Hydraulic Control Optimization System and Method Based on Differential Pressure Direction Variation Pattern Recognition

Abstract

This invention discloses a hydraulic control optimization system and method based on differential pressure direction change pattern recognition, comprising: performing phase space topology reconstruction and singular spectrum analysis based on the pressure signals of the hydraulic cylinder's inlet and outlet chambers to determine the topological mapping features of the differential pressure direction; based on the topological mapping features, extracting spatial features through a multi-scale convolutional network and combining an attention mechanism to determine the pattern reference features; performing real-time differential pressure load pattern recognition using the manifold mapping space; constructing a pattern comparison learning mechanism based on historical load patterns to predict pattern trends and determine hydraulic control target parameters; and dynamically updating the pattern reference features by analyzing the error between the hydraulic cylinder piston rod displacement feedback and the control target parameters through a disturbance observer. This invention achieves accurate real-time identification and dynamic optimization control of differential pressure fluctuation patterns in hydraulic systems, significantly improving the system's operational stability and control accuracy.

Description

Technical Field

[0001] This invention relates to the field of hydraulic system technology, specifically to a hydraulic control optimization system and method based on differential pressure direction change pattern recognition. Background Technology

[0002] Hydraulic systems are widely used in engineering machinery, industrial equipment, and automation control. They achieve mechanical motion and force output through hydraulic cylinders, offering advantages such as high power density, rapid response, and high control precision. However, during hydraulic cylinder operation, cavitation often occurs due to changes in operating conditions, load fluctuations, and the complexity of hydraulic fluid flow, leading to drastic pressure fluctuations within the system. This cavitation phenomenon not only affects the performance of the hydraulic cylinder but also severely damages the service life of hydraulic components over time.

[0003] Traditional hydraulic control strategies typically employ fixed control parameters or simple feedback control mechanisms to adjust operating conditions, failing to effectively identify and respond to the complex, nonlinear pressure fluctuations caused by cavitation. This traditional strategy lacks in-depth analysis of the internal flow state of the hydraulic system, resulting in an inability to accurately predict and quickly respond to abnormal operating conditions, particularly in its insufficient ability to identify differential pressure fluctuation patterns caused by cavitation.

[0004] In existing technologies, the detection and analysis methods for cavitation phenomena mostly employ static threshold settings or single frequency domain analysis. These methods can only identify obvious and stable cavitation phenomena, and cannot perform real-time and accurate pattern recognition and control response for dynamically changing cavitation processes, resulting in poor system stability and inadequate control performance.

[0005] Therefore, accurately identifying the differential pressure direction change pattern caused by cavitation in the hydraulic system and optimizing the hydraulic control strategy based on the real-time identification results to achieve efficient and precise control of the hydraulic cylinder's operating status has become one of the important technical problems that urgently need to be solved in the current hydraulic control field. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a hydraulic control optimization system and method based on differential pressure direction change pattern recognition.

[0007] To achieve the above objectives, the present invention provides a hydraulic control optimization system and method based on differential pressure direction change pattern recognition, comprising:

[0008] The phase space topology reconstruction is performed based on the pressure signals of the hydraulic cylinder's inlet and outlet chambers, and the singular spectrum analysis is performed based on the topology reconstruction results to determine the topological mapping characteristics of the pressure difference direction.

[0009] Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction.

[0010] Based on the model reference characteristics, a manifold mapping space is constructed, and based on the topological mapping characteristics of the real-time differential pressure signal, manifold distance measurement processing is performed to determine the load mode in the current differential pressure direction.

[0011] A pattern comparison and learning mechanism is constructed based on the current load pattern and historical load patterns, and the pattern trend prediction processing is performed based on the pattern comparison and learning results to determine the target parameters of hydraulic control.

[0012] The disturbance observer is processed based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and the mode reference characteristics are updated based on the observer output.

[0013] Furthermore, the topological mapping feature for determining the pressure difference direction includes:

[0014] Based on the physical characteristics of the influence of cavitation in hydraulic cylinders on pressure fluctuations, time delay determination processing is performed to obtain the non-uniform time delay characteristics of topological reconstruction.

[0015] Pressure signal delay coordinate mapping is performed based on non-uniform time delay characteristics to obtain the initial topology of phase space;

[0016] Phase space attractor trajectory density analysis is performed based on the initial topology to determine the characteristics of the attractor region;

[0017] Singular spectral features are extracted based on attractor region characteristics to determine the topological mapping features of the pressure difference direction.

[0018] Furthermore, the logic for generating the non-uniform time-delay features of the topology reconstruction includes:

[0019] Based on the amplitude of the pressure signal change in the hydraulic cylinder inlet chamber, the cavitation initiation condition is identified and processed to obtain the cavitation initiation timing characteristics.

[0020] Based on the timing characteristics of cavitation formation and the physical process of gas-liquid phase transformation, the pressure difference fluctuation transmission path is analyzed to obtain the fluctuation propagation characteristics;

[0021] Based on the wave propagation characteristics, the effective delay interval of the pressure signal is divided into non-uniform intervals to obtain the non-uniform time delay characteristics of topological reconstruction.

[0022] Furthermore, the mode reference feature for determining the direction of pressure difference includes:

[0023] Scale-based hierarchical selection is performed based on the local sensitive regions of the topological mapping features to determine the hierarchical scale of multi-scale convolution.

[0024] Based on the hierarchical scale, the topological mapping features are processed by convolution kernel local spatial feature extraction to obtain convolution local features;

[0025] Spatial correlation analysis is performed based on the strength of the correlation between the spatial location of local convolutional features and changes in operating conditions to obtain spatial correlation weight features;

[0026] Attention fusion processing is performed based on convolutional local features and spatial correlation weight features to determine the pattern baseline features of the pressure difference direction.

[0027] Furthermore, the spatial association weight feature generation logic includes:

[0028] Identify strongly correlated regions and determine initial sensitive regions based on the spatial distribution characteristics of local convolutional features;

[0029] Based on the initial sensitive area, local operating condition differences are identified to determine the regional difference characteristics.

[0030] Spatial association weight features are obtained by non-uniformly distributing association weights based on regional disparity characteristics.

[0031] Further, determine the load pattern in the current pressure differential direction, including:

[0032] Based on the load condition transition characteristics of the model baseline features, the manifold space is constructed to determine the load model manifold characteristics.

[0033] Local feature extraction is performed based on the topological mapping characteristics of the real-time differential pressure signal to determine the real-time topological local features;

[0034] Based on the local mapping distribution characteristics of real-time topological local features in the manifold space, feature mapping similarity measurement is performed to determine the manifold mapping distance features;

[0035] The load pattern is determined by performing optimal matching processing based on the manifold mapping distance characteristics to determine the load pattern in the current pressure difference direction.

[0036] Further, determining the target parameters for hydraulic control includes:

[0037] Based on the working condition transfer characteristics of the current load mode and the historical load mode, load mode transfer path identification processing is performed to determine the characteristics of the mode transfer path.

[0038] Based on the characteristics of the pattern transfer path, a comparative learning sample selection process is performed to determine typical pattern comparison samples.

[0039] Based on the sensitivity characteristics of typical model comparison samples to changes in operating conditions, differential prediction processing of model trends is performed to determine the predictive characteristics of changes in operating conditions.

[0040] Based on the predicted characteristics of changes in operating conditions, hydraulic control target parameters are selected and determined.

[0041] Furthermore, the logic for generating the predicted features of the operating condition change includes:

[0042] Based on the rapid change characteristics of load conditions in typical pattern comparison samples, rapid change sample identification processing is performed to determine rapidly changing sensitive samples.

[0043] Based on rapidly changing sensitive samples, the trend and direction of working condition changes are identified to determine the characteristics of the working condition change direction.

[0044] Based on the characteristics of the direction of operating condition changes, local differential prediction processing of operating condition trends is performed to obtain the predictive features of operating condition changes.

[0045] Furthermore, the step of processing the disturbance observer based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updating the mode reference characteristics based on the observer output, includes:

[0046] Based on the time-domain variation characteristics of the control error, dynamic identification of interference sources is performed to determine the characteristics of the interference sources;

[0047] Based on the characteristics of the interference source and the load response characteristics of the hydraulic actuator, the interference compensation amount is predicted to obtain the interference compensation characteristics.

[0048] The model reference features are updated and fused based on the interference compensation features and the model reference features.

[0049] The hydraulic control optimization system based on differential pressure direction change pattern recognition is implemented based on the aforementioned hydraulic control optimization method based on differential pressure direction change pattern recognition, including:

[0050] Differential pressure topology feature extraction module: Performs phase space topology reconstruction processing based on the pressure signals of the hydraulic cylinder inlet and outlet oil chambers, and performs singular spectrum analysis processing based on the topology reconstruction results to determine the topology mapping features of the differential pressure direction;

[0051] Multi-scale feature fusion module: Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction;

[0052] Load pattern recognition module: Constructs a manifold mapping space based on pattern reference features, and performs manifold distance measurement processing based on the topological mapping features of real-time differential pressure signals to determine the load pattern in the current differential pressure direction;

[0053] Load trend prediction module: Constructs a pattern comparison learning mechanism based on the current load pattern and historical load patterns, and performs pattern trend prediction processing based on the pattern comparison learning results to determine the hydraulic control target parameters;

[0054] Interference Observation and Dynamic Correction Module: The interference observer processes the interference based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updates the mode reference characteristics based on the observer output.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] This invention determines the non-uniform time delay characteristics based on the physical mechanism of cavitation phenomena inside hydraulic cylinders, and combines phase space topology reconstruction and singular spectrum analysis to accurately identify the pressure difference direction change patterns caused by cavitation. This effectively overcomes the problems of poor identification accuracy and insufficient real-time performance caused by relying solely on static thresholds or single frequency domain analysis in existing technologies, and significantly improves the accuracy and reliability of hydraulic system pressure difference fluctuation pattern identification.

[0057] This invention employs a spatial feature extraction method that integrates multi-scale convolutional networks and attention mechanisms. This method can deeply explore the contribution of features from different local regions to the recognition of pressure difference direction change patterns, forming accurate pattern reference features. This significantly enhances the system's sensitivity to different load conditions and dynamic cavitation phenomena, enabling the hydraulic control strategy to respond more precisely to changes in actual working conditions and effectively improving the operational stability and control accuracy of the hydraulic system.

[0058] This invention constructs a load pattern recognition and trend prediction mechanism based on manifold mapping and pattern comparison learning, and combines it with a dynamic correction of the pattern reference features by a disturbance observer for real-time control error. This enables real-time optimization and adaptive adjustment of hydraulic control target parameters, comprehensively solving the problems of poor adaptability to complex dynamic working conditions and response lag in existing hydraulic control strategies, and significantly improving the dynamic performance and actual control effect of the hydraulic system.

[0059] This invention enables accurate real-time identification of pressure difference direction change patterns caused by cavitation in hydraulic systems and dynamic optimization of control strategies, significantly improving the system's operational stability and control accuracy. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of the structure of the method of the present invention;

[0062] Figure 2 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0063] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0064] Example 1

[0065] Please see Figure 1 This invention provides a hydraulic control optimization system and method based on differential pressure direction change pattern recognition, including:

[0066] S101: Perform phase space topology reconstruction processing based on the pressure signals of the hydraulic cylinder's inlet and outlet chambers, and perform singular spectrum analysis processing based on the topology reconstruction results to determine the topology mapping characteristics of the pressure difference direction.

[0067] It should be noted that the pressure signal is obtained through pressure sensors installed at the oil inlet of the hydraulic cylinder's oil inlet chamber and the oil outlet of the oil return chamber;

[0068] In implementation, the topological mapping characteristics for determining the pressure difference direction include:

[0069] S101.1 Based on the physical characteristics of the influence of cavitation in the hydraulic cylinder on pressure fluctuations, time delay determination processing is performed to obtain the non-uniform time delay characteristics of topology reconstruction;

[0070] It should be noted that the determination of the non-uniform time delay characteristics in this embodiment is based on the physical mechanism generated by the actual cavitation phenomenon inside the hydraulic cylinder, rather than traditional simple statistical analysis or mathematical calculation methods.

[0071] Specifically, when cavitation occurs inside a hydraulic cylinder, the formation and collapse of cavitation will cause pressure signal fluctuations in a local area. The formation and collapse of cavitation essentially correspond to a gas-liquid phase transition process, which is manifested as a sudden drop in pressure when bubbles form and a rapid rise in pressure when bubbles collapse. During this process, the pressure fluctuation transmission has a non-uniform propagation delay along the propagation path inside the hydraulic cylinder.

[0072] For example, when performing non-uniform time delay determination processing, the pressure fluctuation curves of the hydraulic cylinder's inlet and outlet chambers are first monitored. Based on the physical characteristics of the pressure fluctuation amplitude during cavitation formation and resolution (i.e., the typical characteristics of sudden drops and rapid rises in the pressure curve), the location of characteristic points of pressure fluctuation is determined. These characteristic points are used as time delay markers, and the non-uniform time delay characteristics are obtained by calculating the actual time difference between adjacent markers. The specific calculation formula is as follows:

[0073] T i =t i+1 -t i i = 1, 2, ..., n-1,

[0074] In the formula, T i Let t represent the i-th non-uniform time delay interval. i The timestamp represents the corresponding feature point of the i-th acupoint (such as the inflection point or abrupt change point of the pressure curve); n represents the total number of acupoint feature points identified.

[0075] It should be understood that the location t of the aforementioned cavitation feature points i The data obtained through actual hydraulic cylinder pressure monitoring experiments can be determined by those skilled in the art based on the actual working conditions of the hydraulic system using the methods described above.

[0076] Specifically, the generation logic of the non-uniform time-delay features of topology reconstruction is as follows:

[0077] S101.1.1 Based on the amplitude of the pressure signal change in the hydraulic cylinder inlet chamber, the cavitation initiation condition is identified and processed to obtain the cavitation initiation timing characteristics.

[0078] It should be noted that the purpose of the cavitation initiation condition identification and processing described in this embodiment is to accurately capture the timing point at which the cavitation phenomenon begins to occur in the oil inlet chamber of the hydraulic cylinder, so as to obtain the timing characteristics of cavitation initiation in the hydraulic system, thereby providing key physical reference for subsequent pressure fluctuation propagation analysis.

[0079] It is understandable that the pressure signal in the oil inlet chamber of the hydraulic cylinder has obvious physical characteristics when cavitation begins to form, which manifests as a rapid drop in local pressure and a typical amplitude change characteristic of subsequent recovery. This characteristic is different from the normal fluctuation or random noise fluctuation of the pressure signal during normal operation.

[0080] For example, in specific implementation, the timing characteristics of cavitation formation can be determined in the following way:

[0081] A pressure sensor is installed inside the oil inlet chamber of the hydraulic cylinder to monitor the pressure signal in the oil inlet chamber in real time, thereby obtaining a real-time sampling sequence of the pressure signal.

[0082] P in(t), t=1,2,…,N,

[0083] Among them, P in (t) represents the pressure amplitude measured in the oil inlet chamber at time t, and N represents the total number of pressure signal samples;

[0084] Cavitation initiation is calculated and determined based on the rate of change of the pressure signal amplitude in the oil inlet chamber. The calculation formula is as follows:

[0085]

[0086] in, The value represents the rate of change of the inlet chamber pressure signal amplitude with time, dt represents the minute change in time t, i.e., the differential variable used to represent differential operations in mathematical analysis, ε represents the threshold for determining cavitation initiation, and t start Indicates the start timestamp of the characteristic point of cavitation emergence;

[0087] It should be noted that the threshold for determining cavitation initiation was obtained by calculating the pressure change rate based on statistical experimental data from stable operation of the hydraulic cylinder and cavitation generation conditions.

[0088] It should also be noted that the feature point identification is achieved by smoothing and filtering the pressure curve, performing derivative calculations on the curve, and determining the location where the pressure change rate is greater than the cavitation initiation threshold as the feature point.

[0089] Based on the above method, the starting time points of multiple cavitation formations were obtained:

[0090]

[0091] Among them, T cav This represents the obtained set of temporal features of cavitation emergence. This represents the start timestamp of the i-th cavitation event, and m represents the total number of cavitation events identified.

[0092] S101.1.2 Based on the timing characteristics of cavitation formation and the physical process of gas-liquid phase transformation, pressure difference fluctuation transmission path analysis is performed to obtain fluctuation propagation characteristics;

[0093] It should be noted that the pressure fluctuation transmission path analysis in this embodiment is based on the obtained cavitation initiation timing characteristics, combined with the pressure disturbance diffusion mechanism that accompanies the fluid inside the hydraulic cylinder from liquid phase to gas phase conversion, to analyze the actual propagation path characteristics of the fluctuation signal along the inside of the hydraulic cylinder.

[0094] It is understandable that when cavitation occurs in the oil inlet chamber, the fluid rapidly vaporizes locally to form bubbles, causing local pressure disturbances. These disturbances will propagate to other areas of the hydraulic system in the form of pressure waves. The propagation path and propagation law are closely related to the changes in the liquid-gas two-phase interface within the system.

[0095] For example, in specific implementation, pressure difference fluctuation transmission path analysis can be performed in the following manner:

[0096] Using the set of temporal feature points T of cavitation emergence cav Determine the location of the initial disturbance source and its corresponding timestamp for cavitation formation;

[0097] Based on the physical laws governing the gas-liquid phase transformation in hydraulic systems, a propagation model for disturbance waves after bubble generation is established:

[0098] P wave (x,t)=P0·e -αx sin(ωt-kx+φ),

[0099] In the formula, P wave (x,t) represents the amplitude of the pressure fluctuation at a distance x from the cavitation source and time t, P0 represents the initial disturbance pressure amplitude, α represents the damping attenuation coefficient of the pressure wave propagating along the hydraulic pipeline, ω represents the fluctuation frequency, k represents the wave number of the pressure wave propagation, and φ represents the initial phase.

[0100] It should be noted that the specific values ​​of the above parameters P0, α, ω, and φ can be determined based on experimental data from actual hydraulic systems. Those skilled in the art can determine these parameters through conventional experiments using the above model and method. k is calculated using wavelength, and its calculation formula is as follows: Where λ represents wavelength;

[0101] It should be further explained that by analyzing the propagation characteristics of disturbance waves generated by each cavitation initiation event through the above model, the laws governing the changes in wave propagation speed, propagation direction, and wave intensity with spatial location can be clarified, thereby obtaining clear wave propagation characteristics and providing a clear physical basis for subsequent topological time delay characteristic analysis.

[0102] S101.1.3 Based on the wave propagation characteristics, the effective delay interval of the pressure signal is divided into non-uniform intervals to obtain the non-uniform time delay characteristics of topological reconstruction;

[0103] It should be noted that the purpose of the non-uniform interval division process in this embodiment is to determine the specific length of each delay interval based on the actual propagation characteristics of the pressure difference fluctuation in the hydraulic cylinder, so that the delay characteristics during topology reconstruction can objectively reflect the non-uniform characteristics of the pressure wave propagation in the actual hydraulic medium.

[0104] It is understandable that, since the propagation of pressure waves in a hydraulic system is affected by the combined effects of medium properties, pipeline structure, and cavitation disturbance characteristics, the delay caused by pressure fluctuations is not evenly distributed at different locations; traditional equal-interval delay classification cannot accurately characterize such non-uniform delay characteristics.

[0105] For example, in a specific implementation, it can be implemented as follows:

[0106] Based on the obtained wave propagation characteristics, the wave propagation model P is analyzed. wave By performing specific calculations and analyses on the relevant parameters in (x,t), the propagation time delay characteristics of pressure fluctuations at different spatial locations are obtained, and the specific time node sequence of pressure fluctuation signals propagating along the hydraulic pipeline is determined:

[0107]

[0108] Among them, T delay This represents the set of effective delay nodes for the propagation of pressure signal fluctuations. The timestamp represents the delay when the pressure signal propagates to the j-th measurement location, and n represents the number of effective delay nodes obtained from the specific measurement.

[0109] Based on the determined set of effective delay nodes, non-equidistant delay intervals are divided. An example of the division method is as follows:

[0110]

[0111] In the formula, Δτ j This represents the length of the j-th non-equidistant delay interval. and These represent the timestamps of two adjacent pressure signal propagation delay nodes, respectively.

[0112] It should be understood that the length of the aforementioned delay interval Δτ j It is calculated based on the characteristics of wave propagation and objectively reflects the non-uniform characteristics of pressure wave propagation inside the actual hydraulic system.

[0113] It should be further explained that the delay interval Δτ is based on non-equidistant division. j It can objectively and accurately construct the non-uniform time-delay characteristics of the topology reconstruction of the hydraulic system, fully reflecting the physical reality of the propagation of differential pressure fluctuations in the system;

[0114] S101.2 Based on the non-uniform time delay characteristics, pressure signal delay coordinate mapping is performed to obtain the initial topology of the phase space;

[0115] It should be noted that the delayed coordinate mapping method used in this embodiment uses the hydraulic cylinder pressure signal and non-uniform time delay characteristics as the basic input to form a topological space structure with physical meaning.

[0116] It is understandable that the basic principle of the pressure signal delay coordinate mapping processing is: using the time delay interval T determined by the non-uniform time delay characteristics. i The pressure signal collected in real time by the hydraulic cylinder is delayed and reconstructed to obtain the trajectory information mapped in the multi-dimensional space and form the initial topology.

[0117] For example, in a specific implementation, it can be obtained in the following ways:

[0118] First, let the sampling sequence of the real-time pressure signal be:

[0119] p(t), t=1,2,…,M,

[0120] Where p(t) represents the amplitude of the pressure signal collected at time t, and M represents the number of sampling points of the pressure signal;

[0121] Furthermore, based on the non-uniform time delay characteristic T i The resulting delay vector is represented as:

[0122]

[0123] In the formula, X(t) represents the delay vector corresponding to the t-th pressure signal, p(t) represents the amplitude of the pressure signal collected at time t, and T i Let m represent the i-th non-uniform time delay interval, and m represent the dimension of the delay vector.

[0124] It should be noted that the delay vector dimension represents the number of signal delay coordinates selected when performing delay coordinate mapping on the pressure signal, that is, the number of delayed signal data points selected when forming the topological space. In other words, the selection of the delay vector dimension determines the dimension of the phase space after delay coordinate mapping. This dimension directly relates to the complexity of the topology and the richness of the pressure signal characteristic information that the phase space can represent. In practical applications, the delay vector dimension m can be gradually increased or decreased through topology reconstruction experiments, and the obtained phase space topology can be evaluated to select the appropriate dimension m that best reflects the pressure difference direction change pattern of the hydraulic system while maintaining computational efficiency. Typically, in actual engineering implementation, the specific value of the dimension m (e.g., between 4 and 10) is determined by the specific characteristics of the hydraulic system and the engineering application requirements.

[0125] For example, in specific implementation, the determined non-uniform time delay characteristic T can be used as a reference. iThe number of elements and the sensitivity required for topology analysis are determined. If the dimension m is too small, it cannot fully reflect the inherent complex topological features of the hydraulic cylinder pressure signal; while if the dimension m is too large, it may introduce redundant information, leading to a decrease in subsequent computational efficiency or a decrease in the reliability of feature extraction.

[0126] It should be understood that, through the above-mentioned delayed coordinate mapping processing method, the initial topological structure that can characterize the pressure fluctuation inside the hydraulic cylinder under non-uniform time delay conditions can be obtained. Compared with the traditional uniform time delay processing method, it can more realistically reflect the irregular transmission characteristics of pressure fluctuation, thereby accurately capturing the multi-mode characteristics of pressure difference direction changes.

[0127] S101.3 Perform phase space attractor trajectory density analysis based on the initial topology to determine the characteristics of the attractor region;

[0128] It should be noted that the purpose of the above phase space attractor trajectory density analysis is to identify dense regions of pressure signal trajectory distribution in the topology, thereby characterizing the pressure fluctuation aggregation characteristics of the hydraulic system under different load modes.

[0129] Understandably, regions with higher density of trajectory points in the initial topology indicate that the pressure fluctuations in the hydraulic system are more stable in that region, reflecting the typical characteristics of the system under specific load conditions; conversely, regions with dispersed trajectory points indicate that the pressure fluctuations are in an unstable or transitional phase when the operating conditions change or the system is subjected to external disturbances.

[0130] For example, in the specific implementation process, a spatial grid density calculation method can be used to uniformly divide the initial topological structure space, map the topological space to a three-dimensional coordinate system (e.g., X, Y, Z coordinates obtained through delayed coordinate reconstruction), and construct uniform grid cells. The specific calculation method is disclosed as follows:

[0131]

[0132] Where, ρ k N represents the density of trajectory points within the k-th spatial grid cell. k V represents the number of trajectory points contained in the k-th spatial grid cell. k Let K represent the volume of the k-th spatial grid cell, where K represents the total number of spatial grid cells.

[0133] It should be noted that the density ρ of trajectory points within the grid cell is calculated. k This allows us to obtain the spatial distribution characteristics of trajectory density in the initial topology. In this embodiment, grid cell regions with density higher than a set threshold (e.g., density higher than a certain multiple of the overall average density of trajectory points, the specific multiple being determined experimentally) are defined as attractor regions.

[0134] It should also be noted that the determined attractor region features effectively reflect the aggregation characteristics of pressure fluctuations in the topological space, providing important feature support for subsequent singular spectrum feature extraction and identification of hydraulic cylinder pressure differential direction changes;

[0135] S101.4 Singular spectrum features are extracted based on attractor region features to determine the topological mapping features of the pressure difference direction;

[0136] It should be noted that the singular spectrum feature extraction method used in this embodiment aims to extract the dominant features of hydraulic cylinder pressure fluctuations in the topological space, so as to achieve effective identification of pressure difference direction change patterns.

[0137] It is understandable that the so-called singular spectrum feature extraction is to obtain singular values ​​that reflect the primary and secondary relationships of the hydraulic cylinder pressure signal feature distribution by performing singular value decomposition on the set of attractor region trajectories, and use these as topological mapping features.

[0138] For example, the specific implementation process is as follows:

[0139] The trajectory point data of the determined attractor region are constructed into a data matrix:

[0140]

[0141] Where D represents the attractor region data matrix, X i Y i and Z i Let represent the coordinates of the i-th trajectory point in the topological space; n represents the total number of trajectory points in the attractor region.

[0142] The singular value decomposition of matrix D is performed using the following formula:

[0143] D=UΣV T ,

[0144] In the formula, U and V represent the orthogonal matrices obtained from the singular value decomposition, and Σ represents the singular value diagonal matrix, which is specifically expressed as:

[0145]

[0146] Where, σ i Let σi be the singular value satisfying σ1≥σ2≥…≥σi r >0;

[0147] It should be noted that V T The transpose of an orthogonal matrix V is obtained by transposing the original matrix V, which is to change the row elements of the original matrix V into column elements and the column elements into row elements.

[0148] It should be understood that the above singular value σ iThe magnitude of the singular values ​​represents the primary and secondary relationships of the topological features of pressure fluctuations; larger singular values ​​correspond to features that have a more significant impact on changes in the direction of pressure difference. In this embodiment, the first p singular values ​​(e.g., the first three singular values) are selected as the topological mapping features for determining the direction of pressure difference. The specific values ​​can be determined through actual experiments.

[0149] It should be further noted that the topological mapping features determined by this singular spectrum feature extraction method can comprehensively and accurately characterize the main feature information of the differential pressure direction change mode of the hydraulic system, thereby providing accurate and reliable data support for the optimization of hydraulic control parameters.

[0150] S102: Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction.

[0151] Specifically, the model reference characteristics for determining the direction of pressure difference include:

[0152] S102.1 Based on the local sensitive regions of the topological mapping features, scale hierarchical selection processing is performed to determine the hierarchical scale of multi-scale convolution;

[0153] It should be noted that the purpose of the scale layering selection process in this embodiment is to reasonably layer the scale of the convolutional kernel in the convolutional network feature extraction process based on the importance of local sensitive regions in the topological mapping feature space, so as to improve the representativeness of the pattern baseline features.

[0154] It is understandable that different regions in the topological mapping feature space have different sensitivities in reflecting the differential pressure change of the hydraulic cylinder. Therefore, it is necessary to select the scale layer to highlight the characteristic differences of key regions in a targeted manner.

[0155] It should also be noted that the convolutional kernel scales used in this embodiment are small scale (3×3), medium scale (5×5) and large scale (7×7), and the number of convolutional kernels at each scale is set to [16, 32 and 64]. The multi-scale convolutional network structure adopts three convolutional layers, and each convolutional layer is followed by max pooling operation for feature dimensionality reduction. The final output feature dimension is 128.

[0156] For example, in specific implementation, the hierarchical scale of multi-scale convolution can be determined in the following way:

[0157] Based on the topological mapping characteristics, the feature space is divided into multiple local regions:

[0158] R = {R1, R2, ..., R} i ,…,R k},

[0159] Each region R i Represents a specific set of locations in a topological mapping;

[0160] According to the local region R i Calculation of regional sensitivity index S(R) based on statistical characteristics of topological feature changes within the region i Example calculation formula is as follows:

[0161]

[0162] In the formula, S(R) i ) represents the sensitivity of the i-th local region. Represents the gradient or rate of change of a topological feature at spatial location x; |R i | Represents region R i The number of spatial points;

[0163] Based on the aforementioned regional sensitivity indicators, scale stratification was performed for all regions.

[0164] If the sensitivity index is greater than the maximum value of the preset sensitivity stratification threshold range, then a small-scale convolution kernel is selected;

[0165] If the sensitivity index falls within the preset sensitivity stratification threshold range, then select a medium-scale convolution kernel;

[0166] If the sensitivity index is less than the minimum value of the preset sensitivity stratification threshold range, then a large-scale convolution kernel is selected;

[0167] It should be noted that the preset sensitivity stratification threshold range is determined by those skilled in the art based on actual needs;

[0168] Therefore, a set of hierarchical scales for multi-scale convolution applicable to different sensitive regions was determined, forming a hierarchical scale set:

[0169] Scale={scale(R1),scale(R2),…,scale(R i ),…,scale(R k )},

[0170] In the formula, scale(R) i ) represents the kernel scale selected for the i-th region;

[0171] S102.2 Based on the hierarchical scale, the topological mapping features are processed by convolution kernel local spatial feature extraction to obtain convolution local features;

[0172] It should be noted that the local spatial feature extraction process of the convolution kernel in this embodiment aims to extract spatial features of the topological mapping features based on the multi-scale hierarchical scale set determined in step S102.1, so as to fully capture the feature differences of different regions and obtain reliable local convolution features.

[0173] It is understandable that the spatial local characteristics of the topological mapping feature space distribution vary significantly at different scales. Therefore, by using convolution kernels of different scales to extract local features, the essential characteristics of pressure difference changes can be objectively reflected.

[0174] For example, in specific implementation, the extraction of local features from convolution can be achieved in the following way:

[0175] A multi-scale convolution processing method is defined, which involves performing convolution operations at different scales on the topological mapping feature space T(x,y). An example of the convolution operation formula is as follows:

[0176] F scale (x,y)=∑ u ∑ v T(xu,yv)·K scale (u,v),

[0177] In the formula, F scale (x,y) represents the convolutional local features extracted at spatial location (x,y) using a convolutional kernel of scale , T(xu,yv) represents the original feature values ​​in the topologically mapped feature space, and K scale (u,v) represents the convolution kernel of the selected scale;

[0178] For different regions, corresponding convolution kernel scales are used to extract local convolution features, forming a set of local convolution features:

[0179]

[0180] In the formula, This represents the convolutional local features extracted from the i-th region using a convolutional kernel of the corresponding scale;

[0181] It should be understood that, through the above-mentioned local spatial feature extraction processing of multi-scale convolution kernels, the obtained set of convolutional local features can effectively reflect the local spatial differences in pressure difference changes in the hydraulic system, thereby providing sufficient and reliable data support for subsequent spatial correlation weight analysis;

[0182] S102.3 Based on the degree of correlation between the spatial location of the local features of convolution and the changes in working conditions, spatial location correlation analysis is performed to obtain spatial correlation weight features;

[0183] It should be noted that the purpose of the spatial location correlation analysis in this embodiment is to determine the sensitivity of each spatial location in the local features of the convolution to the changes in different operating conditions of the hydraulic system, so as to objectively and clearly obtain the spatial correlation weight features corresponding to different regions, thereby improving the operating condition sensitivity of the model reference features.

[0184] It is understandable that, since the local features of different spatial locations respond differently to changes in operating conditions, spatial location correlation analysis is required to clarify the weight distribution characteristics of each region when operating conditions change.

[0185] The logic for generating spatial correlation weight features is as follows:

[0186] S102.3.1 Identify strongly correlated regions and determine initial sensitive regions based on the spatial distribution characteristics of local convolutional features;

[0187] It should be noted that this embodiment first needs to identify which local spatial locations have significantly different characteristics under different operating conditions, so as to use them as initial sensitive areas for subsequent detailed analysis;

[0188] For example, this embodiment uses local feature distribution entropy to objectively identify strongly correlated regions;

[0189] Specifically, calculate the distribution entropy value for each local region:

[0190]

[0191] In the formula, LFDE(R) i Let p represent the distribution entropy of the i-th local region. ij Let represent the probability that the feature value in the i-th local region is located in the j-th feature interval, and Q represent the number of feature value intervals.

[0192] It should be understood that LFDE(R) i p is used to characterize the degree of difference in feature distribution. ij The frequency of characteristic values ​​under different working conditions was determined;

[0193] Based on the distribution entropy of each region, the regions are sorted. Regions with distribution entropy greater than or equal to a preset distribution entropy threshold are identified as regions with significant feature changes and high sensitivity, and are defined as the initial sensitive region set.

[0194] It should be noted that the distribution entropy threshold is set by experimentally determined data and determined by those skilled in the art based on actual needs;

[0195] S102.3.2 Based on the initial sensitive area, perform local operating condition difference identification processing to determine the regional difference characteristics;

[0196] It should be noted that this step aims to clarify the degree of difference between features within the initial sensitive region under different hydraulic system operating conditions, so as to provide an objective basis for subsequent association weight allocation;

[0197] For example, the specific method for identifying and processing local operating condition differences is disclosed as follows:

[0198] Based on the initial set of sensitive regions, extract the convolutional local feature vectors corresponding to each sensitive region under different operating conditions;

[0199] Calculate the Euclidean distance between the feature vectors of the same region under all operating conditions to obtain the regional dissimilarity feature. The specific calculation formula is as follows:

[0200]

[0201] In the formula, D(R) sens This indicates the regional dissimilarity characteristics of the initial sensitive area. Let L represent the local feature value of the k-th position in the initial sensitive region under the i-th working condition, L represent the dimension of the feature values ​​contained in the region, and M represent the total number of hydraulic system working conditions involved.

[0202] S102.3.3 Based on the regional difference characteristics, the association weights are processed by non-uniform distribution to obtain spatial association weight characteristics;

[0203] It should be noted that the purpose of this step is to reasonably and non-uniformly distribute the correlation weights of each initial sensitive region based on the regional difference characteristics clearly calculated in step S102.3.2, so as to obtain spatial correlation weight characteristics that can clearly reflect the differences in the sensitivity of each region to changes in operating conditions.

[0204] For example, the implementation process of the specific non-uniform distribution of association weights is disclosed as follows:

[0205] Based on the regional disparity characteristics of each initial sensitive region, an initial weight mapping function is defined. The initial weight allocation value for each region is determined based on the relative magnitude of the regional disparity characteristics. The specific calculation formula is as follows:

[0206]

[0207] In the formula, This represents the initial weight value of the i-th initial sensitive region. N represents the regional dissimilarity feature of the i-th initial sensitive region. s This represents the total number of initial sensitive regions;

[0208] For each spatial location point (x, y) within each initial sensitive region, a local sensitivity index for changes in local spatial characteristics is defined, as exemplarily defined below:

[0209]

[0210] In the formula, S local (x,y) represents the local sensitivity index of the spatial location (x,y), σ F (x, y) represents the standard deviation of spatial location (x, y) under different operating conditions, μ F (x, y) represents the characteristic mean of spatial location (x, y) under different working conditions;

[0211] Furthermore, based on the initial weight values ​​and local sensitivity indicators of the region, a non-uniform distribution of weights is performed on the spatial location points within the region to obtain the final spatial association weight features. A specific calculation formula example is shown below:

[0212]

[0213] In the formula, W(x, y) represents the final association weight feature value of the spatial location (x, y). S represents the initial weight value of the i-th sensitive region. local (x, y) represents the local sensitivity index of the spatial location (x, y);

[0214] It should be noted that, in the above formula, This represents the sum of local sensitivity indices for all locations within the region, used to achieve a relatively non-uniform distribution of weights;

[0215] S102.4 Attention fusion processing is performed based on convolutional local features and spatial correlation weight features to determine the pattern baseline features of the pressure difference direction;

[0216] It should be noted that this step aims to fuse the convolutional local features and spatial correlation weight features obtained in the previous steps through an attention mechanism, thereby objectively highlighting the differences in the contribution of different spatial location features to the identification of the system pressure difference direction, and finally clearly obtaining the pattern benchmark features of the pressure difference direction.

[0217] For example, this embodiment uses a weighted attention fusion approach for specific implementation, and the specific implementation process is as follows:

[0218] Based on the set of local feature vectors from the convolution and the corresponding spatial correlation weight feature values, an initial spatial feature weighted fusion calculation is performed, as follows:

[0219]

[0220] In the formula, F weighted Ω represents the spatial features obtained after initial weighted fusion. sens F(x, y) represents the set of spatial location points within the sensitive region, F(x, y) represents the convolutional local feature vector of spatial location (x, y), and W(x, y) represents the spatial correlation weight feature value of spatial location (x, y).

[0221] To further highlight the differences in importance of different convolutional feature dimensions, this embodiment introduces a fusion mechanism based on channel attention, the specific calculation method of which is as follows:

[0222] A c =σ(W2·ReLU(W1·F) weighted )),

[0223] In the formula, A c The channel attention weight features are represented, with values ​​between (0,1). W1 and W2 represent the coefficient matrices for calculating the channel attention weights, and σ represents the standard Sigmoid function.

[0224] It should be noted that the dimensions of the coefficient matrices W1 and W2 are d×d / 2 and d / 2×d, respectively, where d is the dimension of the eigenvectors, for example, d=20. The specific dimension of the eigenvectors shall be set by those skilled in the art according to the actual situation.

[0225] Based on the channel attention weight characteristics, the baseline characteristics of the differential pressure direction pattern are obtained, and the calculation formula is as follows:

[0226] F final =A c ⊙F weighted ,

[0227] In the formula, F final This represents the baseline feature of the differential pressure direction mode, and ⊙ represents the multiplication operation of the corresponding dimension element;

[0228] S103: Construct a manifold mapping space based on the model reference characteristics, and perform manifold distance measurement processing based on the topological mapping characteristics of the real-time differential pressure signal to determine the load mode in the current differential pressure direction;

[0229] Specifically, determining the load pattern in the current pressure differential direction includes:

[0230] S103.1 Based on the load condition transition characteristics of the mode reference features, perform manifold space construction processing to determine the load mode manifold characteristics;

[0231] It should be noted that the main purpose of this step is to construct a manifold space that can intuitively reflect the changing law of load conditions based on the pattern reference characteristics obtained in the previous steps, so as to form the manifold characteristics of load patterns and objectively describe the transformation relationship between different load patterns.

[0232] For example, this embodiment specifically implements the construction of the above features through a manifold learning method, and the specific disclosure is as follows:

[0233] The baseline feature set of patterns is used as the data input for manifold learning. The load pattern manifold space is constructed based on the manifold mapping method—Local Linear Embedding (LLE) algorithm. The specific manifold mapping process is as follows:

[0234] For each model baseline feature, a local reconstruction weight matrix is ​​constructed based on its linear relationship with the k nearest neighbor model baseline features. The specific calculation method is as follows:

[0235]

[0236] in, Representation of features Neighborhood features The reconstruction weights, N(i,k) represent the features. The set of k nearest neighbors in the feature space, where k represents the preset local neighborhood size.

[0237] It should be noted that the preset local neighborhood size is set by those skilled in the art based on the actual situation;

[0238] Furthermore, based on the local reconstruction weight matrix, a set of low-dimensional embedded feature vectors in the manifold space is constructed, and this set is used as the load pattern manifold feature.

[0239] S103.2 Based on the topological mapping characteristics of the real-time differential pressure signal, perform local feature extraction processing to determine the real-time topological local features;

[0240] It should be noted that this step mainly involves extracting local features of a specific region from the topology mapping features corresponding to the currently acquired differential pressure signal in real time, in order to form real-time topology local features for subsequent load pattern recognition.

[0241] For example, in this embodiment, the specific method for extracting real-time topological local features is as follows:

[0242] The topological mapping characteristics of the real-time differential pressure signal are obtained, specifically defined as follows:

[0243]

[0244] Among them, T real (t) represents the sequence of topological mapping features measured under real-time operating conditions. Let represent the j-th feature component in the topological mapping space, and m represent the feature dimension of the topological mapping space;

[0245] Furthermore, to clearly capture significant changes in topological mapping features in specific local regions, the embodiment employs Local Principal Component Analysis (LPCA) to extract local features. The specific implementation method is as follows:

[0246] In the topological mapping feature sequence T real Local neighborhood of (t) Within this framework, for example, a data window of length L (e.g., L = 50) is selected to construct a local data matrix:

[0247]

[0248] In the formula, T local This represents the local feature data matrix of the current real-time operating condition, where t0 represents the timestamp corresponding to the current real-time operating condition.

[0249] It should be noted that each row in the current real-time operating condition local feature data matrix represents the feature value of the feature sequence at the corresponding time, the number of columns corresponds to the feature dimension m, and the number of rows is the window length L;

[0250] The current real-time operating condition local feature data matrix is ​​subjected to feature decomposition to extract the principal feature component vector. The specific implementation process is as follows:

[0251] Based on data matrix T local The covariance is calculated to obtain the covariance matrix, and the calculation formula is as follows:

[0252]

[0253] In the formula, C local Represents the covariance matrix. Represents matrix T local The mean L of the column vector represents the length of the data window.

[0254] The covariance matrix is ​​decomposed into eigenvalues ​​using the following formula:

[0255] C local =EΛE T ,

[0256] Where E represents the eigenvector matrix corresponding to the covariance matrix, and Λ represents the diagonal matrix formed by the eigenvalues;

[0257] For example, in this embodiment, the first p principal feature vectors are selected to form a real-time topological local feature vector set F. local :

[0258] F local = [e1, e2, ..., e p ],

[0259] It should be noted that e1, e2, ..., e p This represents an eigenvector ordered from largest to smallest eigenvalue. The specific value of p is determined by those skilled in the art based on actual needs, for example, p = 3.

[0260] S103.3 Based on the local mapping distribution characteristics of real-time topological local features in the manifold space, perform feature mapping similarity measurement processing to determine manifold mapping distance features;

[0261] It should be noted that, based on the real-time topological local features obtained from the aforementioned steps, the similarity of their mapping to the features of different load modes in the manifold space is measured, thereby objectively determining the manifold mapping distance features of the real-time operating conditions corresponding to each load mode.

[0262] For example, in this embodiment, the specific implementation of the similarity measurement between real-time topology local features and load pattern manifold features is as follows:

[0263] Based on the set of local feature vectors of the real-time topology and the set of feature vectors of the load pattern manifold, a weighted distance measurement method based on local mapping similarity is used to obtain the manifold mapping distance between the local features of the real-time topology and the feature vectors of the load pattern manifold. The specific calculation formula is as follows:

[0264]

[0265] In the formula, D map (i) represents the manifold mapping distance between the real-time topology local feature and the i-th load pattern manifold feature, e j This represents the j-th principal component feature vector of the real-time topological local features. Let y represent the eigenvector of the i-th load pattern manifold. i The corresponding mapping feature component in the j-th dimension, α j Indicates the mapping distance weighting coefficient;

[0266] It should be noted that, in the formula, the mapping distance weight coefficient is used to reflect the difference in the contribution of different feature dimensions to load pattern recognition. It is obtained by calculating the proportion of the feature value of the real-time topology local feature in the corresponding dimension to the total dimension value. It is calculated using Euclidean distance, which will not be elaborated on here;

[0267] After the above similarity measurement process, the set of manifold mapping distance features for each load mode under real-time operating conditions is obtained;

[0268] S103.4 Perform optimal matching of load patterns based on manifold mapping distance characteristics to determine the load pattern in the current pressure difference direction;

[0269] It should be noted that, based on the manifold mapping distance characteristics obtained in the previous step, the specific load mode corresponding to the current hydraulic system differential pressure direction is determined, thereby providing a clear mode input basis for the subsequent hydraulic control strategy;

[0270] Specifically, the implementation details are disclosed as follows:

[0271] Based on the manifold mapping distance set corresponding to each predefined load mode in real-time differential pressure direction, the minimum value of the distance set is filtered to obtain the preliminary mode matching result, namely the initial load mode index and the corresponding manifold mapping distance value.

[0272] Understandably, the above process is implemented using the conventional minimum value search method, that is, selecting the element with the smallest value in the set and its corresponding index.

[0273] It should be understood that although the above distance values ​​determine the initial pattern, feature anomalies or measurement errors may occur in the actual system, resulting in a decrease in the reliability of the matching pattern. Therefore, further processing to determine the validity of the pattern matching is required. The specific process is as follows:

[0274] The minimum manifold mapping distance value obtained from the initial pattern matching is compared with a preset distance threshold to determine the validity of the initial pattern matching result;

[0275] If the minimum distance value is less than the preset threshold, the preliminary pattern matching result is considered valid.

[0276] If the minimum distance value is greater than or equal to the preset threshold, the preliminary pattern matching result is deemed invalid.

[0277] It should be noted that the preset threshold was obtained through a series of experiments.

[0278] S104: Construct a pattern comparison learning mechanism based on the current load mode and historical load modes, and perform pattern trend prediction processing based on the pattern comparison learning results to determine the hydraulic control target parameters;

[0279] It should be noted that the above-mentioned model comparison learning mechanism is constructed based on the historical operating condition data of the hydraulic system. Specifically, by analyzing the working condition transfer pattern between the current load mode and the historical load mode, the inherent connection and evolution process between different load modes are explored to clarify the specific trend of load change and thus provide an objective decision-making basis for determining the hydraulic control target parameters.

[0280] Specifically, the target parameters for hydraulic control are determined, including:

[0281] S104.1 Based on the working condition transfer characteristics of the current load mode and the historical load mode, perform load mode transfer path identification processing to determine the mode transfer path characteristics;

[0282] It should be noted that the characteristics of the load mode transfer path are determined based on the transition patterns between load modes in adjacent control cycles during historical operation.

[0283] For example, the continuous load pattern corresponding to the historical operating data of the hydraulic system is obtained;

[0284] The above historical load pattern sequences are combined in pairs to form working condition transition path samples. For example, the historical pattern sequence pattern 1→pattern 2→pattern 3 is decomposed into pattern transition samples of pattern 1→pattern 2 and pattern 2→pattern 3.

[0285] Based on the frequency of occurrence of each transfer path in historical data, the transfer frequency characteristics of each transfer path pattern are obtained.

[0286] Paths whose transfer frequency characteristics exceed the experimentally determined threshold (e.g., accounting for 5% of the total transfer paths) are considered as typical pattern transfer path characteristics.

[0287] S104.2 Based on the characteristics of the pattern transfer path, perform comparative learning sample selection processing to determine typical pattern comparison samples;

[0288] It should be noted that the quantitative standard for selecting typical mode comparison samples is: working condition data with pressure amplitude changes greater than [e.g. 20%] and duration not less than [e.g. 0.5s] are used as typical mode comparison samples;

[0289] For example, based on the typical pattern transition paths in the pattern transition path features, the starting mode and ending mode corresponding to each typical pattern transition path are identified.

[0290] Objective statistical processing of the changes in operating amplitude is performed on the historical operating condition data corresponding to the identified start and end modes.

[0291] It should be noted that the statistical methods mentioned include, but are not limited to, mean calculation, standard deviation calculation, etc.

[0292] Select the start-end pattern pairs with significant amplitude changes and obvious trends as typical pattern comparison samples;

[0293] It should be noted that significant changes and obvious trends in amplitude are determined by comparing with a preset amplitude threshold. When the amplitude is greater than or equal to the preset amplitude threshold, it is judged as significant change and obvious trend. The amplitude threshold is set by technical personnel according to actual needs.

[0294] S104.3 Based on the sensitivity characteristics of working condition changes of typical model comparison samples, perform model trend difference prediction processing to determine the predictive characteristics of working condition changes.

[0295] It should be noted that, from the typical pattern comparison sample, pattern samples that are sensitive to rapid changes in load conditions are identified, and differentiated trend prediction analysis is carried out accordingly.

[0296] The generation logic for the operating condition change prediction features is as follows:

[0297] S104.3.1 Based on the rapid change characteristics of the load conditions of the typical pattern comparison samples, perform rapid change sample identification processing to determine the rapidly changing sensitive samples;

[0298] It should be noted that this step is specifically designed to identify load samples where the rate of change of operating conditions during mode transition is much higher than the normal rate of change.

[0299] In practice, the handling method is as follows:

[0300] The rate of change of hydraulic system pressure signal at adjacent time points in the typical mode comparison sample is calculated using the following formula:

[0301]

[0302] In the formula, P(t+Δt) represents the rate of change of the load pressure signal, where P(t+Δt) represents the pressure value at time t+Δt and Δt represents the sampling period.

[0303] Based on the calculated rate of change, load pattern samples whose absolute value of the rate of change is significantly higher than the experimentally determined threshold (e.g., more than 3 times the normal rate of change under load conditions) are clearly identified as rapidly changing sensitive samples.

[0304] S104.3.2 Based on rapidly changing sensitive samples, identify the trend and direction of working condition changes to determine the characteristics of the working condition change direction;

[0305] It should be noted that the above-mentioned trend direction identification of the load conditions of the identified rapidly changing sensitive samples is to clearly identify whether the load conditions are in an "upward trend", "downward trend" or "stable trend" to ensure the objectivity and effectiveness of the model trend prediction.

[0306] For example, the processing logic disclosed in this embodiment is as follows:

[0307] Based on the local extreme points (inflection points) of the load pressure change curve in rapidly changing sensitive samples, the temporal location and pressure amplitude of the local extreme points are extracted;

[0308] It should be noted that local extrema are determined by the intersection of the first derivative (slope) when it changes from positive to negative or from negative to positive.

[0309] The trend direction is identified by the difference in pressure change amplitude between two adjacent extreme points and the corresponding time difference, and a trend identification factor is obtained.

[0310] Specifically, based on two adjacent local extreme points, the difference in pressure amplitude between the two extreme points is first determined, which is the result of subtracting the pressure of the previous extreme point from the pressure of the next extreme point; then the time difference between the two extreme points is determined, which is the time of the next extreme point from the time of the previous extreme point; finally, the pressure difference is divided by the corresponding time difference to obtain the rate of pressure change over time, which is used to identify the direction of load trend change.

[0311] Based on trend identification factors, determine the directional characteristics of changes in operating conditions;

[0312] If the trend identification factor is greater than the preset positive trend threshold, the change in operating conditions is determined to be an "upward trend".

[0313] If the trend identification factor is less than the preset negative trend threshold, the change in operating condition is determined to be a "downward trend".

[0314] If the trend identification factor falls between the preset positive trend threshold and the preset negative trend threshold, then the change in operating conditions is determined to be a "stable trend".

[0315] Specifically, when the rate of change of pressure over time is positive and the value is greater than the preset positive trend threshold, it indicates that the pressure is increasing rapidly per unit time, and the current load condition can be judged as being in a significant "upward trend".

[0316] When the rate of change of pressure over time is negative and the value is less than the preset negative trend threshold, it indicates that the pressure is decreasing rapidly per unit time, and the current load condition can be clearly determined to be in a clear "downward trend".

[0317] If the rate of change of pressure over time is small and falls between the positive and negative thresholds, it indicates that the pressure change per unit time is not significant, and the current load condition can be clearly determined to be in a "stable trend".

[0318] S104.3.3 Based on the characteristics of the direction of change of working conditions, perform local differential prediction processing of working condition trends to obtain the predictive features of working condition changes;

[0319] It should be noted that, based on the characteristics of the direction of change of operating conditions (i.e., "upward trend", "downward trend" or "stable trend"), the changes of operating conditions under different trend directions are predicted differently in order to obtain accurate predictive features of operating conditions.

[0320] Specifically, the implementation logic is as follows:

[0321] Based on historical load condition operation data, a sample library for predicting load condition changes is constructed, corresponding to three modes: upward trend, downward trend, and stable trend. Each sample library includes several typical load condition prediction samples.

[0322] Based on the characteristics of the current operating condition change direction, a sample library that is consistent with the current trend direction is selected from the operating condition change prediction sample library of the above three mode categories as the basis for this prediction processing.

[0323] The prediction process is as follows:

[0324] The real-time operating data of the current load mode is compared with the typical prediction samples in the selected sample library one by one to perform local difference comparison processing, that is, to explicitly calculate the local differences between the current real-time data and the typical prediction sample data one by one.

[0325] It should be noted that the above-mentioned local difference is calculated as a point-by-point numerical difference calculation, that is, the absolute value of the difference between the corresponding pressure values ​​is calculated for each sampling point, and then the absolute value of the difference is integrated or averaged to obtain the overall local difference value between the current real-time data and the typical prediction sample.

[0326] Based on the local difference values, select one or more typical prediction samples with the smallest local difference values ​​as the basis for predicting the current operating condition change trend.

[0327] The trend evolution analysis of subsequent operating condition changes is carried out based on the typical prediction sample with the smallest local difference value, so as to obtain accurate predictive features of operating condition changes.

[0328] S104.4 Based on the predicted characteristics of working condition changes, hydraulic control target parameters are selected and determined.

[0329] It is understandable that a hydraulic control parameter library has been established, which includes multiple sets of control parameter combinations that are clearly associated with the predictive characteristics of different working conditions. Each set of control parameter combinations includes, but is not limited to, specific hydraulic control target parameters such as target pressure setpoint, target flow setpoint, valve opening reference value, and pump flow distribution ratio.

[0330] For example, the specific method for selecting target parameters is as follows:

[0331] Based on the predicted characteristics of working condition changes, calculate the similarity between the current predicted characteristics and the historical working condition characteristics corresponding to each set of control parameters in the hydraulic control parameter library;

[0332] It should be noted that the similarity calculation includes, but is not limited to, Euclidean distance, cosine similarity, etc.

[0333] Based on the above similarity calculation results, the combination of hydraulic control parameters corresponding to the historical working condition features with the highest similarity (i.e., the value closest to the current predicted features) is used as the initial selection of the current hydraulic control target parameters.

[0334] Based on the specific operating conditions of the current load mode and the actual operating constraints of the hydraulic system (e.g., system maximum allowable pressure, flow limit, actuator displacement speed limit, etc.), the above-mentioned initially selected combination of control parameters is objectively modified to ensure that the determined hydraulic control target parameters can meet the actual control requirements.

[0335] The modified hydraulic control target parameter combination is used in the subsequent hydraulic system control implementation process, including but not limited to the specific control of pumps or valves, to effectively complete the precise position, speed or force control of hydraulic actuators;

[0336] It should be clarified that the specific parameter values ​​in the above hydraulic control parameter library vary depending on the specific working conditions, load type, and hydraulic system structure. The specific values ​​are determined through extensive actual working condition experiments and analysis of historical operating data.

[0337] S105: The interference observer processes the interference based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updates the mode reference characteristics based on the observer output.

[0338] Specifically, the disturbance observer is processed based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and the mode reference characteristics are updated based on the observer output, including:

[0339] S105.1 Based on the time-domain variation characteristics of the control error, dynamic identification of interference sources is performed to determine the characteristics of the interference sources;

[0340] For example, the control error signal between the hydraulic cylinder piston rod displacement feedback and the target displacement is collected in real time;

[0341] The sampling frequency is set to 1000Hz to 5000Hz to fully capture high-frequency fluctuations in the error.

[0342] The time-frequency distribution result is obtained by performing a short-time Fourier transform on the control error signal.

[0343] Based on the time-frequency distribution results, the spectral energy distribution characteristics of the control error signal over time are obtained, and the energy distribution of different frequency bands is analyzed to determine the specific source of interference.

[0344] If the spectral energy increases significantly in the high-frequency band (e.g., frequency greater than 50Hz), it is clearly determined to be external interference (such as caused by rapid load fluctuations).

[0345] If the spectral energy increases significantly in the low-frequency range (e.g., below 20Hz), it is clearly determined to be internal interference (such as valve core hysteresis or pipeline resonance).

[0346] Furthermore, statistical features quantify the characteristics of interference sources. For example, within each time window, the energy ratio of high and low frequency bands (i.e., the proportion of total energy in that frequency band to the total energy of the error signal) is specifically calculated, and this energy ratio is used as the quantified interference source characteristics.

[0347] S105.2 Based on the characteristics of the interference source and the load response characteristics of the hydraulic actuator, the interference compensation amount is predicted to obtain the interference compensation characteristics.

[0348] It should be noted that, in order to predict the compensation amount after the hydraulic actuator is disturbed, this embodiment adopts a recursive least squares prediction method based on load response characteristics.

[0349] For example, based on the characteristics of the interference source, an interference compensation feature prediction model is established using historical interference feature data and hydraulic actuator displacement feedback data. Its structure is as follows:

[0350] y(k)=θ T (k-1)X(k)+e(k),

[0351] In the formula, y(k) represents the predicted value of the disturbance compensation feature required by the hydraulic actuator at time k, X(k) represents the disturbance source feature vector at time k, and θ T (k-1) represents the parameter vector of the prediction model at time k-1, and e(k) represents the prediction error;

[0352] Specifically, to recursively update the prediction model parameters θ(k), this implementation uses the RLS algorithm, and the specific steps are as follows:

[0353] Initialize the parameter vector θ(0) and the covariance matrix P(0). The initial value θ(0) can be set as the zero vector, and the initial covariance matrix P(0) can be set as a large numerical matrix, such as 10. 4 I, to reflect the uncertainty of the initial forecast;

[0354] At each time k, the parameter vector is recursively updated using the following formula:

[0355]

[0356] θ(k)=θ(k-1)+K(k)[y m (k)-X T(k)θ(k-1)]

[0357]

[0358] In the formula, K(k) represents the Kalman gain matrix, y m (k) represents the actual measured compensation requirement value between the feedback displacement of the piston rod of the k-th hydraulic cylinder and the target displacement of the hydraulic control, and λ represents the forgetting factor, which specifically takes a value between 0.95 and 0.99.

[0359] It should be noted that the Kalman gain matrix is ​​used to determine the parameter correction magnitude, and the forgetting factor is used to gradually reduce the influence of historical data on the current parameter estimation, so as to track the changing characteristics of the hydraulic system in real time.

[0360] It should be understood that the RLS algorithm described above enables real-time dynamic correction of the parameters of the interference compensation feature prediction model, thereby clearly obtaining the interference compensation features at each moment, which serve as the basis for subsequent update of the model baseline features.

[0361] S105.3 Update and fuse the interference compensation features and the mode reference features to update the mode reference features;

[0362] It should be noted that, in order to achieve the above-mentioned fusion update process, this embodiment exemplarily adopts a weighted fusion mechanism based on a sliding time window to perform real-time updates of the pattern baseline features.

[0363] For example, let the current control error amplitude be e(t), set its corresponding absolute error amplitude to |e(t)|, and set the error threshold e. max and e min ;

[0364] It should be noted that the error threshold e max and e min Determined based on actual hydraulic control precision requirements;

[0365] When the error magnitude |e(t)|≥e max When this is done, the weight α(t) is explicitly set to a larger value, for example, 0.8 to 1, so that the updated model reference features are more in line with the current real-time interference compensation requirements;

[0366] When the error magnitude |e(t)|≤e min When this is the case, the weight α(t) is explicitly set to a smaller value, for example, 0 to 0.2, to ensure the stability of the model baseline characteristics and avoid frequent fluctuations due to small disturbances;

[0367] When the error magnitude e min <|e(t)| <e max At that time, the weight α(t) is determined by linear interpolation.

[0368] It should be understood that, through the aforementioned weight fusion mechanism based on dynamic adjustment of real-time control error amplitude, the real-time and accurate updating of the mode reference characteristics is achieved, ensuring that the hydraulic control strategy continuously and effectively adapts to changes in actual load conditions, thereby improving the overall control performance of the system.

[0369] Example 2

[0370] like Figure 2 As shown, this embodiment discloses a hydraulic control optimization system based on differential pressure direction change pattern recognition, including:

[0371] Differential pressure topology feature extraction module: Performs phase space topology reconstruction processing based on the pressure signals of the hydraulic cylinder inlet and outlet oil chambers, and performs singular spectrum analysis processing based on the topology reconstruction results to determine the topology mapping features of the differential pressure direction;

[0372] Multi-scale feature fusion module: Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction;

[0373] Load pattern recognition module: Constructs a manifold mapping space based on pattern reference features, and performs manifold distance measurement processing based on the topological mapping features of real-time differential pressure signals to determine the load pattern in the current differential pressure direction;

[0374] Load trend prediction module: Constructs a pattern comparison learning mechanism based on the current load pattern and historical load patterns, and performs pattern trend prediction processing based on the pattern comparison learning results to determine the hydraulic control target parameters;

[0375] Interference Observation and Dynamic Correction Module: The interference observer processes the interference based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updates the mode reference characteristics based on the observer output;

[0376] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0377] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0378] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0379] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0380] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0381] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A hydraulic control optimization method based on differential pressure direction change pattern recognition, characterized in that, include: The phase space topology reconstruction is performed based on the pressure signals of the hydraulic cylinder's inlet and outlet chambers, and the singular spectrum analysis is performed based on the topology reconstruction results to determine the topological mapping characteristics of the pressure difference direction. The topological mapping features for determining the pressure difference direction include: Based on the physical characteristics of the influence of cavitation in hydraulic cylinders on pressure fluctuations, time delay determination processing is performed to obtain the non-uniform time delay characteristics of topological reconstruction. Pressure signal delay coordinate mapping is performed based on non-uniform time delay characteristics to obtain the initial topology of phase space; Phase space attractor trajectory density analysis is performed based on the initial topology to determine the characteristics of the attractor region; Singular spectral features are extracted based on attractor region characteristics to determine the topological mapping features of the pressure difference direction; Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction. Based on the model reference characteristics, a manifold mapping space is constructed, and based on the topological mapping characteristics of the real-time differential pressure signal, manifold distance measurement processing is performed to determine the load mode in the current differential pressure direction. A pattern comparison and learning mechanism is constructed based on the current load pattern and historical load patterns, and the pattern trend prediction processing is performed based on the pattern comparison and learning results to determine the target parameters of hydraulic control. The disturbance observer is processed based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and the mode reference characteristics are updated based on the observer output.

2. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized in that, The generation logic for the non-uniform time-delay features of the topology reconstruction includes: Based on the amplitude of the pressure signal change in the hydraulic cylinder inlet chamber, the cavitation initiation condition is identified and processed to obtain the cavitation initiation timing characteristics. Based on the timing characteristics of cavitation formation and the physical process of gas-liquid phase transformation, the pressure difference fluctuation transmission path is analyzed to obtain the fluctuation propagation characteristics; Based on the wave propagation characteristics, the effective delay interval of the pressure signal is divided into non-uniform intervals to obtain the non-uniform time delay characteristics of topological reconstruction.

3. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized in that, The model reference features for determining the direction of pressure difference include: Scale-based hierarchical selection is performed based on the local sensitive regions of the topological mapping features to determine the hierarchical scale of multi-scale convolution. Based on the hierarchical scale, the topological mapping features are processed by convolution kernel local spatial feature extraction to obtain convolution local features; Spatial correlation analysis is performed based on the strength of the correlation between the spatial location of local convolutional features and changes in operating conditions to obtain spatial correlation weight features; Attention fusion processing is performed based on convolutional local features and spatial correlation weight features to determine the pattern baseline features of the pressure difference direction.

4. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 3, characterized in that, The spatial association weight feature generation logic includes: Identify strongly correlated regions and determine initial sensitive regions based on the spatial distribution characteristics of local convolutional features; Based on the initial sensitive area, local operating condition differences are identified to determine the regional difference characteristics. Spatial association weight features are obtained by non-uniformly distributing association weights based on regional disparity characteristics.

5. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized in that, The determination of the load mode in the current differential pressure direction includes: Based on the load condition transition characteristics of the model baseline features, the manifold space is constructed to determine the load model manifold characteristics. Local feature extraction is performed based on the topological mapping characteristics of the real-time differential pressure signal to determine the real-time topological local features; Based on the local mapping distribution characteristics of real-time topological local features in the manifold space, feature mapping similarity measurement is performed to determine the manifold mapping distance features; The load pattern is determined by performing optimal matching processing based on the manifold mapping distance characteristics to determine the load pattern in the current pressure difference direction.

6. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized in that, The determination of the target parameters for hydraulic control includes: Based on the working condition transfer characteristics of the current load mode and the historical load mode, load mode transfer path identification processing is performed to determine the characteristics of the mode transfer path. Based on the characteristics of the pattern transfer path, a comparative learning sample selection process is performed to determine typical pattern comparison samples. Based on the sensitivity characteristics of typical model comparison samples to changes in operating conditions, differential prediction processing of model trends is performed to determine the predictive characteristics of changes in operating conditions. Based on the predicted characteristics of changes in operating conditions, hydraulic control target parameters are selected and determined.

7. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 6, characterized in that, The generation logic for the predicted operating condition features includes: Based on the rapid change characteristics of load conditions in typical pattern comparison samples, rapid change sample identification processing is performed to determine rapidly changing sensitive samples. Based on rapidly changing sensitive samples, the trend and direction of working condition changes are identified to determine the characteristics of the working condition change direction. Based on the characteristics of the direction of operating condition changes, local differential prediction processing of operating condition trends is performed to obtain the predictive features of operating condition changes.

8. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized in that, The process of processing the disturbance observer based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updating the mode reference characteristics based on the observer output, includes: Based on the time-domain variation characteristics of the control error, dynamic identification of interference sources is performed to determine the characteristics of the interference sources; Based on the characteristics of the interference source and the load response characteristics of the hydraulic actuator, the interference compensation amount is predicted to obtain the interference compensation characteristics. The model reference features are updated and fused based on the interference compensation features and the model reference features.

9. A hydraulic control optimization system based on differential pressure direction change pattern recognition, implemented based on the hydraulic control optimization method based on differential pressure direction change pattern recognition as described in any one of claims 1-8, characterized in that, include: Differential pressure topology feature extraction module: Performs phase space topology reconstruction processing based on the pressure signals of the hydraulic cylinder inlet and outlet oil chambers, and performs singular spectrum analysis processing based on the topology reconstruction results to determine the topology mapping features of the differential pressure direction; Multi-scale feature fusion module: Based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and attention mechanism is used to fuse the spatial features to determine the pattern baseline features of the pressure difference direction; Load pattern recognition module: Constructs a manifold mapping space based on pattern reference features, and performs manifold distance measurement processing based on the topological mapping features of real-time differential pressure signals to determine the load pattern in the current differential pressure direction; Load trend prediction module: Constructs a pattern comparison learning mechanism based on the current load pattern and historical load patterns, and performs pattern trend prediction processing based on the pattern comparison learning results to determine the hydraulic control target parameters; Interference Observation and Dynamic Correction Module: The interference observer processes the interference based on the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameters, and updates the mode reference characteristics based on the observer output.

Images