Hydraulic control optimization system and method based on differential pressure direction change mode recognition

Abstract

The invention discloses a hydraulic control optimization system and method based on differential pressure direction change pattern recognition, and the method comprises the steps: carrying out the phase-space topology reconstruction and singular spectrum analysis according to the pressure signals of an oil inlet cavity and an oil return cavity of a hydraulic cylinder, and determining the topology mapping characteristics of a differential pressure direction; based on the topological mapping features, spatial features are extracted through a multi-scale convolutional network, and mode reference features are determined in combination with an attention mechanism; carrying out real-time differential pressure load mode identification by utilizing a manifold mapping space; constructing a mode comparison learning mechanism based on a historical load mode, predicting a mode trend and determining hydraulic control target parameters; and analyzing errors between displacement feedback of the piston rod of the hydraulic cylinder and control target parameters through the disturbance observer, and dynamically updating mode reference characteristics. Accurate real-time recognition and dynamic optimization control of the differential pressure fluctuation mode of the hydraulic system are achieved, and the operation stability and control precision of the system are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydraulic systems, in particular to a hydraulic control optimization system and method based on pressure difference direction change pattern recognition. BACKGROUND

[0002] Hydraulic systems are widely used in engineering machinery, industrial equipment and automation control fields, and realize mechanical movement and force output through hydraulic cylinders, with the advantages of high output power density, fast response, high control precision, etc. However, during the operation of the hydraulic cylinder, due to the change of working conditions, load fluctuation and the complexity of the flow of hydraulic medium, the cavitation phenomenon often occurs, which causes the internal pressure of the system to fluctuate violently. This cavitation phenomenon not only affects the performance of the hydraulic cylinder, but also seriously damages the service life of the hydraulic components if it exists for a long time.

[0003] The traditional hydraulic control strategy generally adopts fixed control parameters or simple feedback control mechanism for operation state adjustment, and cannot effectively identify and respond to the complex and nonlinear pressure fluctuation characteristics caused by cavitation. This traditional strategy lacks in-depth analysis of the internal flow state of the hydraulic system, which leads to the inability to accurately predict and quickly respond to abnormal working conditions of the hydraulic system in actual operation, especially the insufficient pattern recognition ability of the pressure difference fluctuation caused by cavitation.

[0004] In the prior art, the detection and analysis method for the cavitation phenomenon mostly adopts static threshold setting or single frequency domain analysis method. This method can only identify obvious and stable cavitation phenomenon, and cannot perform real-time and accurate pattern recognition and control response on the dynamic cavitation process, thereby causing poor system operation stability and poor control performance.

[0005] Therefore, how to accurately identify the pressure difference direction change pattern caused by cavitation in the hydraulic system, and optimize the hydraulic control strategy based on the real-time identification result to realize efficient and accurate control of the operation state of the hydraulic cylinder, has become one of the important technical problems to be solved in the current hydraulic control field. SUMMARY

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

[0007] To achieve the above-mentioned purpose, the present application provides a hydraulic control optimization system and method based on pressure difference direction change pattern recognition, comprising:

[0008] The pressure signals of the oil inlet cavity and the oil return cavity of the hydraulic cylinder are subjected to phase space topology reconstruction processing, and the topology mapping features of the pressure difference direction are determined according to the topology reconstruction results and singular spectrum analysis processing;

[0009] According to the topological mapping characteristics, spatial features are extracted through a multi-scale convolution network, and mode reference features in the differential pressure direction are determined through attention mechanism fusion processing based on the spatial features;

[0010] According to the mode reference features, a manifold mapping space is constructed, and the load mode in the current differential pressure direction is determined through manifold distance measurement processing based on the topological mapping characteristics of the real-time differential pressure signal;

[0011] According to the current load mode and the historical load mode, a mode comparison learning mechanism is constructed, and the hydraulic control target parameter is determined through mode trend prediction processing based on the mode comparison learning result;

[0012] According to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, a disturbance observer is processed, and the mode reference features are updated according to the observer output.

[0013] Further, the topological mapping characteristics of the differential pressure direction include:

[0014] According to the physical characteristics of the influence of the air pocket phenomenon in the hydraulic cylinder on pressure fluctuation, time delay determination processing is performed to obtain non-uniform time delay features of topological reconstruction;

[0015] According to the non-uniform time delay features, pressure signal delay coordinate mapping processing is performed to obtain the initial topological structure of the phase space;

[0016] According to the initial topological structure, the trajectory density of the phase space attractor is analyzed to determine the attractor region features;

[0017] According to the attractor region features, singular spectrum features are extracted to determine the topological mapping characteristics of the differential pressure direction.

[0018] Further, the generation logic of the non-uniform time delay features of topological reconstruction includes:

[0019] According to the change amplitude of the hydraulic cylinder oil inlet cavity pressure signal, air pocket initiation working condition recognition processing is performed to obtain air pocket initiation timing features;

[0020] According to the air pocket initiation timing features combined with the physical process of gas-liquid phase state transformation, pressure differential fluctuation transmission path analysis is performed to obtain fluctuation propagation features;

[0021] According to the fluctuation propagation features, the effective delay interval of the pressure signal is divided into non-equidistant intervals to obtain the non-uniform time delay features of topological reconstruction.

[0022] Further, the mode reference features of the differential pressure direction include:

[0023] According to the local sensitive region of the topological mapping characteristics, scale hierarchical selection processing is performed to determine the hierarchical scale of multi-scale convolution;

[0024] According to the hierarchical scale, the topological mapping features are subjected to convolution kernel local spatial feature extraction processing to obtain convolution local features;

[0025] According to the spatial position and the strong and weak correlation degree of the working condition change of the convolution local features, spatial position correlation analysis is performed to obtain spatial correlation weight features;

[0026] According to the convolution local features and the spatial correlation weight features, attention fusion processing is performed to determine the mode reference features of the differential pressure direction.

[0027] Further, the spatial correlation weight feature generation logic comprises:

[0028] According to the spatial distribution characteristics of the convolution local features, a strong correlation region is identified to determine an initial sensitive region;

[0029] According to the initial sensitive region, local working condition difference recognition processing is performed to determine a region difference degree feature;

[0030] According to the region difference degree feature, correlation weight non-uniform distribution processing is performed to obtain the spatial correlation weight features.

[0031] Further, the load mode of the current differential pressure direction is determined, comprising:

[0032] According to the load working condition transition characteristics of the mode reference features, manifold space construction processing is performed to determine a load mode manifold feature;

[0033] According to the topological mapping features of the real-time differential pressure signal, local feature extraction processing is performed to determine real-time topological local features;

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

[0035] According to the manifold mapping distance feature, load mode best matching processing is performed to determine the load mode of the current differential pressure direction.

[0036] Further, the determination of the hydraulic control target parameter comprises:

[0037] According to the working condition transition characteristics of the current load mode and the historical load mode, load mode transition path identification processing is performed to determine a mode transition path feature;

[0038] According to the mode transition path feature, comparative learning sample selection processing is performed to determine a typical mode comparison sample;

[0039] According to the working condition change sensitive characteristics of the typical mode comparison sample, mode trend difference prediction processing is performed to determine a working condition change prediction feature;

[0040] According to the working condition change prediction characteristic, a hydraulic control target parameter selection process is performed to determine the hydraulic control target parameter.

[0041] Further, the generation logic of the working condition change prediction characteristic comprises:

[0042] According to the rapid change sample identification process of the load working condition rapid transition characteristic of the typical mode compared with the sample, a rapid change sensitive sample is determined;

[0043] According to the working condition change trend direction identification process of the rapid change sensitive sample, a working condition change direction characteristic is determined;

[0044] According to the working condition trend local differentiation prediction process of the working condition change direction characteristic, a working condition change prediction characteristic is obtained.

[0045] Further, the disturbance observer process according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, and the updating of the mode reference characteristic according to the observer output, comprises:

[0046] According to the time domain variation characteristic of the control error, a disturbance source dynamic identification process is performed to determine a disturbance source characteristic;

[0047] According to the disturbance source characteristic combined with the load response characteristic of the hydraulic actuator, a disturbance compensation amount prediction process is performed to obtain a disturbance compensation characteristic;

[0048] According to the disturbance compensation characteristic and the mode reference characteristic, an updating fusion process is performed to update the mode reference characteristic.

[0049] The hydraulic control optimization system based on the differential pressure direction change mode recognition is realized based on the hydraulic control optimization method based on the differential pressure direction change mode recognition described above, comprising:

[0050] A differential pressure topology feature extraction module: according to the hydraulic cylinder oil inlet cavity and oil return cavity pressure signals, a phase space topology reconstruction process is performed, and according to the topology reconstruction result, a singular spectrum analysis process is performed to determine the topology mapping characteristic of the differential pressure direction;

[0051] A multi-scale feature fusion module: according to the topology mapping characteristic, a spatial feature is extracted through a multi-scale convolution network, and according to the spatial feature, an attention mechanism fusion process is performed to determine the mode reference characteristic of the differential pressure direction;

[0052] A load mode recognition module: according to the mode reference characteristic, a manifold mapping space is constructed, and according to the topology mapping characteristic of the real-time differential pressure signal, a manifold distance measurement process is performed to determine the load mode of the current differential pressure direction;

[0053] Load trend prediction module: construct a mode comparison learning mechanism according to the current load mode and the historical load mode, and perform mode trend prediction processing according to the mode comparison learning result to determine the hydraulic control target parameter;

[0054] Interference observation and dynamic correction module: interference observer processing is performed according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, and the mode reference feature is updated according to the observer output.

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

[0056] The present application can accurately identify the pressure difference direction change mode caused by cavitation by determining the non-uniform time delay feature based on the physical mechanism of the internal cavitation phenomenon of the hydraulic cylinder, and combining phase space topology reconstruction and singular spectrum analysis, thereby effectively overcoming the problems of poor recognition accuracy and insufficient real-time caused by relying only on static threshold or single frequency domain analysis in the prior art, and significantly improving the accuracy and reliability of the pressure difference fluctuation mode recognition of the hydraulic system.

[0057] The present application can deeply mine the contribution degree of different local area features to the pressure difference direction change mode recognition by adopting the spatial feature extraction method of multi-scale convolution network and attention mechanism fusion, form a precise mode reference feature, significantly enhance the recognition sensitivity of the system to different load working conditions and dynamic cavitation phenomenon, make the hydraulic control strategy can respond to actual working condition changes more finely, effectively improve the operation stability and control precision of the hydraulic system.

[0058] The present application realizes real-time optimization and adaptive adjustment of the hydraulic control target parameter by constructing a load mode recognition and trend prediction mechanism based on manifold mapping and mode comparison learning, and dynamically correcting the mode reference feature by combining the interference observer of real-time control error, fully solves the problems of poor adaptability to complex dynamic working conditions and response lag in the existing hydraulic control strategy, significantly improves the dynamic performance and actual control effect of the hydraulic system.

[0059] The present application realizes the accurate real-time recognition of the pressure difference direction change mode caused by cavitation in the hydraulic system and the dynamic optimization of the control strategy, significantly improves the operation stability and control precision of the system. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0061] Figure 1 is a structural schematic diagram of the method of the present application;

[0062] Figure 2 is a structural schematic diagram of the system of the present application. DETAILED DESCRIPTION

[0063] The technical solutions of the present application will be described clearly and completely below in conjunction with embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0064] Embodiment 1

[0065] Please refer to Figure 1 The present application provides a hydraulic control optimization system and method based on differential pressure direction change pattern recognition, comprising:

[0066] S101: topological reconstruction processing is performed on pressure signals of an oil inlet cavity and an oil return cavity of a hydraulic cylinder, and singular spectrum analysis processing is performed according to a topological reconstruction result to determine a topological mapping feature of a differential pressure direction;

[0067] It should be noted that the pressure signals are obtained by pressure sensors arranged at an oil inlet of the oil inlet cavity and an oil outlet of the oil return cavity of the hydraulic cylinder;

[0068] In implementation, the topological mapping feature of the differential pressure direction comprises:

[0069] S101.1 time lag determination processing is performed on physical characteristics of influences of a cavitation phenomenon in the hydraulic cylinder on pressure fluctuations to obtain a non-uniform time lag feature of topological reconstruction;

[0070] It should be noted that the determination of the non-uniform time lag feature in the present embodiment is based on a physical mechanism generated by the cavitation phenomenon actually existing in the hydraulic cylinder, rather than a traditional simple statistical analysis or mathematical calculation method;

[0071] Specifically, when the cavitation phenomenon occurs in the hydraulic cylinder, generation and collapse of the cavitation will cause pressure signal fluctuation characteristics in a local area; the cavitation formation and collapse process essentially corresponds to a gas-liquid phase change process, which specifically shows that the pressure drops sharply when the bubble is formed, and the pressure rises rapidly when the bubble breaks; in this process, there is a non-uniform propagation delay along the propagation path in the hydraulic cylinder.

[0072] Exemplarily, when the non-uniform time delay determination processing is performed, firstly, the pressure fluctuation curves of the hydraulic cylinder oil inlet cavity and the oil return cavity are monitored, and the characteristic point positions of the pressure fluctuation are determined according to the physical characteristics (i.e., the typical characteristics of local pressure curve sudden drop and rapid rise) of the pressure fluctuation amplitude in the air cavity occurrence and subsidence process; the characteristic points are taken as time delay marker points, and the non-uniform time delay characteristics are obtained by calculating the actual time difference of adjacent marker points, and 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 represents the i th non-uniform time delay interval, t i represents the corresponding time stamp of the i th air cavity characteristic point (such as the inflection point or the mutation point of the pressure curve); and n represents the total number of air cavity characteristic points identified;

[0075] It should be understood that the air cavity characteristic point position t i is obtained by actual pressure monitoring experiment of the hydraulic cylinder, and those skilled in the art can determine it according to the actual working condition of the hydraulic system by referring to the above method;

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

[0077] S101.1.1 Air cavity occurrence condition recognition processing is performed according to the pressure signal change amplitude of the hydraulic cylinder oil inlet cavity, and air cavity occurrence time sequence characteristics are obtained.

[0078] It should be noted that the purpose of the air cavity occurrence condition recognition processing described in the embodiment is to accurately capture the time sequence point at which the air cavity phenomenon in the hydraulic cylinder oil inlet cavity begins to occur, so as to obtain the air cavity occurrence time sequence characteristics in the hydraulic system, thereby providing a key physical reference basis for subsequent pressure fluctuation propagation analysis.

[0079] It can be understood that the pressure signal in the hydraulic cylinder oil inlet cavity has obvious physical characteristics when the air cavity begins to form, which is characterized by local pressure rapid drop and subsequent typical amplitude change characteristics. This feature is different from the conventional fluctuation or random noise fluctuation of the pressure signal in normal operation.

[0080] Exemplarily, in the specific implementation process, the air cavity occurrence time sequence characteristics can be determined by the following method:

[0081] A pressure sensor is arranged in the hydraulic cylinder oil inlet cavity to monitor the pressure signal of the oil inlet cavity in real time, and a real-time sampling sequence of the pressure signal is obtained.

[0082] P in(t), t = 1, 2, …, N,

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

[0084] According to the change rate of the amplitude of the oil inlet cavity pressure signal, the cavitation inception calculation is determined, and the calculation formula is:

[0085]

[0086] wherein, represents the change rate of the amplitude of the oil inlet cavity pressure signal with time, dt represents the small change of time t, i.e. the differential variable used to represent the differential operation in mathematical analysis, ε represents the cavitation inception determination threshold, t start represents the starting time stamp of the cavitation inception characteristic point;

[0087] It should be noted that the cavitation inception determination threshold is obtained by calculating the pressure change rate through the experimental data statistics of the stable operation and cavitation generation conditions of the hydraulic cylinder;

[0088] It should be further noted that the identification of the characteristic point is through the smoothing filtering processing of the pressure curve, the derivative operation of the curve and the determination of the position where the pressure change rate is greater than the cavitation inception determination threshold as the characteristic point;

[0089] According to the above-mentioned manner, a plurality of starting time sequence points of cavitation inception are obtained:

[0090]

[0091] wherein, T cav represents the obtained cavitation inception time sequence characteristic set, represents the starting time stamp of the i-th cavitation inception event, and m represents the total number of identified cavitation inception events;

[0092] S101.1.2 According to the cavitation inception time sequence characteristics and the physical process of gas-liquid phase transformation, the pressure difference fluctuation transmission path analysis is performed to obtain the fluctuation propagation characteristics;

[0093] It should be noted that the pressure difference fluctuation transmission path analysis in the embodiment is based on the obtained cavitation inception time sequence characteristics, combined with the pressure disturbance diffusion mechanism generated during the transformation process of the fluid in the hydraulic cylinder from liquid phase to gas phase, and analyzes the actual propagation path characteristics of the fluctuation signal in the hydraulic cylinder;

[0094] It can be understood that when the cavitation phenomenon in the oil inlet cavity occurs, the local pressure disturbance is caused by the rapid vaporization of the fluid to form bubbles, and this disturbance will propagate to other areas of the hydraulic system in the form of pressure waves, and the propagation path and propagation law are closely related to the change of the liquid-gas two-phase interface in the system;

[0095] Exemplarily, in the specific implementation, the differential pressure fluctuation transmission path analysis can be performed in the following manner:

[0096] The set of time sequence characteristic points T of cavitation initiation is utilized cav The initial disturbance source position and the corresponding timestamp of cavitation formation are determined;

[0097] According to the physical law of gas-liquid phase transformation in the hydraulic system, a disturbance wave propagation model 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 of x from the cavitation source position and at time t, P0 represents the initial disturbance pressure amplitude, a represents the damping attenuation coefficient of the pressure wave along the hydraulic pipeline, ω represents the fluctuation frequency, k represents the pressure wave propagation wave number, and φ represents the initial phase;

[0100] It should be noted that the specific values of the above parameters P0, a, ω and φ can be determined according to the experimental data of the actual hydraulic system, and those skilled in the art can determine these parameters through conventional experiments according to the above model and method, and k is obtained by wavelength calculation, and the calculation formula is Where λ represents the wavelength;

[0101] It should be further noted that by the above model, the disturbance wave propagation characteristics generated by each cavitation initiation event are analyzed, and then the laws of wave propagation speed, propagation direction and wave intensity changing with spatial position are determined, so that the clear wave propagation characteristics are obtained, which provides a clear physical basis for subsequent topology time delay characteristic analysis;

[0102] S101.1.3 The effective delay interval of the pressure signal is divided and processed in a non-equidistant manner according to the wave propagation characteristics, and the non-uniform time delay characteristics of topology reconstruction are obtained;

[0103] It should be noted that the purpose of the non-equidistant division processing in the embodiment is to determine the specific length of each delay interval based on the actual propagation characteristics of the differential pressure fluctuation in the hydraulic cylinder, so that the delay characteristics in topology reconstruction can objectively reflect the non-uniformity characteristics of the pressure wave propagation in the actual hydraulic medium;

[0104] It can be understood that, due to the comprehensive influence of medium properties, pipeline structure and cavitation disturbance characteristics on the propagation of pressure fluctuations in the hydraulic system, the delay caused by pressure fluctuations at different positions is not evenly distributed; traditional evenly spaced delay division cannot accurately represent such non-uniform delay characteristics;

[0105] Exemplarily, in specific implementation, the following method can be used:

[0106] According to the obtained wave propagation characteristics, the wave propagation model P wave (x,t) is calculated and analyzed to obtain the propagation time delay characteristics of pressure fluctuations at different spatial positions, and the specific time node sequence of the propagation of pressure fluctuations along the hydraulic pipeline is determined:

[0107]

[0108] Where T delay represents the effective delay node set of pressure signal wave propagation, represents the delay time stamp when the pressure signal propagates to the jth measurement position, and n represents the number of effective delay nodes measured;

[0109] According to the above determined effective delay node set, non-uniform delay interval division is carried out, and the division method is as follows:

[0110]

[0111] Where Δτ j represents the length of the jth non-uniform delay interval, and respectively represent the time stamps of two adjacent pressure signal propagation delay nodes;

[0112] It should be understood that the above delay interval length Δτ j is calculated and obtained according to the wave propagation characteristics, and objectively reflects the non-uniform characteristics of the propagation of pressure waves in the actual hydraulic system;

[0113] It should be further pointed out that, based on the non-uniform delay interval Δτ j , the topological reconstruction non-uniform time delay characteristics of the hydraulic system can be objectively and accurately constructed, and the physical reality characteristics of the propagation of system pressure difference fluctuations are fully reflected;

[0114] S101.2 According to the non-uniform time delay characteristics, the delay coordinate mapping processing of the pressure signal is carried out to obtain the initial topological structure of the phase space;

[0115] It should be noted that the delay coordinate mapping method adopted in the embodiment is based on the hydraulic cylinder pressure signal and the non-uniform time delay characteristics as the basis input to form a topological space structure with physical meaning.

[0116] It can be understood that the basic principle of the pressure signal delay coordinate mapping processing is to use the time delay interval T determined by the non-uniform time delay characteristics i The real-time collected pressure signal of the hydraulic cylinder is subjected to delay reconstruction processing, and then the trajectory information mapped in the multi-dimensional space is obtained to form an initial topological structure.

[0117] Exemplarily, in specific implementation, the following method can be used to obtain:

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

[0119] p(t), t = 1, 2, …, M,

[0120] Wherein, p(t) represents the amplitude of the pressure signal collected at time t, and M represents the sampling point number of the pressure signal.

[0121] Further, according to the non-uniform time delay characteristics T i The delay vector formed is represented as:

[0122]

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

[0124] It should be noted that the dimension of the delay vector represents the number of selected signal delay coordinates when the pressure signal is subjected to delay coordinate mapping processing, that is, the number of selected delay signal data points when the topological space is formed. It can be understood that the selection of the dimension of the delay vector determines the dimension of the phase space after the delay coordinate mapping, which is directly related to the complexity of the topological structure and the richness of the pressure signal characteristic information that the phase space can embody. In actual application, the dimension m of the delay vector can be gradually increased or decreased through topological structure reconstruction experiment, and the obtained phase space topological structure is evaluated, so as to select a suitable dimension m which can best reflect the differential pressure direction change mode of the hydraulic system and at the same time maintain the calculation efficiency. Generally, in actual engineering implementation, the specific value of the dimension m (for example, between 4 and 10) is determined by the specific hydraulic system characteristics and engineering application requirements.

[0125] Exemplarily, in specific implementation, the non-uniform time delay characteristics T iThe number of the trajectory points and the sensitivity required for the topological analysis are determined, and if the dimension m is too small, the internal complex topological characteristics of the hydraulic cylinder pressure signal cannot be fully reflected, and if the dimension m is too large, redundant information may be introduced, resulting in reduced calculation efficiency or reduced reliability of feature extraction.

[0126] It should be understood that, by the above delay coordinate mapping processing method, the initial topological structure capable of representing the internal pressure fluctuation of the hydraulic cylinder under the condition of non-uniform time lag can be obtained, which can more truly reflect the irregular transmission characteristics of the pressure fluctuation compared with the traditional uniform time lag processing method, so as to accurately capture the multi-mode characteristics of the pressure difference direction change;

[0127] S101.3. According to the initial topological structure, the trajectory density analysis of the phase space attractor is performed to determine the attractor region characteristics;

[0128] It should be noted that the purpose of the above phase space attractor trajectory density analysis is to identify the dense region of the trajectory distribution of the pressure signal in the topological structure, so as to represent the pressure fluctuation aggregation characteristics of the hydraulic system under different load modes;

[0129] It can be understood that the higher the trajectory point density in the initial topological structure, the stronger the stability of the pressure fluctuation of the hydraulic system in the region, reflecting the typical characteristics of the system under specific load conditions; on the contrary, the scattered region of the trajectory points indicates that the pressure fluctuation is in a non-stable or transition stage when the working condition changes or the system is disturbed by the outside world;

[0130] Exemplarily, in the specific implementation process, the 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 (for example, X, Y, and Z coordinates obtained by delay coordinate reconstruction), construct uniform grid cells, and the specific disclosed calculation method is as follows:

[0131]

[0132] Wherein, ρ k represents the density of the trajectory points in the kth spatial grid cell, N k represents the number of trajectory points contained in the kth spatial grid cell, V k represents the volume of the kth spatial grid cell, and K represents the total number of spatial grid cells;

[0133] It should be noted that by calculating the trajectory point density ρ k , the spatial distribution characteristics of the trajectory density in the initial topological structure can be obtained. In this embodiment, the grid cell region with a density higher than a certain threshold (for example, the density is higher than a certain multiple of the average density of the total trajectory points, and the specific multiple is determined by experiment) is determined as the attractor region.

[0134] It also needs to be explained that the determined attractor region feature effectively reflects the gathering characteristics of pressure fluctuations in the topological space, and provides important feature support for subsequent singular spectrum feature extraction and identification of hydraulic cylinder pressure difference direction change.

[0135] S101.4 Singular spectrum feature extraction is performed according to the attractor region feature, and the topological mapping feature of the pressure difference direction is determined.

[0136] It should be noted that the singular spectrum feature extraction method adopted in this embodiment aims to extract the dominant features of the hydraulic cylinder pressure fluctuations in the topological space to effectively identify the pressure difference direction change mode.

[0137] It can be understood that the so-called singular spectrum feature extraction is to perform singular value decomposition on the attractor region trajectory set to obtain singular values reflecting the primary and secondary relationship of the hydraulic cylinder pressure signal feature distribution, which are used as topological mapping features.

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

[0139] The determined attractor region trajectory point data is constructed into a data matrix:

[0140]

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

[0142] The singular value decomposition of the data matrix D is performed, and the specific formula is:

[0143] D=UΣV T ,

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

[0145]

[0146] Wherein, σ i represents the i th singular value, and satisfies σ1≥σ2≥…≥σ r >0.

[0147] It should be noted that V T is a transpose matrix obtained by transposing the orthogonal matrix V, that is, the row elements in the original matrix V are changed into column elements, and the column elements are changed into row elements to obtain the matrix.

[0148] It should be understood that the singular value σ iThe numerical size of the singular value represents the primary and secondary relationship of the pressure fluctuation topological feature, and the feature corresponding to the larger singular value represents the degree of influence on the pressure difference direction change more significantly. In this embodiment, the first p singular values (for example, the first three singular values) are selected as the topological mapping features for determining the pressure difference direction, and the specific numerical value can be obtained through actual experiments;

[0149] Further need to be explained is that the topological mapping features determined by the singular spectrum feature extraction method can comprehensively and accurately represent the main feature information of the hydraulic system pressure difference direction change mode, thereby providing accurate and reliable data support for hydraulic control parameter optimization;

[0150] S102: According to the topological mapping features, the spatial features are extracted through the multi-scale convolution network, and the mode reference features of the pressure difference direction are determined through attention mechanism fusion processing according to the spatial features;

[0151] Specifically, the mode reference features of the pressure difference direction include:

[0152] S102.1 According to the scale layer selection processing of the local sensitive area of the topological mapping features, the hierarchical scale of the multi-scale convolution is determined;

[0153] It should be noted that the purpose of the scale layer selection processing in this embodiment is to reasonably layer the convolution kernel scale in the convolution network feature extraction process according to the importance of the local sensitive area in the topological mapping feature space, so as to improve the representativeness of the mode reference features;

[0154] It can be understood that the sensitivity of different areas in the topological mapping feature space in reflecting the hydraulic cylinder pressure difference change law is different, therefore, the scale layer selection is needed to highlight the feature difference of the key areas;

[0155] It should be further noted that the convolution kernel scales used in this embodiment are small scale (3x3), medium scale (5x5) and large scale (7x7), and the number of convolution kernels of each scale is set to [16, 32 and 64]. The multi-scale convolution network structure adopts three layers of convolution, and each layer of convolution is followed by a maximum pooling operation for feature dimension reduction. The final output feature dimension is 128;

[0156] Exemplarily, in the specific implementation process, the hierarchical scale of the multi-scale convolution can be determined as follows:

[0157] According to the topological mapping features, the feature space is divided into a plurality of local areas:

[0158] R={R1,R2,…,R i ,…,R k},

[0159] wherein each region R i represents a specific position set in the topological map;

[0160] According to the statistical characteristics of the amplitude of the change in the topological features within the local region R i , the region sensitivity index S(R i ) is calculated, and an 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 the topological features at the spatial position x; and |R i | represents the number of spatial points of the region R i .

[0163] According to the above-mentioned region sensitivity index, all regions are selected in scale hierarchy;

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

[0165] If the sensitivity index belongs to the preset sensitivity hierarchical threshold interval, a medium-scale convolution kernel is selected;

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

[0167] It should be noted that the preset sensitivity hierarchical threshold interval is determined by a person skilled in the art according to actual needs;

[0168] Thus, the hierarchical scale of the multi-scale convolution suitable for different sensitive regions is determined, and a hierarchical scale set is formed:

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

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

[0171] S102.2 performs convolution kernel local spatial feature extraction processing on the topological map features according to the hierarchical scale, and obtains convolution local features;

[0172] It should be noted that the convolution kernel local spatial feature extraction process in the embodiment is aimed at extracting spatial features of the topological mapping features according to 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 convolution local features.

[0173] It can be understood that the spatial local characteristics embodied by the topological mapping feature space at different scales are significantly different, and therefore the use of different scale convolution kernels to extract local features can objectively reflect the essential features of the differential pressure change.

[0174] Exemplarily, in the specific implementation process, the extraction of the convolution local features can be realized in the following manner:

[0175] A multi-scale convolution processing method is defined, that is, different scale convolution operations are respectively performed on the topological mapping feature space T(x, y), and an example of a convolution operation formula is as follows:

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

[0177] In the formula, F scale (x, y) represents the convolution local feature extracted at the spatial position (x, y) by the convolution kernel with a scale of scale, T(x-u, y-v) represents the original feature value of the topological mapping feature space, and K scale (u, v) represents the convolution kernel with a selected scale.

[0178] Corresponding convolution kernel scales are used for different regions to respectively extract convolution local features and form a convolution local feature set:

[0179]

[0180] In the formula, F represents the convolution local feature extracted by the corresponding scale convolution kernel in the i-th region.

[0181] It should be understood that the convolution local feature set obtained through the above-mentioned multi-scale convolution kernel local spatial feature extraction process can effectively reflect the local spatial difference characteristics of the differential pressure change of the hydraulic system, thereby providing sufficient and reliable data support for subsequent spatial correlation weight analysis.

[0182] S102.3 performs spatial position correlation analysis according to the strong and weak correlation degree between the spatial position of the convolution local feature and the working condition change, and obtains a spatial correlation weight feature;

[0183] It should be noted that the purpose of the spatial position correlation analysis in this embodiment is to determine the sensitivity of each spatial position in the convolution local feature to the change of different working conditions of the hydraulic system, so as to objectively and clearly obtain the spatial correlation weight features corresponding to different regions, and improve the working condition sensitivity of the mode reference features.

[0184] It can be understood that, since the response degree of the local features of different spatial positions to the working condition change is different, spatial position correlation analysis is needed to determine the weight distribution characteristics of each region when the working condition changes;

[0185] The spatial correlation weight feature generation logic is as follows:

[0186] S102.3.1. Identifying strong correlation regions according to the spatial distribution characteristics of the convolution local features, and determining the initial sensitive regions;

[0187] It should be noted that, in this embodiment, it is first needed to determine which local spatial positions have relatively significant feature differences under different working condition changes, so as to take them as the initial sensitive regions for subsequent fine analysis;

[0188] For example, the local feature distribution entropy is used in this embodiment to objectively identify the strong correlation regions;

[0189] Specifically, the distribution entropy value of each local region is calculated:

[0190]

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

[0192] It should be understood that LFDE(R i ) is used to represent the feature distribution difference degree, and p ij is determined by counting the feature value frequency under different working conditions;

[0193] According to the distribution entropy of each region, the regions are sorted, and the regions with distribution entropy greater than or equal to the preset distribution entropy threshold value are identified as strong correlation regions, and are defined as the initial sensitive region set, which represent that the feature change is significant and the sensitivity is high.

[0194] It should be noted that the distribution entropy threshold value is determined by experiment, which is set by a person skilled in the art according to actual needs;

[0195] S102.3.2. Local working condition difference identification processing according to the initial sensitive regions, and determining the region difference degree feature;

[0196] It should be noted that the purpose of the step is to clearly define the difference between the characteristics of the initial sensitive area under different hydraulic system working conditions, so as to provide an objective basis for subsequent correlation weight allocation;

[0197] Exemplarily, the specific local working condition difference recognition processing method is disclosed as follows:

[0198] According to the initial sensitive area set, the corresponding convolution local feature vector of each sensitive area under different working conditions is extracted;

[0199] The Euclidean distance between the feature vectors of the same area under all working conditions is calculated respectively to obtain the area difference degree feature, and the specific calculation formula is as follows:

[0200]

[0201] In the formula, D(R sens ) represents the area difference degree feature of the initial sensitive area, represents the local feature value of the kth position in the initial sensitive area under the ith working condition, L represents the dimension of the feature value contained in the area, and M represents the total number of hydraulic system working conditions involved;

[0202] S102.3.3. According to the area difference degree feature, the correlation weight non-uniform allocation processing is carried out to obtain the spatial correlation weight feature;

[0203] It should be noted that the purpose of the step is to reasonably and non-uniformly allocate the correlation weight of each initial sensitive area based on the area difference degree feature calculated in step S102.3.2, so as to obtain a spatial correlation weight feature that can clearly reflect the difference in sensitivity of each area to working condition changes;

[0204] Exemplarily, the implementation process of the specific correlation weight non-uniform allocation processing is disclosed as follows:

[0205] According to the area difference degree feature of each initial sensitive area, the initial weight mapping function is defined, and the initial weight allocation value of each area is determined based on the relative size of the area difference degree feature, and the specific calculation formula is as follows:

[0206]

[0207] In the formula, represents the initial weight value of the ith initial sensitive area, represents the area difference degree feature of the ith initial sensitive area, N s represents the total number of initial sensitive areas;

[0208] For each spatial position point (x, y) in each initial sensitive region, a local sensitivity index of local spatial feature variation is defined, and an exemplary definition is as follows:

[0209]

[0210] In the formula, S local (x, y) represents the local sensitivity index of the spatial position (x, y), and σ F (x, y) represents the standard deviation of the spatial position (x, y) under different working conditions, and μ F (x, y) represents the feature mean value of the spatial position (x, y) under different working conditions.

[0211] Further, according to the initial weight value of the region and the local sensitivity index, non-uniform distribution of the weight of each spatial position point in the region is performed, and a final spatial correlation weight feature is obtained, and an exemplary calculation formula is as follows:

[0212]

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

[0214] It should be noted that in the formula, represents the sum of the local sensitivity indexes of all positions in the region, so as to realize relative non-uniform distribution of the weight.

[0215] S102.4 According to the convolution local feature and the spatial correlation weight feature, attention fusion processing is performed to determine the mode reference feature of the pressure difference direction.

[0216] It should be noted that this step aims to fuse the convolution local feature and the spatial correlation weight feature obtained in the foregoing steps through an attention mechanism, so as to objectively highlight the contribution difference of different spatial position features to the system pressure difference direction recognition, and finally obtain the mode reference feature of the pressure difference direction.

[0217] Exemplarily, the embodiment is implemented in a weighted attention fusion manner, and the implementation process is as follows:

[0218] According to the convolution local feature vector set and the corresponding spatial correlation weight feature value, initial spatial feature weighted fusion calculation processing is performed, and the calculation manner is as follows:

[0219]

[0220] In the formula, F weighted represents the spatial features obtained by initial weighted fusion, Ω sens represents a set of spatial position points in the sensitive area, F(x, y) represents a convolution local feature vector of the spatial position (x, y), and W(x, y) represents a spatial correlation weight feature value of the spatial position (x, y);

[0221] To further highlight the importance difference of different convolution feature dimensions, the embodiment introduces a fusion mechanism based on channel attention, and the specific calculation method is as follows:

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

[0223] In the formula, A c represents a channel attention weight feature, which takes a value between 0 and 1, W1 and W2 represent coefficient matrices for channel attention weight calculation, and σ represents a 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, and d is the feature vector dimension, for example, d = 20. The specific feature vector dimension is set by a person skilled in the art according to the actual situation;

[0225] According to the channel attention weight feature, a differential pressure direction mode reference feature is obtained, and the calculation formula is:

[0226] F final = A c ⊙F weighted ,

[0227] In the formula, F final represents the differential pressure direction mode reference feature, and ⊙ represents a multiplication operation on corresponding dimension elements.

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

[0229] Specifically, determining the load mode of the current differential pressure direction includes:

[0230] S103.1: Perform manifold space construction processing according to the load working condition transition characteristics of the mode reference feature to determine the load mode manifold feature;

[0231] It should be noted that the main purpose of this step is to construct a manifold space that can intuitively reflect the load working condition change law according to the mode reference feature obtained in the foregoing step, so as to form the manifold feature of the load mode and objectively describe the conversion relationship between different load modes.

[0232] Exemplarily, the embodiment implements the construction of the above features by a manifold learning method, and the specific disclosure is as follows:

[0233] The mode reference feature set is input as data for manifold learning, and a local linear embedding (LLE) algorithm is used to construct a load mode manifold space. The specific manifold mapping process is as follows:

[0234] For each mode reference feature, a local reconstruction weight matrix is constructed according to the linear relationship between the mode reference feature and the k nearest neighbor features, and the specific calculation method is as follows:

[0235]

[0236] wherein, represents the feature of the reconstruction weight of the neighborhood feature , N(i, k) represents the nearest neighbor k feature set of the feature in the feature space, and k represents the preset local neighborhood size,

[0237] It should be noted that the preset local neighborhood size is set by a person skilled in the art according to the actual situation;

[0238] Further, according to the local reconstruction weight matrix, a low-dimensional embedding feature vector set of the manifold space is constructed, and the set is used as a load mode manifold feature.

[0239] S103.2 performs local feature extraction processing according to the topological mapping feature of the real-time differential pressure signal, and determines a real-time topological local feature.

[0240] It should be noted that this step is mainly to extract local features in a specific region for the topological mapping feature corresponding to the current real-time collected differential pressure signal, so as to form a real-time topological local feature, which is used for subsequent load mode recognition.

[0241] Exemplarily, in the embodiment, the specific extraction method of the real-time topological local feature is as follows:

[0242] The topological mapping feature of the real-time differential pressure signal is obtained, and is specifically defined as:

[0243]

[0244] wherein, T real (t) represents a topological mapping feature sequence measured under a real-time working condition, represents the jth feature component in the topological mapping space, and m represents the feature dimension of the topological mapping space.

[0245] Further, to capture the significant change of the topology mapping feature in a specific local area, the embodiment adopts a local principal component analysis (LPCA) method to specifically extract local features, and the specific implementation method is as follows:

[0246] In the topology mapping feature sequence T real (t) in a local neighborhood , a data window with a length of L (such as L = 50) is exemplarily selected, and a local data matrix is constructed:

[0247]

[0248] In the formula, T local represents a local feature data matrix of the current real-time working condition, and t0 represents a time stamp corresponding to the current real-time working condition.

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

[0250] The local feature data matrix of the current real-time working condition is subjected to feature decomposition to extract a principal feature component vector, and the specific implementation process is as follows:

[0251] According to the data matrix T local , covariance calculation is performed to obtain a covariance matrix, and the calculation formula is as follows:

[0252]

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

[0254] The covariance matrix is subjected to eigenvalue decomposition, and the specific decomposition formula is as follows:

[0255] C local = EΛE T ,

[0256] Wherein, E represents a feature vector matrix corresponding to the covariance matrix, and A represents a diagonal matrix composed of eigenvalues.

[0257] Exemplarily, the embodiment selects the first p principal feature vectors to constitute a real-time topology local feature vector set F local :

[0258] F local = [e1, e2, …, e p ],

[0259] It should be noted that e1, e2, …, e p represents the eigenvector in the order of eigenvalue from large to small, and the specific value of p is determined by the person skilled in the art according to the actual demand, for example, p = 3;

[0260] S103.3 According to the local mapping distribution characteristics of the real-time topological local feature in the manifold space, a feature mapping similarity measurement process is performed to determine the manifold mapping distance feature;

[0261] It should be noted that the mapping similarity between the real-time topological local feature and different load mode features in the manifold space is measured according to the real-time topological local feature obtained in the foregoing steps, and then the manifold mapping distance feature of the real-time working condition corresponding to each load mode is objectively determined;

[0262] Exemplarily, in the embodiment, the similarity measurement between the real-time topological local feature and the load mode manifold feature is specifically implemented as follows:

[0263] According to the real-time topological local feature vector set and the load mode manifold feature vector set, the manifold mapping distance between the real-time topological local feature and the load mode manifold feature vector is obtained based on the weighted distance measurement method of local mapping similarity, and the specific calculation formula is as follows:

[0264]

[0265] In the formula, D map (i) represents the manifold mapping distance between the real-time topological local feature and the i-th load mode manifold feature, e j represents the j-th principal component feature vector of the real-time topological local feature, represents the i-th load mode manifold feature vector y i corresponding to the j-th dimension mapping feature component, α j represents the mapping distance weight coefficient;

[0266] It should be noted that in the formula, the mapping distance weight coefficient is used to reflect the difference in the contribution degree of different feature dimensions to load mode recognition, which is obtained by calculating the proportion of the eigenvalue of the real-time topological local feature in the corresponding dimension to the total dimension value, and in the formula is obtained by Euclidean distance calculation, which will not be described in detail here;

[0267] After the above similarity measurement process, the manifold mapping distance feature set of the real-time working condition to each load mode is obtained;

[0268] S103.4 According to the manifold mapping distance feature, a load mode best matching process is performed to determine the load mode of the current differential pressure direction;

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

[0270] Specifically, the exemplary disclosed implementation is as follows:

[0271] According to the set of manifold mapping distances corresponding to each predefined load mode according to the real-time pressure difference direction, a minimum value screening process is performed on the distance set to obtain a preliminary mode matching result, i.e., an initial load mode index and a corresponding manifold mapping distance value;

[0272] It can be understood that the above process is implemented using a conventional minimum value search process, i.e., selecting the element with the smallest value in the distance set and its corresponding index;

[0273] It should be understood that although the above distance value determines the initial mode, feature abnormalities or measurement errors may occur in the actual system, resulting in reduced reliability of the matching mode, and therefore further mode matching validity judgment processing is required, the specific process implementation is as follows:

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

[0275] If the minimum distance value is less than the preset threshold, it is determined that the preliminary mode matching result is valid;

[0276] If the minimum distance value is greater than or equal to the preset threshold, it is determined that the preliminary mode matching result is invalid;

[0277] It should be noted that the preset threshold is obtained through multiple experiments;

[0278] S104: Construct a mode comparison learning mechanism according to the current load mode and the historical load mode, and perform mode trend prediction processing according to the mode comparison learning result to determine the hydraulic control target parameter;

[0279] It should be noted that the construction of the above mode comparison learning mechanism is based on the historical operating condition data of the hydraulic system. Specifically, by analyzing the operating condition transfer rule between the current load mode and the historical load mode, the internal relationship between different load modes and their evolution process are mined to clearly predict the specific trend of load change, thereby providing an objective decision basis for determining the hydraulic control target parameter;

[0280] Specifically, determining the hydraulic control target parameter includes:

[0281] S104.1According to the working condition transition characteristics of the current load mode and the historical load mode, load mode transition path identification processing is performed to determine mode transition path characteristics;

[0282] It should be noted that the determination of the load mode transition path characteristics is based on the transition rules between adjacent control periods of load modes in the historical operation process;

[0283] Exemplarily, the continuous load modes corresponding to the historical operation data of the hydraulic system are obtained;

[0284] The above historical load mode sequence is combined into working condition transition path samples, for example, the historical mode sequence mode 1→mode 2→mode 3 is decomposed into mode transition samples mode 1→mode 2 and mode 2→mode 3;

[0285] According to the statistical frequency of each transition path in the historical data, the transition frequency characteristics of each mode transition path are obtained;

[0286] The paths with transition frequency characteristics exceeding the threshold value (such as 5% of the overall transition paths) determined by experiments are taken as typical mode transition path characteristics;

[0287] S104.2According to the mode transition path characteristics, a comparative learning sample selection process is performed to determine a typical mode comparison sample;

[0288] It should be noted that the quantitative standard for selecting a typical mode comparison sample is: the pressure amplitude change is greater than [for example, 20%], and the duration is not less than [for example, 0.5s] working condition data, which is taken as a typical mode comparison sample;

[0289] Exemplarily, according to the typical mode transition path in the mode transition path characteristics, the starting mode and the ending mode corresponding to each typical mode transition path are identified;

[0290] The historical working condition data corresponding to the identified starting mode and ending mode are subjected to objective working condition amplitude change statistical processing;

[0291] It should be noted that the statistical method includes but is not limited to mean calculation, standard deviation calculation, etc.;

[0292] The starting-terminating mode pair with significant amplitude change and obvious change trend is selected as a typical mode comparison sample;

[0293] It should be noted that the significant amplitude change and obvious change trend are determined by comparing with a preset change amplitude threshold value, and when greater than or equal to the preset change amplitude threshold value, it is judged as significant change and change trend, and the change amplitude threshold value is set by the technical personnel according to the actual demand;

[0294] S104.3.1, according to the load condition change sensitive characteristics of the typical mode contrast sample, a rapid change sample identification process is performed to determine the rapid change sensitive sample;

[0295] It should be noted that the mode sample sensitive to rapid change of load condition is identified from the typical mode contrast sample, and the differentiated trend prediction analysis is performed accordingly;

[0296] The generation logic of the load condition change prediction feature is as follows:

[0297] S104.3.1, according to the load condition change sensitive characteristics of the typical mode contrast sample, a rapid change sample identification process is performed to determine the rapid change sensitive sample;

[0298] It should be noted that this step is specifically designed to identify load samples with a load condition change rate much higher than the regular change rate during mode transition;

[0299] In specific implementation, the processing mode is as follows:

[0300] The change rate of the hydraulic system pressure signal in the typical mode contrast sample at adjacent time points is calculated, and the specific calculation formula is as follows:

[0301]

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

[0303] According to the change rate calculated above, the load mode sample corresponding to the change rate absolute value significantly higher than the experimentally determined threshold (for example: higher than 3 times of the regular change rate of load condition) is identified as the rapid change sensitive sample;

[0304] S104.3.2, according to the rapid change sensitive sample, a load condition change trend direction identification process is performed to determine the load condition change direction feature;

[0305] It should be noted that the above trend direction identification of the load condition of the identified rapid change sensitive sample is to specifically identify the load condition in the "rising trend", "falling trend" or "stable trend" to ensure the objectivity and effectiveness of the mode trend prediction;

[0306] Exemplarily, the processing logic disclosed in the embodiment is as follows:

[0307] Based on the local extreme point (inflection point) of the load pressure change curve in the rapid change sensitive sample, the time sequence position and pressure amplitude of the local extreme point are extracted;

[0308] It should be noted that the local extreme point is determined by the intersection point of the first derivative (slope) from positive to negative or from negative to positive;

[0309] According to the pressure change amplitude difference between the adjacent two extreme points and the corresponding time difference, the trend direction identification processing is performed to obtain a trend identification factor;

[0310] Specifically, based on the adjacent two local extreme points, first, the difference between the pressure amplitudes corresponding to the front and rear extreme points is determined, that is, the result of subtracting the pressure of the former extreme point from the pressure of the latter extreme point; then the time difference corresponding to the front and rear extreme points is determined, that is, the time of the latter extreme point minus the time of the former extreme point; finally, the pressure difference is divided by the above-mentioned corresponding time difference, that is, the rate of change of pressure with time is obtained, which is used to identify the direction of load trend change;

[0311] According to the trend identification factor, the direction feature of the working condition change is determined;

[0312] If the trend identification factor is greater than the preset positive trend threshold, it is determined that the working condition change is an "upward trend";

[0313] If the trend identification factor is less than the preset negative trend threshold, it is determined that the working condition change is a "downward trend";

[0314] If the trend identification factor is between the preset positive trend threshold and the preset negative trend threshold, it is determined that the working condition change is a "steady trend";

[0315] Specifically, when the rate of change of pressure with time is positive and the value is greater than the preset positive trend threshold, it indicates that the pressure increases rapidly in unit time, and the current load working condition can be determined as being in a significant "upward trend";

[0316] When the rate of change of pressure with time is negative and the value is less than the preset negative trend threshold, it indicates that the pressure decreases rapidly in unit time, and the current load working condition can be determined as being in a significant "downward trend";

[0317] If the rate of change of pressure with time is small and between the positive and negative thresholds, it indicates that the pressure change in unit time is not obvious, and the current load working condition can be determined as being in a "steady trend";

[0318] S104.3.3 According to the working condition change direction feature, a working condition trend local differential prediction processing is performed to obtain a working condition change prediction feature;

[0319] It should be noted that based on the working condition change direction feature (i.e. "upward trend", "downward trend" or "steady trend"), the working condition change under different trend directions is predicted differently to obtain accurate working condition change prediction features;

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

[0321] Based on the historical load working condition operation data, a working condition change prediction sample library corresponding to three mode categories of rising trend, falling trend and stable trend is constructed, and each sample library includes several typical working condition prediction samples;

[0322] According to the current working condition change direction feature, the sample library consistent with the current trend direction is selected from the working condition change prediction sample libraries of the above three mode categories as the basis for this prediction processing;

[0323] The prediction processing process is as follows:

[0324] The real-time operation data of the current load mode is compared with the typical prediction samples in the selected sample library one by one, that is, the local difference between the current real-time data and the typical prediction sample data is calculated one by one;

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

[0326] According to the local difference value, one or more typical prediction samples with the smallest local difference value are selected as the prediction basis of the current working condition change trend;

[0327] According to the typical prediction sample with the smallest local difference value, the trend evolution analysis of the subsequent working condition change is carried out to obtain the accurate working condition change prediction feature;

[0328] S104.4 According to the working condition change prediction feature, the hydraulic control target parameter selection processing is carried out to determine the hydraulic control target parameter;

[0329] It can be understood that a hydraulic control parameter library has been established, which includes a plurality of control parameter combinations associated with different working condition change prediction features, and each control parameter combination includes but is not limited to target pressure set value, target flow set value, valve opening reference value, pump flow distribution ratio and other specific hydraulic control target parameters;

[0330] Exemplarily, the specific disclosed target parameter selection processing method is as follows:

[0331] According to the working condition change prediction feature, the similarity between the current prediction feature and the historical working condition feature corresponding to each control parameter combination in the hydraulic control parameter library is calculated;

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

[0333] According to the similarity calculation result, the hydraulic control parameter combination corresponding to the historical working condition feature with the highest similarity (i.e. the value closest to the current predicted feature) is selected as the initial selection of the current hydraulic control target parameter;

[0334] According to the specific operating conditions of the current load mode, in combination with the actual hydraulic system operating constraint conditions (such as: system maximum allowable pressure, flow limit, actuator displacement speed limit, etc. Constraint conditions), the control parameter combination of the initial selection is objectively corrected to ensure that the determined hydraulic control target parameter can meet the actual control demand;

[0335] According to the corrected hydraulic control target parameter combination, it is used in the subsequent hydraulic system control implementation process, including but not limited to the specific control of the pump or valve, to effectively complete the accurate position, speed or force control of the hydraulic actuator;

[0336] It should be clear that the specific parameter values of the above hydraulic control parameter library are different due to different specific working conditions, load types, and hydraulic system structures. The specific values are determined through a large number of actual working condition experiments and historical operation data analysis;

[0337] S105: According to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, the disturbance observer is processed, and the mode reference feature is updated according to the observer output;

[0338] Specifically, according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, the disturbance observer is processed, and the mode reference feature is updated according to the observer output, including:

[0339] S105.1 According to the time domain variation characteristics of the control error, the disturbance source dynamic identification processing is performed to determine the disturbance source feature;

[0340] Exemplarily, 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-5000Hz to sufficiently capture the high-frequency fluctuation information of the error;

[0342] According to the control error signal, short-time Fourier transform processing is performed to obtain the time-frequency distribution result;

[0343] According to the time-frequency distribution result, the frequency spectrum energy distribution characteristics of the control error signal changing with time are obtained, and the energy distribution of different frequency bands is analyzed to determine the specific source of the disturbance:

[0344] If the spectral energy significantly increases in the high frequency band (e.g. frequency greater than 50Hz), it is determined as external disturbance (e.g. caused by rapid load fluctuation);

[0345] If the spectral energy significantly increases in the low frequency band (e.g. frequency lower than 20Hz), it is determined as internal disturbance (e.g. caused by spool hysteresis or pipe resonance);

[0346] Further, the statistical features quantify the disturbance source features, for example, the high and low frequency band energy proportions (i.e. the proportions of the total energy in the error signal in the total energy) are specifically calculated in each time window, and the energy proportions are taken as the quantified disturbance source features;

[0347] S105.2 According to the disturbance source features, the disturbance compensation amount prediction processing is performed in combination with the load response characteristics of the hydraulic actuator, and the disturbance compensation features are obtained;

[0348] It should be noted that, in order to predict the compensation amount of the hydraulic actuator after being disturbed, the present embodiment adopts a recursive least square prediction method based on the load response characteristics;

[0349] Exemplarily, according to the disturbance source features, a disturbance compensation feature prediction model is established by historical disturbance feature data and hydraulic actuator displacement feedback data, and the structure form 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 of the hydraulic actuator at time k, X(k) represents the disturbance source feature vector at time k, 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, in order to recursively update the prediction model parameters θ(k), the present embodiment adopts the RLS algorithm processing, and the specific steps are as follows:

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

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

[0355]

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

[0357]

[0358] wherein K(k) represents a Kalman gain matrix, y m (k) represents a compensation demand value actually measured between the feedback displacement of the hydraulic cylinder piston rod and the hydraulic control target displacement at the kth moment, and λ represents a forgetting factor, and the specific value is between 0.95 and 0.99;

[0359] It should be noted that the Kalman gain matrix is used to determine the parameter correction amplitude, 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 real-time dynamic correction of the interference compensation characteristic prediction model parameter is realized through the above RLS algorithm, and then the interference compensation characteristic at each moment is obtained, which is used as the basis for updating the subsequent mode reference characteristic;

[0361] S105.3 updates and fuses the mode reference characteristic according to the interference compensation characteristic and the mode reference characteristic;

[0362] It should be noted that, in order to realize the above fusion update processing, the embodiment exemplarily adopts a weighted fusion mechanism based on a sliding time window to realize real-time update of the mode reference characteristic;

[0363] Exemplarily, it is assumed that the amplitude of the current control error is e(t), the corresponding absolute error amplitude is |e(t)|, and the error threshold e max and e min are set.

[0364] It should be noted that the error threshold e max and e min are determined according to the actual hydraulic control accuracy requirement;

[0365] When the error amplitude |e(t)|≥e max , it is determined that the weight α(t) takes a larger value, exemplarily 0.8-1, so that the updated mode reference characteristic is more suitable for the current real-time interference compensation requirement;

[0366] When the error amplitude |e(t)|≤e min , it is determined that the weight α(t) takes a smaller value, exemplarily 0-0.2, so as to ensure the stability of the mode reference characteristic and avoid frequent fluctuations due to small amplitude interference;

[0367] When the error amplitude e min <|e(t)|<e max , the weight α(t) is determined through a linear interpolation method.

[0368] It should be appreciated that through the above weight fusion mechanism based on real-time control error amplitude dynamic adjustment, real-time accurate updating of the mode reference feature is realized, 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] Embodiment 2

[0370] As shown in Figure 2 The embodiment disclosed provides a hydraulic control optimization system based on differential pressure direction change pattern recognition, which includes:

[0371] Differential pressure topology feature extraction module: according to the pressure signals of the hydraulic cylinder oil inlet cavity and oil return cavity, perform phase space topology reconstruction processing, and according to the topology reconstruction result, perform singular spectrum analysis processing to determine the topology mapping feature of the differential pressure direction;

[0372] Multi-scale feature fusion module: according to the topology mapping feature, extract spatial features through a multi-scale convolution network, and according to the spatial features, perform attention mechanism fusion processing to determine the mode reference feature of the differential pressure direction;

[0373] Load pattern recognition module: according to the mode reference feature, construct a manifold mapping space, and according to the topology mapping feature of the real-time differential pressure signal, perform manifold distance measurement processing to determine the current load mode of the differential pressure direction;

[0374] Load trend prediction module: according to the current load mode and historical load mode, construct a mode comparison learning mechanism, and according to the mode comparison learning result, perform mode trend prediction processing to determine the hydraulic control target parameter;

[0375] Interference observer and dynamic correction module: according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, perform interference observer processing, and according to the observer output, update the mode reference feature;

[0376] The above-described embodiments can be implemented in part or in whole through software implemented by a processor. The results of such software can be stored in a memory and / or other storage mediums.

[0377] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, and the division of the units is only one, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0378] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0379] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0380] Part of data in the above formula is calculated by removing dimension, and the formula is obtained by software simulation of a large amount of collected data to be closest to the real situation; the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0381] The above embodiments are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A hydraulic control optimization method based on differential pressure direction change pattern recognition, characterized by, The method comprises the following steps: According to the pressure signals of the inlet and return oil chambers of the hydraulic cylinder, the phase space topology is reconstructed, and the topology mapping feature of the pressure difference direction is determined according to the topology reconstruction result and the singular spectrum analysis processing; According to the topology mapping feature, the spatial feature is extracted through a multi-scale convolution network, and the mode reference feature of the pressure difference direction is determined through attention mechanism fusion processing based on the spatial feature; According to the mode reference feature, a manifold mapping space is constructed, and the load mode of the current pressure difference direction is determined through the topology mapping feature of the real-time pressure difference signal and the manifold distance measurement processing; According to the current load mode and the historical load mode, a mode comparison learning mechanism is constructed, and the hydraulic control target parameter is determined through mode trend prediction processing based on the mode comparison learning result; According to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, the disturbance observer processing is performed, and the mode reference feature is updated according to the observer output.

2. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized by, The determination of the topology mapping feature of the pressure difference direction comprises: According to the physical characteristics of the influence of the cavitation phenomenon in the hydraulic cylinder on the pressure fluctuation, the time lag is determined, and the non-uniform time lag feature of the topology reconstruction is obtained; According to the non-uniform time lag feature, the pressure signal delay coordinate mapping processing is performed, and the initial topology structure of the phase space is obtained; According to the initial topology structure, the trajectory density of the phase space attractor is analyzed, and the attractor region feature is determined; According to the attractor region feature, the singular spectrum feature is extracted, and the topology mapping feature of the pressure difference direction is determined.

3. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 2, characterized by, The generation logic of the non-uniform time lag feature of the topology reconstruction comprises: According to the pressure signal change amplitude of the inlet oil chamber of the hydraulic cylinder, the cavitation initiation working condition recognition processing is performed, and the cavitation initiation time sequence feature is obtained; According to the cavitation initiation time sequence feature, the pressure difference fluctuation transmission path analysis is performed in combination with the physical process of gas-liquid phase state transformation, and the fluctuation propagation feature is obtained; According to the fluctuation propagation feature, the effective delay interval of the pressure signal is divided into non-equidistant intervals, and the non-uniform time lag feature of the topology reconstruction is obtained.

4. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, characterized by, The determination of the mode reference feature of the pressure difference direction comprises: According to the scale hierarchical selection processing of the local sensitive region of the topology mapping feature, the hierarchical scale of multi-scale convolution is determined; According to the hierarchical scale, the local spatial feature extraction processing of the convolution kernel is performed on the topology mapping feature, and the convolution local feature is obtained; According to the spatial position correlation analysis of the convolution local feature and the strong and weak correlation degree of the working condition change, the spatial correlation weight feature is obtained; According to the attention fusion processing of the convolution local feature and the spatial correlation weight feature, the mode reference feature of the pressure difference direction is determined.

5. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 4, characterized by, The spatial correlation weight feature generation logic comprises: According to the strong correlation region identified by the spatial distribution characteristics of the convolution local feature, the initial sensitive region is determined; According to the local working condition difference recognition processing of the initial sensitive region, the region difference degree feature is determined; According to the non-uniform distribution processing of the correlation weight based on the region difference degree feature, the spatial correlation weight feature is obtained.

6. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, wherein, The determination of the load mode of the current pressure difference direction comprises: According to the load working condition transition characteristics of the mode reference feature, the manifold space construction processing is performed, and the load mode manifold feature is determined; The real-time topological mapping characteristics of the differential pressure signal are used for local feature extraction processing to determine the real-time topological local feature. The local mapping distribution characteristics of the real-time topological local feature in the manifold space are used for feature mapping similarity measurement processing to determine the manifold mapping distance feature. The manifold mapping distance feature is used for load mode optimal matching processing to determine the current differential pressure direction load mode.

7. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, wherein, The determination of the hydraulic control target parameter includes: The load mode transfer path feature is determined by using the working condition transfer characteristics of the current load mode and the historical load mode for load mode transfer path identification processing. The typical mode contrast sample is determined by using the mode transfer path feature for contrast learning sample selection processing. The working condition change prediction feature is determined by using the working condition change sensitive characteristics of the typical mode contrast sample for mode trend difference prediction processing. The hydraulic control target parameter is determined by using the working condition change prediction feature for hydraulic control target parameter selection processing.

8. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 7, characterized by, The generation logic of the working condition change prediction feature includes: The rapid change sensitive sample is determined by using the load working condition rapid change characteristics of the typical mode contrast sample for rapid change sample identification processing. The working condition change direction feature is determined by using the rapid change sensitive sample for working condition change trend direction identification processing. The working condition change prediction feature is obtained by using the working condition change direction feature for working condition trend local difference prediction processing.

9. The hydraulic control optimization method based on differential pressure direction change pattern recognition according to claim 1, wherein, The disturbance observer processing is performed according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, and the mode reference feature is updated according to the observer output, including: The disturbance source feature is determined by using the time domain variation characteristics of the control error for disturbance source dynamic identification processing. The disturbance compensation feature is obtained by using the disturbance source feature in combination with the load response characteristics of the hydraulic actuator for disturbance compensation amount prediction processing. The mode reference feature is updated by using the disturbance compensation feature and the mode reference feature for update fusion processing.

10. 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 according to any one of claims 1 to 9, characterized in that, It includes: The differential pressure topological feature extraction module: the topological reconstruction of the phase space is performed according to the pressure signals of the hydraulic cylinder oil inlet cavity and the oil return cavity, and the topological mapping feature of the differential pressure direction is determined by using the singular spectrum analysis processing of the topological reconstruction result; The multi-scale feature fusion module: the spatial feature is extracted by using the multi-scale convolution network according to the topological mapping feature, and the mode reference feature of the differential pressure direction is determined by using the attention mechanism fusion processing according to the spatial feature; The load mode identification module: the manifold mapping space is constructed according to the mode reference feature, and the load mode of the current differential pressure direction is determined by using the topological mapping characteristics of the real-time differential pressure signal for manifold distance measurement processing; The load trend prediction module: the mode contrast learning mechanism is constructed according to the current load mode and the historical load mode, and the hydraulic control target parameter is determined by using the mode trend prediction processing according to the mode contrast learning result; The disturbance observer and dynamic correction module: the disturbance observer processing is performed according to the control error between the hydraulic cylinder piston rod displacement feedback and the hydraulic control target parameter, and the mode reference feature is updated according to the observer output.

Images