Quality measurement method, model training method, electronic device, storage medium and program product

Abstract

The invention provides a quality measurement method, a model training method, electronic equipment, a storage medium and a program product. The method comprises: obtaining neutron counting data of a nuclear material to be measured within a first measurement duration, the first measurement duration being less than a second measurement duration, and the second measurement duration being a duration for performing neutron multiplex measurement on the nuclear material to be measured based on a preset mathematical model; based on the neutron counting data, a neural network model is used for conducting mass measurement on the to-be-measured nuclear material, a measurement result output by the neural network model is obtained, and the measurement result represents the mass of the to-be-measured nuclear material. By means of the scheme, the accuracy of nuclear material quality measurement can be improved under the condition that the measurement time of the to-be-measured nuclear material is shortened.

Description

Quality measurement methods, model training methods, electronic devices, storage media and program products Technical Field

[0001] This application relates to the field of nuclear material testing technology, and in particular to a quality measurement method, a model training method, an electronic device, a storage medium, and a program product. Background Technology

[0002] Non-destructive testing (NDA) is a non-invasive testing method that analyzes relevant parameters of nuclear materials by measuring elemental characteristic signals without direct contact. In the field of nuclear safety assurance, neutron detection technology, due to its non-destructive nature and high accuracy, is one of the main techniques for non-destructive analysis of nuclear materials.

[0003] Neutron multiplicity counting (NMC) technology, as a cutting-edge technology in neutron detection, is an inheritance and innovative development based on neutron coincidence measurement technology. Neutron multiplicity counting can more accurately determine the breeding characteristics of nuclear materials and effectively reduce measurement biases caused by differences in the geometric shape and density inhomogeneity of nuclear materials. It has been widely recognized and applied in important fields such as international nuclear safeguards.

[0004] However, neutron multiplicity measurement techniques still have certain limitations. When using neutron multiplicity measurement techniques to perform neutron measurements on nuclear materials, the measurement time is relatively long, and the measurement results are prone to deviation. Summary of the Invention

[0005] To address the related technical issues, embodiments of this application provide a quality measurement method, a model training method, an electronic device, a storage medium, and a program product.

[0006] The technical solution of this application embodiment is implemented as follows: This application embodiment provides a mass measurement method, the method comprising: acquiring neutron count data of a nuclear material to be tested within a first measurement duration, wherein the first measurement duration is less than a second measurement duration, and the second measurement duration is the duration for neutron multiplicity measurement of the nuclear material to be tested based on a preset mathematical model; and, based on the neutron count data, performing mass measurement of the nuclear material to be tested using a pre-trained neural network model to obtain a measurement result output by the neural network model, wherein the measurement result represents the mass of the nuclear material to be tested.

[0007] In the above scheme, the neural network model includes a one-dimensional convolutional neural network; the neutron counting data is one-dimensional data; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron counting data and obtaining the measurement result output by the neural network model includes: inputting the neutron counting data into the one-dimensional convolutional neural network, using the one-dimensional convolutional neural network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

[0008] In the above scheme, the neural network model includes a two-dimensional convolutional neural network; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data and obtaining the measurement result output by the neural network model includes: converting the neutron count data into two-dimensional data to obtain converted neutron count data; inputting the converted neutron count data into the two-dimensional convolutional neural network, and using the two-dimensional convolutional neural network to measure the mass of the nuclear material under test and obtain the measurement result output by the neural network model.

[0009] In the above scheme, the neural network model includes a long short-term memory network; the neutron count data is data measured periodically at preset time intervals within the first measurement duration; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data, and obtaining the measurement result output by the neural network model, includes: arranging the neutron count data in chronological order to obtain time-series data; inputting the time-series data into the long short-term memory network, and using the long short-term memory network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

[0010] In the above scheme, the neural network model includes a backpropagation neural network; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data and obtaining the measurement result output by the neural network model includes: inputting the neutron count data into the backpropagation neural network, using the backpropagation neural network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

[0011] In the above scheme, the neutron counting data includes one or more of the following: single neutron count rate, double coincidence neutron count rate, triple coincidence neutron count rate, and neutron pulse time series.

[0012] This application embodiment also provides a model training method, the method comprising: acquiring measurement data of sample nuclear material, wherein the measurement data represents neutron distribution; dividing the measurement data into a training set and a test set; training a neural network model to be trained based on the measurement data in the training set to obtain a trained neural network model; testing the trained neural network model based on the measurement data in the test set to obtain test results; and obtaining a pre-trained neural network model when the test results meet preset test conditions.

[0013] This application also provides an electronic device, including: a processor and a memory for storing a computer program that can run on the processor; wherein, when the processor runs the computer program, it performs the steps of the above-described quality measurement method, or performs the steps of the above-described model training method.

[0014] This application embodiment also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described quality measurement method, or implements the steps of the above-described model training method.

[0015] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described quality measurement method or the steps of the above-described model training method.

[0016] The mass measurement method, model training method, electronic device, storage medium, and program product provided in this application's embodiments process the neutron count data of the nuclear material under test using a pre-trained neural network model. It only requires acquiring the neutron count data obtained from neutron multiplicity measurements of the nuclear material under test within a first measurement duration, which is shorter than the duration required for neutron multiplicity measurements based on a preset mathematical model. The neural network model analyzes the relationship between the neutron count data and the mass of the nuclear material, obtaining the mass of the nuclear material under test without relying on a physical model. Because the neural network model has good generalization ability, it can obtain highly accurate measurement results even when dealing with irregular nuclear materials under test, improving the accuracy of nuclear material mass measurement. Therefore, measuring the mass of the nuclear material under test using a neural network can improve the accuracy of nuclear material mass measurement while reducing the measurement time. Attached Figure Description

[0017] Figure 1 is a flowchart illustrating a quality measurement method provided in an embodiment of this application; Figure 2 is a structural diagram illustrating a two-dimensional convolutional neural network provided in an embodiment of this application; Figure 3 is a schematic diagram illustrating a long short-term memory network for processing time-series data provided in an embodiment of this application; Figure 4 is a structural diagram illustrating a long short-term memory layer provided in an embodiment of this application; Figure 5 is a structural diagram illustrating a backpropagation neural network provided in an embodiment of this application; Figure 6 is a flowchart illustrating a model training method provided in an embodiment of this application; Figure 7 is a flowchart illustrating another model training method provided in an embodiment of this application; Figure 8 is a schematic diagram illustrating the comparison between the output of a neural network model and the data labels of sample kernel materials provided in an embodiment of this application; Figure 9 is a structural diagram illustrating a quality measurement device provided in an embodiment of this application; Figure 10 is a structural diagram illustrating a model training device provided in an embodiment of this application; Figure 11 is a structural diagram illustrating an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Nuclear materials are radioactive materials. For example, elements such as plutonium, uranium, and uranium are all nuclear materials. Reliability metrology of nuclear materials is a core aspect of nuclear safety assurance. Non-destructive testing (NDT) of nuclear materials is an important technology for nuclear safety assurance. NDT techniques can obtain relevant information about nuclear materials by detecting the radiation they emit without damaging their packaging containers.

[0020] Neutrons possess extremely strong penetrating power, capable of penetrating the encapsulation container of nuclear materials. Furthermore, neutrons exhibit specificity for nuclear materials; that is, they carry relevant information about the nuclear material, such as its volume, density, and spatial distribution. Therefore, neutron detection technology is a primary method for non-destructive testing of nuclear materials.

[0021] Nuclear materials produce neutrons during fission. Fission occurs in two ways: spontaneous fission and induced fission. For example, plutonium isotopes undergo spontaneous fission, releasing an average of 2 to 3 neutrons per burst. To obtain information about nuclear materials, neutron detection can be performed by bombarding the sample with an external neutron source to actively induce fission, thus releasing neutrons through induced fission.

[0022] During nuclear fission, the neutrons released from nuclear material are typically not uniform and independent, but rather multiple neutrons are released instantaneously. Neutron detectors can capture these released neutrons, achieving neutron detection. The neutrons captured by the neutron detector are detected as a pulse signal. The time stamps of multiple pulse signals can form a pulse time series. This pulse time series can serve as the basis for nuclear material analysis.

[0023] For nuclear materials, neutron multiplication can also occur. During nuclear fission, the material releases multiple neutrons. Some of these neutrons may escape and be captured by a neutron detector, detected as pulse signals. Others may continue to bombard the nuclear material, triggering new fission and producing more neutrons. This process can continue multiple times, transforming the fission process into a fission chain. This phenomenon is called neutron multiplication. Neutron multiplication causes multiple neutrons released from the nuclear material to arrive at the neutron detector in quick succession. In the neutron pulse time series, this manifests as multiple neutrons from the same fission or the same fission chain arriving at nearly the same time, resulting in the pulse signals of these neutrons having their time signatures concentrated within a short time window.

[0024] Neutron detection is easily affected by the background environment (such as ambient radiation), making it inaccurate to analyze nuclear material information solely based on the number of neutrons detected by a neutron detector. To extract accurate information related to nuclear material properties from the background environment, neutron multiplexing measurement technology was developed. Since multiple neutrons originating from the same fission event or fission chain are temporally close, while neutrons from the background environment are temporally independent with uniformly and randomly distributed time intervals, neutron multiplexing measurement technology distinguishes between neutrons originating from the nuclear material and those from the background environment by observing the temporal correlation of multiple neutrons produced by neutron fission and neutron multiplication phenomena, thus obtaining accurate information about the nuclear material. Neutron multiplexing achieved through an external neutron source can be called active neutron multiplexing.

[0025] In active neutron multiplicity measurements, a very short time window (i.e., a coincidence gate) is set to determine the total pulse signal distribution within that time window. To separate neutrons produced by neutron fission and neutron multiplication, another time window (i.e., a background coincidence gate) is set after the coincidence gate to determine the distribution of random pulse signals generated by the background environment within that time window. Thus, by combining the total pulse signal distribution with the random pulse signal distribution, the pulse signal distribution originating from the nuclear material can be determined, i.e., the double coincidence count rate (which can be represented as D) and the triple coincidence count rate (which can be represented as T).

[0026] The double coincidence count rate represents the number of neutron pair events (i.e., double coincidence events) that occur per unit time during the fission process of nuclear material. The double coincidence count rate is related to the number of fission events in nuclear material and is a key indicator for measuring the mass of nuclear material. The triple coincidence count rate represents the number of three neutron combination events (i.e., triple coincidence events) that occur per unit time during the fission process of nuclear material. The triple coincidence count rate is related to the neutron multiplication phenomenon in nuclear material and is a key indicator reflecting this phenomenon. The mass of nuclear material can be calculated using the double and triple coincidence count rates. That is, a mapping relationship exists between the double and triple coincidence count rates and the mass of nuclear material, and this mapping relationship can be characterized by a pre-defined mathematical model. The mass of nuclear material can be solved using the double and triple coincidence count rates and the pre-defined mathematical model of nuclear material mass.

[0027] However, the method of determining the mass of nuclear materials using neutron multiplicity measurement (NMP) still has certain limitations: First, NMP requires a long measurement time. Since the probability of a triple coincidence event in nuclear materials is much lower than that of a double coincidence event, a long-term NMP measurement is needed to accumulate triple coincidence event counts in order to obtain a stable triple coincidence count rate for determining the mass of the nuclear materials. Typically, the NMP measurement time needs to last for more than 1000 seconds (e.g., 30 minutes), making the efficiency of nuclear material mass measurement low.

[0028] Furthermore, the measurement results of neutron multiplex measurement techniques are easily affected by the nuclear material sample. Since different nuclear material samples vary in volume, geometry, density, and composition distribution, they deviate significantly from a geometrically homogeneous point free of neutron absorption. This deviates from the pre-defined mathematical model based on this geometric point, leading to errors in the calculated nuclear material mass. Although correction factors or other calibration methods can be used, these methods lack universality and are complex to implement.

[0029] Based on this, embodiments of this application provide a mass measurement scheme. A pre-trained neural network model processes the neutron count data of the nuclear material under test, and analyzes the relationship between the neutron count data and the mass of the nuclear material using the neural network model. The mass of the nuclear material under test is obtained without relying on a pre-set mathematical model. Because the neural network model has good generalization ability, it can obtain high-accuracy measurement results even when dealing with irregular nuclear materials under test, thus improving the accuracy of nuclear material mass measurement.

[0030] Furthermore, measuring the mass of the nuclear material under test using a neural network only requires acquiring neutron count data obtained from neutron multiplicity measurements of the material within a short period (i.e., the first measurement duration). This duration is shorter than the duration required for neutron multiplicity measurements based on a pre-defined mathematical model (i.e., the second measurement duration). Therefore, measuring the mass of the nuclear material under test using a neural network can reduce the measurement time and improve the efficiency of mass measurement, making it suitable for non-destructive testing tasks of nuclear materials in complex environments.

[0031] First, this application provides a quality measurement method. This method can be applied to electronic devices. The electronic device in this application can be a device for nuclear material analysis, such as a nuclear material non-destructive analyzer. Alternatively, the electronic device can be a device capable of data processing, such as a computer, server, or system. This application does not limit the specific form of the electronic device. The quality measurement method provided in this application will be described below using an electronic device as the execution subject.

[0032] As shown in Figure 1, the mass measurement method provided in this application includes the following steps: Step 101: The electronic device acquires neutron count data of the nuclear material to be tested within a first measurement period; Step 102: Based on the neutron count data, the electronic device uses a pre-trained neural network model to measure the mass of the nuclear material to be tested, and obtains the measurement result output by the neural network model, which represents the mass of the nuclear material to be tested.

[0033] In practical applications, neutron count data represents the number of neutrons. For example, neutron count data can represent either a neutron count or a neutron count rate. A neutron count is the number of neutron events that occur within a certain time length (such as the first measurement duration). A neutron count rate is the number of neutron events that occur per unit of time. The unit of time can be set according to the actual application scenario or requirements. For example, the unit of time can be 1 second. A neutron event can be understood as the event in which neutrons released by nuclear material are detected. Neutron events include single neutron events and multiple neutron events.

[0034] Neutron count data can reflect the distribution of neutrons produced during the fission process of the tested nuclear material. In some implementations, neutron count data includes one or more of the following: single neutron count rate (which can be represented as S), double coincidence neutron count rate, triple coincidence neutron count rate, total coincidence neutron count distribution, background neutron count distribution, and neutron pulse time series.

[0035] The single neutron count rate is the number of individual neutron events per unit time. It can reflect the fission reaction status of the nuclear material being tested. For example, the single neutron count rate is related to factors such as the mass, physical state, and background environment of the nuclear material. The single neutron count rate corresponds to the first moment conforming to the total neutron count distribution. By observing changes in the single neutron count rate, a preliminary judgment can be made as to whether the nuclear material under test has undergone a fission reaction, providing a basis for mass measurement.

[0036] The double coincidence neutron count rate is the net number of neutron pairs per unit time. It reflects the fission activity of the tested nuclear material. For example, the double coincidence neutron count rate is related to factors such as the mass and multiplication phenomena of the tested nuclear material. The double coincidence neutron count rate eliminates background interference. By using the double coincidence neutron count rate, the intensity of the fission reaction in the tested nuclear material can be quantified, providing a basis for measuring the mass of the tested nuclear material.

[0037] The triple coincidence neutron count rate is the net number of neutron clusters formed by three neutrons per unit time. It reflects the neutron multiplication of the nuclear material being tested. For example, the triple coincidence neutron count rate is related to neutron multiplication. The triple coincidence neutron count rate removes background interference. It can provide a basis for measuring the mass of the nuclear material being tested.

[0038] The coincidence neutron count distribution is the probability distribution of neutron events statistically within coincidence events. The total coincidence neutron count includes neutron events from nuclear fission and neutron events from the background environment. The coincidence neutron count distribution can be described by moments (such as the first, second, and third moments). The coincidence neutron count distribution reflects the fission intensity of the nuclear material being tested. The coincidence neutron count distribution is the basis for determining the double coincidence neutron count rate and the triple coincidence neutron count rate.

[0039] The background neutron count distribution is the probability distribution of neutron events generated by the background environment, statistically analyzed within the background coincidence gate. It reflects the interference conditions of the background environment. The background neutron count distribution can be described by moments (such as the first, second, and third moments). It forms the basis for determining the double coincidence neutron count rate and the triple coincidence neutron count rate.

[0040] A neutron pulse time series is a time sequence of neutron-generated pulse signals arranged chronologically. It provides raw time information about neutron events and can serve as the primary data basis for neutron multiplicity analysis. The neutron count rate can be determined using the neutron pulse times, providing a data foundation for mass measurement of the nuclear material under test.

[0041] In step 101, the electronic device can perform a neutron multiplicity measurement on the test nucleus material to obtain neutron count data of the test nucleus material within a first measurement duration. Alternatively, the electronic device can obtain neutron count data of the test nucleus material within the first measurement duration from other devices.

[0042] For example, the electronic device includes a neutron detector and a neutron multiplicity counter. The neutron detector is used to detect the pulse signal generated by neutrons released from the nuclear material under test. The neutron multiplicity counter is used to convert the pulse signal detected by the neutron detector into a neutron pulse time series. The electronic device performs neutron multiplicity measurements on the nuclear material under test within a first measurement duration using the neutron detector and the neutron multiplicity counter, obtaining the neutron pulse sequence of the nuclear material under test within the first measurement duration. In some implementations, the neutron multiplicity counter is also used to obtain the single neutron count rate, double coincidence neutron count rate, and triple coincidence neutron count rate based on the neutron pulse time series. The electronic device can also obtain the single neutron count rate, double coincidence neutron count rate, and triple coincidence neutron count rate of the nuclear material under test within the first measurement duration using the neutron multiplicity counter.

[0043] For example, the electronic equipment establishes a communication connection (wired or wireless) with the nuclear material nondestructive analyzer. The nuclear material nondestructive analyzer performs neutron detection on the nuclear material under test during the first measurement period, obtaining neutron count data of the nuclear material under test during the first measurement period. The electronic equipment obtains the neutron count data of the nuclear material under test from the nuclear material nondestructive analyzer during the first measurement period.

[0044] In this embodiment, the first measurement duration is the time required for neutron multiplicity measurement of the nuclear material under test. In this embodiment, the mass of the nuclear material under test is measured using a neural network model, thus the first measurement duration can be set to a relatively short time window. For example, the first measurement duration can be set to 10 seconds, 50 seconds, 100 seconds, etc. The second measurement duration is the time required for neutron multiplicity measurement of the nuclear material under test based on a preset mathematical model. The second measurement duration is often on the order of kiloseconds or longer. The first measurement duration is shorter than the second measurement duration. The preset mathematical model is a mathematical model representing the mapping relationship between double coincidence rate and triple coincidence rate and the mass of the nuclear material.

[0045] The neutron multiplicity measurement in this application embodiment can be an active neutron multiplicity measurement.

[0046] To improve the accuracy of neutron count data, and thus the accuracy of the mass measurement results of the material under test, in some optional implementations, the neutron count data is obtained by performing multiple neutron multiplicity measurements within a first measurement duration. For example, the neutron count data is obtained by periodically measuring the nuclear material under test at preset time intervals within the first measurement duration.

[0047] In this embodiment, the preset time interval can be set according to actual application scenarios or needs. For example, the preset time interval can be set to 1 second. The first measurement duration includes multiple preset time intervals. The preset time intervals can serve as the period for neutron multiplicity measurement. Correspondingly, the neutron count data of the test material within the first measurement duration includes the neutron count data of the test material within each preset time interval. For example, the electronic device uses the preset time intervals as the period for neutron multiplicity measurement, and periodically performs neutron multiplicity measurement on the test material within the first measurement duration to obtain the neutron count data of the test material within each preset time interval.

[0048] In step 102, the electronic device uses a pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data of the nuclear material under test during the first measurement period, and obtains the measurement result output by the neural network model, that is, the mass of the nuclear material under test.

[0049] To improve the accuracy of measurement results, in some optional implementations, the electronic device can first preprocess the acquired neutron count data, and then input the preprocessed neutron count data into a pre-trained neural network model to obtain the measurement results output by the neural network model.

[0050] For example, electronic devices can perform one or more preprocessing operations on neutron count data, such as noise reduction, normalization, and data augmentation, to obtain preprocessed neutron count data. By preprocessing the neutron count data, the effective information in the neutron count data can be enhanced, and the influence of random errors on the neutron count data can be suppressed. This makes the preprocessed neutron count data more suitable for processing by neural network models, thereby improving the accuracy of the measurement results of the neural network models.

[0051] In this embodiment, the pre-trained neural network model is a neural network model trained based on measurement data of the sample nuclear material. The neural network model is used to measure the quality of the nuclear material. For example, the neural network model includes one or more of the following (one or more can be understood as at least one): one-dimensional convolutional neural network (1D-CNN), two-dimensional convolutional neural network (2D-CNN), long short-term memory network (LSTM), and back propagation neural network (BPNN).

[0052] The electronic device inputs pre-processed neutron count data into a pre-trained neural network model. The model extracts key features from the neutron count data and then uses these features to predict the mass of the nuclear material under test. The result is the measurement output of the pre-trained neural network model, representing the mass of the nuclear material. The pre-trained neural network model can establish a correspondence between neutron count data and nuclear material mass and possesses good generalization ability, exhibiting high predictive performance even when faced with irregular nuclear materials and non-ideal background environments. Therefore, determining the mass of the nuclear material under test through a neural network model not only improves the efficiency of nuclear material mass measurement but also enhances its accuracy and stability.

[0053] In some alternative implementations, the neural network model is a one-dimensional convolutional neural network (CNN). The input data to a one-dimensional CNN is one-dimensional data. For example, if the neutron count data is one-dimensional, it could be a one-dimensional vector containing single neutron count rates, double coincidence neutron count rates, and triple coincidence neutron count rates. Alternatively, the neutron count data could be a neutron pulse time series. In this case, the electronic device directly inputs the neutron count data into the one-dimensional CNN, processes the data using the CNN, and obtains the measurement result output by the one-dimensional CNN.

[0054] If the neutron count data is measured periodically at preset time intervals, the electronic device concatenates the neutron count data (such as S, D, and T) from each preset time interval in chronological order along the time dimension, thus converting the neutron count data into one-dimensional data. For example, taking a first measurement duration comprising 10 preset time intervals, the neutron count data for each preset time interval is 1×3 one-dimensional data including S, D, and T. After the electronic device concatenates the neutron count data from each preset time interval along the time dimension, it obtains 1×30 one-dimensional data. Then, the electronic device uses a one-dimensional convolutional neural network to process the converted neutron count data, obtaining the measurement result output by the one-dimensional convolutional neural network.

[0055] A one-dimensional convolutional neural network (CNN) can include network structures such as one or more convolutional layers, one or more pooling layers, and fully connected layers. Convolutional layers are used to extract local features from neutron count data. Pooling layers are used to reduce the data dimensionality, thereby enhancing the robustness of the CNN. Fully connected layers are used to generate quality measurement results. Electronic devices process neutron count data through the convolutional, pooling, and fully connected layers of the CNN to obtain the measurement results output by the CNN.

[0056] In this embodiment, the neural network model can be a one-dimensional convolutional neural network. By extracting key mass-related features from the neutron count data using a one-dimensional convolutional neural network, the accuracy and efficiency of mass measurement of the nuclear material under test can be improved.

[0057] In some alternative implementations, the neural network model is a two-dimensional convolutional neural network (2D convolutional neural network). The input data to a 2D convolutional neural network is two-dimensional data. When the neutron counting data is two-dimensional, the electronic device directly inputs the neutron counting data into the 2D convolutional neural network, processes the neutron counting data, and obtains the measurement result output by the 2D convolutional neural network.

[0058] If the neutron count data is not two-dimensional, the electronic device converts the neutron count data into two-dimensional data before inputting it into the two-dimensional convolutional neural network. The converted neutron count data is then input into the two-dimensional convolutional neural network, which processes the neutron count data to obtain the measurement result output by the two-dimensional convolutional neural network.

[0059] For example, when neutron count data is measured periodically at preset time intervals, the electronic device concatenates the neutron count data (such as S, D, and T) from each preset time interval in chronological order along the feature dimension, thereby transforming the neutron count data from one-dimensional to two-dimensional data. Taking a first measurement duration comprising 10 preset time intervals as an example, the neutron count data for each preset time interval is 1×3 one-dimensional data including S, D, and T. After concatenating the neutron count data from each preset time interval along the feature dimension, the electronic device obtains 10×3 two-dimensional data. Then, the electronic device uses a two-dimensional convolutional neural network to process the transformed neutron count data, obtaining the measurement result output by the two-dimensional convolutional neural network.

[0060] For example, when the neutron count data is a neutron pulse time series, the electronic device divides the neutron pulse time series into multiple sub-time series segments with the same time interval (e.g., 1 second), and splices these sub-time series segments along the column direction of the two-dimensional data according to their time sequence, thereby reconstructing the one-dimensional neutron pulse time series into two-dimensional data. The electronic device then inputs this two-dimensional data into a two-dimensional convolutional neural network for processing, obtaining the measurement results output by the two-dimensional convolutional neural network.

[0061] For example, a two-dimensional convolutional neural network (CNN) may include an input layer, one or more convolutional layers, one or more pooling layers, one or more fully connected layers, and an output layer. The input layer receives the input neutron count data. The convolutional layers extract local features from the neutron count data. The pooling layers reduce the data dimensionality, thereby enhancing the robustness of the CNN. The fully connected layers map the local features to the output. The output layer outputs the measurement results. For instance, the network structure of a two-dimensional CNN is shown in Figure 2. The network structure includes two convolutional layers, two pooling layers, and two fully connected layers. The convolutional kernels of the convolutional layers are 3×3 with a padding parameter of 1 and a stride of 1. The pooling kernels of the pooling layers are 2×2 with a stride of 2. The first fully connected layer may contain 480 neurons, and the second fully connected layer may contain 64 neurons. The fully connected layers use the ReLU activation function to achieve full connectivity. Two-dimensional convolutional neural networks extract key features related to the mass of nuclear materials from neutron count data step by step through convolutional and pooling layers, and integrate these features through fully connected layers to obtain the measurement results of the mass of the nuclear material to be tested.

[0062] In this embodiment, the neural network model can be a two-dimensional convolutional neural network. By processing the neutron count data using a two-dimensional convolutional neural network, the changing trend of the neutron count data over time can be captured, improving the accuracy and efficiency of the mass measurement of the nuclear material under test.

[0063] In some alternative implementations, the neural network model is a Long Short-Term Memory (LSTM) network. The input data to the LSM network is sequential data. When the neutron count data is sequential data, for example, when the neutron count data is a neutron pulse time series, the electronic device directly inputs the neutron count data into the LSM network, processes the neutron count data using the LSM network, and obtains the measurement result output by the LSM network.

[0064] If the neutron count data is not sequential data, the electronic device converts the neutron count data into sequential data before inputting it into the Long Short-Term Memory (LSTM) network. The converted neutron count data is then input into the LTM network, which processes the data to obtain the measurement results output by the LTM network.

[0065] For example, as shown in Figure 3, the neutron count data is data (such as S, D, T) measured periodically in 1-second intervals (example of a preset time interval) within 100 seconds (example of the first measurement duration). The electronic device combines the neutron count data measured in each 1-second interval in the feature dimension according to the time sequence, thereby converting the neutron count data into time sequence data (i.e., 100×1 second neutron count data). Then, the Long Short-Term Memory (LSTM) network is used to process the converted time sequence data to obtain the measurement results output by the LSM network.

[0066] For example, the Long Short-Term Memory (LSTM) network includes an LSM layer and a fully connected layer. The LSM layer extracts sequence features from the time-series data. The fully connected layer generates a measurement result of the quality of the nuclear material under test based on the sequence features of the time-series data. The structure of the LSM layer is shown in Figure 4. The LSM layer includes a cell state path C, a hidden state path H, a forget gate F, an input gate I, and a candidate memory layer. And output gate O. Time series data includes input data at multiple time steps.

[0067] The cell state path C is used to transfer and accumulate long-term features related to the quality of the nuclear material being tested. The hidden state path H is used to generate temporal features at time step t based on the input data and long-term features at time step t. The forget gate is used to... (The data corresponding to time step t in the time series data) and the time series characteristics of time step t-1. This generates a forgetting gate signal. This forgetting gate signal is used to control the long-term characteristics of the cell state pathway at time step t-1. The process involves selecting key features. The input gate generates an input gate signal based on the input data at time step t and the timing features at time step t-1. This input gate signal controls the candidate memory layer to output candidate features. The candidate memory layer, under the influence of control signals (such as the input gate signal), generates candidate features for time step t based on the input data at time step t and the timing features at time step t-1. The output gate generates an input gate signal based on the input data at time step t and the timing features at time step t-1. This input gate signal controls the output timing features for time step t. .

[0068] The forget gate, input gate, and output gate can use a sigmoid activation function (denoted as ). The candidate memory layer can be implemented using a fully connected layer with the tanh activation function.

[0069] In time step t, the electronic device will input data at time step t. Temporal characteristics with time step t-1 The input data is fed into the forget gate, input gate, and output gate to obtain the forget gate signal, input gate signal, and output gate signal, respectively. Correspondingly, the electronic device will also transmit the input data at time step t. Temporal characteristics with time step t-1 Input the candidate memory layer to generate candidate features.

[0070] Electronic devices will have long-term characteristics Perform element-wise multiplication (one element-wise operator) with the forget gate signal, then perform element-wise addition (another element-wise operator) with the dot product result to obtain the long-term characteristics of time step t. The dot product result is the element-wise multiplication of the input gate signal and the candidate features. The electronic device then uses the tanh function to process the long-term features. The process is performed, and the result is multiplied element-wise with the output gate signal to obtain the timing characteristics at time step t. After obtaining the sequence features of the last time step, the electronic device inputs these features into the fully connected layer to generate the measurement results of the mass of the nuclear material under test.

[0071] In this embodiment, the neural network model can employ a Long Short-Term Memory (LSTM) network. By extracting the temporal dependencies of neutron count data using an LSM network, and capturing the characteristic changes of neutron count data at different times, the neural network model's predictive ability for the mass of the tested nuclear material is improved, thereby enhancing the accuracy and efficiency of mass measurement.

[0072] In some alternative implementations, the neural network model is a backpropagation neural network. The electronic device inputs neutron counting data into the backpropagation neural network, processes the neutron counting data using the backpropagation neural network, and obtains the measurement result output by the backpropagation neural network.

[0073] A backpropagation neural network is a multilayer feedforward neural network. Electronic devices use the backpropagation algorithm to adjust the network parameters of the backpropagation neural network to optimize it. A backpropagation neural network consists of an input layer, one or more hidden layers, and an output layer. Each layer contains one or more neurons. The neurons in each layer are interconnected, establishing a mapping relationship between neutron count data and the mass of the nuclear material, thereby enabling accurate measurement of the mass of the nuclear material being tested.

[0074] For example, the structure of the backpropagation neural network is shown in Figure 5. The backpropagation neural network includes an input layer, a hidden layer, and an output layer. Each layer includes multiple neurons. The input layer includes three neurons x1, x2, and x3, used to input neutron count data. For example, neuron x1 inputs the single neutron count rate, neuron x2 inputs the double coincidence neutron count rate, and neuron x3 inputs the triple coincidence neutron count rate. The hidden layer includes five neurons h1 to h5, used to perform nonlinear transformations and feature integration on the neutron count data to extract key mass-related features from the neutron count data. The output layer includes one neuron o1, used to map the key features extracted by the hidden layer to the measurement result of the mass of the nuclear material to be tested.

[0075] In this embodiment, the neural network model can be a backpropagation neural network. By performing nonlinear transformation and feature integration on the neutron count data using the backpropagation neural network, the mapping relationship between the neutron count data and the mass of the nuclear material under test is captured. This allows for the measurement result of the mass of the nuclear material under test to be obtained from the neutron count data, achieving rapid, accurate, and non-destructive measurement of the mass of the nuclear material under test.

[0076] This application also provides a model training method. This method is applied to an electronic device. The electronic device performing model training and the electronic device performing quality measurement can be the same electronic device or different electronic devices. This application does not limit this. The following description uses the electronic device as the execution subject to illustrate the model training method provided in this application.

[0077] As shown in Figure 6, the model training method provided in this application embodiment includes the following steps: Step 601: The electronic device acquires the measurement data of the sample nuclear material; Step 602: The electronic device divides the measurement data into a training set and a test set; Step 603: The electronic device trains the neural network model to be trained based on the measurement data in the training set to obtain the trained neural network model; Step 604: The electronic device tests the trained neural network model based on the measurement data in the test set to obtain the test results; Step 605: If the test results meet the preset test conditions, the electronic device determines the neural network model as the neural network model used to measure the mass of the nuclear material to be tested.

[0078] In this embodiment, the sample nuclear material is a nuclear material sample used for training the neural network model. The quality of the sample nuclear material is known data. The measurement data is used to represent the neutron distribution of the sample nuclear material during the fission process. The method of obtaining the measurement data is the same as the method of obtaining the neutron count data described above, and will not be repeated here. In some implementations, the measurement data includes one or more of the following: single neutron count rate, double coincidence neutron count rate, triple coincidence neutron count rate, total coincidence neutron count distribution, background neutron count distribution, and neutron pulse time series.

[0079] In practical applications, electronic devices can acquire measurement data and corresponding data labels for multiple sample core materials. For example, an electronic device can acquire measurement data and data labels for sample core materials of different masses. The data labels are used to represent the true mass of the sample core material. In some implementations, the electronic device can also acquire measurement data and data labels for multiple sample core materials through simulation, thereby enriching the training samples for the neural network model.

[0080] To improve the training performance of neural network models, in some optional implementations, as shown in Figure 7, after acquiring the measurement data of the sample core material, the electronic device can preprocess the measurement data. For example, the electronic device can perform one or more preprocessing operations such as noise reduction, normalization, and data augmentation to obtain preprocessed measurement data. This makes the measurement data more suitable for neural network model processing, thereby improving the training effect of the neural network model.

[0081] In step 602, after preprocessing the measurement data, the electronic device can divide the measurement data and data labels of multiple sample core materials into a training set and a test set. The training set is used to train the neural network model. The test set is used to validate the neural network model.

[0082] In step 603, the electronic device trains the neural network model to be trained based on measurement data from the training set. In some optional implementations, the neural network model includes one or more of a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, a long short-term memory network, and a backpropagation neural network.

[0083] For example, when the neural network model is a one-dimensional convolutional neural network and the measurement data is one-dimensional, the electronic device inputs the measurement data from the training set into the one-dimensional convolutional neural network, and uses the measurement data to train the one-dimensional convolutional neural network to obtain the trained one-dimensional convolutional neural network.

[0084] When the neural network model is a two-dimensional convolutional neural network, the electronic device converts the measurement data in the training set into two-dimensional data, obtaining the converted measurement data. Then, the electronic device inputs the converted measurement data into the two-dimensional convolutional neural network, and uses the measurement data to train the two-dimensional convolutional neural network, obtaining the trained two-dimensional convolutional neural network.

[0085] When the neural network model includes a Long Short-Term Memory (LSTM) network, and the measurement data is measured periodically at preset time intervals, the electronic device arranges the measurement data in chronological order to obtain measurement sequence data. Then, the electronic device inputs the measurement sequence data into the LTM network to train it, resulting in a trained LTM network.

[0086] When the neural network model is a backpropagation neural network, the electronic device inputs the measurement data from the training set into the backpropagation neural network, and uses the measurement data to train the long short-term memory network, thus obtaining the trained long short-term memory network.

[0087] In practical applications, the training set includes not only the measurement data of the sample core materials but also the data labels of the sample core materials. As shown in Figure 7, the electronic device trains the neural network model to be trained based on the measurement data and data labels in the training set.

[0088] For example, the electronic device inputs measurement data from the training set into a neural network model, processes the measurement data using the neural network model, and obtains the output of the neural network model. Based on the output of the neural network model and the data labels of the sample kernel materials, the electronic device calculates the model loss of the neural network model. Then, the electronic device updates the model parameters (i.e., network parameters) of the neural network model according to the model loss using a backpropagation algorithm and an optimization algorithm (such as gradient descent), i.e., iteratively optimizing the neural network model to obtain the trained neural network model.

[0089] In this embodiment, the electronic device can calculate the model loss of the neural network model using a loss function. For example, the loss function can be a mean squared error loss function, a mean absolute error loss function, or a smoothed L1 loss function. By training the neural network model, the model extracts key features related to the quality of the sample nuclear material from the measurement data, establishes a mapping relationship between the quality of the nuclear material and the measurement data, and thus achieves the quality measurement of the nuclear material to be tested through the trained neural network model.

[0090] In step 604, in order to evaluate the performance of the trained neural network model, as shown in Figure 7, after obtaining the trained neural network model, the electronic device uses measurement data from the test set to test the trained neural network model to verify the predictive performance of the trained neural network model.

[0091] During testing, the electronic device inputs the measurement data from the test set into a trained neural network model. The trained neural network model then processes the measurement data to obtain the output of the neural network. In addition to the measurement data of the sample core materials, the test set also includes the data labels of the sample core materials. Based on the output of the neural network and the corresponding data labels, the electronic device determines the test result of the neural network model.

[0092] In this embodiment, the test results of the neural network model can be represented by some evaluation metrics. For example, the test results can be evaluation metrics such as mean absolute error, root mean square error, and coefficient of determination. The electronic device calculates evaluation metrics such as mean absolute error, root mean square error, and coefficient of determination based on the output of the neural network and the corresponding data labels to obtain the test results of the neural network model.

[0093] In step 605, the electronic device determines whether the test results of the neural network model meet preset test conditions. For example, the electronic device determines whether the mean absolute error of the neural network model is less than a preset error threshold, or whether the determination coefficient of the neural network model is greater than a preset coefficient threshold.

[0094] If the test results of the neural network model meet the preset test conditions, it indicates that the neural network model has good predictive performance. The electronic device then uses the neural network model as a pre-trained neural network model to measure the mass of the nuclear material under test. If the test results of the neural network model do not meet the preset test conditions, the electronic device returns to the training phase of the neural network model and retrains the neural network model using the measurement data in the training set.

[0095] For example, Figure 8 shows a comparison between the output of the neural network model and the data labels of the sample nuclear material. The horizontal axis of the comparison graph represents the true quality indicated by the data labels, and the vertical axis represents the output of the neural network. The dashed line represents a baseline consistent with the true quality. Dots (including black dots and circles) represent the output of the neural network. The closer a dot is to the dashed line, the more accurate the neural network output. It can be seen that the output obtained by the electronic device using the neural network model to process the measurement data of the sample nuclear material at different measurement durations is near the baseline. The longer the measurement time, the closer the output of the neural network model is to the baseline. Specifically, the output corresponding to a 10-second measurement duration has a smaller difference from the baseline, and the difference further decreases with a 100-second measurement duration. This shows that the neural network model can obtain relatively accurate output results when the measurement duration reaches the ten-second level. Therefore, the neural network model can achieve accurate prediction of nuclear material quality with a significantly reduced measurement duration.

[0096] In this embodiment, a neural network model (i.e., a pre-trained neural network model) for measuring the quality of nuclear materials can be obtained through training and testing of the neural network model. The neural network model has good generalization ability and is suitable for the quality inspection needs of nuclear materials with various physical shapes, thereby realizing rapid and accurate measurement of the quality of nuclear materials.

[0097] To implement the mass measurement method provided in this application embodiment, this application embodiment also provides a mass measurement device. As shown in FIG9, the mass measurement device includes: an acquisition module 91, used to acquire neutron count data of the nuclear material to be tested within a first measurement duration, wherein the first measurement duration is less than a second measurement duration, and the second measurement duration is the duration for neutron multiplicity measurement of the nuclear material to be tested based on a preset mathematical model; and a measurement module 92, used to perform mass measurement of the nuclear material to be tested based on the neutron count data and a pre-trained neural network model, to obtain the measurement result output by the neural network model, wherein the measurement result represents the mass of the nuclear material to be tested.

[0098] In some optional implementations, the neural network model includes a one-dimensional convolutional neural network; the neutron counting data is one-dimensional data; the measurement module 92 is specifically used to: input the neutron counting data into the one-dimensional convolutional neural network, use the one-dimensional convolutional neural network to perform mass measurement on the nuclear material to be tested, and obtain the measurement result output by the neural network model.

[0099] In some optional implementations, the neural network model includes a two-dimensional convolutional neural network; the measurement module 92 is specifically used to: convert the neutron count data into two-dimensional data to obtain the converted neutron count data; input the converted neutron count data into the two-dimensional convolutional neural network, and use the two-dimensional convolutional neural network to perform mass measurement on the nuclear material to be tested, thereby obtaining the measurement result output by the neural network model.

[0100] In some optional implementations, the neural network model includes a long short-term memory network; the neutron count data is data measured periodically at preset time intervals within the first measurement duration; the measurement module 92 is specifically used to: arrange the neutron count data in chronological order to obtain time-series data; input the time-series data into the long short-term memory network, and use the long short-term memory network to perform mass measurement on the nuclear material under test to obtain the measurement result output by the neural network model.

[0101] In some optional implementations, the neural network model includes a backpropagation neural network; the measurement module 92 is specifically used to: input the neutron count data into the backpropagation neural network, use the backpropagation neural network to perform mass measurement on the nuclear material to be tested, and obtain the measurement result output by the neural network model.

[0102] In some optional implementations, the neutron counting data includes one or more of the following: single neutron count rate, double coincidence neutron count rate, triple coincidence neutron count rate, and neutron pulse time series.

[0103] In practical applications, the acquisition module 91 can be implemented by a processor in the quality measurement device combined with a communication interface, and the measurement module 92 can be implemented by a processor in the quality measurement device.

[0104] It should be noted that the quality measuring device provided in this application embodiment is only illustrated by the above-described division of program modules when performing quality measurements. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the quality measuring device and quality measuring method provided in this application embodiment belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.

[0105] To implement the model training method provided in this application embodiment, this application embodiment also provides a model training device. As shown in FIG10, the model training device includes: an acquisition module 1001, used to acquire measurement data of sample nuclear material, wherein the measurement data represents neutron distribution; a partitioning module 1002, used to partition the measurement data into a training set and a test set; a training module 1003, used to train a neural network model to be trained based on the measurement data in the training set, to obtain a trained neural network model; a testing module 1004, used to test the trained neural network model based on the measurement data in the test set, to obtain test results; and, if the test results meet preset test conditions, a pre-trained neural network model is obtained.

[0106] In some optional implementations, the neural network model includes a one-dimensional convolutional neural network; the measurement data is one-dimensional data; the training module 1003 is specifically used to: input the measurement data in the training set into the one-dimensional convolutional neural network, and train the one-dimensional convolutional neural network using the measurement data to obtain the trained one-dimensional convolutional neural network.

[0107] In some optional implementations, the neural network model includes a two-dimensional convolutional neural network; the training module 1003 is specifically used to: convert the measurement data in the training set into two-dimensional data to obtain the converted measurement data; input the converted measurement data into the two-dimensional convolutional neural network, and train the two-dimensional convolutional neural network using the measurement data to obtain the trained two-dimensional convolutional neural network.

[0108] In some optional implementations, the neural network model includes a long short-term memory network; the measurement data is data measured periodically at preset time intervals; the training module 1003 is specifically used to: arrange the measurement data in chronological order to obtain measurement sequence data; input the measurement sequence data into the long short-term memory network, and train the long short-term memory network using the measurement sequence data to obtain a trained long short-term memory network.

[0109] In some optional implementations, the neural network model includes a backpropagation neural network; the training module 1003 is specifically used to: input the measurement data into the backpropagation neural network, and use the measurement data to train the long short-term memory network to obtain the trained long short-term memory network.

[0110] In some optional implementations, the neutron counting data includes one or more of the following: single neutron count rate, double coincidence neutron count rate, triple coincidence neutron count rate, and neutron pulse time series.

[0111] In practical applications, the acquisition module 1001 can be implemented by the processor in the model training device in conjunction with the communication interface, and the partitioning module 1002, training module 1003 and testing module 1004 can be implemented by the processor in the model training device.

[0112] It should be noted that the model training device provided in this application embodiment is only illustrated by the above-described division of program modules during model training. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. Furthermore, the model training device and model training method provided in this application embodiment belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0113] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of this application, the embodiments of this application also provide an electronic device, as shown in FIG11. The electronic device includes: a communication interface 1101, which can interact with other devices (such as a nuclear material non-destructive analyzer); a processor 1102, which is connected to the communication interface 1101 to enable interaction with other devices (such as a nuclear material non-destructive analyzer) and is used to execute the method provided by one or more of the above technical solutions when running a computer program; and a memory 1103, on which the computer program is stored.

[0114] When the electronic device is used for mass measurement, the processor 1102, in conjunction with the communication interface 1101, acquires neutron count data of the nuclear material under test within a first measurement duration, wherein the first measurement duration is less than a second measurement duration, and the second measurement duration is the duration for neutron multiplicity measurement of the nuclear material under test based on a preset mathematical model; the processor 1102 is used to perform mass measurement of the nuclear material under test based on the neutron count data using a pre-trained neural network model, and obtain the measurement result output by the neural network model, wherein the measurement result represents the mass of the nuclear material under test.

[0115] In some optional implementations, the neural network model includes a one-dimensional convolutional neural network; the processor 1102 is specifically used to: input the neutron counting data into the one-dimensional convolutional neural network, use the one-dimensional convolutional neural network to perform mass measurement on the nuclear material to be tested, and obtain the measurement result output by the neural network model.

[0116] In some optional implementations, the neural network model includes a two-dimensional convolutional neural network; the processor 1102 is specifically used to: convert the neutron counting data into two-dimensional data to obtain converted neutron counting data; input the converted neutron counting data into the two-dimensional convolutional neural network, and use the two-dimensional convolutional neural network to perform mass measurement on the nuclear material to be tested, thereby obtaining the measurement result output by the neural network model.

[0117] In some optional implementations, the neural network model includes a long short-term memory network; the neutron count data is data measured periodically at preset time intervals within the first measurement duration; the processor 1102 is specifically used to: arrange the neutron count data in chronological order to obtain time-series data; input the time-series data into the long short-term memory network, use the long short-term memory network to perform mass measurement on the nuclear material under test, and obtain the measurement result output by the neural network model.

[0118] In some optional implementations, the neural network model includes a backpropagation neural network; the processor 1102 is specifically configured to: input the neutron counting data into the backpropagation neural network, use the backpropagation neural network to perform mass measurement on the nuclear material under test, and obtain the measurement result output by the neural network model.

[0119] When the electronic device is used for model training, the processor 1102, in conjunction with the communication interface 1101, acquires measurement data of the sample nuclear material, wherein the measurement data represents the neutron distribution; the processor 1102 is used to divide the measurement data into a training set and a test set; the neural network model to be trained is trained based on the measurement data in the training set to obtain a trained neural network model; the trained neural network model is tested based on the measurement data in the test set to obtain test results; and a pre-trained neural network model is obtained if the test results meet preset test conditions.

[0120] It should be noted that when the processor 1102 of the electronic device is used to run computer programs, the steps of the above method are not repeated here. The specific processing procedures of the processor 1102 and the communication interface 1101 can be understood with reference to the above method.

[0121] Of course, in practical applications, the various components in an electronic device are coupled together through bus system 1104. It can be understood that bus system 1104 is used to achieve communication between these components. In addition to the data bus, bus system 1104 also includes a power bus, a control bus, and a status signal bus. However, for clarity, all buses are labeled as bus system 1104 in Figure 11.

[0122] The memory 1103 in this embodiment is used to store various types of data to support the operation of the electronic device. Examples of such data include any computer program used to operate on the electronic device.

[0123] The methods disclosed in the embodiments of this application can be applied to the processor 1102, or implemented by the processor 1102. The processor 1102 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 1102 or by instructions in the form of software. The processor 1102 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 1102 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the memory 1103. The processor 1102 reads the information in the memory 1103 and completes the steps of the aforementioned method in conjunction with its hardware.

[0124] In an exemplary embodiment, the electronic device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0125] It is understood that the memory 1103 in this embodiment can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0126] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 1103 storing a computer program, which can be executed by a processor 1102 of an electronic device to complete the steps described in the above method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0127] In an exemplary embodiment, this application also provides a computer program product, including a computer program that can be executed by a processor 1102 of an electronic device to perform the steps described in the foregoing method.

[0128] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0129] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0130] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A method for measuring quality, characterized in that, The method includes: acquiring neutron count data of the nuclear material to be tested within a first measurement duration, wherein the first measurement duration is shorter than a second measurement duration, and the second measurement duration is the duration for measuring the neutron multiplicity of the nuclear material to be tested based on a preset mathematical model; and, based on the neutron count data, performing a mass measurement on the nuclear material to be tested using a pre-trained neural network model to obtain a measurement result output by the neural network model, wherein the measurement result represents the mass of the nuclear material to be tested.

2. The method according to claim 1, characterized in that, The neural network model includes a one-dimensional convolutional neural network; the neutron counting data is one-dimensional data; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron counting data, and obtaining the measurement result output by the neural network model, includes: inputting the neutron counting data into the one-dimensional convolutional neural network, using the one-dimensional convolutional neural network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

3. The method according to claim 1, characterized in that, The neural network model includes a two-dimensional convolutional neural network; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data and obtaining the measurement result output by the neural network model includes: converting the neutron count data into two-dimensional data to obtain converted neutron count data; inputting the converted neutron count data into the two-dimensional convolutional neural network, and using the two-dimensional convolutional neural network to measure the mass of the nuclear material under test and obtain the measurement result output by the neural network model.

4. The method according to claim 1, characterized in that, The neural network model includes a long short-term memory network; the neutron count data is data measured periodically at preset time intervals within the first measurement duration; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data, and obtaining the measurement result output by the neural network model, includes: arranging the neutron count data in chronological order to obtain time-series data; inputting the time-series data into the long short-term memory network, and using the long short-term memory network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

5. The method according to claim 1, characterized in that, The neural network model includes a backpropagation neural network; the step of using the pre-trained neural network model to measure the mass of the nuclear material under test based on the neutron count data and obtaining the measurement result output by the neural network model includes: inputting the neutron count data into the backpropagation neural network, using the backpropagation neural network to measure the mass of the nuclear material under test, and obtaining the measurement result output by the neural network model.

6. The method according to any one of claims 1-5, characterized in that, The neutron counting data includes one or more of the following: single neutron count rate, double coincidence neutron count rate, triple coincidence neutron count rate, and neutron pulse time series.

7. A model training method, characterized in that, The method includes: acquiring measurement data of sample nuclear material, wherein the measurement data is used to represent neutron counts; dividing the measurement data into a training set and a test set; training a neural network model to be trained based on the measurement data in the training set to obtain a trained neural network model; testing the trained neural network model based on the measurement data in the test set to obtain test results; and obtaining a pre-trained neural network model when the test results meet preset test conditions.

8. An electronic device, characterized in that, include: A processor and a memory for storing a computer program capable of running on the processor; wherein, when the processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 6, or performs the steps of the method according to claim 7.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6, or implements the steps of the method according to claim 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6, or implements the steps of the method according to claim 7.

Images