Adaptive environmental factor radiation dose evaluation method and system

Abstract

The invention discloses a self-adaptive environmental factor radiation dose evaluation method and system, and relates to the technical field of radiation monitoring, and the method comprises the steps: arranging a sensor to collect evaluation data, and carrying out the preprocessing of the collected data; constructing a radiation dose evaluation model based on the multi-layer perceptron model; performing model training on the radiation dose evaluation model; and inputting the preprocessed evaluation data into the radiation dose evaluation model, and outputting an evaluation result. According to the adaptive environmental factor radiation dose evaluation method provided by the invention, the radiation dose evaluation model is adaptively adjusted according to the environmental parameter change through the multi-layer perceptron model, and the evaluation precision of the radiation dose under different environmental conditions is ensured. Different from a traditional radiation monitoring method using a fixed model, the method can adjust model parameters in real time, and avoids data errors caused by sensor performance fluctuation due to environmental changes.

Description

Technical Field

[0001] The present invention relates to the field of radiation monitoring technology, and in particular to an adaptive environmental factor radiation dose assessment method and system. Background Art

[0002] In the field of radiation monitoring, accurate assessment of radiation dose is crucial for environmental protection, personnel safety and nuclear facility management. Existing radiation monitoring systems usually rely on fixed models to assess radiation dose, which are designed and calibrated according to specific environmental parameters. However, environmental conditions such as temperature, humidity and atmospheric pressure often change, which may affect the measurement accuracy of the sensor, resulting in errors in radiation dose assessment.

[0003] At present, although some radiation monitoring systems have begun to introduce data processing algorithms and machine learning technologies to improve assessment accuracy, most of these systems rely on complex cloud processing and require devices to be continuously connected to the network for data transmission and processing. For portable devices or in environments with unstable network conditions, existing technologies are difficult to ensure the real-time and accuracy of radiation data. In addition, existing technologies lack the ability to flexibly adjust assessment models in the face of complex environmental changes, resulting in inaccurate assessment results. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing radiation monitoring method has the problem that the radiation dose assessment model cannot adapt in real time due to the dynamic changes of environmental parameters, and how to adjust the assessment model in real time without a network connection and avoid the optimization problem of the model relying on complex configuration.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an adaptive environmental factor radiation dose assessment method, comprising:

[0007] Arrange sensors to collect evaluation data and pre-process the collected data;

[0008] Based on the multi-layer perceptron model, a radiation dose assessment model is constructed;

[0009] Performing model training on the radiation dose assessment model;

[0010] The preprocessed evaluation data is input into the radiation dose evaluation model and the evaluation results are output.

[0011] As a preferred solution of the adaptive environmental factor radiation dose assessment method of the present invention, wherein: the assessment data includes environmental parameters and radiation dose;

[0012] The environmental parameters include temperature, humidity and atmospheric pressure;

[0013] The radiation dose mentioned refers to the radiation monitoring dose rate;

[0014] The preprocessing includes data cleaning and data standardization.

[0015] As a preferred solution of the adaptive environmental factor radiation dose assessment method of the present invention, wherein: after the collected data is preprocessed, feature selection and feature construction are performed through the preprocessed data;

[0016] The feature selection uses random forest based on LASSO technology to determine the environmental parameters that have the greatest impact on radiation dose assessment;

[0017] The feature construction includes constructing new features based on business knowledge;

[0018] The new features include the rate of change of environmental parameters and the accumulated radiation.

[0019] As a preferred embodiment of the adaptive environmental factor radiation dose assessment method of the present invention, the rate of change of the environmental parameter is expressed as:

[0020]

[0021] Among them, E t and E t-1 are the parameter values ​​of two consecutive time points, and Δt is the time interval.

[0022] As a preferred embodiment of the adaptive environmental factor radiation dose assessment method of the present invention, the radiation accumulation is expressed as:

[0023]

[0024] Among them, R i is the radiation dose rate at each time point, and n is the window size.

[0025] As a preferred solution of the adaptive environmental factor radiation dose assessment method of the present invention, wherein: the radiation dose assessment model includes an input layer, a hidden layer and an output layer;

[0026] The hidden layer uses the ReLU activation function, and the output layer uses a linear or nonlinear activation function as needed.

[0027] As a preferred solution of the adaptive environmental factor radiation dose assessment method described in the present invention, the model training adopts the Adam optimizer, combined with the learning rate adjustment strategy, uses the mean square error as the loss function, and adds the Dropout layer and L1, L2 regularization technology.

[0028] Another object of the present invention is to provide an adaptive environmental factor radiation dose assessment system, which can solve the problem that existing radiation monitoring equipment cannot adjust the model in real time when environmental conditions change by constructing an adaptive environmental factor radiation dose assessment system, and avoid the situation where the assessment accuracy is reduced due to the influence of the environment on sensor performance.

[0029] To solve the above technical problems, the present invention provides the following technical solutions: an adaptive environmental factor radiation dose assessment system, comprising: a data acquisition module, a model construction module, a model training module and a dose assessment module; the data acquisition module is used to arrange sensors to collect assessment data and pre-process the collected data; the model construction module is used to construct a radiation dose assessment model based on a multi-layer perceptron model; the model training module is used to perform model training on the radiation dose assessment model; the dose assessment module is used to input the pre-processed assessment data into the radiation dose assessment model and output the assessment result.

[0030] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned adaptive environmental factor radiation dose assessment method when executing the computer program.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned adaptive environmental factor radiation dose assessment method.

[0032] Beneficial effects of the present invention: The adaptive environmental factor radiation dose assessment method provided by the present invention adaptively adjusts the radiation dose assessment model according to changes in environmental parameters through a multi-layer perceptron model, thereby ensuring the assessment accuracy of radiation dose under different environmental conditions.

[0033] Different from the traditional radiation monitoring method using a fixed model, the present invention can adjust the model parameters in real time to avoid data errors caused by fluctuations in sensor performance due to environmental changes.

[0034] In addition, through automated feature selection technology, the present invention can effectively screen out the most important environmental parameters for radiation dose assessment, optimize the data processing process, reduce the assessment error caused by inefficient feature selection, and ensure the reliability of the assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0036] Figure 1 An overall flow chart of an adaptive environmental factor radiation dose assessment method provided by an embodiment of the present invention.

[0037] Figure 2 An overall structural diagram of an adaptive environmental factor radiation dose assessment system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0040] Example 1

[0041] Reference Figure 1 , as an embodiment of the present invention, provides an adaptive environmental factor radiation dose assessment method, comprising:

[0042] S1 arranges sensors to collect evaluation data and pre-processes the collected data;

[0043] S2 builds a radiation dose assessment model based on a multi-layer perceptron model;

[0044] S3 performs model training on the radiation dose assessment model;

[0045] S4 inputs the preprocessed evaluation data into the radiation dose evaluation model and outputs the evaluation results.

[0046] The evaluation data include environmental parameters and radiation dose;

[0047] The environmental parameters include temperature, humidity and atmospheric pressure;

[0048] The radiation dose mentioned refers to the radiation monitoring dose rate;

[0049] The preprocessing includes data cleaning and data standardization.

[0050] It should be noted that the temperature (unit: °C) is collected by the temperature sensor, ranging from -40 °C to 125 °C. Humidity (unit: %RH) is collected by the humidity sensor, ranging from 0% to 100%RH. Atmospheric pressure (unit: hPa) is collected by the atmospheric pressure sensor, ranging from 300hPa to 1100hPa. Radiation dose rate (unit: μSv / h) is collected by radiation detectors such as GM tubes, ranging from environmental reference background levels to several hundred mSv / h.

[0051] Data cleaning is done by detecting and correcting missing values ​​and outliers in the data.

[0052] For example, missing temperature data (set as T) can be filled with the average temperature of the surrounding time points, and the missing value is:

[0053] For abnormal values ​​of humidity, the standard deviation based on the median is used to judge and filter: assuming that the humidity is H and the median is median(H), the values ​​exceeding median(H)±2σ are removed.

[0054] Data normalization Normalize temperature T, humidity H, atmospheric pressure P and radiation dose rate R:

[0055] Normalize for temperature:

[0056] Normalize for humidity:

[0057] The above standardization of temperature and humidity, the standardization formula is as follows:

[0058]

[0059] Among them, Z represents the standardized value, the data value after standardization. It is used to convert the original data to a standard normal distribution with a mean of 0 and a standard deviation of 1. This operation can eliminate the impact of dimensional differences between different features and improve data consistency. X represents the original data value, the original feature value collected from the sensor. Specifically, X may be the value of temperature, humidity, atmospheric pressure or radiation dose rate at a certain moment. For example, the temperature data T = 22.5℃ at a certain sampling. μ represents the sample mean, the mean of all collected data, which represents the average level of a certain feature. For temperature data T, μ T represents the average value of all temperature data. For example, if we collected 100 temperature data points, μT It is the average value of these 100 temperature values. For example: the mean value of temperature μ T =23.0℃, σ represents the sample standard deviation, the standard deviation of all collected data, indicating the fluctuation of a certain feature. For temperature data T, σ T Indicates the standard deviation of temperature values, reflecting the degree of temperature dispersion. For example, if the fluctuation of temperature data is small, the standard deviation is small; if the fluctuation is large, the standard deviation is relatively large. For example: the standard deviation of temperature σ T =2.5℃.

[0060] Assume that the temperature data of a certain sampling is X T =22.5℃, the mean of temperature data is μ T =23.0℃, standard deviation is σ T =2.5℃, then the standardized temperature value Z T The calculation is as follows:

[0061]

[0062] At this time, the standardized temperature data Z T =-0.2, indicating that the temperature value is slightly lower than the average temperature by 0.2 standard deviations.

[0063] Feature selection refers to the input of standardized data into the model for feature selection. Random forest will determine the impact of different features (temperature, humidity, atmospheric pressure, etc.) on radiation dose by constructing multiple decision trees.

[0064] MLP model training refers to the input of standardized data into the MLP model as the input features of the model. The standardized data makes each feature in the same order of magnitude, which can improve the training stability of the model and ensure the balanced contribution of different features when the gradient is updated.

[0065] The above data preparation (including data collection, cleaning and standardization) provides data of uniform scale for subsequent steps, enabling the model to better understand the relationship between environmental parameters (temperature, humidity, atmospheric pressure, etc.) and radiation dose, and improve the reliability of radiation dose assessment.

[0066] It should be noted that the sensor includes an environmental sensor that integrates multiple environmental sensors, such as a temperature sensor, a humidity sensor, an atmospheric pressure sensor, etc., to collect environmental data in real time.

[0067] Temperature sensor, using thermocouple or digital temperature sensor, ensures high accuracy and fast response. Measuring range: -40℃ to 125℃ to cover extreme environmental conditions. Accuracy: ±0.5℃ to ensure the accuracy of temperature monitoring. Integration method: miniaturized design, integrated inside the device to reduce external influences.

[0068] The humidity sensor adopts a capacitive humidity sensor with high stability and fast response time. Measuring range: 0% to 100% RH, no condensation. Accuracy: ±2% RH, ensuring the accuracy and reliability of humidity data. Features: Anti-chemical pollution design to improve long-lasting performance in harsh environments.

[0069] Atmospheric pressure sensor uses MEMS technology. Measuring range: 300hPa to 1100hPa, covering pressure changes from sea level to high altitude areas. Accuracy: ±1hPa, ensuring the accuracy of atmospheric pressure measurement. Features: low power consumption and high performance, suitable for portable devices.

[0070] The radiation sensor uses a high-sensitivity GM tube or other radiation detectors to monitor the relevant radiation dose rate in real time.

[0071] GM tubes use highly sensitive Geiger-Müller (GM) tubes, suitable for detecting X-rays and gamma rays. Sensitivity: can detect radiation levels from ambient levels to several hundred mSv / h. Features: has high response speed and good linear output for a wide range of radiation dose rates. Durability: designed to withstand high radiation and high temperature environments, ensuring long-term stable operation.

[0072] In addition to the GM tube, semiconductor detectors such as silicon photodiodes or PIN photodiodes can be integrated as needed for high-precision detection of specific types of radiation.

[0073] Single or multiple detectors can be configured to increase the measurement range and sensitivity, or to distinguish different types of radiation. The design includes a microcontroller and analog-to-digital converter (ADC) for signal amplification, data conversion and preliminary processing. All sensor data is processed synchronously through a centralized processing module to ensure data consistency and timeliness. Sensors are integrated using standard communication protocols (such as I2C, SPI) to ensure accurate data transmission and scalability of the device.

[0074] The collected radiation data and environmental data are processed and analyzed in real time using embedded systems or microprocessors.

[0075] The multi-layer perceptron (MLP) algorithm is used for radiation dose assessment.

[0076] Multilayer Perceptron (MLP) is a feedforward artificial neural network suitable for processing nonlinear problems and performing complex pattern recognition. In the present invention, MLP is used to adaptively adjust the radiation dose assessment model according to changes in environmental parameters.

[0077] Feature selection uses automated feature selection techniques such as random forests and model-based selection (such as LASSO) to determine the most influential environmental parameters for radiation dose assessment.

[0078] Feature construction constructs new features based on business knowledge, such as the rate of change of environmental parameters, radiation accumulation, etc.

[0079] The steps include:

[0080] 1. Data preparation: Extract features from the data collected by sensors, including environmental parameters such as temperature, humidity, atmospheric pressure, and radiation data such as gamma-ray dose rate.

[0081] 2. Model training: Use these features to train a random forest model. A random forest consists of multiple decision trees, each of which uses a random subset of features during training.

[0082] 3. Evaluate feature importance: Evaluate the contribution of each feature in the decision tree to improving the model's predictive power. This is usually based on how often and how deeply the feature is used in the tree split.

[0083] 4. Feature selection: Features with the highest importance score are selected for further analysis and model training.

[0084] Calculating the rate of change of key environmental parameters, such as temperature and humidity, can reveal changes in environmental conditions over short periods of time, which may affect the measurement of radiation levels.

[0085] In the feature construction step, the rate of change of the environmental parameter is expressed as,

[0086]

[0087] Among them, E t and E t-1 are the parameter values ​​of two consecutive time points, and Δt is the time interval.

[0088] The radiation accumulation is expressed as,

[0089]

[0090] Among them, R i is the radiation dose rate at each time point, and n is the window size.

[0091] These feature construction and selection methods provide an exhaustive data basis for model training, aiming to optimize the accuracy and robustness of radiation dose prediction models.

[0092] In order to effectively predict the radiation dose rate, a multi-layer perceptron (MLP) network will be designed. This network will specifically include an input layer, a hidden layer, and an output layer, and the design of each layer is based on the specific requirements of the prediction task.

[0093] The number of nodes in the input layer will directly correspond to the number of features selected. Suppose after the feature selection process, we identify 10 key features, then the input layer will have 10 neurons.

[0094] Data input: Each neuron corresponds to a feature, such as temperature change rate, humidity, radiation accumulation, etc.

[0095] Hidden layers, number of layers and nodes: The model will include two hidden layers. If the input layer has 10 nodes, the first hidden layer can have 20 to 30 nodes, and the second hidden layer can be halved again, using 10 to 15 nodes. Such a design helps to gradually abstract and synthesize the characteristics of the input data, and enhance the model's ability to capture complex patterns in the data.

[0096] Activation function: All hidden layers will use the ReLU (Rectified Linear Unit) activation function. The mathematical expression of the ReLU function is (f(x) = max(0, x)), which can maintain the simplicity of calculation and the fast convergence of the model while increasing nonlinearity.

[0097] Output layer, number of nodes: The output layer will have only one node, which is used to output continuous radiation dose rate prediction values.

[0098] Activation function: The choice of activation function for the output layer will be based on the nature of the prediction task. Since the radiation dose rate is a continuous value, a linear activation function can be chosen, i.e. (f(x) = x). If the prediction task requires limiting the output range, such as within a certain range, consider using a sigmoid or tanh function to compress the output to a specific interval.

[0099] The MLP model architecture extracts important features from the input data through multi-layer processing, and finally generates a prediction of the radiation dose rate through layer-by-layer abstraction. ReLU is selected as the hidden layer activation function to take advantage of its good performance and computational efficiency in the optimization process. The design of the output layer focuses on the precise expression of the prediction. The linear activation function allows the model to output a wide range of continuous values, which is suitable for the prediction of radiation dose rate.

[0100] The optimization algorithm used during training is the Adam optimizer, which combines the features of the Momentum and RMSprop optimizers. It maintains the moving average (first moment estimate) of each parameter and the moving average (second moment estimate) of the unsquared gradient to adjust the learning rate.

[0101] Use adaptive learning rate adjustment technology, specifically learning rate decay and cyclic learning rates (CLR). Learning rate decay: Gradually reducing the learning rate as the number of training rounds increases can help the model converge stably in the later stages of training.

[0102] Cyclic learning rate: Principle: CLR can explore different learning rates at different stages of training by cyclically adjusting the learning rate between the minimum and maximum values, thereby avoiding falling into the local minimum.

[0103] The mean square error (MSE) is used as the loss function, which is suitable for the prediction of continuous values ​​and is expressed as:

[0104]

[0105] Among them, Y i is the actual radiation dose, is the radiation dose predicted by the model, and n is the number of samples.

[0106] To prevent overfitting, the Dropout layer and regularization techniques (L1, L2 regularization) are introduced into the model.

[0107] Dropout layer, principle: randomly "drop" (temporarily remove) some neurons in the network during training to prevent the network from overfitting to specific training data. Set in the hidden layer, the Dropout rate (i.e. the proportion of neurons dropped) is generally set between 0.2 and 0.5.

[0108] Regularization (L1, L2): L1 regularization promotes sparse solutions and can be used for feature selection. L2 regularization: penalizes the square value of the weight and effectively controls the complexity of the model.

[0109] Batch training technique is used, with 64 to 128 samples per batch.

[0110] Cross-validation: K-fold cross-validation is used to evaluate the generalization ability of the model. This method can effectively reduce the accidental errors in model evaluation and improve the accuracy of the evaluation.

[0111] Performance metrics: The following metrics are used to evaluate the performance of the model:

[0112] Mean Squared Error (MSE): Measures the difference between the predicted values ​​and the actual values.

[0113] Root mean square error (RMSE): The square root of the MSE, expressing the error in the same dimension as the original data.

[0114] Mean Absolute Error (MAE): The average of the absolute values ​​of the differences between all individual observations and the arithmetic mean.

[0115] R 2 Statistic: Indicates the proportion of data variability that can be explained by the model. A higher R 2 A higher value indicates a better fit quality.

[0116] Based on the model's performance during training, fine-tune network parameters such as the number of neurons, learning rate, and optimizer parameters to achieve optimal performance.

[0117] Continuously review and optimize the feature set used, which may include adding new features or removing features that contribute less to the model.

[0118] The multi-layer perceptron (MLP) algorithm can effectively adapt to different environmental variables and improve the accuracy of radiation dose assessment. These strategies not only ensure the high performance of the model, but also ensure stability and reliability in actual deployment. This approach makes MLP an ideal choice for processing complex and dynamic environmental data, especially for key fields such as nuclear facility monitoring, environmental protection and public safety.

[0119] After the collected data is preprocessed, feature selection and feature construction are performed through the preprocessed data;

[0120] The feature selection uses random forest based on LASSO technology to determine the environmental parameters that have the greatest impact on radiation dose assessment;

[0121] The feature construction includes constructing new features based on business knowledge;

[0122] The new features include the rate of change of environmental parameters and the accumulated radiation.

[0123] The radiation dose assessment model includes an input layer, a hidden layer and an output layer;

[0124] The hidden layer uses the ReLU activation function, and the output layer uses a linear or nonlinear activation function as needed.

[0125] The model training adopts the Adam optimizer, combined with the learning rate adjustment strategy, uses the mean square error as the loss function, and adds the Dropout layer and L1 and L2 regularization techniques.

[0126] Example 2

[0127] Reference Figure 2 , as an embodiment of the present invention, provides an adaptive environmental factor radiation dose assessment system, comprising:

[0128] Data acquisition module, model building module, model training module and dose assessment module;

[0129] The data acquisition module is used to arrange sensors to collect evaluation data and pre-process the collected data;

[0130] The model building module is used to build a radiation dose assessment model based on the multi-layer perceptron model;

[0131] The model training module is used to perform model training on the radiation dose assessment model;

[0132] The dose assessment module is used to input the preprocessed assessment data into the radiation dose assessment model and output the assessment results.

[0133] Temperature, humidity, atmospheric pressure and radiation dose rate data were cleaned and normalized.

[0134] Make sure there are no missing values ​​or outliers, and standardize the data to have the same scale.

[0135] Using the random forest algorithm, multiple decision trees were constructed to evaluate the influence of each feature on radiation dose prediction.

[0136] Model-based selection was performed by LASSO to further remove irrelevant or less influential features.

[0137] Based on the feature importance scores, the features that have the greatest impact on radiation dose assessment are selected. If temperature and humidity have the highest scores, they will be prioritized for building the model.

[0138] After feature selection and construction, the extracted features such as the rate of change of environmental parameters and radiation accumulation will serve as the input layer of the subsequent multi-layer perceptron (MLP) model.

[0139] These features not only enhance the model's predictive power, but also improve the accuracy of the response to radiation dose changes through abstract processing of multi-layer neural networks. Ultimately, these constructed features will be used as input to the model together with other original features (such as temperature, humidity, and atmospheric pressure) for radiation dose prediction.

[0140] Specific improvements include:

[0141] Feature selection uses methods such as random forest and LASSO to accurately select the environmental parameters (such as temperature, humidity, and atmospheric pressure) that have the greatest impact on radiation dose.

[0142] Feature construction Based on business knowledge, new features are constructed, such as the rate of change of environmental parameters and radiation accumulation. These features can better capture the dynamic impact of environmental factors on radiation dose.

[0143] Customization of network structure,input layer: According to the selected number of features, the corresponding input nodes are designed,including the original environment parameters and the newly constructed features.

[0144] Hidden layers: The number of hidden layers and the number of nodes in each layer are adjusted to suit the complexity of the data. For example, there may be two hidden layers, with 20-30 nodes in the first layer and 10-15 nodes in the second layer.

[0145] Activation function: The hidden layer uses the ReLU activation function, and the output layer uses a linear or nonlinear activation function as required to adapt to the continuous value prediction of radiation dose.

[0146] The optimization algorithm uses the Adam optimizer and combines learning rate adjustment strategies (such as learning rate decay and cyclic learning rate) to improve the convergence speed and stability of the model.

[0147] The regularization technology introduces the Dropout layer and L1 and L2 regularization, which effectively prevents overfitting and enhances the generalization ability of the model.

[0148] The loss function uses the mean square error (MSE) as the loss function, which is suitable for continuous value regression problems.

[0149] Batch training uses small batch training (64 to 128 samples per batch) to improve training efficiency and model stability.

[0150] Cross-validation uses K-fold cross-validation to evaluate the generalization ability of the model and reduce accidental errors.

[0151] The collected data were cleaned and standardized to eliminate dimensional differences and improve the effect of model training.

[0152] Based on the training results, fine-tune the network parameters, such as the number of neurons, learning rate, etc., to achieve the best performance.

[0153] Continuously review and optimize the set of features used to ensure that the model always utilizes the most informative features.

[0154] Customized processing in feature engineering, network structure, training methods and strategies makes it more suitable for radiation dose assessment tasks. It improves the model's ability to capture the complex relationship between environmental parameters and radiation dose, and enhances the accuracy and reliability of the assessment.

[0155] Example 3

[0156] An embodiment of the present invention is different from the first two embodiments in that:

[0157] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0158] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0159] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0160] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0161] Example 4

[0162] As an embodiment of the present invention, an adaptive environmental factor radiation dose assessment method is provided. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0163] 100 data points were generated, each containing temperature, humidity, atmospheric pressure, and radiation dose. Some of the data point parameters are shown in Table 1.

[0164] Table 1 Data point parameter table

[0165] Temperature(℃) humidity(%) Atmospheric pressure (hPa) Radiation dose rate (μSv / h) 22.1 45 1013 0.12 23.4 50 1012 0.15 21.7 47 1014 0.11 24.1 53 1011 0.16 22.8 48 1013 0.13

[0166] Standardize the data to make data of different dimensions comparable. Assume that the average values ​​of temperature, humidity and atmospheric pressure are μ1, μ2, μ3 respectively, and the standard deviations are σ1, σ2, σ3. Use the formula: Standardized data = (original data - average value) / standard deviation.

[0167] For example, the normalization process for a temperature of 22.1°C is: (22.1-μ1) / σ1.

[0168] The data is divided into training and testing sets, 80% and 20% respectively.

[0169] The model is constructed by including an input layer with 3 neurons corresponding to temperature, humidity, and atmospheric pressure. Hidden layer, two hidden layers with 6 and 3 neurons respectively. Output layer, the output layer has 1 neuron, which is used to predict radiation dose. Activation function, the hidden layer uses the ReLU activation function (if the input is positive, the output is equal to the input; if the input is negative, the output is 0), and the output layer uses the linear activation function.

[0170] ReLU (Rectified Linear Unit) is an activation function used in neural networks, expressed as:

[0171] [ReLU(x)=max(0,x)]

[0172] The output of the ReLU function is the larger of the input value and 0. If the input value is positive, the output is the input value itself; if the input value is negative, the output is 0.

[0173] The ReLU activation function has the following main functions in neural networks:

[0174] 1. Introducing nonlinearity: In neural networks, one of the main functions of activation functions is to introduce nonlinearity so that the network can handle more complex nonlinear problems. If there is no activation function (for example, only linear functions are used), no matter how many layers the network has, its essence is still linear and cannot solve complex nonlinear problems.

[0175] 2. Simple calculation: ReLU is very simple to calculate, and only needs to determine whether the input is greater than 0, so the calculation speed is fast. Compared with other activation functions (such as sigmoid and tanh), ReLU is more efficient and suitable for large-scale neural network training.

[0176] 3. Alleviate the vanishing gradient problem: In deep neural networks, the vanishing gradient problem can make training difficult. The characteristics of ReLU make the gradient of positive input always 1, which is not easy to cause the gradient to vanish. The vanishing gradient problem usually occurs in sigmoid and tanh functions because their output gradients are very small when they are close to 0 and 1.

[0177] Assume the following set of input values:

[0178] [x=[-2,-1,0,1,2]]

[0179] The output value after using the ReLU activation function is:

[0180] [ReLU(x)=[0,0,0,1,2]]

[0181] As you can see, all negative inputs become 0, while positive inputs remain unchanged.

[0182] In neural networks, the ReLU activation function is often used for neurons in the hidden layer. By using the ReLU activation function, the hidden layer can better handle complex data patterns and improve the model's learning ability and prediction accuracy.

[0183] Therefore, the ReLU activation function has become a very commonly used activation function in neural networks by introducing nonlinearity, simple calculation and alleviating the gradient vanishing problem. It performs well in many practical applications and helps neural networks learn and handle complex tasks more effectively.

[0184] Perform 100 training cycles according to the above algorithm. (100 training cycles are performed to give the model enough time and opportunities to adjust parameters and gradually learn and capture complex patterns in the data. At the same time, this number can also balance the training time and performance of the model to prevent underfitting and overfitting problems. According to experience, this number of cycles usually enables the model to achieve good results in a reasonable time.)

[0185] Use the Adam optimizer, which is an algorithm that can quickly adjust model parameters. It is a common algorithm in the field of deep learning.

[0186] Dropout is a regularization technique used in neural network training. Its main purpose is to prevent the model from overfitting. Overfitting means that the model performs well on the training data, but performs poorly on new data because it "remembers" the noise and details in the training data too much and does not learn the true characteristics of the data. Dropout alleviates this problem by randomly ignoring (discarding) some neurons during training.

[0187] The basic idea of ​​Dropout is to randomly select some neurons and temporarily ignore them (i.e. set their output to 0) in each training iteration. Doing so helps:

[0188] 1. Reduce the interdependence between neurons: force different parts of the neural network to work independently, thereby learning more robust features.

[0189] 2. Simulated model integration: In each training iteration, a different sub-network (i.e., a random subset of the original network) is actually trained. These sub-networks work together during testing, which is equivalent to training multiple models and taking the average, thereby improving generalization ability.

[0190] Use a portion of the data instead of all the data for each training to speed up training and improve the stability of the model.

[0191] The data is divided into several parts for training and testing to ensure the generalization ability of the model. The mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the accuracy and stability of the model.

[0192] Use the trained model to predict a set of new input data. The predicted data is shown in Table 2.

[0193] Table 2 Experimental data prediction table

[0194] Temperature(℃) humidity(%) Atmospheric pressure (hPa) 23.5 49 1012

[0195] The predicted result is a radiation dose rate of 0.14μSv / h.

[0196] Example 5

[0197] As an embodiment of the present invention, an adaptive environmental factor radiation dose assessment method is provided. In order to verify the beneficial effects of the present invention, another simulation experiment is used for scientific demonstration.

[0198] Assume that we have the following simulated data set, which includes 500 samples, each of which includes environmental parameters and the corresponding radiation dose rate, as shown in Table 3.

[0199] Table 3 Sample data table

[0200] Temperature(℃) humidity(%) Atmospheric pressure (hPa) Radiation dose rate (μSv / h) 21.2 42 1010 0.1 22.5 44 1011 0.12 24 50 1012 0.15 23.6 49 1010 1014 21.9 43 1012 0.11

[0201] Temperature, humidity and atmospheric pressure are normalized using the Z-score normalization method. The dataset is divided into 80% training set and 20% test set.

[0202] Z-score normalization, also known as standard deviation normalization or mean normalization, is a normalization method that transforms data into data with zero mean and unit variance.

[0203] With this normalization, the transformed data will have zero mean (0) and unit variance (1), making data of different features comparable and facilitating model training.

[0204] This example uses 500 data points. Z-score normalization is used to transform the data into a form with zero mean and unit variance. The number and structure of hidden layer nodes may vary, depending on the complexity and size of the dataset. More training cycles (such as 200) are required to ensure that the model is fully trained on a larger dataset. It can better process and predict radiation doses, with stronger generalization ability and higher prediction accuracy.

[0205] The model construction includes,Input layer: 3 nodes (temperature, humidity, atmospheric pressure). Hidden layer,-First hidden layer: 10 nodes. Second hidden layer, 5 nodes. Output layer: 1 node, using linear activation function.

[0206] Perform 200 training cycles. Use Adam optimizer, initial learning rate is 0.001, and learning rate decay is applied. Use MSE. Mean square error (MSE) is the same as in Example 1.

[0207] After each epoch (training cycle), evaluate and record the MSE on the training set and the test set. Compare the MSE of the training set and the test set to monitor overfitting.

[0208] The comparison between the prediction and the results is shown in Table 4.

[0209] Table 4 Data results analysis table

[0210]

[0211]

[0212] The model's test MSE gradually decreases during the training process, indicating that the model gradually learns the patterns in the data and its generalization ability improves. The final test MSE is 0.004, indicating that the model has good prediction accuracy. At the same time, the difference between the training MSE and the test MSE is not large, indicating that the model is not obviously overfitting.

[0213] The MLP model can effectively process and predict complex radiation dose data. Through the appropriate number of layers and node selection, optimization algorithm and loss function application, MLP can not only improve the accuracy of prediction, but also avoid overfitting through appropriate training and verification strategies, so that the model has good generalization ability. This model is very suitable for environmental monitoring and radiation safety management, and can help relevant personnel make more accurate judgments and decisions.

[0214] In machine learning and neural network training, epoch (training cycle) is a key concept. The end of an epoch means that the model has completed a complete traversal and training of the entire training data set.

[0215] Specifically, when an epoch ends, the following processes have been completed:

[0216] 1. Data forward propagation (Forward Propagation):

[0217] All training data is passed through the neural network from the input layer to the output layer.

[0218] The model calculates a prediction based on the current parameters (weights and biases).

[0219] 2. Calculate loss (LossCalculation):

[0220] Calculate the error (loss) between the predicted result of the model and the actual value. Commonly used loss functions include mean square error (MSE), cross entropy, etc.

[0221] For example, using the mean square error (MSE) loss function, the calculation formula is:

[0222]

[0223] Among them, (y i ) is the actual value, is the predicted value and (n) is the sample size.

[0224] 3. Backward Propagation:

[0225] Based on the calculated loss, the gradient of the loss with respect to the model parameters is calculated through the back-propagation algorithm.

[0226] The back-propagation process starts from the output layer, calculates the gradients backwards layer by layer, and uses these gradients to adjust the model parameters.

[0227] 4. Update parameters (ParameterUpdate):

[0228] Use optimization algorithms (such as gradient descent, Adam, etc.) to update the model parameters according to the calculated gradients to minimize the value of the loss function.

[0229] 5. Complete an epoch:

[0230] When the model completes the above steps and processes the entire training dataset, it means that an epoch is over.

[0231] The model parameters have been updated and are ready to be trained and adjusted again in the next epoch.

[0232] Training a neural network usually requires multiple epochs so that the model can gradually learn the patterns and features in the data and gradually optimize the parameters so that the value of the loss function decreases. The following are the main reasons for requiring multiple epochs:

[0233] 1. Optimize model parameters step by step:

[0234] In each epoch, the model parameters are updated according to the current gradient. After multiple epochs of training, the model parameters are gradually optimized, thereby improving the accuracy of the model.

[0235] 2. Avoid overfitting and underfitting:

[0236] If the training epochs are too few, the model may not be able to fully learn the features in the data, resulting in underfitting.

[0237] If too many epochs are trained, the model may overfit to the noise in the training data, resulting in overfitting.

[0238] By adjusting the number of epochs, a balance can be found so that the model performs well on both training data and test data.

[0239] 3. Gradual adjustment of learning rate:

[0240] Optimization algorithms usually gradually adjust the learning rate during training to make model parameter updates smoother and more stable.

[0241] Multiple epochs help the model adapt better to these adjustments and thus converge to the optimal solution faster.

[0242] The training process is as follows:

[0243] Initialize model parameters (weights and biases).

[0244] Start the first epoch, traverse the entire training data set, calculate the loss and update the parameters.

[0245] The first epoch ends and the current loss value is recorded.

[0246] Start the second epoch, traverse the entire training dataset again, calculate the loss and update the parameters.

[0247] At the end of the second epoch, record the current loss value and compare it with the loss value of the first epoch to see if it has decreased.

[0248] Continue for multiple epochs until the predetermined number of training times is reached or the loss value no longer decreases significantly.

[0249] In this way, the model can gradually optimize parameters, improve prediction accuracy, and eventually perform well on the test data. The end of an epoch means that the model has completed a complete traversal and training of the entire training data set. Through multiple epochs of training, the model can gradually optimize parameters and improve prediction accuracy, so that it can perform better in practical applications.

[0250] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An adaptive environmental factor radiation dose assessment method, characterized in that: include: Arrange sensors to collect evaluation data and pre-process the collected data; Based on the multi-layer perceptron model, a radiation dose assessment model is constructed; Performing model training on the radiation dose assessment model; The preprocessed evaluation data is input into the radiation dose evaluation model and the evaluation results are output.

2. The adaptive environmental factor radiation dose assessment method according to claim 1, characterized in that: The evaluation data include environmental parameters and radiation dose; The environmental parameters include temperature, humidity and atmospheric pressure; The radiation dose mentioned refers to the radiation monitoring dose rate; The preprocessing includes data cleaning and data standardization.

3. The adaptive environmental factor radiation dose assessment method according to claim 2, characterized in that: After the collected data is preprocessed, feature selection and feature construction are performed through the preprocessed data; The feature selection uses random forest based on LASSO technology to determine the environmental parameters that have the greatest impact on radiation dose assessment; The feature construction includes constructing new features based on business knowledge; The new features include the rate of change of environmental parameters and the accumulated radiation.

4. The adaptive environmental factor radiation dose assessment method according to claim 3, characterized in that: The rate of change of the environmental parameter is expressed as, Among them, E t and E t-1 are the parameter values ​​of two consecutive time points, and Δt is the time interval.

5. The adaptive environmental factor radiation dose assessment method according to claim 4, characterized in that: The radiation accumulation is expressed as, Among them, R i is the radiation dose rate at each time point, and n is the window size.

6. The adaptive environmental factor radiation dose assessment method according to claim 5, characterized in that: The radiation dose assessment model includes an input layer, a hidden layer and an output layer; The hidden layer uses the ReLU activation function, and the output layer uses a linear or nonlinear activation function as needed.

7. The adaptive environmental factor radiation dose assessment method according to claim 6, characterized in that: The model training adopts the Adam optimizer, combined with the learning rate adjustment strategy, uses the mean square error as the loss function, and adds the Dropout layer and L1 and L2 regularization techniques.

8. A system using the adaptive environmental factor radiation dose assessment method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, model building module, model training module and dose assessment module; The data acquisition module is used to arrange sensors to collect evaluation data and pre-process the collected data; The model building module is used to build a radiation dose assessment model based on a multi-layer perceptron model; The model training module is used to perform model training on the radiation dose assessment model; The dose assessment module is used to input the pre-processed assessment data into the radiation dose assessment model and output the assessment result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the adaptive environmental factor radiation dose assessment method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the adaptive environmental factor radiation dose assessment method according to any one of claims 1 to 7 are implemented.

Images