Vibration inversion method based on long and short time memory in generalized orthogonal domain
By using long-term memory networks and orthogonal basis functions to unfold in the generalized orthogonal domain, the vibration response data features are extracted, and the problem of identifying complex dynamic loads in the prior art is solved, and the recognition accuracy and ability of vibration inversion are improved.
Patent Information
- Application Number
- CN202411452854.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing vibration inversion methods are difficult to identify more complex dynamic loads, and problems such as non-convergence of calculations and error amplification hinder the application of dynamic load recognition technology in engineering.
Using the vibration inversion method based on long and short-term memory in the generalized orthogonal domain, the vibration signal is obtained and converted to the generalized orthogonal domain is constructed, and the data features are extracted using the orthogonal basis function to perform dynamics. Calibration and inversion to identify complex dynamic loads.
The vibration inversion recognition accuracy of deep learning methods is improved, and the dynamic loads with more complex forms can be recognized, which reduces the difficulty of model fitting and improves recognition capabilities.
Smart Images

Figure CN118964813A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of vibration inversion and deep learning, and specifically relates to a vibration inversion method based on long short-term memory in a generalized orthogonal domain. Background Art
[0002] Vibration inversion, also known as dynamic load identification, is an identification technology that inversely calculates unknown dynamic loads through structural vibration response information. The essence of traditional inversion methods is a mathematical inverse problem, which inversely calculates the vibration source through the measured structural dynamic response and the dynamic characteristics information of the structure. Since the transfer function matrix composed of the structural dynamic characteristics information is often ill-conditioned, it will bring many problems to vibration inversion, such as non-convergence of calculations and error amplification. At the same time, traditional methods rely heavily on accurate dynamic models of structures. These problems hinder the specific application of dynamic load identification technology in engineering. In recent years, the research on dynamic load identification methods based on deep learning has become increasingly popular. On the one hand, due to the strong modeling ability of deep learning methods, the mapping between vibration response and dynamic load can be directly established, thereby avoiding the operation of mathematical inverse problems; on the other hand, deep learning methods have strong anti-interference ability, and the identification results are more accurate and have strong anti-noise ability. At present, the research on vibration inversion methods based on deep learning is in its infancy. The identification objects are relatively simple dynamic loads such as sine and half-sine waves, and more complex dynamic loads cannot be identified. Summary of the invention
[0003] An object of the present invention is to provide a vibration inversion method based on long short-term memory in a generalized orthogonal domain, which can solve the technical problem that the vibration inversion method in the prior art cannot identify dynamic loads with more complex forms.
[0004] According to a first aspect of the present invention, a vibration inversion method based on long short-term memory in a generalized orthogonal domain is provided, the method comprising:
[0005] Acquiring a first vibration signal for training a long short-term memory model;
[0006] Converting the first vibration signal into a generalized orthogonal domain to obtain an orthogonal basis function matrix composed of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions;
[0007] Constructing input data of a long short-term memory model according to the orthogonal basis function coefficients and the orthogonal basis function matrix;
[0008] Performing forward propagation training and dynamic calibration on the long short-term memory model according to the input data of the long short-term memory model and the orthogonal basis function coefficients;
[0009] acquiring a second vibration signal;
[0010] Calculating the orthogonal basis function coefficients corresponding to the second vibration signal according to the long short-term memory model after the dynamic calibration;
[0011] The dynamic load corresponding to the second vibration signal is determined according to the orthogonal basis function coefficients corresponding to the second vibration signal and the orthogonal basis function matrix.
[0012] Optionally, converting the first vibration signal into a generalized orthogonal domain to obtain an orthogonal basis function matrix consisting of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions includes:
[0013] Determining the expansion order of the first vibration signal converted into the generalized orthogonal domain based on the frequency domain characteristics of the orthogonal basis function;
[0014] The first vibration signal is expanded according to the expansion order in a generalized orthogonal basis function manner to obtain an orthogonal basis function matrix composed of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions.
[0015] Optionally, determining the expansion order of the first vibration signal converted to a generalized orthogonal domain based on the frequency domain characteristics of the orthogonal basis function includes:
[0016] determining a maximum frequency of the first vibration signal;
[0017] Finding a target orthogonal basis function whose main frequency is greater than the highest frequency of the first vibration signal for the first time according to the order of the orthogonal basis function;
[0018] The expansion order is determined according to the order of the target orthogonal basis functions.
[0019] Optionally, determining the expansion order according to the order of the target orthogonal basis functions comprises:
[0020] The order of the target orthogonal basis functions is taken as the expansion order.
[0021] Optionally, determining the expansion order according to the order of the target orthogonal basis functions comprises:
[0022] The order of the target orthogonal basis function is increased by a preset magnitude as the expansion order.
[0023] Optionally, constructing input data of a long short-term memory model according to the orthogonal basis function coefficients and the orthogonal basis function matrix includes:
[0024] The input data of the long short-term memory model is constructed according to the following formula:
[0025] ;
[0026] in, represents the input data of the long short-term memory model, the operator represents element-wise multiplication, represents the orthogonal basis function matrix, Indicates vibration signal, represents the order of orthogonal basis functions.
[0027] Optionally, the generalized orthogonal basis function is a Legendre generalized orthogonal basis function.
[0028] Optionally, the forward propagation process of the long short-term memory model is as follows:
[0029] ;
[0030] ;
[0031] ;
[0032] ;
[0033] ;
[0034] ;
[0035] Among them, the operator represents element-wise multiplication, Represents the long-term information of the model, Indicates short-term information. represents the Sigmoid activation function, represents the information transmitted by the forget gate, Represents the transmission information of the input gate, Represents long-term information that is temporarily stored by the model. Represents the transmission information of the output gate, Represents the final output of long short-term memory, forget gate, input gate, unit state , and the weight matrices of the output gates are , , , , the bias vectors are , , , .
[0036] Optionally, determining the dynamic load corresponding to the second vibration signal according to the orthogonal basis function coefficients corresponding to the second vibration signal and the orthogonal basis function matrix includes:
[0037] The orthogonal basis function coefficients corresponding to the second vibration signal are fitted with the orthogonal basis function matrix to inversely perform the dynamic load corresponding to the second vibration signal. The calculation process is as follows:
[0038] ;
[0039] in, is the orthogonal basis function fitting operator, is the orthogonal basis function matrix used for fitting, is the orthogonal basis function coefficient corresponding to the second vibration signal, For dynamic load.
[0040] Optionally, the long short-term memory model includes a bidirectional LSTM layer, two unidirectional LSTM layers, a fully connected layer including 3 layers of perceptrons, and an output layer, and the number of neurons in each layer of the fully connected layer is 128, 64, and 32, respectively.
[0041] The beneficial effects of the present invention are as follows: based on the long short-term memory neural network, the present invention introduces the orthogonal basis function theory, and performs dynamic load identification in a generalized orthogonal domain. Compared with the existing methods, the characteristics of the vibration response data are extracted by expanding the orthogonal basis function, and dynamic loads with more complex forms can be identified. The method of the present invention extracts the data characteristics of the vibration signal into a generalized orthogonal domain, reduces the fitting difficulty of the deep learning model, thereby improving the recognition accuracy and enhancing the vibration inversion capability of the deep learning method. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a flow chart of a vibration inversion method based on long short-term memory in a generalized orthogonal domain in one embodiment of the present invention.
[0043] Figure 2 It is a flow chart of a vibration inversion method based on long short-term memory in a generalized orthogonal domain in another embodiment of the present invention.
[0044] Figure 3 It is a schematic diagram of a dynamic model of a cantilever beam according to an embodiment of the present invention.
[0045] Figure 4 It is a schematic diagram of the multi-frequency superposition dynamic load identification result in one embodiment of the present invention.
[0046] Figure 5 It is an absolute error diagram of multi-frequency superposition dynamic load identification in one embodiment of the present invention.
[0047] Figure 6 It is a schematic diagram of multi-frequency superposition dynamic load identification results in another embodiment of the present invention.
[0048] Figure 7It is an absolute error diagram of multi-frequency superposition dynamic load identification in another embodiment of the present invention.
[0049] Figure 8 It is a schematic diagram of the identification results of a steady random dynamic load in one embodiment of the present invention.
[0050] Fig. 9 It is an absolute error diagram of stationary random dynamic load identification in one embodiment of the present invention.
[0051] Fig.10 It is a schematic diagram of the identification results of steady-state random dynamic loads in another embodiment of the present invention.
[0052] Fig.11 It is an absolute error diagram for stationary random dynamic load identification in another embodiment of the present invention. DETAILED DESCRIPTION
[0053] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.
[0054] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
[0055] The techniques, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0056] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0057] In the specification and claims of the present invention, if the term "first" or "second" is involved, it may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "plurality" means two or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally means that the objects connected before and after are in an "or" relationship.
[0058] like Figure 1 As shown, this embodiment introduces a vibration inversion method based on long short-term memory in a generalized orthogonal domain, including steps 1100-1700.
[0059] Step 1100: Acquire a first vibration signal for training a long short-term memory model.
[0060] Step 1200: convert the first vibration signal into a generalized orthogonal domain to obtain an orthogonal basis function matrix composed of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions.
[0061] The vibration signal is expanded in the generalized orthogonal domain to obtain a set of orthogonal basis function coefficients, each of which corresponds to an orthogonal basis function, and all orthogonal basis functions constitute an orthogonal basis function matrix. Specifically, the vibration signal can be converted to the generalized orthogonal domain according to the following formula:
[0062] ;
[0063] in, represents the vibration signal to be transformed into the generalized orthogonal domain, are the orthogonal basis function coefficients generated during the transformation process, To expand the order, is the normalized orthogonal basis function.
[0064] When Legendre generalized orthogonal basis functions are used, ,in is the generalized Legendre orthogonal basis function before normalization, is the normalized generalized Legendre orthogonal basis function. The specific expansion formula is as follows:
[0065] ;
[0066] in, , The above formula indicates that in the time interval In the figure, there are specific relations and recursive relations for the nth-order Legendre orthogonal basis functions. is the nth order Legendre orthogonal basis function. It can be seen that the orthogonal basis function order The choice of has a great influence on the expansion of vibration data. If the order is too small, the accuracy will be too low, and if the order is too large, it may cause divergence.
[0067] Specifically, the step 1200 includes steps 1210-1220.
[0068] Step 1210: Determine the expansion order of the first vibration signal converted into the generalized orthogonal domain based on the frequency domain characteristics of the orthogonal basis functions.
[0069] In this embodiment, step 1210 includes steps 1211-1213.
[0070] Step 1211: Determine the highest frequency of the first vibration signal.
[0071] Step 1212: Find a target orthogonal basis function whose main frequency is greater than the highest frequency of the first vibration signal for the first time according to the order of the orthogonal basis function.
[0072] Step 1213: Determine the expansion order according to the order of the target orthogonal basis functions.
[0073] After the target orthogonal basis functions are obtained, the order of the target orthogonal basis functions can be used as the first vibration signal to be converted to the expansion order of the generalized orthogonal domain.
[0074] Furthermore, in order to obtain better fitting accuracy, the order of the target orthogonal basis function can be increased by a preset amplitude and then converted into the expansion order of the generalized orthogonal domain as the first vibration signal. For example, the preset amplitude can be 20%.
[0075] Step 1220: Expand the first vibration signal in a generalized orthogonal basis function manner according to the expansion order to obtain an orthogonal basis function matrix consisting of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions.
[0076] In the vibration inversion process, the frequency domain range covered by the vibration response generally includes the frequency domain range of the vibration source. Therefore, the order of the orthogonal basis function can be determined by the relationship between the highest frequency of the vibration response and the frequency domain characteristics of the orthogonal basis function.
[0077] Step 1300: construct input data of a long short-term memory model according to the orthogonal basis function matrix.
[0078] Specifically, the input data of the long short-term memory model is constructed according to the following formula:
[0079] ;
[0080] in, represents the input data of the long short-term memory model, the operator represents element-wise multiplication, represents the orthogonal basis function matrix, represents the first vibration signal, represents the order of orthogonal basis functions.
[0081] In this embodiment, the long short-term memory model includes a bidirectional LSTM (Long Short-Term Memory) layer, two unidirectional LSTM layers, a fully connected layer including three layers of perceptrons, and an output layer. The number of neurons in each layer of the fully connected layer is 128, 64, and 32, respectively.
[0082] Step 1400: Perform forward propagation training and dynamic calibration on the long short-term memory model according to the input data of the long short-term memory model and the orthogonal basis function coefficients.
[0083] The output of the LSTM model is an orthogonal basis function. The LSTM model is trained with a large amount of data and calibrated dynamically, so that the model has the ability to identify the coefficients of the orthogonal basis functions.
[0084] Step 1500: Acquire a second vibration signal.
[0085] The second vibration signal is a real vibration response, and the dynamic load corresponding to the second vibration signal needs to be calculated.
[0086] Step 1600: Calculate the orthogonal basis function coefficients corresponding to the second vibration signal according to the dynamically calibrated long short-term memory model.
[0087] The second vibration signal Input data assembled with orthogonal basis function matrices Long short-term memory model with incoming calibration number , calculate the orthogonal basis function coefficients corresponding to the second vibration signal , the expression formula is as follows:
[0088] .
[0089] Step 1700: Determine the dynamic load corresponding to the second vibration signal according to the orthogonal basis function coefficients corresponding to the second vibration signal and the orthogonal basis function matrix.
[0090] Specifically, the step 1700 includes: fitting the orthogonal basis function coefficients corresponding to the second vibration signal with the orthogonal basis function matrix to inversely calculate the dynamic load corresponding to the second vibration signal. The calculation process is as follows:
[0091] ;
[0092] in, is the orthogonal basis function fitting operator, is the orthogonal basis function matrix used for fitting, is the orthogonal basis function coefficient corresponding to the second vibration signal, For dynamic load.
[0093] The present invention is based on a long short-term memory neural network, introduces the orthogonal basis function theory, and performs dynamic load identification in a generalized orthogonal domain. Compared with the existing methods, the characteristics of the vibration response data are extracted by expanding the orthogonal basis function, which can realize the identification of dynamic loads with more complex forms.
[0094] In this embodiment, the forward propagation process of the long short-term memory model is as follows:
[0095] ;
[0096] ;
[0097] ;
[0098] ;
[0099] ;
[0100] ;
[0101] Among them, the operator represents element-wise multiplication, Represents the long-term information of the model, Indicates short-term information. represents the Sigmoid activation function, represents the information transmitted by the forget gate, Represents the transmission information of the input gate, Represents long-term information that is temporarily stored by the model. Represents the transmission information of the output gate, Represents the final output of long short-term memory, forget gate, input gate, unit state , and the weight matrices of the output gates are , , , , the bias vectors are , , , .
[0102] This example introduces a multi-frequency superposition dynamic load vibration inversion calculation example. The object is a cantilever beam structure with a length, width, and height of 1m, 0.05m, and 0.01m respectively. Its elastic modulus and density are 71GPa and 2770kg / m 3 The beam structure is divided into 10 units and 11 nodes. The dynamic model is shown in Figure 3 Apply dynamic load to the cantilever beam node 5# And apply dynamic load at the cantilever beam node 11# , nodes 4# and 10# are selected as the collection points of vibration response. The two-point excitation situation is as follows: both excitations are periodic dynamic loads with multiple frequencies superimposed. The frequencies of the first load signal are 4Hz, 8Hz, 12Hz, and 16Hz, and the frequencies of the second load signal are 5Hz, 10Hz, 15Hz, and 20Hz. According to the selection criteria of the order of the orthogonal basis function, a 66-order orthogonal polynomial is intercepted. The sampling rate is 10000Hz, and the analysis time is 1s. The LSTM model contains a bidirectional LSTM layer, two unidirectional LSTM layers, and a fully connected layer containing 3 layers of perceptrons. The number of neurons in each layer of the fully connected layer is 128, 64, and 32 respectively. Finally, an output layer is included to output the coefficients of the orthogonal polynomial. The results of multi-frequency superposition dynamic load identification are shown in Figures 4 to 7 ,in Figure 4 for Dynamic load identification results, Figure 5 for Absolute error diagram of dynamic load identification, Figure 6 for Dynamic load identification results, Figure 7 for Absolute error diagram of dynamic load identification. From the identification results, it can be seen that in the generalized orthogonal domain, the vibration inversion method based on long short-term memory has good adaptability to the problem of multi-frequency superposition periodic dynamic load identification, the error of the identification result is acceptable, and the load curve has a high degree of fit.
[0103] This embodiment introduces a steady random dynamic load vibration inversion example. This example controls the frequency range of the random dynamic load signal within 0-50Hz, and selects a 152-order orthogonal polynomial based on the interception criterion. The load application point and vibration response pick-up point are the same as those in the above multi-frequency superposition dynamic load vibration inversion example. The construction of the long short-term memory model is the same as that in the above multi-frequency superposition dynamic load vibration inversion example. The identification results of the steady random dynamic load are shown in Figures 8 to 11 ,in Figure 8 for Dynamic load identification results, Fig. 9 for Absolute error diagram of dynamic load identification, Fig.10 for Dynamic load identification results, Fig.11 for Absolute error diagram of dynamic load identification. From the identification results, it can be seen that this method also has good adaptability to the inversion problem of multi-point stationary random dynamic loads. The identification results are good, the curve fit is high, and the error is within the acceptable range of engineering.
[0104] These two examples illustrate that the method proposed in the present invention has good applicability to vibration inversion problems with more complex dynamic load forms and is an effective and reliable vibration inversion method.
[0105] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will be appreciated by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.
[0106] Those of ordinary skill in the art will appreciate that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0108] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0109] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.
[0110] In addition, each functional module in the embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0111] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the energy-saving signal sending / receiving method of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.
[0112] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.
[0113] It should be understood that the size of the sequence number of each step in the content of the invention and the embodiments of the present invention does not absolutely mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. For the purpose of example and description, the foregoing description of the implementation of the present disclosure has been given. The foregoing description is not exhaustive and is not intended to limit the present disclosure to the exact form disclosed. Various deformations and modifications may exist according to the above teachings, or various deformations and modifications may be obtained from the practice of the present disclosure. These embodiments are selected and described to illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can use the present disclosure in various embodiments and various modifications suitable for the specific purpose conceived.
Claims
1. A vibration inversion method based on long short-term memory in a generalized orthogonal domain, characterized in that: The method comprises: Acquiring a first vibration signal for training a long short-term memory model; Converting the first vibration signal into a generalized orthogonal domain to obtain an orthogonal basis function matrix composed of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions; Constructing input data of a long short-term memory model according to the orthogonal basis function matrix; Performing forward propagation training and dynamic calibration on the long short-term memory model according to the input data of the long short-term memory model and the orthogonal basis function coefficients; acquiring a second vibration signal; Calculating the orthogonal basis function coefficients corresponding to the second vibration signal according to the long short-term memory model after the dynamic calibration; The dynamic load corresponding to the second vibration signal is determined according to the orthogonal basis function coefficients corresponding to the second vibration signal and the orthogonal basis function matrix.
2. The method according to claim 1, characterized in that The first vibration signal is converted into a generalized orthogonal domain to obtain a set of orthogonal basis function coefficients and an orthogonal basis function matrix composed of corresponding orthogonal basis functions, including: Determining the expansion order of the first vibration signal converted into the generalized orthogonal domain based on the frequency domain characteristics of the orthogonal basis function; The first vibration signal is expanded according to the expansion order in a generalized orthogonal basis function manner to obtain an orthogonal basis function matrix composed of a set of orthogonal basis function coefficients and corresponding orthogonal basis functions.
3. The method according to claim 2, characterized in that The step of determining the expansion order of the first vibration signal converted to a generalized orthogonal domain based on the frequency domain characteristics of the orthogonal basis function comprises: determining a maximum frequency of the first vibration signal; Finding a target orthogonal basis function whose main frequency is greater than the highest frequency of the first vibration signal for the first time according to the order of the orthogonal basis function; The expansion order is determined according to the order of the target orthogonal basis functions.
4. The method according to claim 3, characterized in that Determining the expansion order according to the order of the target orthogonal basis functions comprises: The order of the target orthogonal basis functions is taken as the expansion order.
5. The method according to claim 3, characterized in that: Determining the expansion order according to the order of the target orthogonal basis functions comprises: The order of the target orthogonal basis function is increased by a preset magnitude as the expansion order.
6. The method according to claim 3, characterized in that The input data of the long short-term memory model is constructed according to the orthogonal basis function matrix, including: The input data of the long short-term memory model is constructed according to the following formula: ; in, represents the input data of the long short-term memory model, the operator represents element-wise multiplication, represents the orthogonal basis function matrix, represents the first vibration signal, represents the order of orthogonal basis functions.
7. The method according to claim 3, characterized in that The generalized orthogonal basis functions are Legendre generalized orthogonal basis functions.
8. The method according to claim 1, characterized in that The forward propagation process of the long short-term memory model is as follows: ; ; ; ; ; ; Among them, the operator represents element-wise multiplication, Represents the long-term information of the model, Indicates short-term information. represents the Sigmoid activation function, represents the information transmitted by the forget gate, Represents the transmission information of the input gate, Represents long-term information that is temporarily stored by the model. Represents the transmission information of the output gate, Represents the final output of long short-term memory, forget gate, input gate, unit state , and the weight matrices of the output gates are , , , , the bias vectors are , , , .
9. The method according to claim 1, characterized in that: The determining the dynamic load corresponding to the second vibration signal according to the orthogonal basis function coefficients corresponding to the second vibration signal and the orthogonal basis function matrix includes: The orthogonal basis function coefficients corresponding to the second vibration signal are fitted with the orthogonal basis function matrix to inversely perform the dynamic load corresponding to the second vibration signal. The calculation process is as follows: ; in, is the orthogonal basis function fitting operator, is the orthogonal basis function matrix used for fitting, is the orthogonal basis function coefficient corresponding to the second vibration signal, For dynamic load.
10. The method according to claim 1, characterized in that The long short-term memory model includes a bidirectional LSTM layer, two unidirectional LSTM layers, a fully connected layer including three layers of perceptrons, and an output layer. The number of neurons in each layer of the fully connected layer is 128, 64, and 32, respectively.
Citation Information
Patent Citations
Frequency domain order determination method for generalized orthogonal polynomial dynamic load identification
CN110135095A
BiLSTM-based composite characteristic optical fiber sensing disturbance signal mode identification method
CN111104891A
Dynamic load identification method, apparatus and device, and computer storage medium
CN117113079A
Dynamic load identification method based on orthogonal basis function and convolutional neural network
CN118013218A
