CN117272829A - MC-GRU-based cold chain temperature prediction method - Google Patents
MC-GRU-based cold chain temperature prediction method Download PDFInfo
- Publication number
- CN117272829A CN117272829A CN202311330166.5A CN202311330166A CN117272829A CN 117272829 A CN117272829 A CN 117272829A CN 202311330166 A CN202311330166 A CN 202311330166A CN 117272829 A CN117272829 A CN 117272829A
- Authority
- CN
- China
- Prior art keywords
- data
- temperature
- time
- cold chain
- gru
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 230000000694 effects Effects 0.000 claims abstract description 4
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 9
- 230000007613 environmental effect Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000003442 weekly effect Effects 0.000 claims description 3
- 230000007774 longterm Effects 0.000 abstract description 6
- 230000005540 biological transmission Effects 0.000 abstract description 5
- 230000008859 change Effects 0.000 description 6
- 238000010606 normalization Methods 0.000 description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241000894006 Bacteria Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 244000005700 microbiome Species 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a cold chain temperature prediction method based on MC-GRU, which comprises the following steps: step 1, acquiring historical temperature data of a cold chain transportation carriage environment, preprocessing the data, and then dividing a training set and a testing set; step 2, establishing a temperature prediction model based on a multi-channel gating circulating unit; step 3, training a temperature prediction model based on the multi-channel gating circulating unit; and 4, testing the effect of the temperature prediction model based on the multi-channel gating circulating unit. According to the invention, the GRU model is used as a basic module, a plurality of GRU models are defined as an auxiliary network, and the time step of each GRU model is gradually reduced, so that the characteristics of a short-term mode and a long-term mode of the cold chain compartment temperature can be simultaneously captured, and the prediction error of a long-term interpretation variable due to the increase of information transmission is avoided, thereby improving the prediction precision of the cold chain compartment temperature.
Description
Technical Field
The invention belongs to the technical field of cold chain logistics, adopts a supervision algorithm and a time sequence prediction technology, and particularly relates to a cold chain temperature prediction method based on MC-GRU.
Background
The proper temperature in the compartment is the basis for ensuring that the refrigerated goods maintain good quality. The traditional cold chain transportation only measures the instant temperature of food when the refrigerator car reaches a destination, but in the cold chain transportation process, the environment state of the food cannot be obtained in time, and when the food is in an environment higher than the optimal temperature, the growth of microorganisms such as bacteria can be caused, so that the quality and safety of the food are seriously affected; in an environment below the optimum temperature, unnecessary energy consumption is increased. There is a need for an early prediction of changes in the cold chain transport environment. The method has the advantages that the change rule of the cold chain environment is mastered, the cold chain transportation environment can be effectively managed, the state of the cold chain environment is analyzed through monitoring data, the change trend of the environment in the cold-storage carriage is predicted, the alarm reminding is carried out in time, the probability of accidents can be reduced, the quality of transported goods is guaranteed, the energy consumption is reduced, and the method has important significance in improving the cold chain logistics level in China.
The cold chain environmental change has the characteristics of time sequence, instability and nonlinearity and is influenced by various factors. The traditional method for determining the running state of the refrigerating unit by monitoring data in real time through a sensor has hysteresis in the feedback of data acquisition and the effectiveness of regulation measures, and the change trend of environmental factors cannot be prejudged, so that the environment state in the refrigerated carriage is remedied when the environment state is lower than a critical state, and economic loss is often caused. The deep learning has the characteristics of nonlinearity, self-adaption, simple structure and the like, and the defects of the traditional prediction mode can be well perfected.
The patent application with publication number of CN114004164A discloses a motor rotor temperature prediction method for control, which models by data generated in actual operation of a motor, considers more uncertain factors into time sequence data, and the obtained model has stronger robustness and adaptability. However, the long-term memory network has slow training speed and cannot be parallel, the prediction speed of the model is not high, and the characteristics in a longer time sequence cannot be extracted due to the structural limitation of the long-term memory network.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides the MC-GRU-based cold chain temperature prediction method which can simultaneously capture the characteristics of a short-term mode and a long-term mode of the cold chain compartment temperature by defining a plurality of GRU models as an auxiliary network and gradually reducing the time step of each GRU model and avoid the prediction error of a long-term interpretation variable due to information transmission increase.
The aim of the invention is realized by the following technical scheme: a cold chain temperature prediction method based on MC-GRU comprises the following steps:
step 1, acquiring historical temperature data of a cold chain transportation carriage environment, preprocessing the data, and then dividing a training set and a testing set; the method comprises the following specific steps:
step 1.1, acquiring first characteristic data TH of a cold chain transport carriage by using a temperature sensor 1 The system comprises a temperature sensor number, a collecting point position, environmental temperature data and a collecting time sequence;
step 1.2, processing the first characteristic data, and removing the repeated data to obtain second characteristic data TH 2 The method comprises the steps of acquiring time sequence and environmental temperature data; if a large amount of continuous sensor data is missing and the data cannot be supplemented, the segment of data is directly discarded; if only single-point sensor data is missing or data of a single time interval is missing, repairing the data by adopting a linear interpolation method; the formula of linear interpolation is:
wherein x is a+i For the data missing at the time a+i to be filled, x a 、x a+j The original sensor data at the time a and the time a+j are valid data values at the adjacent time a+i respectively;
step 1.3, the time sequence obtained in the step 1.2 is corresponding to the temperature data, and the time is used as a label of the temperature data; acquiring temperature data and labels thereof each of which is daily, weekly and monthly, and normalizing the temperature data to obtain third characteristic data TH 3 As a training set;
Step 1.4, third characteristic data TH 3 Dividing a training set and a testing set;
step 2, establishing a predictive model based on the temperature of the multi-channel gating circulating unit;
and step 3, training a temperature prediction model based on the multi-channel gating circulating unit.
Step 4, testing the effect of the temperature prediction model based on the multi-channel gating circulating unit; the method comprises the following specific steps:
step 4.1, verifying the performance of the model, taking the test set in the step 1 as input data, and inputting the model to obtain a prediction result;
and 4.2, comparing a model prediction result with a real temperature, and evaluating the performance of the model by adopting the following four indexes: and (3) determining coefficients according to the root mean square error, the average absolute error and the mean square error, ending training if the four indexes meet the preset requirements, obtaining a final prediction model, otherwise, adjusting parameters of the prediction model to retrain.
The specific method of the step 2 is as follows: the temperature prediction model based on the multi-channel gating circulating unit comprises a plurality of gating circulating units and a full connection layer; defining a plurality of gating cycle units as an auxiliary network, and gradually reducing the time step of each gating cycle unit; taking predicted values of all the gating circulating units as input of a full-connection layer, and taking output of the full-connection layer as a final predicted result;
the structure of the gating cycle unit is as follows:
reset gate r t The method comprises the steps of controlling how much information at the previous moment can be reserved in a candidate state at the current moment, transmitting the input at the current moment and the hidden state at the previous moment to a sigmoid function, and calculating the formula:
r t =σ(W r ·[h t-1 ,x t ]+b r )
wherein σ () represents a sigmoid activation function, W r To reset the weight parameters corresponding to the gates, x t Input at time t, h t-1 Represents the hidden state at time t-1, b r Is an offset term;
Updating door z t The function of the method is to filter a part of information of the hidden state at the last moment and keep useful information, and the calculation formula is as follows:
z t =σ(W z ·[h t-1 ,x t ]+b r )
wherein W is z Representing the weight parameter, x corresponding to the update gate t Input at time t, h t-1 The hidden state at the time t-1 is represented;
candidate state h at the present moment t Is from the hidden state h of the previous moment t-1 And reset gate r t Performing matrix point-by-point multiplication, and inputting x with current time t Converted from the tanh activation function, the calculation formula is as follows:
h t =tanh(W h ·[r t *h t-1 ,x t ]+b h )
wherein tanh () represents tanh activation function, W h Weight parameter representing candidate hidden layer, b h Is a bias term;
hidden state h at the current time t Is from the hidden state h of the previous moment t-1 And candidate state h at the current time t Respectively pass through the updating gates z t After the action, the calculation formula is as follows:
h t =(1-z t )*h t-1 +z t *h t
will h t As a temperature predicted value, the full connection layer is input.
The specific steps of the step 3 are as follows:
step 3.1, preliminarily setting model training parameters such as model training times, batch times, learning rate and the like according to the size and the structure of the training set in the step 1;
and 3.2, training a temperature prediction model based on the multi-channel gating circulating unit by taking the training set as input data.
The beneficial effects of the invention are as follows: the invention can extract the association relation between the features on a large scale by using the deep neural network model, takes the GRU model as a basic module, and gradually reduces the time step of each GRU model by defining a plurality of GRU models as auxiliary networks, thereby simultaneously capturing the features of the short-term mode and the longer-term mode of the cold chain compartment temperature, avoiding the prediction error of the longer-term interpretation variable due to the increase of information transmission, and further improving the prediction precision of the cold chain compartment temperature. The invention has technical support function for the application of extended deep learning in cold chain temperature prediction.
Drawings
FIG. 1 is a flow chart of a MC-GRU based cold chain temperature prediction method of the present invention;
FIG. 2 is a schematic diagram of a temperature prediction model based on a multi-channel gated loop unit according to the present invention;
fig. 3 is a block diagram of a gate control loop unit according to the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
As shown in FIG. 1, the cold chain temperature prediction method based on MC-GRU comprises the following steps:
step 1, acquiring historical temperature data of a cold chain transportation carriage environment, preprocessing the data, and then dividing a training set and a testing set; the method comprises the following specific steps:
step 1.1, acquiring first characteristic data TH of a cold chain transport carriage by using a temperature sensor 1 The system comprises a temperature sensor number, a collecting point position, environmental temperature data and a collecting time sequence;
step 1.2, processing the first characteristic data, and removing the repeated data to obtain second characteristic data TH 2 The method comprises the steps of acquiring time sequence and environmental temperature data; if a large amount of continuous sensor data is missing and the data cannot be supplemented, the segment of data is directly discarded; if only single-point sensor data is missing or data of a single time interval is missing, repairing the data by adopting a linear interpolation method; the formula of linear interpolation is:
wherein x is a+i For the data missing at the time a+i to be filled, x a 、x a+j The original sensor data at the time a and the time a+j are valid data values at the adjacent time a+i respectively;
step 1.3, the time sequence obtained in the step 1.2 is corresponding to the temperature data, and the time is used as a label of the temperature data; acquiring temperature data and labels thereof each of which is daily, weekly and monthly, and normalizing the temperature data to obtain third characteristic data TH 3 As a training set;
in order to ensure rapid convergence and accuracy in the neural network training process, each time series feature needs to be standardized. During the pretreatment, the time series is normalized to [0,1], specifically using a min-max normalization method. The calculation formula for the min-max normalization is as follows:
wherein x is norm Refers to normalized data, x refers to a characteristic sequence which is required to be normalized currently, x min And x max Representing the maximum and minimum values in the current feature sequence.
When the normalized data value is required to obtain a predicted value after training, performing inverse normalization operation, wherein an inverse normalization formula is as follows:
x real =(x max -x min )·x pnorm +x min
wherein x is real Refers to real dimension data, x after inverse normalization of predicted value pnorm For normalized evaluation value, x min And x max Representing the maximum and minimum values in the current feature sequence.
Step 1.4, dividing the third characteristic data TH3 into a training set and a testing set according to a certain proportion, wherein the specific dividing proportion can be determined by a user;
step 2, establishing a temperature prediction model based on a multi-channel gating cycle unit (Multichannel Gated Recurrent Units, MC-GRU); the specific method comprises the following steps: the temperature prediction model based on the multi-channel gating cycle unit comprises a plurality of gating cycle units and a full connection layer, as shown in fig. 2; when the traditional GRU model establishes the change relation between the unit of each time step and the temperature of the carriage, the unit of the longer time step continuously transmits along with the information, so that more units representing the shorter time step cannot accurately represent the change relation between the unit of the shorter time step and the temperature of the cold chain carriage, the autocorrelation of the capture interpretation variable of the prediction model is gradually reduced, and the uncertainty of the prediction result is increased. Thus, defining a plurality of gating loop units as an auxiliary network and progressively reducing the time step of each gating loop unit (the progressive reduction of circles in the rounded boxes in fig. 2 represents progressive reduction of the GRU model time step, which can be set directly in the GRU); the predicted values of all the gating circulating units are used as the input of a full-connection layer, so that the correlation between the predicted values and the temperatures at different historical moments can be enhanced, the increase of prediction model errors caused by information transmission of longer-term explanatory variables is avoided, and the prediction accuracy of the cold chain compartment temperature is improved; taking the output of the full connection layer as a final prediction result;
the gating circulation unit (Gated Recurrent Unit, GRU) is used as a variant network of the LSTM, long-term dependency relationship between long-time sequence information can be enhanced, GRU is reduced by one gating compared with the LSTM, the gating circulation unit has a simpler network structure, comprises a reset gate and an update gate, information transmission is carried out through a hidden state, model network parameters are fewer, and the calculation complexity of a model is reduced. Fig. 3 is a block diagram of a computing unit of the gate control loop unit (Gated Recurrent Unit, GRU), which has the following structure:
reset gate r t The method comprises the steps of controlling how much information at the previous moment can be reserved in a candidate state at the current moment, transmitting the input at the current moment and the hidden state at the previous moment to a sigmoid function, and calculating the formula:
r t =σ(W r ·[h t-1 ,x t ]+b r )
wherein σ () represents a sigmoid activation function, W r To reset the weight parameters corresponding to the gates, x t Input at time t, h t-1 Represents the hidden state at time t-1, b r Is a bias term;
updating door z t The function of the method is to filter a part of information of the hidden state at the last moment and keep useful information, and the calculation formula is as follows:
z t =σ(W z ·[h t-1 ,x t ]+b r )
wherein W is z Representing the weight parameter, x corresponding to the update gate t Input at time t, h t-1 The hidden state at the time t-1 is represented;
candidate state h at the present moment t Is from the hidden state h of the previous moment t-1 And reset gate r t Performing matrix point-by-point multiplication, and inputting x with current time t Converted from the tanh activation function, the calculation formula is as follows:
h t =tanh(W h ·[r t *h t-1 ,x t ]+b h )
wherein tanh () represents tanh activation function, W h Weight parameter representing candidate hidden layer, b h Is a bias term;
hidden state h at the current time t Is from the hidden state h of the previous moment t-1 And candidate state h at the current time t Respectively pass through the updating gates z t After the action, the calculation formula is as follows:
h t =(1-z t )*h t-1 +z t *h t
will h t As a temperature predicted value, the full connection layer is input.
Step 3, training a temperature prediction model based on the multi-channel gating circulating unit; the method comprises the following specific steps:
step 3.1, preliminarily setting model training parameters such as model training times, batch times, learning rate and the like according to the size and the structure of the training set in the step 1;
specifically, the MC-GRU model employs an adaptive moment estimation (Adaptive moment estimation, adam) optimizer. The Adam optimizer has the advantage that it uses the second order matrix information of the gradient, can converge to local optima faster, and adjusts the learning rate adaptively according to the parameters of the predictive model. At the same time, when the parameters converge to optimal values, the predictive model can be aided in updating the parameters with a slight gradient. For an optimal objective function of the MC-GRU prediction model, defining the optimal objective function of the MC-GRU model by applying a mean square error function:
where N represents the number of samples, y (t) represents the actual value of the cold chain car temperature, y MC-GRU And (t) is the output of the MC-GRU model at time step t.
Meanwhile, the learning rate is an important parameter in the temperature prediction model, and can influence the prediction performance of the model. A large number of experiments prove that the model is easy to fall into a local optimal solution in the training process due to the large learning rate; conversely, a smaller learning rate affects the model convergence rate, making it difficult to achieve global optimization. To solve this problem, an exponential decay learning rate method is applied to Adam to increase its convergence rate as follows:
learn i =learn 0 ×decay i/step
wherein, learn 0 Represents the initial learning rate, i is the iteration number, learn i For a learning rate of i iterations, decay is the decay rate. When step is set to 100 and decay is set to 0.96, i.e., 100 iterations of the model, the initial learning rate is reduced by a factor of 0.96.
And 3.2, training a temperature prediction model based on the multi-channel gating circulating unit by taking the training set as input data.
Step 4, testing the effect of the temperature prediction model based on the multi-channel gating circulating unit; the method comprises the following specific steps:
and 4.1, verifying the performance of the model, taking the test set in the step 1 as input data, and inputting the input data into the model to obtain a prediction result.
And 4.2, comparing a model prediction result with a real temperature, and evaluating the performance of the model by adopting the following four indexes: root mean Square error (Root Mean Square Error, RMSE), mean absolute error (Mean Absolute Error, MAE), mean Square error (Mean Square Error, MSE), coefficient (R-Square, R 2 ) And if the four indexes meet the preset requirements, finishing training to obtain a final prediction model, otherwise, adjusting parameters of the prediction model to retrain.
Specifically, the calculation formulas of the four indexes are as follows:
wherein,and y i Respectively representing a predicted value and a true value of the temperature; n represents the total number of test samples; />Representing the average of the temperature true values. The closer the values of RMSE, MAE and MSE are to 0, R 2 The closer to 1 the value of (c) represents the better the performance of the model.
The invention establishes a cold chain temperature prediction method based on MC-GRU, the characteristics of a short-term mode of the cold chain temperature are extracted by taking a traditional GRU network as a basic network, an auxiliary network consists of GRU models with different time steps so as to capture the characteristics of a long-term mode of the cold chain temperature, and the characteristics of different time modes of the cold chain temperature are fused to be used as the characteristics of the whole network model. The cold chain temperature data at the historical moment is input into the MC-GRU model to learn the periodic variation mode of the cold chain temperature, and the correlation between the past explanatory variables is enhanced, so that the accuracy of the prediction model is improved. The model not only plays the advantages of fewer GRU network parameters and low calculation complexity, but also solves the problem that the autocorrelation of the captured interpretation variable of the model is gradually reduced, and further improves the accuracy of cold chain carriage temperature prediction. The model provides a new thought for cold chain temperature prediction, and can meet the requirements of practical application.
Those of ordinary skill in the art will recognize that the embodiments described herein are for the purpose of aiding the reader in understanding the principles of the present invention and should be understood that the scope of the invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific modifications and combinations from the teachings of the present disclosure without departing from the spirit thereof, and such modifications and combinations remain within the scope of the present disclosure.
Claims (3)
1. The cold chain temperature prediction method based on MC-GRU is characterized by comprising the following steps of:
step 1, acquiring historical temperature data of a cold chain transportation carriage environment, preprocessing the data, and then dividing a training set and a testing set; the method comprises the following specific steps:
step 1.1, acquiring first characteristic data TH of a cold chain transport carriage by using a temperature sensor 1 The system comprises a temperature sensor number, a collecting point position, environmental temperature data and a collecting time sequence;
step 1.2, processing the first characteristic data, and removing the repeated data to obtain second characteristic data TH 2 The method comprises the steps of acquiring time sequence and environmental temperature data; if a large amount of continuous sensor data is missing and the data cannot be supplemented, the segment of data is directly discarded; if only a single point sensor data is missingIf the data of a single time interval is lost, repairing the data by adopting a linear interpolation method; the formula of linear interpolation is:
wherein x is a+i For the data missing at the time a+i to be filled, x a 、x a+j The original sensor data at the time a and the time a+j are valid data values at the adjacent time a+i respectively;
step 1.3, the time sequence obtained in the step 1.2 is corresponding to the temperature data, and the time is used as a label of the temperature data; acquiring temperature data and labels thereof each of which is daily, weekly and monthly, and normalizing the temperature data to obtain third characteristic data TH 3 As a training set;
step 1.4, third characteristic data TH 3 Dividing a training set and a testing set;
step 2, establishing a temperature prediction model based on a multi-channel gating circulating unit;
step 3, training a temperature prediction model based on the multi-channel gating circulating unit;
step 4, testing the effect of the temperature prediction model based on the multi-channel gating circulating unit; the method comprises the following specific steps:
step 4.1, verifying the performance of the model, taking the test set in the step 1 as input data, and inputting the model to obtain a prediction result;
and 4.2, comparing a model prediction result with a real temperature, and evaluating the performance of the model by adopting the following four indexes: and (3) determining coefficients according to the root mean square error, the average absolute error and the mean square error, ending training if the four indexes meet the preset requirements, obtaining a final prediction model, otherwise, adjusting parameters of the prediction model to retrain.
2. The MC-GRU-based cold chain temperature prediction method according to claim 1, characterized in that the specific method of step 2 is as follows: the temperature prediction model based on the multi-channel gating circulating unit comprises a plurality of gating circulating units and a full connection layer; defining a plurality of gating cycle units as an auxiliary network, and gradually reducing the time step of each gating cycle unit; taking predicted values of all the gating circulating units as input of a full-connection layer, and taking output of the full-connection layer as a final predicted result;
the structure of the gating cycle unit is as follows:
reset gate r t The method comprises the steps of controlling how much information at the previous moment can be reserved in a candidate state at the current moment, transmitting the input at the current moment and the hidden state at the previous moment to a sigmoid function, and calculating the formula:
r t =σ(W r ·[h t-1 ,x t ]+b r )
wherein σ () represents a sigmoid activation function, W r To reset the weight parameters corresponding to the gates, x t Input at time t, h t-1 Represents the hidden state at time t-1, b r Is a bias term;
updating door z t The function of the method is to filter a part of information of the hidden state at the last moment and keep useful information, and the calculation formula is as follows:
z t =σ(W z •[h t-1 ,x t ]+b r )
wherein W is z Representing the weight parameter, x corresponding to the update gate t Input at time t, h t-1 The hidden state at the time t-1 is represented;
candidate state h at the present moment t Is from the hidden state h of the previous moment t-1 And reset gate r t Performing matrix point-by-point multiplication, and inputting x with current time t Converted from the tanh activation function, the calculation formula is as follows:
h t =tanh(W h ·[r t *h t-1 ,x t ]+b h )
wherein tanh () represents tanh activation function, W h Weight parameter representing candidate hidden layer, b h Is a bias term;
hidden state at current timeState h t Is from the hidden state h of the previous moment t-1 And candidate state h at the current time t Respectively pass through the updating gates z t After the action, the calculation formula is as follows:
h t =(1-z t )*h t-1 +z t *h t
will h t As a temperature predicted value, the full connection layer is input.
3. The MC-GRU-based cold chain temperature prediction method according to claim 1, characterized in that step 3 comprises the following specific steps:
step 3.1, preliminarily setting model training times, batch number and learning rate according to the size and structure of the training set in the step 1;
and 3.2, training a temperature prediction model based on the multi-channel gating circulating unit by taking the training set as input data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311330166.5A CN117272829A (en) | 2023-10-13 | 2023-10-13 | MC-GRU-based cold chain temperature prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311330166.5A CN117272829A (en) | 2023-10-13 | 2023-10-13 | MC-GRU-based cold chain temperature prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117272829A true CN117272829A (en) | 2023-12-22 |
Family
ID=89212166
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311330166.5A Pending CN117272829A (en) | 2023-10-13 | 2023-10-13 | MC-GRU-based cold chain temperature prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117272829A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117875186A (en) * | 2024-01-16 | 2024-04-12 | 西南石油大学 | GRU-based sulfur deposition prediction method for high-sulfur gas reservoir gathering and transportation pipeline |
-
2023
- 2023-10-13 CN CN202311330166.5A patent/CN117272829A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117875186A (en) * | 2024-01-16 | 2024-04-12 | 西南石油大学 | GRU-based sulfur deposition prediction method for high-sulfur gas reservoir gathering and transportation pipeline |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
CN111967688B (en) | Power load prediction method based on Kalman filter and convolutional neural network | |
CN108898215B (en) | Intelligent sludge bulking identification method based on two-type fuzzy neural network | |
CN111079989B (en) | DWT-PCA-LSTM-based water supply amount prediction device for water supply company | |
CN103345559B (en) | The dynamic evolutionary modeling method of aluminium electrolysis process art electrolysis cells energy consumption | |
CN108764539B (en) | Upstream and downstream water level prediction method for cascade power station | |
CN109146162B (en) | A kind of probability wind speed forecasting method based on integrated Recognition with Recurrent Neural Network | |
CN117272829A (en) | MC-GRU-based cold chain temperature prediction method | |
CN111030889B (en) | Network traffic prediction method based on GRU model | |
CN112130086B (en) | Method and system for predicting remaining life of power battery | |
CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
CN111160659A (en) | Power load prediction method considering temperature fuzzification | |
CN116500454A (en) | Lithium ion battery health state estimation method, system, equipment and medium based on multi-feature input time sequence model | |
CN115840917A (en) | Motor temperature rise prediction method | |
CN114596726B (en) | Parking berth prediction method based on interpretable space-time attention mechanism | |
CN113988415B (en) | Medium-and-long-term power load prediction method | |
CN110245398B (en) | Soft measurement deep learning method for thermal deformation of air preheater rotor | |
CN118095570A (en) | Intelligent load prediction method and system for transformer area, electronic equipment, medium and chip | |
CN113743008A (en) | Fuel cell health prediction method and system | |
CN113159395A (en) | Deep learning-based sewage treatment plant water inflow prediction method and system | |
CN118017482A (en) | Flexible climbing capacity demand analysis method based on prediction error feature extraction | |
CN117538783A (en) | Lithium ion battery state of charge estimation method based on time domain fusion converter | |
CN116646929A (en) | PSO-CNN-BILSTM-based short-term wind power prediction method | |
CN116976490A (en) | Multivariable time sequence prediction method based on maximized mutual information entropy improvement | |
CN116627194A (en) | WOA-LSTM temperature control method based on OLED thermal experiment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |