Radioactive waste treatment scheme generation method and device based on artificial intelligence
By constructing an initial database and automatically generating radioactive waste disposal solutions using a large language model, the problem of low efficiency in existing technologies is solved, enabling the generation of efficient disposal solutions even for inexperienced personnel.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA INST FOR RADIATION PROTECTION
- Filing Date
- 2025-12-10
- Publication Date
- 2026-05-01
AI Technical Summary
In existing technologies, the development of radioactive waste disposal solutions relies on experienced personnel, which leads to inefficiency and increases the burden of research and experimentation on staff.
Using an AI-based large language model, a radioactive waste disposal plan is automatically generated by constructing an initial database, extracting feature data, fine-tuning the model, and inputting current experimental data.
It enables inexperienced personnel to automatically generate processing solutions, improving processing efficiency and reducing reliance on experienced personnel.
Smart Images
Figure CN121960730A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for generating radioactive waste treatment schemes based on artificial intelligence. Background Technology
[0002] Radioactive waste treatment technology refers to all operational practices that make radioactive waste suitable for final disposal, such as collection, sorting, concentration, incineration, compression, decontamination, solidification, packaging, storage, and transportation. The goal of waste treatment is to minimize the volume of radioactive waste to reduce storage, transportation, and disposal costs; and to recover or reuse it as much as possible to reduce emissions into the environment.
[0003] Traditional radioactive waste treatment technologies rely on experienced personnel to develop solutions based on existing treatment protocols and the specific characteristics of the waste to be treated. However, for less experienced personnel, developing a radioactive waste treatment technology requires the assistance of experienced staff and reference to relevant experimental data. This approach not only slows down the development of the technology but also increases the research and experimental workload for staff.
[0004] The above problems urgently need to be addressed. Summary of the Invention
[0005] This invention discloses a method and apparatus for generating radioactive waste treatment schemes based on artificial intelligence, aiming to solve the technical problems existing in the prior art.
[0006] The present invention adopts the following technical solution: On one hand, the present invention provides a method for generating radioactive waste treatment schemes based on artificial intelligence, comprising: acquiring historical experimental data and historical treatment schemes of historical radioactive waste, and constructing an initial database; based on the initial database, extracting feature data from the historical experimental data, and based on the feature data, retrieving the same feature points in historical treatment schemes that have the same feature data, and constructing an index database of feature data and feature points of treatment schemes; based on the index database, fine-tuning a large language model, wherein the fine-tuned large language model records the features of the index database, and when the current experimental data input to the fine-tuned large language model matches multiple historical experimental data, filtering multiple historical treatment schemes and constructing candidate treatment schemes; and determining a target treatment scheme based on the matching degree of the candidate treatment schemes.
[0007] Optionally, the historical experimental data includes various types of data, including at least: the name of the object being treated, the nature of the object being treated, experimental nodes in the treatment process, feedback from the treatment process, and the status of secondary waste generation; the historical treatment scheme includes various types of data, including at least: the name of the treatment technology, the conditions of the treatment technology, the environment of the treatment technology, the non-compliance of the treated object, and the technological advantages.
[0008] Optionally, based on the initial database, feature data is extracted from the historical experimental data, and based on the feature data, the same feature points are retrieved from historical processing schemes that have the same feature data, and an index database of feature data and processing scheme feature points is constructed. This includes: preprocessing the initial database, filtering historical experimental data that lacks processing object names or processing experimental nodes to obtain a preprocessed database; based on the preprocessed database, extracting features of each type of data in the historical experimental data to obtain multiple features corresponding to the historical experimental data, and constructing feature data; based on the same features corresponding to the same type of data in the multiple feature data, filtering feature points in historical processing schemes that have the same features, and matching the feature points with the features and data types in the feature data to obtain the index database.
[0009] Optionally, based on the index database, the large language model is fine-tuned, including: performing field feature adaptation on the feature data and feature points in the index database; using low-rank adaptation technology to freeze the weights in the large language model, and training the low-rank matrix in the large language model based on the field features; adjusting the learning rate, training epochs, and loss function of the large language model, and outputting the fine-tuned large language model.
[0010] Optionally, when the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened to construct candidate processing schemes, including: inputting the current experimental data into the fine-tuned large language model; the fine-tuned large language model extracting current features of multiple data from the current experimental data; matching multiple historical experimental data based on multiple current features, corresponding to multiple historical processing schemes; and constructing candidate processing schemes based on the multiple historical processing schemes.
[0011] Optionally, determining the target processing scheme based on the matching degree of the candidate processing schemes includes: matching multiple current features with multiple feature data in the historical experimental data; calculating the matching degree of the current feature with multiple historical experimental data based on feature matching weights, wherein the feature matching weight of the processing object name is the highest; and determining the historical processing scheme corresponding to the historical experimental data with the highest matching degree as the target processing scheme by arranging them in descending order of matching degree.
[0012] Optionally, after determining the target processing scheme based on the matching degree of the candidate processing schemes, the method further includes: extracting feature data from the current experimental data and feature points from the corresponding target processing scheme to construct a current feature set; placing the current feature set into an index database to perform real-time fine-tuning of the large language model.
[0013] According to another aspect of the present invention, an artificial intelligence-based radioactive waste treatment scheme generation device is also provided, comprising: a historical data module, used to acquire historical experimental data and historical treatment schemes of historical radioactive waste, and construct an initial database; a feature extraction module, used to extract feature data from the historical experimental data based on the initial database, and based on the feature data, retrieve the same feature points in historical treatment schemes that have the same feature data, and construct an index database of feature data and treatment scheme feature points; a model adjustment module, used to fine-tune a large language model based on the index database, wherein the fine-tuned large language model records the features of the index database, and when the current experimental data input to the fine-tuned large language model matches multiple historical experimental data, multiple historical treatment schemes are selected to construct candidate treatment schemes; and a scheme generation module, used to determine a target treatment scheme based on the matching degree of the candidate treatment schemes.
[0014] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium storing a plurality of instructions adapted for loading and execution by a processor of any one of the methods for generating an artificial intelligence-based radioactive waste treatment scheme.
[0015] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of any one of the methods for generating a radioactive waste treatment scheme based on artificial intelligence.
[0016] The technical solution adopted in this invention can achieve at least one of the following beneficial effects: In this embodiment of the invention, an initial database is constructed by acquiring historical experimental data and historical treatment schemes for historical radioactive waste. Based on the initial database, feature data is extracted from the historical experimental data, and based on the feature data, identical feature points are retrieved from historical treatment schemes that correspond to the same feature data, thus constructing an index database of feature data and treatment scheme feature points. Based on the index database, a large language model is fine-tuned, wherein the fine-tuned large language model records the features of the index database. When the current experimental data input into the fine-tuned large language model matches multiple historical experimental data, multiple historical treatment schemes are selected to construct candidate treatment schemes. Based on the matching degree of the candidate treatment schemes, a target treatment scheme is determined, achieving the goal of automatically generating treatment schemes. This enables inexperienced workers to perform subsequent radioactive waste treatment based on the automatically generated treatment schemes and their own experience, without relying entirely on experienced personnel, thus improving the technical effect of processing efficiency. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below, forming part of the present invention. The illustrative embodiments of the present invention and their descriptions explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of a method for generating a radioactive waste treatment scheme based on artificial intelligence in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of a radioactive waste treatment scheme generation device based on artificial intelligence in Embodiment 2 of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. In the description of this invention, it should be noted that the term "or" is generally used to include the meaning of "and / or," unless otherwise expressly indicated.
[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or a magnetic connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, in the description of this application, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. In the description of this invention, "a plurality of" means at least two, such as two, three, or more, unless otherwise explicitly specified.
[0020] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0021] To address the problems existing in related technologies, this application provides a method and apparatus for generating radioactive waste treatment schemes based on artificial intelligence.
[0022] Example 1 This embodiment provides a method for generating radioactive waste treatment solutions based on artificial intelligence, such as... Figure 1 As shown, Figure 1 This is a flowchart of a method for generating a radioactive waste treatment scheme based on artificial intelligence according to Embodiment 1 of the present invention. The method includes: Step S102: Obtain historical experimental data and historical treatment plans for historical radioactive waste, and construct an initial database; Optionally, historical experimental data primarily refers to various historical data accumulated by the user based on previous experiments, as well as valid data retrieved from the internet using generative artificial intelligence's search capabilities. Specifically, historical data accumulated from previous experiments refers to various data sources accumulated by the user's institution during the process, including experimental subjects, data, procedures, and conclusions. Valid data retrieved from the internet refers to content related to radioactive waste treatment technology selected through generative artificial intelligence software by searching major Chinese and foreign academic paper websites.
[0023] In some preferred embodiments, historical experimental data includes a variety of data, including at least: the name of the object to be treated, the nature of the object to be treated, experimental nodes of the treatment process, feedback of the treatment process, and the status of secondary waste generation; historical treatment schemes include a variety of data, including at least: the name of the treatment technology, the conditions of the treatment technology, the environment of the treatment technology, the non-compliance of the treated object, and the technical advantages.
[0024] Optionally, the information collected in the initial database should be gathered from two aspects: For the treatment object (i.e., historical experimental data on radioactive waste), it should at least include: the name of the treatment object, its characteristics (physical and chemical properties), experimental nodes in the treatment process, feedback from the treatment process, and secondary waste status. For the treatment technology object (i.e., historical treatment schemes), it should at least include: the name of the treatment technology, the applicable scenarios for the treatment technology (including conditions, environment, etc.), the non-compliance of the treated object, and the technological advantages.
[0025] Step S104: Based on the initial database, extract feature data from historical experimental data, and based on the feature data, retrieve the same feature points in historical processing schemes that have the same feature data, and construct an index database of feature data and feature points of processing schemes. Optionally, based on the natural language text content of historical experimental data, relevant conditions and special requirements are extracted from the historical experimental data; a finely tuned large language model is then used to generate the desired processing technology. The natural language text here can be divided into two categories: radioactive waste treatment technology and radioactive waste itself. In practical work, one may encounter a waste to be treated where a suitable treatment technology cannot be found; conversely, one may encounter a good non-nuclear industry treatment technology, but a suitable treatment object cannot be found. In such cases, based on the transformation of the natural language text content of historical experimental data into feature data, the most closely matching feature data can be used to find the appropriate processing technology.
[0026] In some preferred embodiments, based on an initial database, feature data is extracted from historical experimental data. Based on the feature data, identical feature points are retrieved from historical processing schemes that correspond to the same feature data, and an index database of feature data and processing scheme feature points is constructed. This includes: preprocessing the initial database, filtering historical experimental data that lacks processing object names or experimental nodes of the processing process to obtain a preprocessed database; based on the preprocessed database, extracting features of each type of data in the historical experimental data to obtain multiple features corresponding to the historical experimental data, and constructing feature data; based on the identical features corresponding to the same type of data in the multiple feature data, filtering feature points in historical processing schemes that correspond to historical experimental data with the same features, and matching the feature points with the features and data types in the feature data to obtain the index database.
[0027] Step S106: Based on the index database, fine-tune the large language model. The fine-tuned large language model records the features of the index database. When the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened to construct candidate processing schemes. Optionally, a large language model capable of online self-learning can be deployed on the internet, embedded within WPS, and its information should be readable. The performance of the internet-connected computer should be sufficient to support the large language model's operating parameters. The large language model should be selected based on the specific task, choosing a suitable open-source LLM model structure, such as Deepseek or other self-controllable large language models. Factors such as the scale of the large language model should be considered during the design process.
[0028] Optionally, the large language model is fine-tuned to integrate various types of accumulated data and effective data retrieved from the internet. This generates a process that generates current data based on user input, deeply understanding and analyzing the user's natural language text and comprehending the user's preferences and needs.
[0029] In some preferred embodiments, the large language model is fine-tuned based on the index database, including: adapting the feature data and feature points in the index database to field features; using low-rank adaptation technology to freeze the weights in the large language model and training the low-rank matrix in the large language model based on the field features; adjusting the learning rate, training epochs and loss function of the large language model, and outputting the fine-tuned large language model.
[0030] Optionally, feature data and feature points should first be adapted to field features. This involves representing core fields according to field identifiers. For example, the unique identifier for an experiment ID might be input as text plus numbers, such as "Nuclear Waste Treatment - 20230510-001". During the field feature adaptation stage, this would be directly converted to a string, i.e., "20230510-001". Similarly, for key historical experimental parameters, including quantitative indicators (such as temperature, pressure, radioactivity, and material ratios; standardized units are required, such as "Activity: 10⁻⁶"), this adaptation would be performed. 4 The fields "Bq" and "Temperature: 500℃" can be identified as key-value pairs. This method improves the efficiency of subsequent retrieval and matching.
[0031] Optionally, after field feature adaptation, it is necessary to unify parameter units (e.g., convert "mSv / h" and "μSv / h" to the same order of magnitude); and remove invalid data (e.g., records with ≥30% missing experimental parameters or vague scheme descriptions); and also perform word segmentation on unstructured text (e.g., treatment schemes, effect evaluations) (using domain dictionaries, such as nuclear industry terminology databases) to generate keyword tags.
[0032] Optionally, the above data can be constructed into a comprehensive index database. Specifically, a dual indexing mechanism of "vector index + keyword index" is adopted. The vector index concatenates "process object name + processing experimental node + processing feedback + secondary waste status" into text, encodes it into a 768-dimensional vector through a domain pre-trained model (such as SciBERT, ERNIE-3.0 Science), and stores it in a vector database (such as FAISS, Milvus), supporting semantic similarity retrieval. The keyword index constructs an inverted index based on "process object label", supporting exact matching (such as filtering historical experimental data related to "high radioactive waste" and "vitrification").
[0033] Optionally, randomly extract historical experimental data from the index database and construct "input-output" pairs in the above format. It is recommended that the sample size be ≥200 (when experimental data is scarce, similar samples can be generated through "data augmentation" methods searched on the Internet). Divide the index database into training set (80%), validation set (10%), and test set (10%) to ensure that the test set covers different experimental scenarios.
[0034] Optionally, the Deepseek large language model can be used as the base model, and low-rank adaptation (LoRA) can be used for fine-tuning. The weights of the base model are frozen, and only the low-rank matrix (rank value r=8~32) is trained to reduce the amount of computation (no need for full fine-tuning, GPU memory ≥24GB is sufficient). Meanwhile, the key parameters of the large language model are adjusted as follows: learning rate: 5e-5 to 1e-4; batch size: adjusted according to GPU memory, using 8 to 16; training epochs: 3 to 5 epochs (to avoid overfitting, stop training when the validation set accuracy is achieved); loss function: cross-entropy loss (for text generation tasks) + contrastive learning loss (to strengthen the mapping between "similar experiments → similar solutions").
[0035] In some preferred embodiments, when the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened to construct candidate processing schemes, including: inputting the current experimental data into the fine-tuned large language model; extracting current features of multiple data from the current experimental data using the fine-tuned large language model; matching multiple historical experimental data based on multiple current features, corresponding to multiple historical processing schemes; and constructing candidate processing schemes based on multiple historical processing schemes.
[0036] Optionally, the text content generated by the large language model may still differ from the actual usable content. In the early stages, various candidate processing solutions can be generated by the large language model. At this point, experienced experts and practitioners need to evaluate the solutions and refer to past experimental experiences to finally determine the processing techniques or solutions that may be applicable.
[0037] Optionally, the user inputs the type and characteristics of the waste to be addressed, interacts with the computer based on the information required for radioactive waste treatment, and the large language model understands the input based on prior training and existing data, outputting content that meets the user's expectations. Furthermore, in the early stages, experienced experts and practitioners evaluate the information and reference past experimental experience, fine-tuning the data before outputting the results.
[0038] Optionally, the current experimental data is used as input into a large language model. First, keyword coarse screening is performed to quickly narrow down the scope. Based on the "processing object + experimental scenario" of the current experiment, a preliminary matching historical experiment is selected through an inverted index (e.g., filtering records related to "high-level radioactive waste" and "solidification"), retaining the Top 50 records. Then, feature fine screening is performed using semantic + parameter similarity. The standardized data of the current experiment is encoded into a vector, and the cosine similarity is calculated with the 50 historical data vectors selected in the first stage (the similarity threshold is set to ≥0.7, which can be increased to ≥0.8 in professional scenarios); at the same time, the matching degree is calculated (e.g., if the current temperature is 450℃ and the historical experimental temperature is 480℃, the matching degree = (min(450,480) / max(450,480)) = 93.75%). Only records with "similarity ≥0.7 and matching degree ≥70%" are retained, and finally, the Top 10 records are retained, effectively avoiding redundancy caused by too many candidate processing schemes.
[0039] Finally, historical solutions are filtered and prioritized. The filtering rules are to remove records where the "object being processed does not meet the conditions" (such as solutions where "cracking occurs due to excessive temperature") and to exclude solutions that conflict with the current constraints, resulting in multiple candidate processing solutions.
[0040] Step S108: Determine the target processing solution based on the matching degree of the candidate processing solutions.
[0041] In some preferred embodiments, determining the target processing scheme based on the matching degree of the candidate processing schemes includes: matching multiple current features with multiple feature data in historical experimental data; calculating the matching degree between the current feature and multiple historical experimental data based on feature matching weights, wherein the feature matching weight of the processing object name is the highest; and determining the historical processing scheme corresponding to the historical experimental data with the highest matching degree as the target processing scheme by arranging them from largest to smallest.
[0042] Optionally, among the final selected candidate treatment schemes, priority ranking (in descending order of weight) is as follows: degree of achievement of experimental objectives (e.g., the weight of a historical treatment scheme that is "up to standard" > "close to target"); matching degree of key parameters (weight 40%); consistency of experimental scenario (weight 30%); matching degree of condition constraints (weight 20%); and implementation complexity of the scheme (e.g., the weight of a scheme that "does not require special equipment" > "requires customized equipment" (weight 10%).
[0043] The process ultimately yields multiple candidate processing schemes and their ranking. After generating processing techniques, the selected scheme is adjusted and optimized. The resulting processing techniques or schemes are then tested and verified using an experimental setup to ultimately determine which technique to use for waste treatment. Through this effective process, the resulting data is compiled into a new dataset and resubmitted to the large language model system for learning and data enrichment.
[0044] In some preferred embodiments, after determining the target processing scheme based on the matching degree of the candidate processing schemes, the method further includes: extracting feature data from the current experimental data and feature points from the corresponding target processing scheme to construct a current feature set; placing the current feature set into an index database to perform real-time fine-tuning of the large language model.
[0045] Optionally, the improved target processing scheme can be tested and verified, and the test results can be organized as learning materials and sent back to the large language model for further learning.
[0046] Based on the generated content, select the most likely feasible solution to build an experimental setup, and conduct experiments to verify the solution. Organize the overall experimental process and results according to the information collection requirements for the corresponding radioactive waste treatment object and technology. Simultaneously, evaluate the overall experimental process, describe areas for improvement, and feed these into a large language model to enhance the model's ability to understand user intent.
[0047] Through steps S102 to S108, the large language model technology is applied to the generation process of radioactive waste treatment technology. This fully leverages the excellent deep learning and text generation capabilities of the large language model, making the natural language text information generated during user interaction with the model more accurate and tailored to user needs. Furthermore, the generation capabilities of the large language model can provide users with more informative and personalized processing solutions, enhancing their understanding and acceptance of the generated results.
[0048] Example 2 This embodiment also provides an artificial intelligence-based radioactive waste treatment scheme generation device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0049] According to an embodiment of the present invention, an apparatus embodiment for implementing the above-described method for generating an artificial intelligence-based radioactive waste treatment scheme is also provided. Figure 2 This is a schematic diagram of a radioactive waste treatment scheme generation device based on artificial intelligence in Embodiment 2 of the present invention, as shown below. Figure 2 As shown, the above-mentioned device includes: a historical data module 201, a feature extraction module 202, a model adjustment module 203, and a scheme generation module 204, wherein: Historical data module 201 is used to acquire historical experimental data and historical treatment plans for historical radioactive waste and to build an initial database. The feature extraction module 202 is connected to the historical data module 201. It is used to extract feature data from historical experimental data based on the initial database, and to retrieve the same feature points in historical processing schemes with the same feature data based on the feature data, and to build an index database of feature data and feature points of processing schemes. The model adjustment module 203 is connected to the feature extraction module 202 and is used to fine-tune the large language model based on the index database. The fine-tuned large language model records the features of the index database. When the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened and candidate processing schemes are constructed. The scheme generation module 204, connected to the model adjustment module 203, is used to determine the target processing scheme based on the matching degree of the candidate processing schemes.
[0050] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.
[0051] It should be noted that the historical data module 201, feature extraction module 202, model adjustment module 203, and scheme generation module 204 mentioned above correspond to steps S102 to S108 in the embodiments. The instances and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run on a computer terminal.
[0052] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.
[0053] The aforementioned artificial intelligence-based radioactive waste treatment scheme generation device may further include a processor and a memory. The aforementioned historical data module 201, feature extraction module 202, model adjustment module 203, and scheme generation module 204 are all stored in the memory as program modules, and the processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.
[0054] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.
[0055] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program runs, it controls the device containing the non-volatile storage medium to execute any of the aforementioned artificial intelligence-based radioactive waste treatment scheme generation methods.
[0056] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.
[0057] Optionally, during program execution, the device containing the non-volatile storage medium may be controlled to perform the following functions: acquire historical experimental data and historical treatment schemes for historical radioactive waste, and construct an initial database; based on the initial database, extract feature data from the historical experimental data, and based on the feature data, search for the same feature points in historical treatment schemes that have the same feature data, and construct an index database of feature data and treatment scheme feature points; based on the index database, fine-tune the large language model, wherein the fine-tuned large language model records the features of the index database, and when the current experimental data input to the fine-tuned large language model matches multiple historical experimental data, multiple historical treatment schemes are selected to construct candidate treatment schemes; and the target treatment scheme is determined based on the matching degree of the candidate treatment schemes.
[0058] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described methods for generating an artificial intelligence-based radioactive waste treatment scheme.
[0059] According to an embodiment of this application, an embodiment of a computer program product is also provided. Optionally, in this embodiment, the computer program product includes a computer program that, when executed by a processor, implements the steps of any of the above-described artificial intelligence-based radioactive waste treatment scheme generation methods.
[0060] Optionally, when the aforementioned computer program product is executed on a data processing device, it is suitable to execute an initialization program with the following method steps: acquiring historical experimental data and historical treatment schemes for historical radioactive waste, and constructing an initial database; based on the initial database, extracting feature data from the historical experimental data, and based on the feature data, retrieving the same feature points from historical treatment schemes that have the same feature data, and constructing an index database of feature data and treatment scheme feature points; based on the index database, fine-tuning a large language model, wherein the fine-tuned large language model records the features of the index database, and when the current experimental data input to the fine-tuned large language model matches multiple historical experimental data, filtering multiple historical treatment schemes and constructing candidate treatment schemes; and determining the target treatment scheme based on the matching degree of the candidate treatment schemes.
[0061] This invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring historical experimental data and historical treatment schemes for historical radioactive waste, and constructing an initial database; based on the initial database, extracting feature data from the historical experimental data, and based on the feature data, searching for identical feature points in historical treatment schemes that correspond to the same feature data, and constructing an index database of feature data and feature points of treatment schemes; based on the index database, fine-tuning a large language model, wherein the fine-tuned large language model records the features of the index database, and when the current experimental data input to the fine-tuned large language model matches multiple historical experimental data, filtering multiple historical treatment schemes and constructing candidate treatment schemes; and determining the target treatment scheme based on the matching degree of the candidate treatment schemes.
[0062] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.
[0063] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0064] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.
[0065] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0066] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.
[0067] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0068] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for generating radioactive waste treatment schemes based on artificial intelligence, characterized in that, include: Acquire historical experimental data and historical treatment plans for radioactive waste, and construct an initial database; Based on the initial database, feature data is extracted from the historical experimental data, and based on the feature data, the same feature points are retrieved in the historical processing schemes that have the same feature data, and an index database of feature data and processing scheme feature points is constructed. Based on the index database, the large language model is fine-tuned, wherein the fine-tuned large language model records the features of the index database. When the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened to construct candidate processing schemes. Based on the matching degree of the candidate processing schemes, the target processing scheme is determined.
2. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 1, characterized in that, The historical experimental data includes various types of data, including at least: the name of the object being treated, the nature of the object being treated, experimental nodes in the treatment process, feedback from the treatment process, and the status of secondary waste generation; The historical processing scheme includes various types of data, including at least: the name of the processing technology, the conditions of the processing technology, the environment of the processing technology, the non-compliance of the processed object, and the advantages of the technology.
3. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 2, characterized in that, Based on the initial database, feature data is extracted from the historical experimental data. Then, based on this feature data, identical feature points are retrieved from historical processing schemes that correspond to the same feature data. An index database of feature data and processing scheme feature points is constructed, including: The initial database is preprocessed by filtering historical experimental data that lacks processing object names or processing experimental nodes to obtain a preprocessed database. Based on the preprocessed database, features of each type of data in the historical experimental data are extracted to obtain multiple features corresponding to the historical experimental data and construct feature data. Based on the same features corresponding to the same type of data in multiple feature data, feature points in historical processing schemes corresponding to historical experimental data with the same features are selected, and the feature points are matched with the features and data types in the feature data to obtain an index database.
4. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 3, characterized in that, Based on the aforementioned index database, the large language model is fine-tuned, including: Perform field feature adaptation on the feature data and feature points in the index database; The low-rank adaptation technique is used to freeze the weights in the large language model, and the low-rank matrix of the large language model is trained based on the field features; Adjust the learning rate, training epochs, and loss function of the large language model to output the fine-tuned large language model.
5. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 4, characterized in that, When the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are filtered to construct candidate processing schemes, including: Input the current experimental data into the finely tuned large language model; The fine-tuned large language model extracts current features of various data from the current experimental data; Based on multiple current features, multiple historical experimental data are matched to correspond to multiple historical processing schemes; Based on the aforementioned historical processing schemes, candidate processing schemes are constructed.
6. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 5, characterized in that, The determination of the target processing scheme based on the matching degree of the candidate processing schemes includes: Matching multiple current features with various feature data in the historical experimental data; Based on feature matching weights, the matching degree between the current feature and multiple historical experimental data is calculated, wherein the feature matching weight of the processing object name is the highest; Sort the historical experimental data by matching degree from highest to lowest, and determine the historical processing scheme corresponding to the historical experimental data with the highest matching degree as the target processing scheme.
7. The method for generating a radioactive waste treatment scheme based on artificial intelligence according to claim 1, characterized in that, After determining the target processing scheme based on the matching degree of the candidate processing schemes, the method further includes: Extract the feature data from the current experimental data and the feature points from the corresponding target processing scheme to construct the current feature set; The current feature set is placed into the index database, and the large language model is fine-tuned in real time.
8. A device for generating radioactive waste treatment solutions based on artificial intelligence, characterized in that, include: The historical data module is used to acquire historical experimental data and historical treatment plans for historical radioactive waste and to build an initial database. The feature extraction module is used to extract feature data from the historical experimental data based on the initial database, and to retrieve the same feature points in the historical processing schemes that have the same feature data based on the feature data, and to construct an index database of feature data and feature points of processing schemes. The model adjustment module is used to fine-tune the large language model based on the index database. The fine-tuned large language model records the features of the index database. When the current experimental data input into the fine-tuned large language model is matched with multiple historical experimental data, multiple historical processing schemes are screened to construct candidate processing schemes. The scheme generation module is used to determine the target processing scheme based on the matching degree of the candidate processing schemes.
9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by an artificial intelligence-based method for generating a radioactive waste treatment scheme as described in any one of claims 1 to 7.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for generating a radioactive waste treatment scheme based on artificial intelligence as described in any one of claims 1 to 7.