Workload analysis method, device and system based on load spectrum

Abstract

The application provides a kind of based on load spectrum's job load analysis method, device and system, the method comprises: obtaining the measured load spectrum of the job machine to be evaluated;Measured load spectrum is input to analysis model, and the load job fragment corresponding to each job intensity level of the job machine to be evaluated output by analysis model is obtained.The application provides based on load spectrum's job load analysis method, device and system, based on measured load spectrum as the input of analysis model, the result output is the load job fragment belonging to different job intensity levels identified in long sequence measured load spectrum.By mining and analyzing typical characteristics, the instantaneous job intensity corresponding to long distance and large data volume load spectrum can be effectively identified, the problem that load spectrum job intensity is not clear due to serious load spectrum data redundancy and load spectrum cutting does not meet the requirements of heavy load working condition is solved, and the efficiency and accuracy of spectrum analysis are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent agricultural technology, and particularly relates to a work load analysis method, device and system based on a load spectrum. BACKGROUND

[0002] Reliability is an important foundation to support the development of large-scale, intelligent and unmanned agricultural equipment. As an important method of agricultural equipment reliability test and verification means, load spectrum technology has become a research hotspot in the direction of positive design and verification of agricultural equipment, because it can directly process the basic load data of field work.

[0003] At present, the load spectrum has developed rapidly thanks to the popularization and application of intelligent measurement and control technology. However, the introduction of advanced technology has caused the redundancy of multi-parameter and multi-dimensional load spectrum data due to the variety of measurable parameters and the excessively high sampling frequency during load spectrum measurement. In addition, even when a tractor is working in the same field, it cannot avoid encountering various soil state parameters and ground relief characteristics, which cause different plowing depths in a single work of the machine set. Therefore, it cannot guarantee that the work load corresponding to the load spectrum is always the same, that is, the work intensity is different.

[0004] If the artificial method is used to process each section of the load spectrum to verify whether it is in high work intensity or in the same work intensity, it is obviously not desirable because the tractor works in a complex field terrain with transient resistance, and it needs to pay attention to every frame of data of each sensor. Obviously, the accuracy is insufficient, the workload is too large and it is not universal. If the traditional theoretical method is used to identify the typical load spectrum corresponding to the high-intensity work of the tractor in the large sample of the load spectrum, the mechanism of the agricultural implement, PTO shaft, electro-hydraulic suspension system and other mechanisms needs to be simplified to clarify its load transfer mechanism and characteristics, and simulation analysis is carried out combined with real load spectrum data that can reflect fatigue. This has two problems: first, the model is large, and the simplification of the key mechanism will cause the model deviation to increase. Second, the real load spectrum data is generally prepared based on the idea of equivalent damage, and many load characteristics that can represent the working conditions and work intensity are lost in the processing process. It can be seen that the existing research method has great limitations in actual engineering application, and cannot guarantee its accuracy and applicability. SUMMARY

[0005] The present application provides a work load analysis method, device and system based on a load spectrum to solve the defect that the accuracy and applicability of load spectrum analysis in the prior art cannot be considered at the same time.

[0006] The present application provides a work load analysis method based on a load spectrum, comprising:

[0007] obtaining a measured load spectrum of an evaluation work machine;

[0008] inputting the measured load spectrum into the analysis model to obtain a load operation segment corresponding to each operation intensity level of the to-be-evaluated operation machine output by the analysis model;

[0009] The measured load spectrum is a sampling value corresponding to at least one test parameter of the to-be-evaluated operation machine within a preset period of time; and the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and an operation intensity level corresponding to the sample load spectrum segment.

[0010] According to the load spectrum-based operation load analysis method provided in the present application, the analysis model comprises a feature extraction layer, a segment recognition layer and a segment classification layer.

[0011] The measured load spectrum is input into the feature extraction layer to obtain an encoding combination output by the feature extraction layer.

[0012] The encoding combination is input into the segment recognition layer to obtain the segment feature information output by the segment recognition layer.

[0013] The segment feature information is input into the segment classification layer to obtain the operation intensity level and the load operation segment corresponding to the operation intensity level output by the segment classification layer.

[0014] According to the load spectrum-based operation load analysis method provided in the present application, the encoding combination is input into the segment recognition layer to obtain the segment feature information output by the segment recognition layer, which comprises:

[0015] The encoding combination is subjected to feature extraction to obtain a global feature map.

[0016] The global feature map is subjected to feature correlation by using an attention mechanism to obtain the segment feature information.

[0017] According to the load spectrum-based operation load analysis method provided in the present application, the global feature map is subjected to feature correlation by using an attention mechanism to obtain the segment feature information, which comprises:

[0018] The global feature map is input into a recurrent neural network model to obtain an intermediate hidden state.

[0019] According to a multi-head attention mechanism, the weight of each intermediate hidden state is determined.

[0020] Based on the weight of each intermediate hidden state, the segment feature information output by the recurrent neural network model is obtained.

[0021] The recurrent neural network model comprises one of a long short-term memory network (LSTM), a gated recurrent unit (GRU) and an LSTM variant.

[0022] According to the load spectrum-based work load analysis method provided in the application, the measured load spectrum is input into the feature extraction layer, and an encoding combination output by the feature extraction layer is obtained, comprising:

[0023] The encoding combination is determined based on a sampling value of each test parameter in the measured load spectrum and a standard value of the test parameter at each level.

[0024] The application further provides a load spectrum-based work load analysis device, comprising:

[0025] A data acquisition module is configured to acquire a measured load spectrum of a work machine to be evaluated.

[0026] An analysis module is configured to input the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model.

[0027] The measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period; and the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a work intensity level annotated corresponding to the sample load spectrum segment.

[0028] The application further provides a load spectrum-based work load analysis system, comprising a work machine provided with a sensing module and a remote terminal, wherein the remote terminal is configured to execute the load spectrum-based work load analysis method according to any one of the above.

[0029] The sensing module is configured to send a measured load spectrum collected from the work machine to the remote terminal when it is determined that the work machine is in a working state.

[0030] The remote terminal is configured to input the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model.

[0031] The measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine within a preset time period; and the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a work intensity level annotated corresponding to the sample load spectrum segment.

[0032] The application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the load spectrum-based job load analysis method according to any one of the above when executing the program.

[0033] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the load spectrum-based job load analysis method according to any one of the above.

[0034] The application further provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor to implement the load spectrum-based job load analysis method according to any one of the above.

[0035] The application provides a load spectrum-based job load analysis method, device and system, which takes a measured load spectrum as an input of an analysis model, and outputs a long-sequence measured load spectrum in which load job segments belonging to different job intensity levels are identified. The method can effectively identify the instantaneous job intensity corresponding to a long-distance and large-data load spectrum by mining and analyzing typical features, and solves the problems of unclear load spectrum job intensity identification and non-compliance of load spectrum cutting with heavy load working condition requirements due to serious load spectrum data redundancy, greatly improving the efficiency and accuracy of spectrum analysis. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Figure 1 is a flowchart of the load spectrum-based job load analysis method provided by the application;

[0038] Figure 2 is a structural schematic diagram of the load spectrum-based job load analysis device provided by the application;

[0039] Figure 3 is a structural schematic diagram of the load spectrum-based job load analysis system provided by the application;

[0040] Figure 4 is a structural schematic diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0041] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0042] It should be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0043] The terms "comprise" and "include" indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0044] Figure 1 is a flowchart of the load spectrum-based work load analysis method provided by the present application. As shown in Figure 1 The load spectrum-based work load analysis method provided by the embodiment of the present application comprises the following steps: step 101, obtaining a measured load spectrum of a work machine to be evaluated.

[0045] The measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period.

[0046] It should be noted that the execution subject of the load spectrum-based work load analysis method provided by the embodiment of the present application is a load spectrum-based work load analysis device.

[0047] The application scenario of the load spectrum-based work load analysis method provided by the embodiment of the present application is to analyze a series of continuous multi-parameter multi-dimensional load spectrum of the work machine under any working condition, and identify the load segments of different work intensities.

[0048] The work machine includes but is not limited to a tractor, a tillage machine, a rotary tillage machine, and the like agricultural machinery.

[0049] The load spectrum-based work load analysis method provided by the embodiment of the present application is suitable for a user to analyze the load data of the work machine of the user continuously running under any working condition through an electronic device.

[0050] The electronic device described in the embodiments of the present application can be implemented in various forms. For example, the electronic device described in the embodiments of the present application can include a mobile terminal such as a mobile phone, a smart phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a PMP (Portable Multimedia Player), a navigation device, a smart bracelet, a smart watch, and the like, and a stationary terminal such as a digital TV, a desktop computer, and the like. Hereinafter, it is assumed that the electronic device is a mobile terminal. However, those skilled in the art will appreciate that the configuration according to the embodiments of the present application can also be applied to a stationary type terminal, except for elements particularly used for mobile purposes.

[0051] Specifically, in step 101, the work load analysis device based on load spectrum receives a measured load spectrum of the running state of each braking component of the work machine to be evaluated over time when the work machine performs work within a preset period.

[0052] The measured load spectrum refers to the sampling of test parameters by the sensing module provided in each braking component, and the integration of the sampling values of each test parameter changing over time into the measured load spectrum under the corresponding working condition of the work machine to be evaluated.

[0053] The embodiments of the present application do not make specific limitations on the acquisition method of the measured load spectrum.

[0054] Optionally, the work load analysis device based on load spectrum can receive the measured load spectrum transmitted by the user terminal.

[0055] Optionally, the work load analysis device based on load spectrum can receive the sensing data transmitted in real time by the sensing module of the currently running work machine to be evaluated, and then form the measured load spectrum by changing the multi-dimensional sensing data over time.

[0056] Step 102, input the measured load spectrum into the analysis model to obtain the load work segment corresponding to each work intensity level of the work machine to be evaluated output by the analysis model.

[0057] The analysis model is trained based on the sample load spectrum, the sample load spectrum segment contained in the sample load spectrum, and the work intensity level annotated by the sample load spectrum segment.

[0058] It should be noted that the analysis model can be a neural network model, and the structure and parameters of the neural network include but are not limited to the number of layers of the input layer, the hidden layer and the output layer of the neural network, and the weight parameters of each layer. The embodiments of the present application do not make specific limitations on the type and structure of the neural network.

[0059] For example, the analysis model can be a neural network model composed of an input layer, a hidden layer and an output layer, wherein: For example, the analysis model can be a neural network model composed of an input layer, a hidden layer and an output layer, wherein:

[0060] The input layer is at the front end of the entire network, directly measures the load spectrum and digitizes and encodes it.

[0061] The hidden layer can have one or more layers and operates on the input vector by weighted summation of neurons in the hidden layer.

[0062] The output layer is the last layer and is used to decode the vector obtained by the weighted summation to analyze the load spectrum segment corresponding to different work intensity from the long sequence load spectrum.

[0063] It should be noted that the sample data includes a sample load spectrum corresponding to the sample data, sample load spectrum segments at different time sequences in the sample load spectrum, and work intensity levels marked in the sample load spectrum segments, respectively. The sample data is divided into a training set and a test set according to a certain proportion.

[0064] The proportion of the training set and the test set in the sample data includes but is not limited to 9:1, 8:2, etc., and the embodiments of the present application do not make specific limitations.

[0065] The embodiments of the present application do not make specific limitations on the labeling method of the training set.

[0066] Optionally, a long sequence sample load spectrum or a plurality of continuous short sequence sample load spectrum segments obtained by cutting can be obtained under a specified work intensity level.

[0067] Exemplarily, according to the standard for tractor rotary tillage operation in the industry, 3 parameters such as tillage depth, tillage depth stability, and soil crushing rate are selected as quantitative indicators of work intensity levels. The tillage depth, tillage depth stability, and soil crushing rate parameters corresponding to the sample load spectrum can be used to refer to the evaluation indicators of the work intensity level for weighted integration, and the work intensity level is mapped to 3 levels of large, medium, and small corresponding to three score intervals (for example, the interval is divided into 100-80 points, 80-60 points, and 60-0 points). The evaluation indicators and integral methods are shown in Table 1.

[0068] Table 1 Work intensity evaluation indicators and scoring methods

[0069]

[0070] Optionally, the sample load spectrum can also be obtained by controlling the work machine to perform different work intensity levels at different time sequences during the acquisition process to form sample load spectrum segments corresponding to different work intensity levels at corresponding time sequences.

[0071] Specifically, in step 102, the weight coefficients between the layers of the constructed analysis model are initialized by the job load analysis device based on the load spectrum, and then a group of sample load spectra in the training set and the sample load spectrum segments with the labeled job intensity levels are input to the neural network under the current weight coefficients, and the outputs of the nodes in the input layer, the hidden layer and the output layer are calculated in turn. The accumulated error between the final output result of the output layer and the actual connection position state type is corrected according to the gradient descent method, and the weight coefficients between the nodes in the input layer and the hidden layer are corrected. According to the above process, the weight coefficients of the input layer and the hidden layer can be obtained by traversing all the samples in the training set.

[0072] According to the weight coefficients of the input layer and the hidden layer of the neural network, the analysis model in step 102 is restored by the job load analysis device based on the load spectrum, and a group of measured load spectra in the test set is input to the trained analysis model, so that the load job segments in different job intensity levels in the job process corresponding to the measured load spectrum can be obtained.

[0073] In the embodiment of the application, the measured load spectrum is taken as the input of the analysis model, and the output result is the load job segments belonging to different job intensity levels identified in the long sequence of measured load spectra. By mining and analyzing the typical characteristics, the instantaneous job intensity corresponding to the long-distance and large-data load spectrum can be effectively identified, the problems of unclear load spectrum job intensity identification and load spectrum cutting not meeting the requirements of heavy load working conditions caused by serious load spectrum data redundancy are solved, and the efficiency and accuracy of spectrum analysis are greatly improved.

[0074] On the basis of any of the above embodiments, the analysis model comprises a feature extraction layer, a segment identification layer and a segment classification layer.

[0075] The measured load spectrum is input to the feature extraction layer to obtain the code combination output by the feature extraction layer.

[0076] The code combination is input to the segment identification layer to obtain the segment feature information output by the segment identification layer.

[0077] The segment feature information is input to the segment classification layer to obtain the job intensity level output by the segment classification layer and the load job segment corresponding to the job intensity level.

[0078] Specifically, the analysis model built in the job load analysis device based on the load spectrum is composed of an input layer, a hidden layer and an output layer. The hidden layer is used for feature extraction and association of the measured load spectrum, and is used for identifying the load job segments with obvious working characteristics and the job intensity levels corresponding to the segments.

[0079] The structure of the hidden layer in the embodiment of the application is not specifically limited.

[0080] Preferably, the hidden layer comprises at least three layers, namely a feature extraction layer, a segment recognition layer and a segment classification layer.

[0081] The feature extraction layer is used for feature extraction on the measured load spectrum, so as to represent the multi-dimensional and continuous sensing data in the measured load spectrum by means of unstructured encoding combination.

[0082] The segment sorting layer is used for feature reuse and dense connection on the encoding combination in the channel dimension corresponding to each test parameter, so as to optimize the transmission mechanism and utilization efficiency of the features, and intercept the segment feature information corresponding to the time period with obvious work characteristics.

[0083] The segment feature information is composed of the feature vectors of each dimension of the work machine in the time period with obvious work characteristics.

[0084] The answer conversion layer is used for global average pooling of the segment feature information into a fully connected network, and the last fully connected layer uses the Softmax classifier to obtain the normalized probability distribution of multiple farming loads and work intensities, so as to output the work intensity grade analyzed by the segment feature information, and store the load work segment corresponding to the segment feature information in time sequence on the measured load spectrum into the classification corresponding to the work intensity grade.

[0085] The embodiment of the present application extracts features from the measured load spectrum based on the feature extraction layer, processes the encoding combination output by the feature extraction layer through the segment recognition layer, and then classifies the intercepted multiple segment feature information to the corresponding work intensity grade through the segment classification layer, so as to output the load work segment corresponding to the work intensity grade. By mining and analyzing typical features, the instantaneous work intensity corresponding to the long-distance and large-data load spectrum can be effectively identified, the problems of unclear load spectrum work intensity identification and load spectrum interception not meeting the requirements of heavy load working conditions caused by serious load spectrum data redundancy are solved, the efficiency and accuracy of spectrum analysis are greatly improved, and the spectrum analysis device has more universality.

[0086] On the basis of any of the above embodiments, the encoding combination is input into the segment recognition layer, and the segment feature information output by the segment recognition layer is obtained, including: feature extraction on the encoding combination to obtain a global feature map.

[0087] Specifically, the work load analysis device based on the load spectrum takes the time sequence of the continuous load sequence as the core, extracts the internal features in the load sequence based on the context information contained in the encoding combination, and obtains a global feature map containing each test parameter dimension.

[0088] The extraction process of the global feature map in the embodiment of the present application is not specifically limited.

[0089] Exemplarily, the work load analysis based on the load spectrum disassembles the original load sequence corresponding to the encoding combination in the time and amplitude dimensions, and the output tensor generated is sequentially processed by a batch normalization (BN) layer, a Dropout layer and a Reshape layer, and finally a global feature map of the load data is generated.

[0090] The BN layer can normalize each layer output, and then restore the data features before normalization through the parameters learned in the BN.

[0091] The Dropout layer randomly reduces some neurons in each batch of training, and is used to reduce the overfitting of the neural network.

[0092] The Reshape layer can rearrange a multi-dimensional matrix of a certain dimension to construct a new matrix of the same number of elements but different dimension sizes, so as to finally form a global feature map representing global features.

[0093] The global feature map is associated with features by using an attention mechanism to obtain segment feature information.

[0094] Specifically, the work load analysis device based on the load spectrum analyzes the correlation of the internal features of the global feature map to generate an attention matrix, and different weights are assigned to each feature according to the importance of the features to automatically extract the most obvious feature representation, so as to obtain segment feature information with obvious work features.

[0095] The embodiment of the application can improve the prediction accuracy and calculation efficiency by deeply mining the internal local features contained in the front and rear loads in the load change process of the encoding combination, and using the attention mechanism to extract the most obvious segment feature information from the obtained global feature map and transmit it to the classifier. The work load analysis device based on the load spectrum can accurately and efficiently identify the load work segments under each work intensity level in the long sequence load spectrum of the tractor.

[0096] On the basis of any of the above embodiments, the global feature map is input into a recurrent neural network model to obtain intermediate hidden states by using an attention mechanism to associate features and obtain segment feature information.

[0097] According to the multi-head attention mechanism, the weight of each intermediate hidden state is determined.

[0098] Based on the weight of each intermediate hidden state, the segment feature information output by the recurrent neural network model is obtained.

[0099] The recurrent neural network model includes one of a long short-term memory network (LSTM), a gated recurrent unit (GRU) and a variant of the LSTM.

[0100] It should be noted that the recurrent neural network model provided in the load spectrum-based operation load analysis device will bring the output result of the previous time to the next hidden layer for training each time.

[0101] The embodiment of the application does not specifically limit the recurrent neural network model.

[0102] Optionally, the recurrent neural network model can be a long short-term memory (LSTM) network, which introduces a memory cell for recording additional information. In order to control the memory cell, three types of gates need to be set.

[0103] The output gate is used to output entries from the cell.

[0104] The input gate is used to determine when to read data into the cell.

[0105] The forget gate is used to reset the content of the cell through a mechanism. The motivation of this design is the same as that of the gated recurrent unit, which can determine when to remember or ignore the input in the hidden state through a dedicated mechanism.

[0106] Optionally, the recurrent neural network model can be a gated recurrent unit (GRU), which combines the forget gate and the input gate into an "update gate". It combines the neuron state and the hidden layer state, and also makes some adaptive changes.

[0107] Optionally, the recurrent neural network model can be a variant of LSTM.

[0108] Exemplarily, the recurrent neural network model can be a ConvLSTM network, which is composed of a memory cell and three gates (input gate, forget gate, and output gate). ConvLSTM uses convolution operations instead of fully connected operations in the original LSTM, and relies on the local feature capturing characteristics of the CNN network to improve the learning and extraction ability of the training network for internal features of long sequence loads. Therefore, ConvLSTM can be expressed as

[0109] i t =σ(W xi *X t +W hi *H t-1 +W ci ⊙C t-1 +b i )

[0110] f t =σ(W xf *Xt +W hf *H t-1 +W cf ⊙C t-1 +b f )

[0111] C t =f t ⊙C t-1 +i t ⊙tanh(W xc *X t +W hc *H t-1 +b c )

[0112] o t =σ(W xo *X t +W ho *H t-1 +W co ⊙C t-1 +b f )

[0113] H t =o t ⊙tanh(C t )

[0114] where * and ⊙ denote convolution and Hadamard product operations, respectively. X t is the input tensor contained in the global feature map. H t is the hidden state tensor. C t is the memory cell tensor, W x *and W h *are two-dimensional convolution kernels corresponding to the input state and the hidden state, respectively, and b i , b f , b o , and b c are bias terms.

[0115] Specifically, the work load analysis device based on the load spectrum inputs the global feature map into a recurrent neural network model for processing, takes each feature vector as a node, and can obtain the corresponding hidden state of each feature dimension. The hidden state of the node is used to predict the hidden state of the next node output, and the hidden state of the last node is output as the corresponding intermediate hidden state of the global feature map. The intermediate hidden state corresponding to each time is continuously recycled, the weights and biases of the recurrent neural network model are obtained, and the trained model parameters are used to configure the recurrent neural network model to perform time series association. Subsequently, the load spectrum work intensity features output by the ConvLSTM are automatically weighted and distributed by combining the multi-head attention module, and the most obvious segment feature information is transmitted to the classifier.

[0116] The embodiment of the application inputs the global feature map into the recurrent neural network model, recursively cycles through the intermediate hidden state, and obtains key features from multiple angles and multiple levels through multiple parallel calculations by the multi-head attention mechanism, thereby strengthening the internal relationship of any feature in long-distance and large-data load, greatly improving the recognition of segment feature information with obvious work characteristics in the load spectrum, and improving the calculation speed and load correlation of the model, and improving the efficiency of load spectrum analysis.

[0117] On the basis of any of the above embodiments, the measured load spectrum is input into the feature extraction layer to obtain the encoding combination output by the feature extraction layer, including: determining the encoding combination based on the sampling value of each test parameter in the measured load spectrum and the standard value of the test parameter at each level.

[0118] It should be noted that before step 102, evaluation indexes for test parameters need to be developed according to the to-be-evaluated work machine and relevant industry standards.

[0119] The evaluation indexes divide each test environment parameter involved in the to-be-evaluated work machine into multiple levels according to relevant industry standards, and the data of multiple test environment parameters at each level corresponds to one running evaluation condition.

[0120] For example, taking typical tillage working conditions such as rotary tillage and plowing of tractors as examples, the driving state, power output shaft, and suspension mechanism of the tractor are sampled by the sensing module arranged on the tractor, and the test parameters and matching sensors involved are shown in Table 2.

[0121] Table 2 Test system test parameters and sensors

[0122]

[0123]

[0124] Specifically, the operation load analysis device based on a load spectrum sequentially extracts features of each test parameter in the measured load spectrum according to evaluation indexes to obtain an encoding combination in a continuous running state.

[0125] The encoding combination is a set of feature values of the sampling values of the test parameters.

[0126] The encoding mode includes but is not limited to One-Hot encoding, label-encoding, and self-defined encoding, and the present embodiment does not make specific limitation on the encoding.

[0127] The present embodiment encodes the standard values and the sampling values of the test parameters in each level to obtain the corresponding encoding combination. The high-dimensional sparse feature vector is converted into a low-dimensional dense feature vector, and the load spectrum can be quickly and accurately analyzed through dimension reduction processing of the data.

[0128] Figure 2 FIG. 1 is a structural schematic diagram of the operation load analysis device based on a load spectrum provided by the present application. As shown in FIG. 1, the device includes a data acquisition module 210 and an analysis module 220, wherein: Figure 2

[0129] The data acquisition module 210 is configured to acquire a measured load spectrum of an evaluated working machine.

[0130] The analysis module 220 is configured to input the measured load spectrum into an analysis model to obtain load operation segments corresponding to each working intensity level of the evaluated working machine output by the analysis model.

[0131] The measured load spectrum is a sampling value corresponding to at least one test parameter of the evaluated working machine in a preset time period. The analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a working intensity level corresponding to the sample load spectrum segment.

[0132] Specifically, the data acquisition module 210 and the analysis module 220 are sequentially electrically connected.

[0133] The data acquisition module 210 receives a measured load spectrum of each brake component of the evaluated working machine in a preset time period.

[0134] ​The analysis module 220 initializes the weight coefficients between the layers of the constructed analysis model, then inputs a group of sample load spectra in the training set and the sample load spectrum segments with labeled work intensity levels to the neural network under the current weight coefficients, and sequentially calculates the outputs of the nodes in the input layer, the hidden layer and the output layer. The cumulative error between the final output result of the output layer and the actual connection position state type is corrected according to the gradient descent method, and the weight coefficients between the nodes in the input layer and the hidden layer are corrected. According to the above process, until all the samples in the training set are traversed, the weight coefficients of the input layer and the hidden layer can be obtained.

[0135] Then, according to the weight coefficients of the input layer and the hidden layer of the neural network, the analysis model is restored, and a group of measured load spectra in the test set is input to the trained analysis model, and the load work segments in different work intensity levels corresponding to the measured load spectrum in the work process can be obtained.

[0136] Optionally, the analysis module 220 comprises a feature extraction unit, a segment recognition unit and a segment classification unit, wherein:

[0137] The feature extraction unit is configured to input the measured load spectrum to the feature extraction layer and obtain the encoding combination output by the feature extraction layer.

[0138] The segment recognition unit is configured to input the encoding combination to the segment recognition layer and obtain the segment feature information output by the segment recognition layer.

[0139] The segment classification unit is configured to input the segment feature information to the segment classification layer, and obtain the work intensity level and the load work segment corresponding to the work intensity level output by the segment classification layer.

[0140] Optionally, the segment recognition unit comprises a feature extraction subunit and a feature correlation subunit, wherein:

[0141] The feature extraction subunit is configured to perform feature extraction on the encoding combination and obtain a global feature map.

[0142] The feature correlation subunit is configured to perform feature correlation on the global feature map by using an attention mechanism and obtain the segment feature information.

[0143] Optionally, the feature correlation subunit comprises an intermediate hidden state obtaining micro unit, a weight updating micro unit and an attention distribution micro unit, wherein:

[0144] The intermediate hidden state obtaining micro unit is configured to input the global feature map to a recurrent neural network model and obtain an intermediate hidden state.

[0145] The weight updating micro unit is configured to determine the weight of each intermediate hidden state according to a multi-head attention mechanism.

[0146] The attention distribution micro unit is used to obtain the segment feature information output by the recurrent neural network model based on the weight of each intermediate hidden state.

[0147] The recurrent neural network model comprises one of a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a variant of the LSTM.

[0148] Optionally, the feature extraction unit is specifically configured to determine the code combination based on the sampling value of each test parameter in the measured load spectrum and the standard value of the test parameter at each level.

[0149] The load spectrum-based work load analysis device provided in the embodiments of the present application is used to execute the load spectrum-based work load analysis method provided in the present application, and the implementation manner and the same beneficial effects can be achieved as the implementation manner of the load spectrum-based work load analysis method provided in the present application, and thus details are not repeated here.

[0150] The embodiments of the present application take the measured load spectrum as the input of the analysis model, and the output result is the load work segments belonging to different work intensity levels identified in the long sequence of the measured load spectrum.

[0151] Figure 3 is a structural schematic diagram of the load spectrum-based work load analysis system provided in the present application. Figure 3 As shown in the above any embodiment, the system comprises a work machine 310 provided with a sensing module 311 and a remote terminal 320, and the remote terminal 320 is used to execute the load spectrum-based work load analysis method.

[0152] The sensing module 311 is used to send the measured load spectrum collected by the work machine 310 to the remote terminal 320 in a case where it is determined that the work machine 310 is in a working state.

[0153] The remote terminal 320 is used to input the measured load spectrum into an analysis model to obtain the load work segments corresponding to each work intensity level of the work machine 310 to be evaluated output by the analysis model.

[0154] The measured load spectrum is the sampling value corresponding to at least one test parameter of the work machine 310 within a preset time period. The analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a work intensity level annotated by the sample load spectrum segment.

[0155] Specifically, the work load analysis system based on load spectrum analyzes and monitors a plurality of work machines 310 in which the at least one remote terminal 320 is accommodated.

[0156] Each work machine 310 is in communication connection with the remote terminal 320 through a built-in sensing module 311.

[0157] The sensing module 311 can be in long-distance transmission communication through wireless communication technology (Wi-Fi), Bluetooth or serial communication, and transmit the measured load spectrum of the work machine 310 in which the sensing module 311 is located under the corresponding working condition to the remote terminal 320, so that the remote terminal 320 analyzes the measured load spectrum by using an analysis model that combines the ConvLSTM network structure and the Multi-Head Attention mechanism, maintains the time sequence of the continuously changing load sequence as the basic principle, deeply mines the internal local features contained in the front and rear loads in the load change process, and strengthens the matching relationship of the internal features in the long-distance large data load, so as to store the load work segments belonging to the corresponding work intensity level separately.

[0158] For example, the work load analysis system based on load spectrum can be used for three work machines 310 equipped with the same sensing module 311 and selected as the control group analysis model of LSTM and ConvLSTM.

[0159] The accuracy of the MHA-ConvLSTM analysis model provided by the embodiment of the application reaches 97.69%, which is 11.96% and 6.71% higher than that of the control group model, and has a large increase in precision, recall rate and F1 value compared with the control group, the F1 value reaches 97.83%, and the operation time is only 0.5289s, which has good real-time performance.

[0160] Therefore, the overall performance of the MHA-ConvLSTM large work load segment recognition model is better than that of the control group model, and meets the application requirements.

[0161] In addition, the identification effect is verified by using a series of load spectrum with different operation intensity, and the series of load spectrum is decomposed into three operation intensity corresponding load spectrum segments by artificial processing, and the number of segments is 399, 198 and 198, and the corresponding data amount is 10000, 5000 and 5000. The model identifies 98388 large load, high intensity load segments, and does not identify medium and low intensity load segments. The error rate is stably below 3%, and the complete automatic processing flow of load operation intensity identification takes no more than 0.8s. In the actual development and application of the load spectrum, the model can greatly simplify the artificial processing flow, and effectively improve the accuracy of the load spectrum compilation and the processing efficiency.

[0162] The embodiment of the application takes the measured load spectrum as the input of the analysis model, and the output result is the load operation segment belonging to different operation intensity levels identified in the long sequence of measured load spectrum. By mining and analyzing typical characteristics, the instantaneous operation intensity corresponding to the long distance and large data amount of the load spectrum can be effectively identified, the problems of unclear load spectrum operation intensity identification and load spectrum cutting not meeting the requirements of heavy load working condition caused by serious load spectrum data redundancy are solved, and the efficiency and accuracy of spectrum analysis are greatly improved.

[0163] Figure 4 An example of an entity structure diagram of an electronic device is shown as Figure 4 As shown, the electronic device can include a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 complete mutual communication through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the load spectrum based operation load analysis method, which includes: obtaining a measured load spectrum of a to-be-evaluated operation machine; inputting the measured load spectrum into an analysis model to obtain the load operation segment corresponding to each operation intensity level of the to-be-evaluated operation machine output by the analysis model; wherein the measured load spectrum is the sampling value corresponding to at least one test parameter of the to-be-evaluated operation machine in a preset period; the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and an operation intensity level labeled corresponding to the sample load spectrum segment.

[0164] Further, the logic instructions in the memory 430 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. 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 methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various media that can store program codes.

[0165] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the load spectrum based work load analysis method provided by the above-mentioned methods, the method comprising: obtaining a measured load spectrum of a work machine to be evaluated; inputting the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model; wherein the measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period; and the analysis model is obtained by training based on a sample load spectrum, sample load spectrum segments contained in the sample load spectrum, and work intensity levels labeled corresponding to the sample load spectrum segments.

[0166] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the load spectrum based work load analysis method provided by the above-mentioned methods, the method comprising: obtaining a measured load spectrum of a work machine to be evaluated; inputting the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model; wherein the measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period; and the analysis model is obtained by training based on a sample load spectrum, sample load spectrum segments contained in the sample load spectrum, and work intensity levels labeled corresponding to the sample load spectrum segments.

[0167] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0168] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0169] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A load spectrum-based job load analysis method characterized by, The method comprises: obtaining a measured load spectrum of a work machine to be evaluated; inputting the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model; wherein the measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period; the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a work intensity level annotated corresponding to the sample load spectrum segment; the analysis model comprises a feature extraction layer, a segment recognition layer, and a segment classification layer; inputting the measured load spectrum into the feature extraction layer to obtain an encoding combination output by the feature extraction layer; inputting the encoding combination into the segment recognition layer to obtain segment feature information output by the segment recognition layer; inputting the segment feature information into the segment classification layer to obtain the work intensity level and the load work segment corresponding to the work intensity level output by the segment classification layer; the inputting the encoding combination into the segment recognition layer to obtain the segment feature information output by the segment recognition layer comprises: performing feature extraction on the encoding combination to obtain a global feature map; performing feature association on the global feature map using an attention mechanism to obtain the segment feature information; the performing feature association on the global feature map using an attention mechanism to obtain the segment feature information comprises: inputting the global feature map into a recurrent neural network model to obtain an intermediate hidden state; determining the weight of each intermediate hidden state according to a multi-head attention mechanism; based on the weight of each intermediate hidden state, obtaining the segment feature information output by the recurrent neural network model; wherein the recurrent neural network model comprises one of a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a variant of LSTM.

2. The load spectrum based job load analysis method according to claim 1, characterized by, the inputting the measured load spectrum into the feature extraction layer to obtain the encoding combination output by the feature extraction layer comprises: based on the sampling value of each test parameter in the measured load spectrum and the standard value of the test parameter at each level, determining the encoding combination.

3. A load spectrum-based job load analysis apparatus characterized by comprising: The method comprises: a data acquisition module configured to obtain a measured load spectrum of a work machine to be evaluated; an analysis module configured to input the measured load spectrum into an analysis model to obtain load work segments corresponding to each work intensity level of the work machine to be evaluated output by the analysis model; wherein the measured load spectrum is a sampling value corresponding to at least one test parameter of the work machine to be evaluated within a preset time period; the analysis model is obtained by training based on a sample load spectrum, a sample load spectrum segment contained in the sample load spectrum, and a work intensity level annotated corresponding to the sample load spectrum segment; the analysis model comprises a feature extraction layer, a segment recognition layer, and a segment classification layer; inputting the measured load spectrum into the feature extraction layer to obtain an encoding combination output by the feature extraction layer; inputting the encoding combination into the segment recognition layer to obtain segment feature information output by the segment recognition layer; inputting the segment feature information into the segment classification layer to obtain the work intensity level and the load work segment corresponding to the work intensity level output by the segment classification layer; the inputting the encoding combination into the segment recognition layer to obtain the segment feature information output by the segment recognition layer comprises: performing feature extraction on the encoding combination to obtain a global feature map; performing feature association on the global feature map using an attention mechanism to obtain the segment feature information; the performing feature association on the global feature map using an attention mechanism to obtain the segment feature information comprises: inputting the global feature map into a recurrent neural network model to obtain an intermediate hidden state; determining the weight of each intermediate hidden state according to a multi-head attention mechanism; based on the weight of each intermediate hidden state, obtaining the segment feature information output by the recurrent neural network model; wherein the recurrent neural network model comprises one of a long short-term memory network (LSTM), a gated recurrent unit (GRU), and a variant of LSTM. the inputting the measured load spectrum into the feature extraction layer to obtain the encoding combination output by the feature extraction layer comprises: based on the sampling value of each test parameter in the measured load spectrum and the standard value of the test parameter at each level, determining the encoding combination. Input the encoding combination into the segment recognition layer to obtain the segment feature information output by the segment recognition layer; Input the segment feature information into the segment classification layer to obtain the work intensity level output by the segment classification layer and the load work segment corresponding to the work intensity level; The inputting of the encoding combination into the segment recognition layer to obtain the segment feature information output by the segment recognition layer comprises: Performing feature extraction on the encoding combination to obtain a global feature map; Performing feature association on the global feature map by using an attention mechanism to obtain the segment feature information; The performing of feature association on the global feature map by using an attention mechanism to obtain the segment feature information comprises: Inputting the global feature map into a recurrent neural network model to obtain an intermediate hidden state; Determining the weight of each intermediate hidden state according to a multi-head attention mechanism; Obtaining the segment feature information output by the recurrent neural network model based on the weight of each intermediate hidden state; The recurrent neural network model comprises one of a long short-term memory network (LSTM), a gated recurrent unit (GRU) and a variant of the LSTM.

4. A load spectrum-based work load analysis system comprising a work machine provided with a sensor module and a remote terminal, characterized by, The remote terminal is configured to perform the load analysis method based on a load spectrum according to any one of claims 1 to 2. The sensing module is configured to send a measured load spectrum collected by the working machine to the remote terminal when it is determined that the working machine is in a working state. The remote terminal is configured to input the measured load spectrum into an analysis model to obtain the load work segments corresponding to each work intensity level of the working machine to be evaluated output by the analysis model. The measured load spectrum is a sampling value corresponding to at least one test parameter of the working machine within a preset time period; and the analysis model is obtained by training based on a sample load spectrum, sample load spectrum segments contained in the sample load spectrum and work intensity levels labeled corresponding to the sample load spectrum segments.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the load analysis method based on a load spectrum according to any one of claims 1 to 2 when executing the program.

6. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the load analysis method based on a load spectrum according to any one of claims 1 to 2 when executed by the processor.

7. A computer program product comprising a computer program, characterized in that, The computer program implements the load analysis method based on a load spectrum according to any one of claims 1 to 2 when executed by the processor.

Images