing manual processing and secondary conversion work, and improving the efficiency of engineering applications.
[0145] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.
Claims
1. A method for predicting the performance of marine turbocharger compressors using a performance map, characterized in that, include: S1. Obtain a sparse performance feature point dataset of the compressor at multiple speeds, where each feature point data includes at least the speed, mass flow rate, pressure ratio and isentropic efficiency. S2. Group the data point set according to rotational speed to obtain multiple subsets of rotational speed line data; S3. Determine the target rotational speed to be predicted. Select two rotational speed data subsets that are close to the target rotational speed from multiple rotational speed data subsets. Construct the feature point data of the target rotational speed through linear interpolation. Based on the feature point data of the target speed, calculate the similarity index of each known speed line data subset relative to the target speed. The similarity index includes at least the speed distance difference, the flow rate difference corresponding to the pressure ratio peak position, the pressure ratio peak amplitude difference, the flow rate difference corresponding to the surge endpoint, the flow rate difference corresponding to the blockage endpoint, the local slope difference of the speed line, and the local curvature difference of the speed line. S4. Based on the similarity index, generate adaptive weights, perform weighted calculations on all known speed line candidate curves for pressure ratio and isentropic efficiency, and predict the target speed for pressure ratio and isentropic efficiency. S5. Perform joint constraint solution on the pressure ratio prediction curve and isentropic efficiency prediction curve of the target speed to construct the objective function, and apply constraint conditions at the same time to form the target speed line prediction result that satisfies all constraint conditions. S6. Based on the obtained target speed curve prediction results, generate a compressor performance map.
2. The marine turbocharger compressor performance map prediction method according to claim 1, characterized in that, S3 include: S31. By using linear interpolation of adjacent speed lines, estimate the peak pressure ratio position, peak pressure ratio amplitude, surge endpoint, and blockage endpoint at the target speed, respectively. ； The characteristic point data for the target rotational speed includes the peak pressure ratio location, peak pressure ratio amplitude, surge endpoint, or blockage endpoint. , This refers to the feature point data adjacent to the rotational speed line; For the target speed, , The speed is the speed of the adjacent speed line; S32. Each subset of speed curve data is compared item by item with the feature points of the target speed to construct a subset of speed curve data. Differences related to target speed , , respectively representing the speed distance difference, the flow rate difference corresponding to the pressure ratio peak position, the pressure ratio peak amplitude difference, the flow rate difference corresponding to the surge end, the flow rate difference corresponding to the blockage end, the local slope difference of the speed line, and the local curvature difference of the speed line; S33, Based on the subset of speed line data Differences related to target speed Obtain the similarity index of each known speed curve data subset relative to the target speed. : ； These are the weighting coefficients for the corresponding difference terms.
3. The marine turbocharger compressor performance map prediction method according to claim 1, characterized in that, In S4, adaptive weights are generated based on the similarity index. : ； Represents the sensitivity coefficient. Indicates the number of subsets of speed curve data; Candidate pressure ratio curves for all known speed lines With isentropic efficiency candidate curve By performing weighted calculations, the pressure ratio prediction curve for the target rotational speed is obtained. With isentropic efficiency prediction curve : ； 。 4. The marine turbocharger compressor performance map prediction method according to claim 3, characterized in that, The method for obtaining the pressure ratio curve and isentropic efficiency curve of the adjacent speed line is as follows: Within each subset of speed lines, the data points are sorted by mass flow rate from smallest to largest. Interpolation is performed on the pressure ratio sequence and isentropic efficiency sequence of each subset of speed lines to obtain the corresponding pressure ratio curve and isentropic efficiency curve. The pressure ratio peak value is determined based on each pressure ratio curve, and a post-peak monotonically non-increasing constraint is applied to the high flow rate region after the pressure ratio peak value to obtain a candidate pressure ratio curve that is monotonically non-increasing after the peak value. An efficiency limit constraint is applied to each of the isentropic efficiency curves to obtain isentropic efficiency candidate curves that satisfy the efficiency limit.
5. The marine turbocharger compressor performance map prediction method according to claim 4, characterized in that, For each subset of rotational speed lines, perform piecewise cubic Hermite conformal interpolation on the pressure ratio sequence and the isentropic efficiency sequence.
6. The marine turbocharger compressor performance map prediction method according to claim 4, characterized in that, The constraints include post-peak monotonicity constraint, efficiency limit constraint, surge boundary continuity constraint, blockage boundary continuity constraint, speed-line smoothness constraint, and speed-crossing smoothness constraint.
7. The marine turbocharger compressor performance map prediction method according to claim 6, characterized in that, The objective function is: ； in, For the total loss, To fit the weighting coefficients for efficiency, The robust interpolation pressure ratio curve is obtained from robust conventional interpolation (such as PCHIP). The robust interpolation isentropic efficiency curve is obtained by robust conventional interpolation (such as PCHIP); This is the loss due to the smoothing constraint within the speed range line; Loss due to smoothness constraints across rotational speeds; The loss is due to the continuity constraint of the surge boundary; The loss is due to the blockage of boundary continuity constraints; , , , Represents the pre-trained constraint weight coefficients; Post-peak monotonically non-increasing constraint The efficiency limit constraint is ; The position index of the pressure ratio amplitude in the subset of speed curve data sorted by mass flow rate. The pressure ratio amplitude, It is the position index where the pressure ratio amplitude reaches its maximum value; This represents the isentropic efficiency within a subset of the rotational speed data. This represents the maximum value of the isentropic efficiency.
8. The method for predicting the performance of marine turbocharger compressors according to claim 1, characterized in that, The performance map includes at least a pressure ratio table, an efficiency table, and confidence indices corresponding to the pressure ratio table and efficiency table. The confidence indices include: prediction uncertainty based on the difference between adjacent speed lines, back-substitution error of the original measurement points, and consistency of indices for multiple interpolation or fitting strategies. When the confidence index falls below a set threshold, the system automatically switches to a conservative reconstruction strategy, which includes strengthening the constraint strength, limiting the extrapolation interval, or changing the weighted prediction model.
9. The marine turbocharger compressor performance map prediction method according to claim 1, characterized in that, The compressor performance map is a matrix table composed of a speed vector and an auxiliary coordinate parameter vector. The auxiliary coordinate parameters have a value range of [0, 1] and are used to represent the position index of each speed line.
10. A marine turbocharger compressor performance map prediction device, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the marine turbocharger compressor performance map prediction method as described in any one of claims 1 to 7.