Agricultural machinery working condition acquisition device and method

Abstract

The agricultural machinery working condition acquisition device comprises a data acquisition module, and the data acquisition module is in signal connection with a sensor module, a controller module, a GPS positioning system module and a camera module in sequence. The sensor module is in signal connection with a temperature sensor unit, an air pressure sensor unit, an oil quantity sensor unit and a speed sensor unit in sequence, and the controller module is in signal connection with a stroke controller unit and an engine control unit in sequence. Through mutual assistance of data, working states and stopping states of different operations of the tractor are identified, all working data after the tractor is started are obtained, invalid data in abnormal working states are removed, related algorithm analysis is carried out, and actual oil consumption data in the operation process are obtained. The device can truly reflect different oil consumption data consumed by different operation methods in the actual working process of the agricultural machinery.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural machinery working condition collection, and in particular relates to an agricultural machinery working condition collection device and method. Background Art

[0002] With the continuous upgrading of modern agricultural technology, tractors, as an important agricultural machinery, their data collection and analysis have become an important means to improve agricultural production efficiency and optimize agricultural production management. The tractor data collection device can collect tractor operation data, including working hours, fuel consumption, equipment operation status, etc., and make scientific decisions and management through data analysis, model building and other means, which can improve agricultural production efficiency and management level.

[0003] In the prior art, when the tractor data acquisition device collects data, it is difficult to monitor the fuel consumption data accurately according to the actual operation process, so that a lot of invalid data is mixed in the data collection for algorithm analysis. The obtained fuel consumption data has a certain deviation from the actual fuel consumption data during the operation process, and cannot provide accurate help for improving the production efficiency of agricultural machinery.

[0004] Therefore, it is necessary to invent an agricultural machinery operating condition collection device and method to solve the above-mentioned problems, which can be used to truly reflect the fuel consumption during the operation of an agricultural tractor. Summary of the invention

[0005] In view of the above problems, the present invention provides an agricultural machinery operating condition collection device and method to solve the problems raised in the above background technology.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an agricultural machinery working condition collection device and method, comprising a data collection module, wherein the data collection module is respectively connected to a sensor module, a controller module, a GPS positioning system module and a camera module, wherein the sensor module is sequentially connected to a temperature sensor unit, an air pressure sensor unit, an oil level sensor unit and a speed sensor unit, wherein the controller module is sequentially connected to a travel controller unit and an engine control unit;

[0007] The data acquisition module is connected to the cloud platform via a network signal.

[0008] Preferably, the sensor module is a core component of the tractor data acquisition module, which can collect data information of the tractor's operating environment and mechanical equipment in real time; the temperature sensor monitors the temperature of the agricultural machinery engine system in real time; the air pressure sensor monitors the air pressure of the agricultural machinery engine system in real time; the oil level sensor monitors the oil level change of the agricultural machinery tank system in real time; the speed sensor monitors the movement speed of the agricultural machinery in real time. The data collected by the sensor acquisition module can be used for comprehensive monitoring and data analysis of the tractor's working status.

[0009] Preferably, the controller module is another core component of the tractor data acquisition module, which has the functions of controlling, converting and processing data; the travel controller unit controls the travel data of the agricultural machinery, the engine controller unit controls the working data of the agricultural machinery engine, and the controller module transmits the data of the travel control unit and the engine controller unit, so that the travel data and engine data of the agricultural machinery can be processed and analyzed.

[0010] Preferably, the cloud platform is a new technology that combines tractor data collection and analysis with cloud computing.

[0011] Preferably, the GPS positioning system module provides accurate longitude and latitude data of the agricultural machinery in real time to locate the position of the agricultural machinery in the operating area.

[0012] Preferably, the camera module remotely checks on-site operation conditions.

[0013] A method for collecting agricultural machinery working conditions, characterized in that the method is as follows:

[0014] Step 1: Connect the temperature sensor unit and the air pressure sensor unit to the agricultural tractor engine system through the sensor module, access the engine-related working status, collect data, connect the agricultural tractor fuel tank system through the oil volume sensor unit of the sensor module, access the real-time consumption of the fuel tank, collect data, connect the speed control system of the agricultural tractor through the speed sensor unit of the sensor module, access the real-time operating speed of the tractor speed control system, collect data, and upload to the data acquisition module;

[0015] Step 2: Connect the travel control unit to the agricultural tractor travel control system through the controller module, access the working status related to the tractor work travel, collect data, connect the engine controller unit to the agricultural tractor engine system through the controller module, access the working status related to the tractor engine control work, collect data, and upload the data collection module;

[0016] Step 3: Provide accurate positioning latitude and longitude data of agricultural machinery in real time through the GPS positioning system module and upload it to the data collection module;

[0017] Step 4: According to the characteristics of agricultural operations and the actual operation process, the working state and stop state of the tractor for different operations are identified through the mutual assistance of data from the temperature sensor unit, the air pressure sensor unit, the oil level sensor unit, the speed sensor unit, the stroke controller unit and the engine control unit, and all the data of the tractor after it is started are obtained. The invalid data of the abnormal working state is eliminated and the relevant algorithm analysis is performed to obtain the actual fuel consumption data during the operation process;

[0018] Step 5: Cloud platform computing realizes refined and centralized management of tractor data, which can truly reflect the different fuel consumption data consumed by different operating methods during the actual working process of agricultural tractors;

[0019] Step 6: Perform real-time video monitoring through the camera module.

[0020] Technical effects and advantages of the present invention:

[0021] 1. The present invention identifies the working state and stop state of different operations of the tractor according to the characteristics of agricultural operations and in combination with the actual operation process through the mutual assistance of data from the temperature sensor unit, the air pressure sensor unit, the oil level sensor unit, the speed sensor unit, the stroke controller unit and the engine control unit, obtains all the data of the tractor after it is started, eliminates invalid data in abnormal working states and performs relevant algorithm analysis to obtain the actual fuel consumption data during the operation process. The device can truly reflect the different fuel consumption data consumed by different operation methods during the actual operation of agricultural machinery.

[0022] 2. The present invention constructs a working condition recognition model based on an autoencoder and a long short-term memory network (LSTM). This method can automatically identify complex working conditions and reduce human intervention. The use of a throttle control unit and a fuel injection controller can further optimize the accuracy of fuel consumption data.

[0023] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a three-dimensional schematic diagram of the present invention;

[0026] Figure 2 It is a flow chart of the working condition identification algorithm of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Embodiment 1: The present invention provides Figure 1 The agricultural machinery working condition collection device and method shown in the figure include a data collection module, the data collection module is respectively connected to a sensor module, a controller module, a GPS positioning system module and a camera module, the sensor module is sequentially connected to a temperature sensor unit, an air pressure sensor unit, an oil level sensor unit and a speed sensor unit, and the controller module is sequentially connected to a travel controller unit and an engine control unit;

[0029] The data acquisition module is connected to the cloud platform via a network signal.

[0030] As a specific embodiment of the present invention, the sensor module is a core component of the tractor data acquisition module, which can collect data information of the tractor operating environment and mechanical equipment in real time; the temperature sensor monitors the temperature of the agricultural machinery engine system in real time; the air pressure sensor monitors the air pressure of the agricultural machinery engine system in real time; the oil level sensor monitors the oil level change of the agricultural machinery tank system in real time; the speed sensor monitors the movement speed of the agricultural machinery in real time. The data collected by the sensor acquisition module can be used for comprehensive monitoring and data analysis of the tractor working status.

[0031] As a specific embodiment of the present invention, the controller module is another core component of the tractor data acquisition module, which has the functions of controlling, converting and processing data; the travel controller unit controls the travel data of the agricultural machinery, and the engine controller unit controls the working data of the agricultural machinery engine. The controller module transmits the data of the travel control unit and the engine controller unit, so that the travel data and engine data of the agricultural machinery are processed and analyzed. The above-mentioned travel controller unit and engine controller unit are conventional control components and belong to the known technology. The travel controller can collect and analyze a large amount of real-time data and information through sensors and monitoring equipment, such as soil moisture, meteorological data, etc.

[0032] As a specific implementation of the present invention, the cloud platform is a new technology that combines tractor data collection and analysis with cloud computing.

[0033] As a specific implementation of the present invention, the GPS positioning system module provides accurate positioning longitude and latitude data of agricultural machinery in real time to locate the position of the agricultural machinery operation area.

[0034] As a specific implementation of the present invention, the camera module remotely views on-site operation conditions.

[0035] As a specific embodiment of the present invention, a method for collecting agricultural machinery working conditions:

[0036] The sensor module is the core component of the tractor data acquisition device. It integrates various sensors to collect data information about the tractor's operating environment and mechanical equipment in real time. The temperature sensor unit and the air pressure sensor unit are connected to the agricultural tractor engine system through the sensor module to access the engine's related working status and collect data. The oil volume sensor unit of the sensor module is connected to the agricultural tractor's fuel tank system to access the fuel tank's related real-time consumption and collect data. The speed sensor unit of the sensor module is connected to the speed control system of the agricultural tractor to access the real-time operating speed of the tractor's speed control system, collect data, and upload the data acquisition module.

[0037] The controller is another core component of the tractor data acquisition device, which has the functions of controlling, converting and processing data. The travel control unit is connected to the agricultural tractor travel control system through the controller module, and the working status related to the tractor working travel is accessed, and data is collected. The engine controller unit is connected to the agricultural tractor engine system through the controller module, and the working status related to the tractor engine control work is accessed, and data is collected and uploaded to the data acquisition module;

[0038] The GPS positioning system module provides accurate latitude and longitude data of agricultural machinery in real time, uploads it to the data acquisition module, and uses the camera module to assist in real-time video monitoring to locate the position and working status of the tractor;

[0039] According to the characteristics of agricultural operations and in combination with the actual operation process, the working state and stop state of the tractor for different operations are identified through the mutual assistance of data from the temperature sensor unit, the air pressure sensor unit, the oil volume sensor unit, the speed sensor unit, the stroke controller unit and the engine control unit, and all the data of the tractor after it is started are obtained, and invalid data in the abnormal working state are eliminated. The invalid data includes the fuel consumption data generated by the action displacement in the non-working state, the fuel consumption data generated during the tractor maintenance and care, and the fuel consumption data when the oil volume is reduced due to the abnormal consumption of the tractor tank; the relevant algorithm analysis is performed to obtain the actual fuel consumption data during the operation process;

[0040] Tractor fuel consumption calculation formula:

[0041] The fuel consumption per 100 kilometers of the tractor is calculated based on the tractor's fuel consumption data after invalid data is removed and the mileage after invalid mileage data is removed. The calculation formula is:

[0042] Fuel consumption per 100 kilometers = fuel consumption (liters) ÷ mileage (kilometers) × 100;

[0043] Under the same operation mode of the tractor, the tractor has different loads, travels on different road conditions, and works at different ambient temperatures. By controlling a single variable, the corresponding actual fuel consumption can be obtained;

[0044] Through the mutual assistance of data from the temperature sensor unit, air pressure sensor unit, oil level sensor unit, speed sensor unit, travel controller unit and engine control unit, different operating methods in the actual working process of the tractor are identified, and different real fuel consumption data are obtained according to different types of operations. The cloud platform computing realizes the refined and centralized management of tractor data, which can truly reflect the different fuel consumption data consumed by different operating methods in the actual working process of agricultural tractors;

[0045] Tractor fuel consumption calculation formula:

[0046] The fuel consumption per 100 kilometers of the tractor is calculated based on the tractor's fuel consumption data after invalid data is removed and the mileage after invalid mileage data is removed. The calculation formula is:

[0047] Fuel consumption per 100 kilometers = fuel consumption (liters) ÷ mileage (kilometers) × 100;

[0048] Under different operating modes of the tractor, the tractor has the same load, travels on the same road conditions, and works at the same ambient temperature, and the actual fuel consumption of the tractor under different operating modes can be obtained.

[0049] Embodiment 2: By constructing a working condition recognition model based on an autoencoder and a long short-term memory network (LSTM), the algorithm can learn to extract representative and important features from multi-dimensional data (temperature, air pressure, oil volume, speed) collected by sensors; the autoencoder can extract core features from the data by compressing and reconstructing the data, and the LSTM can capture long-term dependencies in the time series, thereby more accurately identifying different working conditions of agricultural tractors; this method can automatically identify complex working conditions, reduce human intervention, and enable the throttle control unit and the fuel injection controller to further optimize the accuracy of fuel consumption data;

[0050] Step 1: Data preprocessing

[0051] First, the multi-dimensional data collected by the sensor module; then the data of each sensor is scaled to prevent data of different dimensions from having an adverse effect on model training; the data is sorted in chronological order to construct a time series data set so that the subsequent LSTM model can capture time dependencies;

[0052] Data collection: The sensor module continuously collects multi-dimensional data such as temperature, air pressure, oil level, and speed. The temperature sensor (T) is used to record the temperature of the engine when it is working; the air pressure sensor (P) is used to record the air pressure at the engine air intake, with a range of; the oil level sensor (F) is used to record the real-time oil level in the tractor tank; the speed sensor (S) is used to record the real-time driving speed of the tractor;

[0053] Data standardization: In multi-dimensional data, the dimensions of temperature (℃), air pressure (kPa), fuel volume (L), and speed (km / h) are different. Directly inputting the data into the model will lead to poor training results. Therefore, it is necessary to standardize the data of each sensor and unify them into the same scale range.

[0054] The min-max normalization method is used to scale the data to the range of [0, 1]. The normalization formula is as follows:

[0055]

[0056] Where x is the original data point, that is, the original value to be normalized; x min is the minimum value of the original data, indicating the theoretical minimum value of the data set or the lowest value in the actual measurement range; max is the maximum value of the original data, indicating the theoretical maximum value of the data set or the highest value in the actual measurement range; x′ is the normalized data value, ranging from [0, 1], which is used to unify data of different dimensions to the same scale to facilitate subsequent algorithm processing;

[0057] The standardized data will be used to train the machine learning model to avoid the impact of different dimensions on the model;

[0058] Time series construction: Since the working conditions of agricultural tractors have strong time correlation, the LSTM model requires time series data to capture the temporal dependence of various tractor parameters. In order to construct a time series dataset, a time window is set, that is, continuous time step data of the time window is input each time to predict the next state of the tractor.

[0059] The numerical examples are as follows:

[0060] Data collection: During a certain period of time, the sensor module collected the following multi-dimensional data: temperature sensor (T) is 80°C; air pressure sensor (P) is 100kPa; fuel level sensor (F) is 120L; speed sensor (S) is 35km / h;

[0061] Data standardization: The theoretical minimum and maximum values ​​of the sensor are: temperature range 0℃ to 120℃; air pressure range 80kPa to 120kPa; oil volume range 0L to 200L; speed range 0km / h to 50km / h;

[0062] Using min-max normalization, the normalized data results are: temperature T′=(80-0) / (120-0)=0.6667; air pressure P′=(100-80) / (120-80)=0.5; oil volume F′=(120-0) / (200

[0063] -0)=0.6; speed S′=(35-0) / (50-0)=0.7;

[0064] The standardized data is: X = [0.6667, 0.5, 0.6, 0.7]

[0065] Time series construction: The time window is set to 3 time steps of data, and the input is:

[0066] X 1 =[0.6667,0.5,0.6,0.7]

[0067] X 2 =[0.7,0.55,0.62,0.72]

[0068] X 3 =[0.65,0.52,0.61,0.71]

[0069] Use this data to construct a time series dataset and predict the data for the next time step.

[0070] Step 2: Feature extraction

[0071] The autoencoder mainly consists of two parts: an encoder and a decoder. The autoencoder compresses the high-dimensional sensor data into a low-dimensional feature representation through the encoder and extracts the core features in the data. The decoder then reconstructs these low-dimensional features into the original data to ensure that the extracted features can retain the information of the original data as much as possible.

[0072] 2.1 Encoder part

[0073] The role of the encoder is to compress the input high-dimensional data into a low-dimensional feature representation;

[0074] The output of the encoder is a low-dimensional feature vector Z, whose formula is:

[0075] Z=f(W e ·X+b e )

[0076] Where X represents the sensor input data vector, which contains multi-dimensional data of T (temperature), P (air pressure), F (fuel volume), and S (speed); W e represents the weight matrix of the encoder, which is used to map the input data to a low-dimensional space; b e represents the bias term of the encoder; f represents the activation function, and a nonlinear activation function (ReLU) is used to increase the expressiveness of the model; Z represents the low-dimensional feature representation, which represents the features of the compressed input data; these features will be used in the subsequent LSTM network for time series modeling;

[0077] 2.2 Decoder

[0078] The role of the decoder is to reconstruct the low-dimensional feature representation Z into original data as close as possible to the input X to ensure that the extracted features can retain the key information in the sensor data; the output formula of the decoder is:

[0079]

[0080] in Represents the reconstructed output data, which is expected to be as close as possible to the original input X; W d represents the weight matrix of the decoder, which is used to map low-dimensional features back to high-dimensional space; b d represents the bias term of the decoder; g represents the activation function, which uses the same activation function as the encoder; Z represents the low-dimensional feature obtained from the encoder;

[0081] 2.3 Loss Function

[0082] The goal of the autoencoder is to minimize the The difference between them is used to ensure the effectiveness of feature extraction; the loss function here is the mean square error (MSE), and its formula is:

[0083]

[0084] Where L is the loss function, which represents the difference between the input and the reconstructed output; n is the dimension of the input data; X i represents the i-th data point of the original input; represents the i-th data point after reconstruction;

[0085] By minimizing the loss function L, the autoencoder can learn the key features in the sensor data and retain the original information as much as possible;

[0086] 2.4 Numerical Examples

[0087] Encoder part: The input multi-dimensional sensor data is X = [T, P, F, S], and a simple 2-dimensional encoder is used to compress the data.

[0088] The output of the encoder is:

[0089] Z=f(W e ·X+b e )

[0090] W e =[[0.2, 0.3, 0.5, 0.1], [0.4, 0.2, 0.3, 0.6]]

[0091] b e =[0.1,0.2]

[0092] Use ReLU as the activation function.

[0093] The calculation results are:

[0094] Z=f([0.2·0.6667+0.3·0.5+0.5·0.6+0.1·0.7]

[0095] +0.1[0.4 0.6667+0.2 0.5+0.3 0.6+0.6 0.7]+0.2)

[0096] Calculate each element:

[0097] Z 1 =f(0.6133+0.1)=0.7133

[0098] Z 2 =f(0.91334+0.2)=1.11334

[0099] Get the encoder output:

[0100] Z = [0.7133, 1.11334]

[0101] Decoder part: The decoder uses the following weight matrix:

[0102] W a =[[0.3,0.4],[0.2,0.1],[0.5,0.2],[0.3,0.6]]

[0103] b d =[0.1,0.2,0.3,0.4]

[0104] Decoder output:

[0105]

[0106] calculate:

[0107]

[0108] The reconstructed data is:

[0109]

[0110] Loss function (MSE): Original data X = [0.6667, 0.5, 0.6, 0.7], then the MSE loss function is:

[0111]

[0112] calculate:

[0113]

[0114] Step 3: Time Series Modeling

[0115] The convolutional features are modeled in time series through the LSTM network; LSTM is suitable for processing time series data and capturing the state changes of agricultural tractors under different working conditions; the LSTM network gradually learns the time dependency of sensor data and can understand the changes of various working condition data of tractors over time; the features extracted from the autoencoder are used as the input of LSTM, and the working condition state is modeled through multi-layer LSTM to predict the working condition changes in the future;

[0116] 3.1 LSTM unit design

[0117] The LSTM network is a special version of the recurrent neural network (RNN) used to capture temporal dependencies and is particularly suitable for processing time series data. It learns the long-term and short-term dependencies in sensor data by designing memory units, thereby understanding the changes in the tractor's operating conditions. The core structure of the LSTM unit represents

[0118] Forget Gate is used to determine the historical information that needs to be forgotten; the formula is

[0119] f t =σ(W f ·[h t-1 , x t ]+b f )

[0120] where f tRepresents the output of the forget gate, with a value between 0 and 1, indicating how much past information is retained; W f represents the weight matrix of the forget gate; h t-1 represents the hidden state of the previous time step; x t b represents the feature vector of the current time step input; f represents the bias term of the forget gate; σ represents the activation function Sigmoid, which compresses the result to between 0 and 1;

[0121] The input gate is used to determine the current information that needs to be updated; the formula is expressed as

[0122] i t =σ(W i ·[h t-1 , x t ]+b i )

[0123] where i t Represents the output of the input gate, which controls the proportion of the current input that is retained; W i represents the weight matrix of the input gate; b i represents the bias term of the input gate; x t represents the feature vector of the current time step input; h t-1 represents the hidden state of the previous time step;

[0124] CandidateCellState is used to generate the current candidate memory content; the formula is expressed as

[0125]

[0126] in represents the candidate memory state of the current time step; W C The weight matrix representing the candidate memory state; b C represents the bias term of the candidate memory state; tanh represents the hyperbolic tangent function, which compresses the result to between -1 and 1; h t-1 represents the hidden state of the previous time step; x t The feature vector representing the input at the current time step;

[0127] Memory update (CellStateUpdate) is used to update the memory state through the forget gate and input gate; the formula is expressed as

[0128]

[0129] Where C t Represents the memory state of the current time step; C t-1 Represents the memory state of the previous time step; i tRepresents the output of the input gate, which controls the proportion of the current input that is retained; represents the candidate memory state of the current time step; f t Represents the output of the forget gate;

[0130] The output gate (OutputGate) is used to determine the hidden state of the output; the formula is expressed as

[0131] o t =σ(W o ·[h t-1 , x t ]+b o )

[0132] The hidden state is calculated as

[0133] h t =o t tanh(C t )

[0134] Among them t represents the output of the output gate; σ represents the Sigmoid activation function; W o represents the weight matrix of the output gate [h t-1 , x t ] represents the hidden state h of the previous time step t-1 With the input x at the current time step t The concatenation vector of o Represents the bias term of the output gate; C t Represents the memory state of the current time step; tanh represents the hyperbolic tangent function; C t Represents the memory state of the current time step;

[0135] 3.2 Input and Output Sequences

[0136] Input sequence: In the present invention, the feature map extracted from the autoencoder is used as the input of LSTM. The LSTM network models the time series data to understand the changes in the working conditions of the tractor;

[0137] Output sequence: The output of the LSTM network will be used to predict the working state of the tractor in the future; the state output of the LSTM network is h t ,LSTM can capture the impact of past data on the future and help predict the next state of the tractor;

[0138] Time series modeling is performed through multi-layer LSTM, and the prediction results are finally output:

[0139] y t =LSTM(x 1 , x 2 , …, xt )

[0140] where y t is the output of the current time step (here it can characterize the state of the fuel-consuming tractor: prediction of normal operation, standby or low-load operation), indicating the current working state of the tractor; x 1 , x 2 , …, x t is the time series of sensor input features;

[0141] 3.3 Algorithm Improvement

[0142] Traditional LSTM only considers the information transfer from the past to the present, while in some cases, future state information is also helpful for current predictions; by introducing bidirectional LSTM, the model can consider the past and future time step information at the same time, thereby more accurately modeling the changes in the tractor's working conditions;

[0143] Bidirectional LSTM consists of two LSTM networks, one that processes the time series from front to back and the other from back to front; their output is the concatenation of the hidden states in both directions to form the final feature representation;

[0144] Hidden state of the forward LSTM for:

[0145]

[0146] Hidden state of backward LSTM for:

[0147]

[0148] The final output of the bidirectional LSTM is:

[0149]

[0150] in: The hidden state of the forward LSTM captures the past information of time step t; The hidden state of the backward LSTM captures the future information at time step t; h t : The output of the bidirectional LSTM contains comprehensive features of the past and the future (here, the state of the fuel-consuming tractor can be characterized: prediction of normal operation, standby or low-load operation);

[0151] 3.4 Numerical examples

[0152] LSTM unit design: The features Z = [0.7133, 1.11334] extracted from the autoencoder are input into the LSTM.

[0153] Forget gate calculation:

[0154] f t =σ(W f ·[h t-1 , Z]+b f )

[0155] in:

[0156] W f =[0.3,0.2]

[0157] b f =0.1

[0158] The hidden state h at the previous time step t-1 =0.5

[0159] calculate:

[0160] f t =σ(0.3·0.5+0.2·0.7133+0.1)=σ(0.35+0.14266+0.1)

[0161] =σ(0.59266)=0.6448

[0162] Input gate calculation: input gate weight W i =[0.2,0.4], bias b i =0.2, then:

[0163] i t =σ(W i ·[h t-1 , Z]+b i )=σ([0.2·0.5+0.4·0.7133]+0.2)

[0164] i t =σ(0.1+0.28532+0.2)=σ(0.58532)=0.6422

[0165] Output gate calculation: Calculate the output gate o t , the weight matrix and bias are:

[0166] o t =σ(W o ·[h t-1 , Z]+b o )

[0167] Hidden state update: Finally calculate the hidden state of the current time step:

[0168] h t =o t tanh(C t )

[0169] Step 4: Fuel consumption data optimization

[0170] 4.1 Through the above working condition identification model, the tractor fuel consumption data under different working conditions can be automatically identified to reduce the impact of ineffective working fuel consumption; combined with the actual fuel consumption data under different working conditions, the tractor fuel consumption management and working efficiency can be further optimized;

[0171] This implementation combines the autoencoder and LSTM algorithms, which have advantages in processing multi-dimensional sensor data and time series data, and can efficiently realize the automatic identification of agricultural tractor working conditions;

[0172] Specifically, through the above formula

[0173] y t =LSTM(x 1 , x 2 , ..., x t )

[0174] The long short-term memory network (LSTM) model is used to predict the operating state at future moments and calculate the final fuel consumption based on the prediction results; each time step y output by LSTM t Each one will correspond to a fuel consumption rate, and these fuel consumption rates will be recorded.

[0175] The fuel consumption calculation formula is: Fuel consumption = fuel consumption rate × time interval

[0176] Add up the fuel consumption of all time steps to get the total fuel consumption of the entire time series:

[0177]

[0178] Where T is the total number of time steps, fuel consumption rate t is the fuel consumption rate at the tth time step;

[0179] 4.2 Numerical examples

[0180] There are two main operating conditions of the tractor:

[0181] Normal operation means that the tractor is working in the field, with a high speed and heavy engine load; standby or low-load operation means that the tractor is in standby or low-speed state, with relatively low fuel consumption;

[0182] In these two states, the fuel consumption data will change with the operating conditions; when it is recognized as normal operation, the fuel consumption is 10L / hour; when it is recognized as standby, the fuel consumption is 3L / hour;

[0183] There is the following LSTM prediction output (each time step is 1 hour):

[0184] Time step 1: y1 The forecast is "normal operation", and the fuel consumption rate is 10L / hour;

[0185] Time step 2: y 2 The prediction is "standby", and the fuel consumption rate is 3L / hour;

[0186] Time step 3: y 3 The forecast is "normal operation", and the fuel consumption rate is 10L / hour;

[0187] Time step 4: y 4 The prediction is "standby", and the fuel consumption rate is 3L / hour;

[0188] Calculate the fuel consumption at each time step

[0189] Time step 1: Fuel consumption = 10L / hour × 1 hour = 10L

[0190] Time step 2: Fuel consumption = 3L / hour × 1 hour = 3L

[0191] Time step 3: Fuel consumption = 10L / hour × 1 hour = 10L

[0192] Time step 4: Fuel consumption = 3L / hour × 1 hour = 3L

[0193] The total fuel consumption is thus:

[0194] Total fuel consumption = 10+3+10+3=26L.

[0195] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions 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 invention.

Claims

1. An agricultural machinery working condition collection device, comprising a data collection module, characterized in that: The data acquisition module is respectively connected to a sensor module, a controller module, a GPS positioning system module and a camera module, the sensor module is sequentially connected to a temperature sensor unit, an air pressure sensor unit, an oil level sensor unit and a speed sensor unit, and the controller module is sequentially connected to a travel controller unit and an engine control unit; The data acquisition module is connected to the cloud platform via a network signal.

2. The agricultural machinery working condition collection device according to claim 1, characterized in that: The sensor module is the core component of the tractor data acquisition module, which can collect data information about the tractor's operating environment and mechanical equipment in real time; the temperature sensor monitors the temperature of the agricultural machinery engine system in real time; The air pressure sensor monitors the air pressure of the agricultural machinery engine system in real time; The oil level sensor monitors the oil level changes in the agricultural machinery tank system in real time; the speed sensor monitors the movement speed of the agricultural machinery in real time, and the data collected by the sensor acquisition module can be used for comprehensive monitoring and data analysis of the tractor's working status.

3. The agricultural machinery working condition collection device according to claim 1, characterized in that: The controller module is another core component of the tractor data acquisition module, which has the functions of controlling, converting and processing data; the travel controller unit controls the travel data of the agricultural machinery, and the engine controller unit controls the working data of the agricultural machinery engine. The controller module transmits the data of the travel control unit and the engine controller unit, so that the travel data and engine data of the agricultural machinery can be processed and analyzed.

4. The agricultural machinery working condition collection device according to claim 1, characterized in that: The cloud platform is a technology that combines tractor data collection and analysis with cloud computing.

5. The agricultural machinery working condition collection device according to claim 1, characterized in that: The GPS positioning system module provides accurate positioning longitude and latitude data of the agricultural machinery in real time to locate the position of the agricultural machinery in the operating area.

6. The agricultural machinery working condition collection device according to claim 1, characterized in that: The camera module remotely checks the on-site operation status.

7. A method for collecting agricultural machinery working conditions, characterized in that: The method is as follows: Step 1: Connect the temperature sensor unit and the air pressure sensor unit to the agricultural tractor engine system through the sensor module, access the engine-related working status, collect data, connect the agricultural tractor fuel tank system through the oil volume sensor unit of the sensor module, access the real-time consumption of the fuel tank, collect data, connect the speed control system of the agricultural tractor through the speed sensor unit of the sensor module, access the real-time operating speed of the tractor speed control system, collect data, and upload to the data acquisition module; Step 2: Connect the travel control unit to the agricultural tractor travel control system through the controller module, access the working status related to the tractor work travel, collect data, connect the engine controller unit to the agricultural tractor engine system through the controller module, access the working status related to the tractor engine control work, collect data, and upload the data collection module; Step 3: Provide accurate positioning latitude and longitude data of agricultural machinery in real time through the GPS positioning system module and upload it to the data collection module; Step 4: According to the characteristics of agricultural operations and the actual operation process, the data of the temperature sensor unit, the air pressure sensor unit, the oil level sensor unit, the speed sensor unit, the stroke controller unit and the engine control unit are mutually assisted, and the working state and stop state of the tractor for different operations are identified according to the working condition identification algorithm, and all the data of the tractor after starting are obtained, and the invalid data of the abnormal working state is eliminated and the relevant algorithm analysis is performed to obtain the actual fuel consumption data during the operation process; Step 5: Cloud platform computing realizes refined and centralized management of tractor data, which can truly reflect the different fuel consumption data consumed by different operating methods during the actual working process of agricultural tractors; Step 6: Perform real-time video monitoring through the camera module.

8. The method for collecting agricultural machinery working conditions according to claim 7, characterized in that: The specific steps of the working condition identification algorithm mentioned in step 4 are as follows: By building a working condition recognition model based on autoencoders and long short-term memory networks (LSTMs), the algorithm can learn to extract representative and important features from multi-dimensional data such as temperature, air pressure, oil volume, and speed collected by sensors; the autoencoder can extract core features from the data by compressing and reconstructing the data, while the LSTM can capture long-term dependencies in time series, thereby more accurately identifying different working conditions of agricultural tractors; Step 1: Data preprocessing First, the multi-dimensional data collected by the sensor module; then the data of each sensor is scaled to prevent data of different dimensions from having an adverse effect on model training; the data is sorted in chronological order to construct a time series data set so that the subsequent LSTM model can capture time dependencies; Data collection: The sensor module will continuously collect multi-dimensional data such as temperature, air pressure, oil level, and speed; The temperature sensor is used to record the temperature T of the engine when it is working; the air pressure sensor is used to record the air pressure P at the engine air intake; the oil level sensor is used to record the real-time oil level F in the tractor tank; the speed sensor is used to record the real-time driving speed S of the tractor; Data standardization: In multi-dimensional data, the dimensions of temperature, air pressure, fuel volume, and speed are different. Directly inputting them into the model will lead to poor training results. Therefore, it is necessary to standardize the data of each sensor and unify them into the same scale range. The min-max normalization method is used to scale the data to the range of [0, 1]. The normalization formula is as follows: Where x is the original data point, that is, the original value to be normalized; x min is the minimum value of the original data, indicating the theoretical minimum value of the data set or the lowest value in the actual measurement range; x max is the maximum value of the original data, indicating the theoretical maximum value of the data set or the highest value in the actual measurement range; x ′ is the normalized data value, ranging from [0, 1], which is used to unify data of different dimensions to the same scale to facilitate subsequent algorithm processing; The standardized data will be used to train the machine learning model to avoid the impact of different dimensions on the model; Time series construction: Since the working conditions of agricultural tractors have strong time correlation, the LSTM model requires time series data to capture the temporal dependence of various tractor parameters. In order to construct a time series dataset, a time window is set, that is, continuous time step data of the time window is input each time to predict the next state of the tractor. Step 2: Feature extraction The autoencoder mainly consists of two parts: encoder and decoder. The autoencoder compresses the high-dimensional sensor data into a low-dimensional feature representation through the encoder and extracts the core features in the data. These low-dimensional features are then reconstructed into original data through the decoder to ensure that the extracted features can retain the information of the original data as much as possible; 2.1 Encoder part The role of the encoder is to compress the input high-dimensional data into a low-dimensional feature representation; The output of the encoder is a low-dimensional feature vector Z, whose formula is: Z=f(W e ·X+b e ) Where X represents the sensor input data vector, which contains multi-dimensional data of temperature T, air pressure P, oil volume F, and speed S; W e represents the weight matrix of the encoder, which is used to map the input data to a low-dimensional space; b e Represents the bias term of the encoder; f represents the activation function, and the nonlinear activation function ReLU is used to increase the expressiveness of the model; Z represents the low-dimensional feature representation, which represents the features of the compressed input data; These features will be used in the subsequent LSTM network for time series modeling; 2.2 Decoder The role of the decoder is to reconstruct the low-dimensional feature representation Z into the original data as close as possible to the input X to ensure that the extracted features can retain the key information in the sensor data; The output formula of the decoder is: in Represents the reconstructed output data, which is expected to be as close as possible to the original input X; W d represents the weight matrix of the decoder, which is used to map low-dimensional features back to high-dimensional space; b d represents the bias term of the decoder; g represents the activation function, which uses the same activation function as the encoder; Z represents the low-dimensional feature obtained from the encoder; 2.3 Loss Function The goal of the autoencoder is to minimize the The difference between them is used to ensure the effectiveness of feature extraction; the loss function here is the mean square error MSE, and its formula is: Where L is the loss function, which represents the difference between the input and the reconstructed output; n is the dimension of the input data; X i represents the i-th data point of the original input; represents the i-th data point after reconstruction; By minimizing the loss function L, the autoencoder can learn the key features in the sensor data and retain the original information as much as possible; Step 3: Time Series Modeling Time series modeling of convolutional features through LSTM network; LSTM is suitable for processing time series data and capturing the state changes of agricultural tractors under different working conditions. The LSTM network gradually learns the time dependency of sensor data and can understand the changes of various working condition data of tractors over time. The features extracted from the autoencoder are used as the input of LSTM, and the working condition state is modeled through multi-layer LSTM to predict the working condition changes in the future. 3.1 LSTM unit design The LSTM network is a special version of the recurrent neural network RNN ​​used to capture time dependencies and is particularly suitable for processing time series data; It designs memory units to learn long-term and short-term dependencies in sensor data, thereby understanding the changes in tractor working conditions; the core structure of the LSTM unit represents The forget gate is used to determine the historical information that needs to be forgotten; the formula is f t =σ(W f ·[h t-1 ,x t ]+b f ) where f t Represents the output of the forget gate, with a value between 0 and 1, indicating how much past information is retained; W f represents the weight matrix of the forget gate; h t-1 represents the hidden state of the previous time step; x t b represents the feature vector of the current time step input; f represents the bias term of the forget gate; σ represents the activation function Sigmoid, which compresses the result to between 0 and 1; The input gate is used to determine the current information that needs to be updated; the formula is expressed as i t =σ(W i ·[h t-1 ,x t ]+b i ) where i t Represents the output of the input gate, which controls the proportion of the current input that is retained; W i represents the weight matrix of the input gate; b i represents the bias term of the input gate; x t represents the feature vector of the current time step input; h t-1 represents the hidden state of the previous time step; The candidate memory state is used to generate the current candidate memory content; the formula is expressed as in represents the candidate memory state of the current time step; W C The weight matrix representing the candidate memory state; b C represents the bias term of the candidate memory state; tanh represents the hyperbolic tangent function, which compresses the result to between -1 and 1; h t-1 represents the hidden state of the previous time step; x t The feature vector representing the input at the current time step; Memory update is used to update the memory state through the forget gate and the input gate; the formula is expressed as Where C t Represents the memory state of the current time step; C t-1 Represents the memory state of the previous time step; i t Represents the output of the input gate, which controls the proportion of the current input that is retained; represents the candidate memory state of the current time step; f t represents the output of the forget gate; The output gate is used to determine the hidden state of the output; the formula is expressed as the t =σ(W o ·[h t-1 ,x t ]+b o ) The hidden state is calculated as h t =o t ·tanh(C t ) Among them t represents the output of the output gate; σ represents the Sigmoid activation function; W o represents the weight matrix of the output gate [h t-1 , x t ] represents the hidden state h of the previous time step t-1 With the input x at the current time step t The concatenation vector of o represents the bias term of the output gate; tanh represents the hyperbolic tangent function; C t Represents the memory state of the current time step; 3.2 Input and Output Sequences Input sequence: The feature map extracted from the autoencoder is used as the input of LSTM. The LSTM network models the time series data to understand the changes in the tractor's working conditions. Output sequence: The output of the LSTM network will be used to predict the working state of the tractor in the future; the state output of the LSTM network is h t ,LSTM can capture the impact of past data on the future and help predict the next state of the tractor: prediction of normal operation, standby or low-load operation; Time series modeling is performed through multi-layer LSTM, and the prediction results are finally output: y t =LSTM(x1,x2,...,x t ) where y t is the output of the current time step, indicating the current working state of the tractor; x1, x2, …, x t is the time series of sensor input features; 3.3 Algorithm Improvement Traditional LSTM only considers the information transfer from the past to the present, while in some cases, future state information is also helpful for current predictions; by introducing bidirectional LSTM, the model can consider the past and future time step information at the same time, thereby more accurately modeling the changes in the tractor's working conditions; Bidirectional LSTM consists of two LSTM networks, one that processes the time series from front to back and the other from back to front; their output is the concatenation of the hidden states in both directions to form the final feature representation; Hidden state of the forward LSTM for: Hidden state of backward LSTM for: The final output of the bidirectional LSTM is: in: The hidden state of the forward LSTM captures the past information of time step t; The hidden state of the backward LSTM captures the future information at time step t; h t : The output of the bidirectional LSTM contains the comprehensive features of the past and the future; LSTM represents the long short-term memory network constructed above; Step 4: Fuel consumption data optimization Through the above working condition identification model, the fuel consumption data of the tractor under different working conditions can be automatically identified to reduce the impact of ineffective working fuel consumption; combined with the actual fuel consumption data under different working conditions, the fuel consumption management and working efficiency of the tractor can be further optimized; Specifically, through the above formula y t =LSTM(x1,x2,…,x t ) The long short-term memory network (LSTM) model is used to predict the operating state at future moments, and the final fuel consumption is calculated based on the prediction results. Each time step y output by LSTM t Each one will correspond to a fuel consumption rate, and these fuel consumption rates will be recorded; The fuel consumption calculation formula is: fuel consumption = fuel consumption rate × time interval; Add up the fuel consumption of all time steps to get the total fuel consumption of the entire time series: Where T is the total number of time steps, fuel consumption rate t is the fuel consumption rate at the tth time step; This implementation process combines the autoencoder and LSTM algorithms, which has advantages in processing multi-dimensional sensor data and time series data, and can efficiently realize the automatic identification of agricultural tractor working conditions.

Images