Intelligent decision-making method and device, electronic device, and readable storage medium
By obtaining decision knowledge information for new decision tasks and adjusting network parameters of existing decision models, the problem of poor accuracy of agents in new decision tasks is solved, and rapid generalization and high-accuracy decision-making effects are achieved.
Patent Information
- Application Number
- CN202410249830.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-05
- Publication Date
- 2025-02-11
- Estimated Expiration
- 2044-03-05
AI Technical Summary
Agents based on decision transformers have poor accuracy in new decision tasks and are difficult to generalize well.
By obtaining decision-making knowledge information for new decision tasks, determining corresponding model parameters, and using these parameters to adjust the network parameters of the decision model of known decision tasks, so that the decision model has knowledge of new decision tasks.
Improves the decision accuracy of the agent in new decision-making tasks, and achieves rapid generalization to new tasks without retraining the model.
Smart Images
Figure CN118153629B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure belongs to the field of intelligent control technology, and more specifically, relates to an intelligent decision-making method and device, an electronic device, and a readable storage medium. Background Art
[0002] Decision transformers have played a huge role in industrial control, autonomous driving, gaming and other fields. By expanding the model size and making full use of large offline data sets, the effectiveness of decision transformers has been further verified, making them a general agent that can efficiently solve various decision tasks. Although the decision transformer-based agent has high accuracy when making decisions on known tasks, it cannot generalize well to decision tasks outside the distribution, and has poor accuracy when performing new decision tasks. Summary of the invention
[0003] The purpose of the present disclosure is to provide an intelligent decision-making method and device, an electronic device, and a readable storage medium to improve the accuracy of decision-making of an intelligent agent in a new decision-making task.
[0004] A first aspect of the embodiments of the present disclosure provides an intelligent decision-making method, including:
[0005] Determining a model parameter corresponding to the first task based on the decision knowledge information of the first task;
[0006] The network parameters of the decision model are adjusted based on the model parameters; the decision model is a decision model for the second task; and the decision model after the network parameters are adjusted is used for task decision.
[0007] A second aspect of the embodiments of the present disclosure provides an intelligent decision-making device, including:
[0008] A parameter calculation module, used to determine the model parameters corresponding to the first task based on the decision knowledge information of the first task;
[0009] A parameter injection module is used to adjust the network parameters of the decision model based on the model parameters; the decision model is the decision model of the second task; the decision model after the network parameters are adjusted is used for task decision.
[0010] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned intelligent decision-making method when executing the computer program.
[0011] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned intelligent decision-making method are implemented.
[0012] The beneficial effects of the intelligent decision-making method and device, electronic device, and readable storage medium provided by the embodiments of the present disclosure are:
[0013] In the embodiment of the present disclosure, the first task is a new decision-making task, and the second task is a known decision-making task. When applying the decision model of the second task to the first task, the decision knowledge information of the first task is first obtained, and then the model parameters corresponding to the first task are determined according to the decision knowledge information of the first task, and then the network parameters of the second task are adjusted according to the model parameters corresponding to the first task. Through the above process, the knowledge of the new decision task is injected into the decision model, so that the decision model has the knowledge of the new decision task, and the accuracy of the decision-making of the decision model (agent) in the new decision task is improved. Compared with the traditional method of adding samples of new decision tasks and retraining decision models, the embodiment of the present disclosure does not require retraining of the model and can be quickly generalized to new tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0015] Figure 1 A flowchart of an intelligent decision-making method provided by an embodiment of the present disclosure;
[0016] Figure 2 A schematic diagram of a visual decision trajectory provided by an embodiment of the present disclosure;
[0017] Figure 3 A schematic diagram of text guidance information provided by an embodiment of the present disclosure;
[0018] Figure 4 A structural block diagram of an intelligent decision-making device provided in one embodiment of the present disclosure;
[0019] Figure 5 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0020] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.
[0021] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below in conjunction with the accompanying drawings.
[0022] Please refer to Figure 1 , Figure 1 A flowchart of an intelligent decision-making method provided in an embodiment of the present disclosure, the method comprising:
[0023] S101: Determine a model parameter corresponding to the first task based on decision knowledge information of the first task.
[0024] In this embodiment, the first task is a new decision-making task, and the decision-making knowledge information of the first task can be extracted from an offline data set of the first task.
[0025] Taking the autonomous driving scenario as an example, the offline dataset of autonomous driving includes: (1) Video data: usually shot by the vehicle camera, video data can provide rich visual information for perceiving roads, vehicles, pedestrians, traffic signs, etc. (2) Sensor data: including data collected by sensors such as radar, LiDAR and GPS. (3) Vehicle status data: including vehicle speed, acceleration, steering angle, braking status, etc. (4) Annotation data: Annotation data usually annotates objects in the video frame, such as lane lines, vehicles, pedestrians, traffic signs, etc. (5) Road and map data: including high-precision maps, road network structure, traffic signals and other data.
[0026] Correspondingly, the autonomous driving scenarios include detection tasks for pedestrians, vehicles, traffic lights, traffic signs, detection tasks for lane lines and drivable areas, path planning tasks, vehicle control tasks (including vehicle steering, acceleration, braking) and many other tasks.
[0027] If the first task is a path planning task, a video of the autonomous driving scene can be obtained from the offline data set, and task guidance information, action information, and element information related to the first task can be extracted from the video as the decision knowledge information of the first task. Among them, the task guidance information is the subtasks of the first task described in natural language, such as avoiding vehicles, entering lanes, avoiding pedestrians, etc.; the action information is the specific actions that should be taken at present, such as acceleration, steering, deceleration, etc.; the element information includes the positions of other vehicles, lanes, pedestrians, etc.
[0028] The decision knowledge information of the first task is used as a decision guide to determine model parameters suitable for the first task.
[0029] S102: adjusting network parameters of a decision model based on model parameters; the decision model is a decision model for a second task; and the decision model after the network parameters are adjusted is used for task decision making.
[0030] In this embodiment, the second task is a known decision-making task. The decision model of the second task is adjusted according to the model parameters of the first task so that the decision model has the knowledge of the first task, thereby improving the accuracy of the decision model in the first task.
[0031] The adjusted decision model can be used for the task decision of the first task. Specifically, the historical decision trajectory before the current action is input into the adjusted decision model, and the decision result of the current action can be output. The historical decision trajectory before the current action includes the state and action sequence before the current action.
[0032] It can be concluded from the above that when the decision model of the second task is applied to the first task, the present embodiment first obtains the decision knowledge information of the first task, then determines the model parameters corresponding to the first task according to the decision knowledge information of the first task, and then adjusts the network parameters of the second task according to the model parameters corresponding to the first task. Through the above process, the knowledge of the new decision task is injected into the decision model, so that the decision model has the knowledge of the new decision task, and improves the accuracy of the decision model (agent) in the new decision task. Compared with the traditional method of adding samples of new decision tasks and retraining decision models, the disclosed embodiment does not need to retrain the model and can be quickly generalized to new tasks.
[0033] In one embodiment of the present disclosure, the intelligent decision-making method further includes:
[0034] Get the mission video for the first mission.
[0035] The visual decision trajectory and text guidance information of the first task are obtained based on the task video.
[0036] The visual decision trajectory and the guidance information of this paper are determined as the decision knowledge information of the first task.
[0037] In this embodiment, an embodiment of extracting decision knowledge information of the first task from an offline database is given. Specifically, firstly, a task video of the first task is obtained from the offline database; then, multiple frames of images in the task video are extracted, and the multiple frames of images in the task video are arranged in chronological order to obtain the visual decision trajectory of the first task; summary information is extracted from the visual decision trajectory using a large visual language model (such as GPT-4V), and the summary information is determined as the task guidance information of the first task; and element information and action guidance information in the task video that are manually annotated or automatically extracted are received to provide rich and fine-grained semantic guidance.
[0038] Among them, the visual decision trajectory is image information, such as Figure 2 As shown in , it is a schematic diagram of the visual decision trajectory. Task guidance information, element information and action guidance information are text guidance information (natural language guidance information), such as Figure 3 The figure shows a schematic diagram of natural language guidance information. Combining the visual decision trajectory and text guidance information as multimodal decision knowledge information enables the decision model to better acquire the relevant domain knowledge of the first task, thereby better understanding the first task and achieving generalization of the first task.
[0039] In one embodiment of the present disclosure, determining a model parameter corresponding to the first task based on decision knowledge information of the first task includes:
[0040] Feature extraction is performed on decision-making knowledge information to obtain decision features.
[0041] The decision features are input into the hypernetwork model to obtain the model parameters.
[0042] In this embodiment, the model parameters of the decision model are generated by a hypernetwork model. Specifically, the decision features are extracted from the decision knowledge information of the first task, and then the decision features are input into the hypernetwork model to obtain the model parameters suitable for the first task.
[0043] By using a hypernetwork model to generate model parameters, knowledge between different decision-making tasks can be shared and can be quickly generalized to new tasks.
[0044] In one embodiment of the present disclosure, feature extraction is performed on decision knowledge information to obtain decision features, including:
[0045] Feature extraction is performed based on the visual decision trajectory and text guidance information in the decision knowledge information of the multimodal pre-training model, and the extracted features are concatenated to obtain the decision features.
[0046] In this embodiment, the text guidance information and the visual decision trajectory can be aligned through a multimodal pre-trained model (such as a CLIP model) to obtain a decision feature containing multimodality and provide more fine-grained guidance.
[0047] In an embodiment of the present disclosure, there are multiple pieces of decision knowledge information, and determining the model parameters corresponding to the first task based on the decision knowledge information of the first task includes:
[0048] Feature extraction is performed on multiple pieces of decision-making knowledge information to obtain multiple decision features.
[0049] Multiple decision features are input into the hypernetwork model respectively to obtain multiple model parameters.
[0050] Multiple model parameters are fused to obtain fused model parameters.
[0051] The fused model parameters are determined as the model parameters corresponding to the first task.
[0052] In this embodiment, multiple task videos of the first task can be obtained, and multiple pieces of decision knowledge information of the first task can be obtained based on the multiple task videos. Multiple model parameters are obtained based on the multiple pieces of decision knowledge information, and then the multiple model parameters are fused, and the fused model parameters are determined as the model parameters corresponding to the first task. In this way, the features of multiple pieces of decision knowledge information can be integrated, which is conducive to obtaining accurate model parameters.
[0053] In one embodiment of the present disclosure, multiple model parameters are fused to obtain fused model parameters, including:
[0054] Feature extraction is performed on the historical decision trajectories in the preset sample set to obtain historical trajectory features.
[0055] The correlations between multiple decision features and historical trajectory features are calculated respectively to obtain multiple correlation coefficients.
[0056] Based on multiple correlation coefficients, multiple model parameters are weighted and summed to obtain fused model parameters.
[0057] In this embodiment, the convolutional neural network can be used to extract features from the second historical decision trajectory in the preset sample set to obtain historical trajectory features. Multiple correlation coefficients are obtained by calculating the correlation between multiple decision features and historical trajectory features, and multiple model parameters are weighted and summed based on the multiple correlation coefficients. The stronger the correlation between the decision feature and the historical trajectory feature, the greater the weight of the corresponding model parameter. In this way, the influence of multiple decision knowledge information on the model parameters can be continuously and dynamically adjusted according to the state of the intelligent agent, and the adaptability is stronger.
[0058] The correlation coefficient can be obtained by calculating the cosine similarity between the decision feature and the historical trajectory feature, or by calculating the Euclidean distance between the decision feature and the historical trajectory feature.
[0059] Taking the calculation of correlation coefficient based on cosine similarity as an example, if the number of decision features is n, they are
[0060] { guide_1, guide_2, …, guide_n}, the correlation coefficient between the i-th decision feature guide_i and the historical trajectory feature traj is:
[0061] ,
[0062] The n model parameters are weighted and summed based on the correlation coefficient, specifically:
[0063] ,
[0064] in, are the model parameters after fusion, is the i-th model parameter.
[0065] In one embodiment of the present disclosure, the network parameters are parameters corresponding to each attention module in the decision model.
[0066] In this embodiment, the decision model is specifically a Transformer decision model, which includes multiple attention blocks. The model parameters corresponding to the first task are used as parameters corresponding to each attention block, which is expressed as follows:
[0067] .
[0068] Since the model parameters generated by the decision knowledge information of different tasks have different activation degrees for each attention block, for example, task A tends to activate block_1, and task B tends to activate block_2, the network structure of the Transformer decision model is also different, making the Transformer decision model more sensitive to different tasks.
[0069] In one embodiment of the present disclosure, adjusting the network parameters of the decision model based on the model parameters includes:
[0070] Divide the model parameters into M equal parts.
[0071] The M equal parts of model parameters are respectively determined as the parameters of the M attention modules of the decision model to activate each attention module.
[0072] In this embodiment, the model parameters can be divided into M equal parts through the reshape operation. Where M is the number of attention modules in the decision model. Taking [p_1,…, p_n, p_n+1,…, p_2n,p2n+1,…,p_3n], M=3 as an example, the model parameters are divided into three parts: [p_1,…, p_n], [p_n+1,…, p_2n], and [p2n+1,…,p_3n], and then the parameters of the three parts are used as the parameters of the M attention modules respectively.
[0073] Corresponding to the intelligent decision-making method in the above embodiment, Figure 4 This is a block diagram of the structure of an intelligent decision-making device provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 4 The intelligent decision-making device 20 includes: a parameter calculation module 21 and a parameter injection module 22.
[0074] The parameter calculation module 21 is used to determine the model parameters corresponding to the first task based on the decision knowledge information of the first task.
[0075] The parameter injection module 22 is used to adjust the network parameters of the decision model based on the model parameters. The decision model is the decision model of the second task. The decision model after the network parameters are adjusted is used for task decision.
[0076] In one embodiment of the present disclosure, the parameter calculation module 21 is further used to:
[0077] Get the mission video for the first mission.
[0078] The visual decision trajectory and text guidance information of the first task are obtained based on the task video.
[0079] The visual decision trajectory and the guidance information of this paper are determined as the decision knowledge information of the first task.
[0080] In one embodiment of the present disclosure, the parameter calculation module 21 is specifically used for:
[0081] Feature extraction is performed on decision-making knowledge information to obtain decision features.
[0082] The decision features are input into the hypernetwork model to obtain the model parameters corresponding to the first task.
[0083] In one embodiment of the present disclosure, there are multiple pieces of decision-making knowledge information, and the parameter calculation module 21 is specifically used to:
[0084] Feature extraction is performed on multiple pieces of decision-making knowledge information to obtain multiple decision features.
[0085] Multiple decision features are input into the hypernetwork model respectively to obtain multiple model parameters.
[0086] Multiple model parameters are fused to obtain fused model parameters.
[0087] The fused model parameters are determined as the model parameters corresponding to the first task.
[0088] In one embodiment of the present disclosure, the parameter calculation module 21 is specifically used for:
[0089] Feature extraction is performed on the historical decision trajectories in the preset sample set to obtain historical trajectory features.
[0090] The correlations between multiple decision features and historical trajectory features are calculated respectively to obtain multiple correlation coefficients.
[0091] Based on multiple correlation coefficients, multiple model parameters are weighted and summed to obtain fused model parameters.
[0092] In one embodiment of the present disclosure, the network parameters are parameters corresponding to each attention module in the decision model.
[0093] In one embodiment of the present disclosure, the parameter injection module 22 is specifically used to:
[0094] Divide the model parameters into M equal parts.
[0095] The M equal parts of model parameters are respectively determined as the parameters of the M attention modules of the decision model to activate each attention module.
[0096] See also Figure 5 , Figure 5 Schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 5 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303 and one or more memories 304. The processors 301, input devices 302, output devices 303 and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of each module / unit in the above-mentioned device embodiments, such as Figure 4 The functions of modules 21 to 22 are shown.
[0097] It should be understood that in the embodiment of the present disclosure, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0098] The input device 302 may include a touch panel, a fingerprint sensor (for collecting fingerprint information of the user and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.
[0099] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.
[0100] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of the intelligent decision-making method provided in the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device described in the embodiments of the present disclosure, which will not be repeated here.
[0101] In another embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, wherein the program instructions are executed by a processor to implement all or part of the processes in the above-mentioned embodiment method, and may also be completed by instructing the relevant hardware through the computer program, wherein the computer program may be stored in a computer-readable storage medium, and wherein the computer program may be executed by the processor to implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the contents contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signal and telecommunication signal.
[0102] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.
[0103] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.
[0104] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0105] In the several embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or it can be an electrical, mechanical or other form of connection.
[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present disclosure.
[0107] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0108] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. An intelligent decision-making method, characterized in that: include: Get the mission video of the first mission; Obtaining a visual decision trajectory and text guidance information for a first task based on the task video; Determine the visual decision trajectory and the guidance information of this article as decision knowledge information for the first task; Determining a model parameter corresponding to the first task based on the decision knowledge information of the first task; Adjusting network parameters of the decision model based on the model parameters; The decision model is a decision model for the second task; The decision model after the network parameters are adjusted is used for task decision making; The determining the model parameters corresponding to the first task based on the decision knowledge information of the first task includes: Extracting features from the decision-making knowledge information to obtain decision-making features; Inputting the decision feature into a hypernetwork model to obtain model parameters corresponding to the first task; The decision-making knowledge information is a plurality of pieces, and determining the model parameters corresponding to the first task based on the decision-making knowledge information of the first task includes: Extracting features from the plurality of pieces of decision-making knowledge information to obtain a plurality of decision-making features; Inputting the plurality of decision features into the hypernetwork model respectively to obtain a plurality of model parameters; Fusing multiple model parameters to obtain fused model parameters; Determining the fused model parameters as the model parameters corresponding to the first task; The step of fusing multiple model parameters to obtain fused model parameters includes: Extract features of historical decision trajectories in a preset sample set to obtain historical trajectory features; Respectively calculating the correlations between the plurality of decision features and the historical trajectory features to obtain a plurality of correlation coefficients; Based on multiple correlation coefficients, multiple model parameters are weighted and summed to obtain fused model parameters.
2. The intelligent decision-making method according to claim 1, characterized in that: The network parameters are parameters corresponding to each attention module in the decision model.
3. The intelligent decision-making method according to claim 1, characterized in that: The adjusting the network parameters of the decision model based on the model parameters includes: Dividing the model parameters into M equal parts; The M equal parts of model parameters are respectively determined as the parameters of the M attention modules of the decision model to activate each attention module.
4. An intelligent decision-making device, characterized in that: include: Parameter calculation module for: Get the mission video of the first mission; Obtaining a visual decision trajectory and text guidance information for a first task based on the task video; Determine the visual decision trajectory and the guidance information of this article as decision knowledge information for the first task; Determining a model parameter corresponding to the first task based on the decision knowledge information of the first task; A parameter injection module, used to adjust the network parameters of the decision model based on the model parameters; the decision model is a decision model for the second task; the decision model after the network parameters are adjusted is used for task decision; The parameter calculation module is also specifically used for: Extract features from decision-making knowledge information to obtain decision-making features; Input the decision features into the hypernetwork model to obtain model parameters corresponding to the first task; There are multiple pieces of decision-making knowledge information, and the parameter calculation module is also used to: Extract features from multiple pieces of decision-making knowledge information to obtain multiple decision-making features; Inputting multiple decision features into the hypernetwork model respectively to obtain multiple model parameters; Fusing multiple model parameters to obtain fused model parameters; Determine the fused model parameters as the model parameters corresponding to the first task; The parameter calculation module is also specifically used for: Extract features of historical decision trajectories in a preset sample set to obtain historical trajectory features; The correlations between multiple decision features and historical trajectory features are calculated respectively to obtain multiple correlation coefficients; Based on multiple correlation coefficients, multiple model parameters are weighted and summed to obtain fused model parameters.
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.
6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.