Material data processing method based on deep learning and model training

Abstract

The invention discloses a material data processing method based on deep learning and model training, and particularly relates to the field of data processing. According to the invention, the quantity of drinking water, tents, emergency food and iodine is collected from a material data storage library, and the quantity of injured persons and the quantity of survival persons in a disaster-affected area are obtained from a site after an earthquake occurs locally, so that data preprocessing is carried out; the method comprises the following steps: preprocessing data, inputting the preprocessed data into a convolutional neural network model for training, inputting previous seismic data into the trained model for application, verifying whether a preset accuracy rate is reached or not, transmitting an output result to a client, correcting by the client according to an output instruction, and performing inventory prediction and demand analysis on the model. Whether the model can be used for subsequent events is judged according to the analysis result and the output instruction, so that the efficiency of classifying the relief goods when an earthquake occurs is greatly improved, and the error rate is reduced.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a material data processing method based on deep learning and model training. Background Art

[0002] Earthquake disasters are one of the most serious sudden disasters in the world and are often referred to as the worst of all disasters.

[0003] However, it still has some shortcomings in actual use. For example, in the current research on the structure of emergency rescue materials, there is less research on the classification of emergency rescue materials. They are often mainly classified based on the characteristics of ordinary materials manually, which is inefficient and has a high error rate. Therefore, material data processing methods based on deep learning and model training came into being. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a material data processing method based on deep learning and model training to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A1: Obtain material-related data from the material system and then pre-process the relevant data;

[0007] A2: Input the processed data into the convolutional neural network model for training;

[0008] A3: Apply the trained model to perform inventory forecasting and demand analysis;

[0009] A4: Compare the values ​​of inventory forecast and demand analysis by threshold, output instructions based on the comparison results, and finally evaluate the model through the output instructions.

[0010] Preferably, in the acquisition of relevant data, data of stored items are collected from a material data repository, including the quantity of drinking water, tents, emergency food and iodine; and the number of injured persons and the number of survivors in the disaster-stricken area are obtained from the local site after the earthquake.

[0011] The relevant data is pre-processed, and the received data is subjected to data verification processing to check whether it is consistent with the quantity in the inventory. If it is inconsistent, an inventory quantity inconsistency instruction is sent to the client, and correction is performed according to the instruction;

[0012] The data of the collected items after preprocessing, the number of injured people in the disaster-stricken areas, and the number of survivors in the disaster-stricken areas are transferred to the convolutional neural network model for training.

[0013] Preferably, the neural network model is trained by inputting data of items collected after preprocessing, the number of injured persons in the disaster-stricken area, and the number of survivors in the disaster-stricken area;

[0014] The time when the area enters the rescue period after the disaster occurs is time zero; from this time on, the time unit is hours, and each time is recorded as t (t = 1, 2, 3...n);

[0015] The specific calculation method for the required quantity of rescue materials and daily materials is as follows:

[0016] Among them, Q(t) represents the amount of emergency relief supplies required in the disaster-stricken area from the occurrence of the earthquake to time t, q represents the time and location adjustment coefficient of the earthquake disaster, γ represents the number of injured people in the disaster-stricken area, μ represents the number of survivors in the disaster-stricken area, and α k represents the per capita demand coefficient of rescue supplies k, β j represents the per capita demand coefficient of daily material j; for per capita demand, it should be noted that since the materials themselves are divided into consumable materials and equipment materials; the former refers to materials that are consumed daily and need to be continuously supplied during the rescue process, while the latter refers to materials that are supplied once and can be continuously used during the rescue process. In order to distinguish the difference between the two, in the actual calculation process, the per capita demand α k and β j In the selection, the unit of equipment materials is the overall per capita demand, while the unit of consumable materials is the per capita demand per unit time.

[0017] Preferably, when the selected material is drinking water, Among them, α 水 It is expressed as the per capita demand coefficient of drinking water for rescue supplies, and w is the daily supply of drinking water for each person; drinking water refers to water that is directly drunk by the human body; it is generally believed that people's demand for drinking water far exceeds their demand for food, especially in the case of spatial isolation in emergencies, the guarantee of drinking water is a basic life project. After an earthquake disaster, the demand for drinking water by a large number of disaster-stricken people in the disaster-stricken areas is often the most prominent; ensuring the smooth and stable supply of drinking water is an important task in emergency rescue of sudden earthquake disasters; and as daily and consumable rescue supplies, the demand for drinking water in sudden earthquake disasters is related to the total number of survivors and per capita demand in the disaster-stricken areas. At the same time, due to the continuous consumption of drinking water, it needs to be continuously supplied.

[0018] When the selected supplies are emergency food, Among them, α 应急食品It is expressed as the per capita demand coefficient of emergency food for rescue supplies, and r is expressed as the daily food supply per person; emergency food refers to food that is easy to carry and eat to ensure basic body functions in an emergency; typical emergency foods usually supplied in disaster-stricken areas include instant noodles and finished grains. Similar to drinking water, emergency food is also a daily and consumable material. In sudden earthquake disasters, the demand for emergency food is related to the total number of survivors and per capita demand in the disaster-stricken area. At the same time, since emergency food will be consumed continuously, it needs to be supplied continuously.

[0019] If the selected material is a tent, where α 帐篷 It is expressed as the per capita demand coefficient of tents for rescue supplies, and p is the area of ​​the tents. Tents are temporary shelters for shelter from wind and rain. In earthquake disasters, due to the damage to buildings and the possibility of aftershocks, the temporary residence of the victims mainly depends on tents. Later on-site emergency command, medical rescue and other work also have certain demands for tents. As equipment-type rescue supplies, tents can be used continuously after a one-time supply, and tents for emergency rescue supplies are uniformly stipulated by the state.

[0020] When the selected material is iodine tincture, according to previous research results, the approximate value of the demand for iodine tincture is selected, then α 碘酒 =0.08, where α 碘酒 It is expressed as the per capita demand coefficient of rescue supplies iodine tincture; iodine tincture is a typical drug used to disinfect skin infections. Emergency rescue medical work during the critical rescue period after an earthquake mainly involves hemostasis and disinfection of injured people. Therefore, iodine tincture, as a rescue consumable, is directly related to the number of injured people in the disaster area and needs to be continuously supplied.

[0021] Preferably, the trained model is input into past earthquake data for application and verification of whether the preset accuracy is achieved. If the verified data accuracy is higher than the preset accuracy, the correct model instruction is output, the model is maintained, and the model is used for inventory forecasting and demand analysis, and the analysis results are input into the client. If the verified data accuracy is lower than the preset accuracy, the model error instruction is output, and professional personnel are required to make corrections.

[0022] Preferably, when the model is used for prediction and demand analysis, the analysis result is received. If the prediction value output after the analysis is greater than a preset prediction threshold, a prediction success instruction is output, and the model can be applied to subsequent events; if the prediction value output after the analysis is less than the preset prediction threshold, a prediction failure instruction is output, and the model cannot be applied to subsequent events, and the model is handed over to professionals for correction.

[0023] Preferably, a neural network is designed: the basic network structure is divided into an input layer, a hidden layer and an output layer; the previous step has clarified the two elements of the output layer and the fifteen elements of the input layer; the structural design mainly determines the number of hidden layers and the number of elements; theoretical research and practical experiments show that the methods to improve accuracy mainly include increasing the number of hidden layers and the number of elements; the higher the number of hidden layers, the better the operation effect; but too high a number of hidden layers often makes the network system redundant; increasing the number of hidden layer elements can also improve the operation accuracy, and it is easier to implement in operation; based on the above advantages and disadvantages, it is selected to start from the number of hidden layer elements, that is, a three-layer structure is used to adjust the number of hidden layer elements; according to the mapping network existence theorem, when the transfer function is appropriately selected, the three-layer structure can approximate any rational function, thereby meeting the prediction requirements, and the selection of the number of hidden layer elements generally adopts the trial method, that is, the reference number values ​​of hidden layer elements provided by different formulas are included in the trial range; the errors of different numbers of elements under the same circumstances are compared one by one, and the minimum value is finally selected.

[0024] Optimized neural network learning: Before learning, the values ​​of the input layer elements need to be processed so that they are between 0 and 1. This is mainly because the input layer elements represent different properties and have different dimensions. Standard normalization processing can make the elements of each input layer comparable to each other. It also takes into account the requirements of the value range of the transfer function. The function expression of the normalization processing is:

[0025] where x c is the input training data, x min is the minimum value of the input training data, x min is the maximum value of the input training data.

[0026] Technical effects and advantages of the present invention:

[0027] The present invention provides a material data processing method based on deep learning and model training. The present invention collects the quantities of drinking water, tents, emergency food and iodine from a material data storage repository and obtains the number of injured persons and the number of survivors in the disaster-stricken area from the local earthquake site. After processing, the data are input into an aggregated neural network model for training. After the training is completed, the model is put into use. Whether the model meets the standard is determined according to the accuracy rate. Inventory forecasting and demand analysis are then performed on the model. Finally, whether the model can be used in the future is determined according to the analysis results and output instructions. Therefore, the error rate can be reduced by applying the model to process material data, and the efficiency of classifying rescue materials in response to earthquakes is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0030] See also Figure 1 As shown, the present invention provides Figure 1 As shown, the present invention provides a material data processing method based on deep learning and model training, comprising the following steps:

[0031] A1: Obtain material-related data from the material system and then pre-process the relevant data;

[0032] In the acquisition of relevant data, data on stored items are collected from the material data repository, including the quantity of drinking water, tents, emergency food and iodine; and the number of injured persons and the number of survivors in the disaster-stricken area are obtained from the local site after the earthquake.

[0033] The relevant data is pre-processed, and the received data is subjected to data verification processing to check whether it is consistent with the quantity in the inventory. If it is inconsistent, an inventory quantity inconsistency instruction is sent to the client, and correction is performed according to the instruction;

[0034] The data of the collected items after preprocessing, the number of injured people in the disaster-stricken areas, and the number of survivors in the disaster-stricken areas are transferred to the convolutional neural network model for training.

[0035] A2: Input the processed data into the convolutional neural network model for training;

[0036] In the neural network model, data of items collected after preprocessing, the number of injured persons in the disaster-stricken area, and the number of surviving persons in the disaster-stricken area are input for training;

[0037] The time when the area enters the rescue period after the disaster occurs is time zero; from this time on, the time unit is hours, and each time is recorded as t (t = 1, 2, 3...n);

[0038] The specific calculation method for the required quantity of rescue materials and daily materials is as follows:

[0039] Among them, Q(t) represents the amount of emergency relief supplies required in the disaster-stricken area from the occurrence of the earthquake to time t, q represents the time and location adjustment coefficient of the earthquake disaster, γ represents the number of injured people in the disaster-stricken area, μ represents the number of survivors in the disaster-stricken area, and α k represents the per capita demand coefficient of rescue supplies k, β j represents the per capita demand coefficient of daily material j; for per capita demand, it should be noted that since the materials themselves are divided into consumable materials and equipment materials; the former refers to materials that are consumed daily and need to be continuously supplied during the rescue process, while the latter refers to materials that are supplied once and can be continuously used during the rescue process. In order to distinguish the difference between the two, in the actual calculation process, the per capita demand α k and β j In the selection, the unit of equipment materials is the overall per capita demand, while the unit of consumable materials is the per capita demand per unit time.

[0040] Among them, when the selected material is drinking water, Among them, α 水 It is expressed as the per capita demand coefficient of drinking water for rescue supplies, and w is the daily supply of drinking water for each person; drinking water refers to water that is directly drunk by the human body; it is generally believed that people's demand for drinking water far exceeds their demand for food, especially in the case of spatial isolation in emergencies, the guarantee of drinking water is a basic life project. After an earthquake disaster, the demand for drinking water by a large number of people in the disaster-stricken areas is often the most prominent; ensuring the smooth and stable supply of drinking water is an important task in emergency rescue of sudden earthquake disasters; and as daily and consumable rescue supplies, the demand for drinking water in sudden earthquake disasters is related to the total number of survivors and per capita demand in the disaster-stricken areas. At the same time, since drinking water will be consumed continuously, it needs to be continuously supplied.

[0041] When the selected supplies are emergency food, Among them, α 应急食品 It is expressed as the per capita demand coefficient of emergency food for rescue supplies, and r is expressed as the daily food supply per person; emergency food refers to food that is easy to carry and eat to ensure basic body functions in an emergency; typical emergency foods usually supplied in disaster-stricken areas include instant noodles and finished grains. Similar to drinking water, emergency food is also a daily and consumable material. In sudden earthquake disasters, the demand for emergency food is related to the total number of survivors and per capita demand in the disaster-stricken area. At the same time, since emergency food will be consumed continuously, it needs to be supplied continuously.

[0042] When the selected material is a tent, where α 帐篷It is expressed as the per capita demand coefficient of tents for rescue supplies, and p is the area of ​​the tents. Tents are temporary shelters for shelter from wind and rain. In earthquake disasters, due to the damage to buildings and the possibility of aftershocks, the temporary residence of the victims mainly depends on tents. Later on-site emergency command, medical rescue and other work also have certain demands for tents. As equipment-type rescue supplies, tents can be used continuously after a one-time supply, and tents for emergency rescue supplies are uniformly stipulated by the state.

[0043] When the selected material is iodine tincture, according to previous research results, the approximate value of the demand for iodine tincture is selected, then α 碘酒 =0.08; where α 碘酒 It is expressed as the per capita demand coefficient of rescue supplies iodine tincture; iodine tincture is a typical drug used to disinfect skin infections. Emergency rescue medical work during the critical rescue period after an earthquake mainly involves hemostasis and disinfection of injured people. Therefore, iodine tincture, as a rescue consumable, is directly related to the number of injured people in the disaster area and needs to be continuously supplied.

[0044] A3: Apply the trained model to perform inventory forecasting and demand analysis;

[0045] The trained model is input into past earthquake data for application and verification to see if it reaches the preset accuracy. If the verified data accuracy is higher than the preset accuracy, the correct model instruction is output, the model is maintained, and the model is used for inventory forecasting and demand analysis, and the analysis results are input into the client. If the verified data accuracy is lower than the preset accuracy, the model error instruction is output, and correction is required by professionals.

[0046] A4: Compare the values ​​of inventory forecast and demand analysis by threshold, output instructions based on the comparison results, and finally evaluate the model based on the output instructions.

[0047] When the model is used for inventory forecasting and demand analysis, the analysis results are received. If the forecast value output after the analysis is greater than the preset forecast threshold, a forecast success instruction is output, and the model can be applied to subsequent events; if the forecast value output after the analysis is less than the preset forecast threshold, a forecast failure instruction is output, and the model cannot be applied to subsequent events, and the model is handed over to professionals for correction.

[0048] Designing a neural network: The basic network structure is divided into an input layer, a hidden layer, and an output layer. The two elements of the output layer and the fifteen elements of the input layer have been identified in the previous step. The structural design mainly involves determining the number of hidden layers and the number of elements. Theoretical research and practical experiments show that the methods to improve accuracy mainly include increasing the number of hidden layers and the number of elements. The higher the number of hidden layers, the better the calculation effect. However, too high a number of hidden layers often makes the network system redundant. Increasing the number of hidden layer elements can also improve the calculation accuracy, and it is easier to implement in operation. Based on the above advantages and disadvantages, the number of hidden layer elements is selected as the starting point, that is, a three-layer structure is used to adjust the number of hidden layer elements. According to the existence theorem of the mapping network, when the transfer function is appropriately selected, the three-layer structure can approximate any rational function, thereby meeting the prediction requirements. The number of hidden layer elements is generally selected by trial and error, that is, the reference number of hidden layer elements provided by different formulas is included in the trial range. The errors of different numbers of elements under the same circumstances are compared one by one, and the minimum value is finally selected.

[0049] Neural network learning: Before learning, the values ​​of the input layer elements need to be processed so that they are between 0 and 1. This is mainly because the input layer elements represent different properties and have different dimensions. Standard normalization can make the elements of each input layer comparable to each other. It also takes into account the requirements of the value range of the transfer function. The function expression of the normalization processing is:

[0050] where x c is the input training data, x min is the minimum value of the input training data, x min is the maximum value of the input training data.

[0051] It should be noted that the preset values ​​in the present invention are determined according to specific circumstances, that is, this embodiment does not specifically limit the specific values.

[0052] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A material data processing method based on deep learning and model training, characterized in that: include: A1: Obtain material-related data from the material system and then pre-process the relevant data; A2: Input the processed data into the convolutional neural network model for training; A3: Apply the trained model to perform inventory forecasting and demand analysis; A4: Compare the values ​​of inventory forecast and demand analysis by threshold, output instructions based on the comparison results, and finally evaluate the model through the output instructions.

2. The material data processing method based on deep learning and model training according to claim 1 is characterized in that: In the acquisition of relevant data, data on stored items are collected from the material data repository, including the quantity of drinking water, tents, emergency food and iodine; and the number of injured persons and the number of survivors in the disaster-stricken area are obtained from the local site after the earthquake. The relevant data is pre-processed, and the received data is subjected to data verification processing to check whether it is consistent with the quantity in the inventory. If it is inconsistent, an inventory quantity inconsistency instruction is sent to the client, and correction is performed according to the instruction; The data of the collected items after preprocessing, the number of injured people in the disaster-stricken areas, and the number of survivors in the disaster-stricken areas are transferred to the convolutional neural network model for training.

3. The material data processing method based on deep learning and model training according to claim 1 is characterized in that: In the neural network model, data of items collected after preprocessing, the number of injured persons in the disaster-stricken area, and the number of surviving persons in the disaster-stricken area are input for training; The time when the area enters the rescue period after the disaster occurs is time zero; from this time on, the time unit is hours, and each time is recorded as t (t = 1, 2, 3...n); The specific calculation method for the required quantity of rescue materials and daily materials is as follows: Among them, Q(t) represents the amount of emergency relief supplies required in the disaster-stricken area from the occurrence of the earthquake to time t, q represents the time and location adjustment coefficient of the earthquake disaster, γ represents the number of injured people in the disaster-stricken area, μ represents the number of survivors in the disaster-stricken area, and α k represents the per capita demand coefficient of rescue supplies k, β j represents the per capita demand coefficient of daily material j; for per capita demand, it should be noted that since the materials themselves are divided into consumable materials and equipment materials; the former refers to materials that are consumed daily and need to be continuously supplied during the rescue process, while the latter refers to materials that are supplied once and can be continuously used during the rescue process; in order to distinguish the difference between the two, in the actual calculation process, the per capita demand α k and β j In the selection, the unit of equipment materials is the overall per capita demand, while the unit of consumable materials is the per capita demand per unit time.

4. The material data processing method based on deep learning and model training according to claim 1 is characterized in that: When the selected material is drinking water, Among them, α 水 It is represented by the per capita demand coefficient of drinking water for rescue supplies, and w is the daily supply of drinking water for each person; When the selected supplies are emergency food, Among them, α 应急食品 It is expressed as the per capita demand coefficient of emergency food for rescue supplies, and r is expressed as the daily food supply per person; If the selected material is a tent, where α 帐篷 It is expressed as the per capita demand coefficient of tents for rescue supplies, and p is the area of ​​the tents; When the selected material is iodine tincture, according to previous research results, the approximate value of the demand for iodine tincture is selected, then α 碘酒 =0.08, where α 碘酒 Expressed as the per capita demand coefficient for rescue supplies iodine.

5. The material data processing method based on deep learning and model training according to claim 1 is characterized in that: The trained model is input into past earthquake data for application and verification to see whether the preset accuracy is achieved. If the verified data accuracy is higher than the preset accuracy, the correct model instructions are output, the model is maintained, and the model is used for inventory forecasting and demand analysis, and the analysis results are input into the client. If the verified data accuracy is lower than the preset accuracy, the model error instructions are output, and professional personnel are required to make corrections.

6. The material data processing method based on deep learning and model training according to claim 1 is characterized in that: In the process of performing prediction and demand analysis on the model, the analysis result is received, and if the prediction value output after analysis is greater than a preset prediction threshold, a prediction success instruction is output, and the model can be applied to subsequent events; If the predicted value output after analysis is less than the preset prediction threshold, a prediction failure instruction is output, the model cannot be applied to subsequent events, and the model is handed over to professionals for correction.

Images