Calibration method for load flow signal in hydraulic system

Abstract

The application provides a calibration method for a load flow signal in a hydraulic system, comprising determining characteristic quantities affecting changes in a load flow signal of an engine-driven pump in the hydraulic system; pre-processing collected flow signal sample data; classifying the pre-processed flow signal sample data to obtain a plurality of similar sample sets; establishing a light gradient boosting tree calibration model for each similar sample set and the characteristic quantities corresponding to the similar sample set; and calibrating the flow signal sample data using the light gradient boosting tree calibration model to obtain calibrated sampling signal values. The above method has high calibration accuracy, good generalization performance and network stability, and is conducive to the regulation and risk assessment of the load flow signal of the engine-driven pump in the hydraulic system, and has higher practical application value.

Description

Technical Field

[0001] This invention relates to the field of electromechanical hydraulic system technology, specifically to a calibration method for load flow signals in a hydraulic system. Background Technology

[0002] The aircraft hydraulic system is one of the most important airborne systems, driving the aircraft's power output. Its performance, stability, and reliability directly affect aircraft safety. In the aircraft hydraulic system, high-pressure oil flow serves as the load for the engine-driven pump (EDP). The rated flow rate is typically designed based on the EDP's structure and the requirements of the drive unit. However, due to factors such as operating environment, component aging, and maintenance, the flow rate fluctuates, which can significantly impact the drive unit. Therefore, measuring flow rate is a crucial criterion for hydraulic system stability. Real-time monitoring of hydraulic system pipeline flow provides strong support for system characteristic analysis and fault diagnosis.

[0003] Since the flow resistance caused by the flow sensor has a significant impact on the system characteristics, the accuracy of data acquisition from the flow sensor using traditional methods is not very high. Therefore, in actual operation, it is necessary to consider parameters that are easier to acquire, such as EDP system pressure, temperature, and high-pressure filtration differential, which have strong coupling and correlation with the load flow. By using these parameters to calibrate the load flow signal, the accuracy of judging the stability of the hydraulic system can be improved.

[0004] Currently, traditional calibration methods mainly rely on linear fitting and neural network calibration. Linear fitting typically uses polynomial linear calibration models, which are simple in structure but have limitations. For example, low-order polynomials cannot provide good calibration results for models with many influencing factors; furthermore, the fitting curves of high-order polynomials are not smooth, often exhibiting large fluctuations in local waveforms, making computation more time-consuming. Neural network calibration models, on the other hand, utilize neural network algorithms with advantages such as high tolerance to sample data, strong nonlinear mapping capabilities, and self-adaptation and self-organization. They have been widely used in signal calibration, but their training requires a large number of samples and high computing power. In airborne embedded environments, limitations in system size, power consumption, and computing power make it difficult to meet the computational processing requirements of neural networks. Summary of the Invention

[0005] To effectively calibrate the signal to be calibrated and improve the accuracy and speed of calibrating the load flow signal of the engine-driven pump in a hydraulic system, this invention designs a calibration method for the load flow signal in a hydraulic system. The calibration model established by this method can achieve high calibration accuracy, has good generalization performance and network stability, and is beneficial for regulating and risk assessment of the load flow signal of the engine-driven pump in a hydraulic system, thus having greater practical application value.

[0006] The technical solution to achieve the purpose of the invention is as follows: A calibration method for load flow signal in a hydraulic system, comprising the following steps:

[0007] Step 1: Determine the characteristic quantities that affect the load flow rate changes of the engine-driven pump in the hydraulic system;

[0008] Step 2: Preprocess the collected flow signal sample data;

[0009] Step 3: Classify the preprocessed flow signal sample data to obtain multiple similar sample sets;

[0010] Step 4: For each similar sample set and its corresponding feature quantity, establish a lightweight gradient boosting tree calibration model for that similar sample set;

[0011] Step 5: Use a lightweight gradient boosting tree calibration model to calibrate the flow signal sample data and obtain the calibration sample signal value.

[0012] Furthermore, in step 1 above, the method for determining the characteristic quantities affecting the change of the load flow signal of the hydraulic system engine drive pump is as follows: based on the Pearson correlation coefficient, the method for obtaining the characteristic quantities affecting the load flow signal.

[0013] Furthermore, methods for obtaining characteristic quantities affecting the load flow signal based on the Pearson correlation coefficient include:

[0014] Step 1.1: Obtain n influencing factors affecting the change in load flow signal;

[0015] Step 1.2: Use the Pearson correlation coefficient r to perform correlation analysis on the n influencing factors;

[0016] Step 1.3: Based on the correlation analysis results of each influencing factor, determine that n1 of the n influencing factors are the characteristic quantities of the sampled signal, where n≥n1.

[0017] Furthermore, in step 2 above, the collected flow signal sample data is preprocessed, including identifying outliers in the flow signal sample data and removing or correcting the outliers.

[0018] Furthermore, the outlier removal method is as follows: when the sampled data satisfies max(|Y采 -Y 标 When |)>ε1, the sampled data is discarded, where Y 采 For sampled data, Y 标 ε1 is the preset sampling signal calibration value; ε1 is the discard threshold.

[0019] Furthermore, the outlier correction method is as follows: when the sampled data satisfies ε1>max(|Y 采 -Y 标 When |)>ε2, then according to the formula Y 采 =Y 标 +(|Y 采 -Y 标 |) / φ corrects the sampled data;

[0020] Among them, Y 采 For sampled data, Y 标 ε1 is the preset sampling signal calibration value; ε2 is the discard threshold; φ is the correction threshold; and φ is the correction coefficient.

[0021] Furthermore, in step 3 above, the preprocessed flow signal sample data is classified to obtain multiple similar sample sets, including:

[0022] Step 3.1: Classify the preprocessed flow signal sample data and its corresponding feature quantities according to the K-means clustering algorithm;

[0023] Step 3.2: Determine the number of clusters and select the cluster center locations;

[0024] Step 3.3: Based on the number of clusters and the location of the cluster centers, multiple similar sample sets are formed, and these multiple similar sample sets are the optimal clustering multiple similar sample sets.

[0025] Furthermore, in step 4 above, the method for establishing a lightweight gradient boosting tree calibration model for each similar sample set and its corresponding feature quantity is as follows: the similar sample set and its corresponding feature quantity are used as the input of the model, the preset sampling signal calibration value is used as the output of the model, the model is trained and iterated to obtain the lightweight gradient boosting tree calibration model.

[0026] Furthermore, step 4 also includes an evaluation step for the lightweight gradient boosting tree calibration model.

[0027] Preferred evaluation methods for lightweight gradient boosting tree calibration models include:

[0028] Calculate the F1 score of the similar sample set based on the precision P and recall R of each similar sample set;

[0029] Calculate the average Macro-F1 based on the F1 scores of all similar sample sets;

[0030] The lightweight gradient boosting tree calibration model is evaluated based on the average Macro-F1.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: The calibration method for load flow signals in hydraulic systems designed in this invention obtains the characteristic quantities affecting the load flow signals of engine-driven pumps in hydraulic systems by performing correlation analysis on the factors affecting the changes in load flow signals of engine-driven pumps in hydraulic systems; after preprocessing the sample data of load flow signals of engine-driven pumps in hydraulic systems, a similar sample set is constructed using the K-means algorithm on the preprocessed data, so that each type of sample set has similar characteristic quantities; finally, by establishing a calibration model based on a lightweight gradient boosting tree for each type of similar sample set, the signal to be calibrated can be effectively calibrated, which can greatly improve the accuracy and speed of calibration of load flow signals of engine-driven pumps in hydraulic systems. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0033] Figure 1 This is a flowchart illustrating the calibration method for the load flow signal in the hydraulic system of the present invention.

[0034] Figure 2 This is a flowchart of the K-means clustering algorithm in a specific implementation;

[0035] Figure 3 A flowchart illustrating the modeling of the lightweight gradient boosting tree calibration model in a specific implementation;

[0036] Figure 4 This is a schematic diagram illustrating the relationship between the sum of squared errors and the number of clusters k in a specific implementation method. Detailed Implementation

[0037] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0038] This specific embodiment discloses a calibration method for load flow signals in a hydraulic system, see [link to relevant documentation]. Figure 1 As shown, the calibration method for the load flow signal includes the following steps:

[0039] Step 1: Determine the characteristic quantities that affect the load flow rate changes of the engine-driven pump in the hydraulic system;

[0040] Step 2: Preprocess the collected flow signal sample data;

[0041] Step 3: Classify the preprocessed flow signal sample data to obtain multiple similar sample sets;

[0042] Step 4: For each similar sample set and its corresponding feature quantity, establish a lightweight gradient boosting tree calibration model for that similar sample set;

[0043] Step 5: Use a lightweight gradient boosting tree calibration model to calibrate the flow signal sample data and obtain the calibration sample signal value.

[0044] In an optional embodiment, since the Pearson correlation coefficient r can measure the strength of the correlation between two variables among the influencing factors, the feature quantities to be used as sampling signals are selected from the influencing factors based on the specific value of the calculated Pearson correlation coefficient. Therefore, in step 1 above, the method for determining the feature quantities affecting the change of the load flow signal of the hydraulic system engine drive pump is: a method for obtaining the feature quantities affecting the load flow signal based on the Pearson correlation coefficient.

[0045] Specifically, methods for obtaining characteristic quantities affecting load flow signals based on the Pearson correlation coefficient include:

[0046] Step 1.1: Obtain n influencing factors affecting the change in load flow signal;

[0047] Step 1.2: Use the Pearson correlation coefficient r to perform correlation analysis on the n influencing factors;

[0048] Step 1.3: Based on the correlation analysis results of each influencing factor, determine that n1 of the n influencing factors are the characteristic quantities of the sampled signal, where n≥n1.

[0049] For example, if we define the Pearson correlation coefficient r as ranging from [-1, +1], when r > 0, it indicates a positive correlation between the two variables; when r < 0, it indicates a negative correlation; and when r = 0, it indicates no correlation. Therefore, the larger the absolute value of the Pearson correlation coefficient r, the stronger the correlation between the two variables. The formula for calculating the Pearson correlation coefficient r is: Where r is the Pearson correlation coefficient between the sampled signal and a certain influencing factor; n is the number of influencing factors; x is the Pearson correlation coefficient value of a certain influencing factor; and P is the average value of the sampled signal.

[0050] When judging the influencing factors of the characteristic quantity of the sampled signal in step 1.3, it is set that when |r|>0.7, the influencing factor can be regarded as the characteristic quantity of the sampled signal.

[0051] In an optional embodiment, step 2 above involves preprocessing the collected traffic signal sample data, including identifying outliers in the traffic signal sample data and removing or correcting them, as well as supplementing missing data and smoothing outliers to reduce noise data or random fluctuations.

[0052] The outlier removal method is as follows: when the sampled data satisfies max(|Y 采 -Y 标 When |)>ε1, the sampled data is discarded, where Y 采 For sampled data, Y 标 ε1 is the preset sampling signal calibration value; ε1 is the discard threshold.

[0053] The outlier correction method is as follows: when the sampled data satisfies ε1>max(|Y 采 -Y 标 When |)>ε2, then according to the formula Y 采 =Y 标 +(|Y 采 -Y 标 |) / φ corrects the sampled data;

[0054] Among them, Y 采 For sampled data, Y 标 ε1 is the preset sampling signal calibration value; ε2 is the discard threshold; φ is the correction threshold; and φ is the correction coefficient.

[0055] In an optional embodiment, in step 3 above, the preprocessed traffic signal sample data is classified to obtain multiple similar sample sets, including:

[0056] Step 3.1: Classify the preprocessed flow signal sample data and its corresponding feature quantities according to the K-means clustering algorithm;

[0057] Step 3.2: Determine the number of clusters and select the cluster center locations;

[0058] Step 3.3: Based on the number of clusters and the location of the cluster centers, multiple similar sample sets are formed, and these multiple similar sample sets are the optimal clustering multiple similar sample sets.

[0059] Specifically, for the detailed process of obtaining the number of clusters and selecting the cluster center location in step 3.2, please refer to [link to relevant documentation]. Figure 2As shown, the optimal number of clusters is determined by observing the changes in the error sum of squares curve for each similar sample set: as the number of clusters gradually increases, the intra-cluster cohesion of each cluster will gradually increase. Before reaching the optimal number of clusters, the error sum of squares curve is relatively steep; when the optimal number of clusters is reached, the error sum of squares curve becomes relatively stable. Therefore, the number of clusters where the error sum of squares curve changes abruptly is selected as the number of similar sample sets, thereby forming multiple similar sample sets with optimal clustering for the overall acquired signal samples and their corresponding features.

[0060] In an optional embodiment, in step 4 above, the method for establishing a lightweight gradient boosting tree calibration model for each similar sample set and its corresponding feature quantity is as follows: the similar sample set and its corresponding feature quantity are used as the input of the model, the preset sampling signal calibration value is used as the output of the model, the model is trained and iterated to obtain a lightweight gradient boosting tree calibration model.

[0061] Specifically, in the process of establishing a lightweight gradient boosting tree calibration model: First, initialize the similar sample set for each class of predictions and its corresponding feature quantities; then, define an objective function to represent the bias of the calibration model, where the objective function Obj is defined as:

[0062] Ω(f t ) represents the regularization term. In the formula: T t ω is the number of leaf nodes; j Let be the node weights on leaf j; γ and λ are pre-defined hyperparameters.

[0063] See Figure 3 As shown, the objective function is simplified using the Taylor algorithm through iterative optimization. The leaf nodes of the decision tree are traversed, and the optimal split point of the leaf node is found using the histogram algorithm, thus constructing a new decision tree. The new decision tree is then used to predict and obtain a set of similar samples from the sampled signal.

[0064] In an optional embodiment, step 4 further includes evaluating the lightweight gradient boosting tree calibration model. The evaluation method for the lightweight gradient boosting tree calibration model includes:

[0065] Calculate the F1 score of the similar sample set based on the precision P and recall R of each similar sample set;

[0066] Calculate the average Macro-F1 based on the F1 scores of all similar sample sets;

[0067] The lightweight gradient boosting tree calibration model is evaluated based on the average Macro-F1.

[0068] Specifically, the precision (P) and recall (R) of each similar sample set can be calculated by statistically analyzing the TP (true positive, predicted positive, and actual positive), FP (false positive, predicted positive, and actual negative), FN (false negative, predicted negative, and actual positive), and TN (true negative, predicted negative, and actual negative) values. This yields the F1 score for each similar sample set. The average F1 score across all similar sample sets, Macro – F1, is then calculated as follows:

[0069]

[0070] The lightweight gradient boosting tree calibration model is evaluated based on the calculated average value Macro-F1. Macro-F1 ranges from [0,1]. The closer the calculated average value Macro-F1 is to 1, the better the performance of the lightweight gradient boosting tree calibration model.

[0071] The present invention discloses a calibration method for load flow signals in hydraulic systems. This method involves performing correlation analysis on factors affecting changes in the load flow signals of engine-driven pumps in hydraulic systems to derive characteristic quantities affecting these signals. After preprocessing the sample data of the load flow signals, a similar sample set is constructed using the K-means algorithm, ensuring that each sample set has similar characteristic quantities. Finally, a calibration model based on a lightweight gradient boosting tree is established for each type of similar sample set, enabling effective calibration of the signal to be calibrated. This significantly improves the accuracy and speed of calibrating the load flow signals of engine-driven pumps in hydraulic systems.

[0072] The following examples illustrate the calibration method for the load flow signal in the aforementioned hydraulic system. Figures 1 to 4 As shown;

[0073] Step 1: Identify the factors affecting the load flow signal change of the hydraulic system engine drive pump: suction port pressure, return port pressure, high-pressure oil filter differential pressure, suction port temperature, return port temperature, and high-pressure port temperature. Calculate the Pearson correlation coefficient r between each influencing factor and the sampled signal (see Table 1). It is concluded that the characteristic quantities affecting the load flow signal change of the hydraulic system engine drive pump are the high-pressure oil filter differential pressure, return port pressure, and housing temperature, totaling three influencing factors.

[0074] Table 1. Statistical results of Pearson correlation coefficient.

[0075]

[0076] Step 2: Perform data preprocessing on all load flow signal sample data collected by the hydraulic system engine drive pump, identify, correct and supplement abnormal points in the sampled data, or discard abrupt signal sample values ​​with large deviations and correct the collected signal samples with slight offsets caused by noise.

[0077] Step 3: Use the K-means clustering algorithm to classify all sampled signal samples and their corresponding temperature values. See [link to implementation details] for more information. Figure 2 During clustering, the optimal number of clusters is determined by observing the changes in the sum of squared errors curve. (See also...) Figure 4 It can be seen that when the number of clusters k is less than 3, the curve of the sum of squared errors is relatively steep; when the number of clusters is greater than 3, the curve of the sum of squared errors becomes relatively stable. Therefore, the optimal number of clusters k for this acquired signal sample is 3, thus dividing the overall acquired signal sample and its corresponding temperature value into 3 similar sample sets.

[0078] Step 4: Establish a calibration model based on a lightweight gradient boosting tree for the three similar sample sets of the hydraulic system engine drive pump load flow signal: First, initialize the similar sample set of each prediction and its corresponding feature quantity; second, iteratively optimize the objective function to construct a new decision tree, and create decision trees in a loop until the stopping condition of 120 iterations is met.

[0079] Step 5: Calculate the Macro-F1 score. Calculate TP, FP, FN and TN for the three similar sample sets respectively, and find the precision and recall for each similar sample set. Then calculate the F1 score and Macro-F1 score. Evaluate the performance of the model classifier based on the Macro-F1 score.

[0080] Step 6: Use the signal to be calibrated and its characteristic temperature value as input to the calibration model based on lightweight gradient boosting tree trained by the similar sample set selected in Step 5. The output of the model is the calibration value corresponding to the calibration signal.

[0081] The calibration method for load flow signals in hydraulic systems disclosed in this specific embodiment has high calibration accuracy, good generalization performance and network stability, and is beneficial for regulating and risk assessment of load flow signals of engine-driven pumps in hydraulic systems, thus having greater practical application value.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0083] Furthermore, this specific embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned calibration method for load flow signals.

[0084] This specific embodiment also discloses a computer-readable storage medium storing a calibration method for performing the above-described load flow signal.

[0085] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method of calibrating a load flow signal in a hydraulic system, characterized by, The method comprises the following steps: Step 1, determining the characteristic quantity affecting the change of the load flow signal of the engine-driven pump of the hydraulic system; Step 2, preprocessing the collected flow signal sample data; Step 3, classifying the preprocessed flow signal sample data to obtain a plurality of similar sample sets; Step 4: For each similar sample set and its corresponding features, establish a lightweight gradient boosting tree calibration model for that similar sample set, including: initializing the similar sample set and its corresponding features for each class of predictions; defining an objective function to represent the bias of the calibration model, where the objective function Obj is defined as: , For regularization terms, In the formula: This represents the number of leaf nodes; For leaves Node weights; Given hyperparameters; the objective function is simplified by iterative optimization using the Tarr algorithm, the leaf nodes of the decision tree are traversed, and the optimal split point of the leaf node is found using the histogram algorithm, thus constructing a new decision tree; the new decision tree is used to predict and obtain a similar sample set of the sampled signal; the similar sample set and its corresponding features are used as the input of the model, and the preset sampled signal calibration value is used as the output of the model. The model is trained and iterated to obtain a lightweight gradient boosting tree calibration model. Step 5, calibrating the flow signal sample data by using a light gradient boosting tree calibration model to obtain calibrated sampling signal values.

2. The method of calibrating a load flow signal in a hydraulic system of claim 1, wherein: In step 1, the method for determining the characteristic quantity affecting the change of the load flow signal of the engine-driven pump of the hydraulic system is as follows: according to the Pearson correlation coefficient, a method for obtaining the characteristic quantity affecting the load flow signal is obtained.

3. The method of calibrating a load flow signal in a hydraulic system of claim 2, wherein: The method for obtaining the characteristic quantity affecting the load flow signal according to the Pearson correlation coefficient comprises: Step 1.1, obtaining n influencing factors affecting the change of the load flow signal; Step 1.2, performing correlation analysis on the n influencing factors by using the Pearson correlation coefficient r; Step 1.3, determining n1 influencing factors in the n influencing factors as the characteristic quantity of the sampling signal according to the correlation analysis results of the influencing factors, wherein n≥n1.

4. The method of calibrating a load flow signal in a hydraulic system of claim 1, wherein: In step 2, the preprocessing of the collected flow signal sample data comprises identifying abnormal points in the flow signal sample data and eliminating or correcting the abnormal points.

5. The method of calibrating a load flow signal in a hydraulic system of claim 4, wherein: The method for removing abnormal points is: when the sampling data satisfies the condition , the sampling data is removed, wherein is the sampling data, is a preset sampling signal calibration value; is a discard threshold.

6. The method of calibrating a load flow signal in a hydraulic system of claim 4, wherein: The method for correcting the abnormal point is: when the sampling data satisfies the condition , then the sampling data is corrected according to the formula . wherein, is a sampling data, is a preset sampling signal calibration value; is a rejection threshold; is a correction threshold; is a correction coefficient.

7. The method of calibrating a load flow signal in a hydraulic system of claim 1, wherein: In step 3, the preprocessed flow signal sample data is classified to obtain a plurality of similar sample sets, which comprises: Step 3.1, classifying the preprocessed flow signal sample data and the corresponding characteristic quantity according to the K-means clustering algorithm; Step 3.2, determining the number of clustering clusters and selecting the cluster center position; Step 3.3, forming a plurality of similar sample sets according to the number of clustering clusters and the cluster center position, and the plurality of similar sample sets are optimal clustering similar sample sets.

8. The method of calibrating a load flow signal in a hydraulic system of claim 1, wherein: Step 4 further comprises an evaluation step of the light gradient boosting tree calibration model.

9. The method of calibrating a load flow signal in a hydraulic system of claim 8, wherein: The evaluation method of the light gradient boosting tree calibration model comprises: According to the precision P and recall R of each similar sample set, the F1 value of the similar sample set is calculated; According to the F1 values of all similar sample sets, the average value Macro-F1 is calculated; According to the average value Macro-F1, the light gradient boosting tree calibration model is evaluated.

Images