Multi-objective decision support system and method used in embedded environment

Abstract

The invention relates to the technical field of decision support systems, in particular to a multi-target decision support system and method used in an embedded environment, sensing data and user feedback are received through an input interface module, and weights of targets such as repair time, quality and resource consumption are dynamically adjusted by a weight adjustment module based on a machine learning algorithm, so that the decision support effect is improved. The scheme optimization module converts a multi-target problem into single target optimization through weighting and scaling functions by using the adjusted weight, generates a unique Pareto optimal solution, and outputs the solution in real time in a visual instruction or machine executable command form through the output module; the method comprises the steps of data collection, dynamic weight adjustment, scheme optimization, scheme output and feedback learning. According to the method, the problems of poor adaptability and complex calculation of the traditional multi-target decision are solved, the multi-target priority can be balanced in real time in resource-limited scenes such as battlefield repair and industrial maintenance, and the method has the advantages of low delay, lightweight calculation and adaptive adjustment.

Description

Technical Field

[0001] This invention belongs to the field of decision support system technology, specifically relating to a multi-objective decision support system and method for embedded environments. Background Technology

[0002] In extreme environments, the maintenance and repair of critical equipment must be carried out rapidly and efficiently under harsh conditions and limited resources. On-site decision-makers must balance several conflicting objectives during maintenance, such as minimizing repair time, maximizing repair quality, and minimizing resource consumption (e.g., spare parts, tools, energy). These objectives are often contradictory: for example, the fastest repair may use more resources or produce lower quality, while a high-quality repair may take longer or be more resource-intensive.

[0003] Traditional multi-objective decision support methods typically rely on pre-fixed objective weights or require human experts to evaluate a range of possible solutions. In static methods, the relative importance of time, quality, and resource usage is predetermined and unchanging, lacking flexibility in dynamic battlefield conditions. Other multi-objective optimization methods may generate a whole set of Pareto optimal solutions (multiple trade-off options), from which the user must manually choose. These traditional methods are unsuitable for real-time embedded environments: they either lack adaptability or impose a heavy computational and cognitive burden on the user (e.g., evaluating multiple alternatives). Therefore, there is a need for an intelligent decision support system that can automatically adapt to constantly changing operational priorities and conditions, quickly providing a single, clear recommendation with minimal computational overhead.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] To address the aforementioned technical problems in existing technologies, this invention provides a multi-objective decision support system and method for embedded environments. It solves the significant shortcomings of traditional multi-objective decision support methods in embedded real-time environments: static fixed weights cannot dynamically adapt to changes in battlefield conditions, resource constraints, or task priority adjustments; generating multiple Pareto optimal solutions requires manual evaluation and selection by the user, leading to low decision-making efficiency and heavy cognitive burden; complex algorithms are difficult to run on resource-constrained embedded hardware, and computation time cannot meet real-time requirements; and insufficient handling of data uncertainty results in unreasonable weight allocation between high-confidence and low-confidence objectives.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: In a first aspect, an intelligent multi-objective decision support system for embedded environments is characterized by comprising: Input interface module: Used to receive data from sensing devices and feedback from operators; Weight adjustment module: used to adjust the weights of multiple decision objectives based on the data information and the feedback information using machine learning algorithms; Solution optimization module: used to evaluate potential remediation actions with weights adjusted by the weight adjustment module, transform the multi-objective problem into a single composite objective optimization problem through weighted and scalar functions based on the evaluation results, and output a single solution as the Pareto optimal trade-off; Output module: Used to convert the single solution into executable instructions or a visual form and then output it.

[0007] Furthermore, the weight adjustment module uses a dynamic weighted average algorithm to dynamically adjust the weights based on changes in sensor data or operator instructions, thereby gradually adjusting the target priority.

[0008] Furthermore, the weight adjustment module reduces the weight of targets with high uncertainty in the prediction results through an uncertainty-weighted algorithm.

[0009] Furthermore, the scheme optimization module assigns a composite score to each option using a weighted and scalar quantization function. The formula for calculating the assigned composite score is as follows:

[0010] in, For repair time, For the sake of repair quality, To repair resource consumption, As time weight, For quality weight, This refers to resource weights.

[0011] Furthermore, the input interface module includes: Sensor interface unit: used for real-time acquisition of equipment diagnostic data, environmental data, and resource availability data; User feedback interface unit: used to receive operator evaluations of the repair results and optimize weight adjustment strategies.

[0012] Furthermore, the scheme optimization module and the weighting module work together in real time. When the weighting module adjusts the weights based on sensor data or user feedback, the scheme optimization module immediately recalculates the optimal solution. Furthermore, the output module supports human-computer interaction interfaces or machine-executable instructions.

[0013] Secondly, an intelligent multi-objective decision support method for embedded environments includes: S1. Data Collection: Receive data information from sensing devices and feedback information from operators; S2. Dynamic weight adjustment: Based on machine learning algorithms, the weights of multiple repair decision objectives are adjusted according to the data information and feedback information. S3. Solution Optimization: Based on the adjusted weights, the multi-objective problem is transformed into a single-objective optimization through weighted and scalarization methods, generating a unique Pareto optimal solution. S4. Solution Output: Output the solution in real time to guide remediation decisions.

[0014] Furthermore, the dynamic weight adjustment step includes: A dynamic weighted average algorithm can be used to gradually adjust the priority of targets based on real-time feedback; alternatively, an uncertainty-based weighted algorithm can be used to reduce the weight of uncertain targets.

[0015] Furthermore, it also includes feedback learning steps: Collect repair results data and user feedback, update the machine learning model, and optimize subsequent weight adjustment strategies.

[0016] Compared with the prior art, the multi-objective decision support system and method for embedded environments provided by the present invention collects sensor data and user feedback in real time through an input interface module. The weight adjustment module dynamically adjusts the weights of multiple objectives such as repair time, quality, and resource consumption based on machine learning algorithms such as dynamic weighted averaging and uncertainty weighting. The scheme optimization module uses the adjusted weights to transform the multi-objective problem into a single composite objective optimization through weighting and scalarization functions, generating a unique Pareto optimal solution, and outputs it in real time in the form of visual instructions or machine-executable commands through an output module. Attached Figure Description

[0017] Figure 1 This is an architecture diagram of the intelligent multi-objective decision support system provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the intelligent multi-objective decision support method provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the weight adjustment module provided in Embodiment 1 of the present invention; Figure 4 This is a graphical diagram illustrating the weight identification of a single Pareto optimal solution by the scheme optimization module provided in Embodiment 1 of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0019] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0020] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0021] Example 1 See Figure 1 , Figure 1 This is a structural diagram of an intelligent multi-objective decision support system for embedded environments proposed in this invention, which may specifically include: A1. Input Interface Module: Used to receive data from sensing devices and feedback from operators; specifically includes: A11. Sensor Interface Unit: Responsible for collecting real-time data from automated sensing devices related to repair decisions. Specifically, it covers sensor types including: diagnostic sensors that provide key parameters such as the degree of equipment damage and system health status; environmental sensors that monitor factors affecting the repair environment, such as temperature and weather conditions; and resource sensors that track resource status such as the quantity of available spare parts, tool status, and power reserves. Analog signals or physical quantities collected by various sensors are digitally processed in real-time through the sensor interface and then transmitted to the processing unit for subsequent analysis.

[0022] A12. User Feedback Unit: An input / output component supporting human-computer interaction, primarily for maintenance technicians, commanders, and other human operators to interact with the system. Operators can input two types of key information through the component: priority adjustment instructions, such as explicitly requesting the system to prioritize repair speed over quality in the current task; and historical repair result evaluation, such as quantitative or qualitative evaluation of the actual effectiveness of a particular repair action. Actual effectiveness includes: repair time, resource consumption, and the degree of equipment performance recovery.

[0023] The user feedback unit supports diverse terminal forms, including portable handheld devices, tablets, and other mobile terminals, as well as direct integration into the control systems of vehicles and equipment to achieve seamless interaction in embedded scenarios.

[0024] A2. Weight Adjustment Module: Used to adjust the weights of multiple decision objectives based on machine learning algorithms, according to the data information and feedback information; it can run a dynamic weighted average algorithm to gradually transfer weights to the objectives that need to be emphasized, or use an uncertainty-based weighted algorithm to reduce the weights of uncertain objectives, and allocate weights according to the confidence level of the objective measurement to ensure that the system optimality standard is consistent with the current operation objectives and conditions.

[0025] A21. The weight adjustment module uses a Dynamic Weighted Average (DWA) algorithm to dynamically adjust weights based on changes in sensor data or operator instructions, thereby gradually adjusting target priorities. For example... Figure 3 How does the decision support system shown dynamically adjust the target weights over time? Figure 3 The solid line represents time weight, the dashed line represents quality weight, and the dotted line represents resource weight. The weight evolution of the system at four key time points (t1 to t4) is as follows: initial state The system assigns equal weight to the three objectives of time, quality, and resources, reflecting a balanced decision-making strategy. User intervention (t2): When the operator inputs the command "prioritize speed", the system immediately increases the time weight. At the same time, reduce quality proportionally and resource weight ; Sensor feedback (t3): When sensor data shows insufficient resources, the system automatically increases the resource weight. And fine-tune the time weights. To balance speed and resource consumption; Task completed The system recalibrates the weights based on task feedback data to restore a balanced state in order to meet subsequent decision-making needs.

[0026] A22. The weight adjustment module uses an uncertainty-weighted (UW) algorithm to reduce the weight of targets with high uncertainty in the prediction results. For example... Figure 3 As shown, it also includes an uncertainty-based weight adjustment (UW) algorithm. A semi-transparent region surrounding each weight represents the range of uncertainty, with a wider region indicating higher uncertainty in the target outcome prediction. The system automatically adjusts the weights based on these uncertainty levels, giving greater influence to more reliable targets and enhancing decision robustness.

[0027] A3. Solution Optimization Module: This module evaluates potential remediation actions using weights adjusted by the weight adjustment module. Based on the evaluation results, it transforms the multi-objective problem into a single composite objective optimization problem using a weighted sum scalarization function, and outputs a single solution as a Pareto optimal tradeoff. The multi-objectives include time, quality, and resources. For example, it calculates a weighted sum score for each candidate remediation operation, or uses other suitable aggregations such as utility functions. The solution optimization module assigns a composite score to each option using a weighted sum scalarization function. The formula for calculating the assigned composite score is:

[0028] in, For repair time, For the sake of repair quality, To repair resource consumption, As time weight, For quality weight, Resource weight The optimization module uses the adjusted weights output by the weighting module to evaluate possible remediation strategies or actions. This is achieved by effectively transforming the multi-objective problem into a single composite objective, such as calculating a weighted sum score for each candidate remediation operation, or employing other suitable aggregation methods like utility functions. Since the problem has been scalarized with weights, the optimization module can efficiently search for the optimal solution using algorithms or heuristics. Traditional search or optimization techniques such as linear programming, heuristic best-first search, and genetic algorithms with a single fitness function can all be applied. The module's output is the single best solution (remediation plan or decision) determined under the current weighting criteria. This solution corresponds to a Pareto optimality under the given weights, meaning that under this weighting configuration, no other available solution can improve one objective without worsening another.

[0029] in addition, Figure 4 This diagram illustrates how the system identifies a single optimal solution on the Pareto front based on dynamically adjusted weights. The curve represents the Pareto front, showing all possible optimal trade-offs between maintenance time (T) and maintenance quality (Q). Points A, B, C, and D on the front each represent a Pareto optimal solution, implying that one objective cannot be improved without sacrificing another. The contour lines represent the system using a weighted objective function that evaluates these solutions.

[0030] Weight vector ( =( , The slope of these contour lines is determined by the weights at point B. Initially, point B might yield a solution, but as the weights are adjusted (as indicated by the arrow), the optimal solution moves along the Pareto front to a new location, point C. Point C is marked as the optimal solution under the current weights; it lies at the point of tangency between the Pareto front and the contour lines reflecting the weighted objective function vector. Single-point selection simplifies the decision-making process, eliminating the need for the operator to evaluate multiple possible solutions. This illustrates how different weight configurations lead the system to select different Pareto optimal solutions, highlighting the importance of dynamic weight adjustments.

[0031] A4. Output module: Used to convert the single solution into executable instructions or a visual form and then output it.

[0032] Once the solution optimization module completes its decision, the output module is responsible for formatting the recommended solution and delivering it to the user or other systems in a timely and clear manner. The output module is a crucial link in the system's transformation of abstract decisions into concrete execution instructions. Through multimodal output and real-time response mechanisms, it ensures the efficient implementation of decision results, making it suitable for multi-objective optimization scenarios with stringent requirements for timeliness and accuracy. Specifically, it includes: A41. Visual Display: Present textual instructions to the user through the screen, such as: Use method Y to replace component X, estimated time 5 minutes, use 2 spare parts; the content includes operation method, object, time estimate and resource requirements, and key information is presented intuitively.

[0033] A42. Audio Alarm: Conveys information in the form of sound, suitable for emergency scenarios or situations requiring rapid user response; such as task priority changes and resource criticality alerts.

[0034] A43. Equipment control commands: Send execution commands to semi-autonomous repair robots or automated systems to directly trigger repair actions; for example, the robot performs component replacement operations according to the command.

[0035] After all components are integrated, the system can operate autonomously in real time. In actual implementation, the sensor and user interface modules provide a continuous data stream to the processing unit. The processing unit updates the weights and calculates decisions by running event-driven or continuous loop modes. The output is refreshed whenever a new decision is generated. The hardware and software are optimized for low latency so that the system can generate updated suggestions with minimal latency from receiving new data or feedback.

[0036] Example 2 See Figure 2 , Figure 2 This is a flowchart illustrating an intelligent multi-objective decision support method for embedded environments proposed in this invention. Specific steps may include: S1. Data Collection: Receive data information from sensing devices and feedback information from operators; The system continuously or periodically collects input data through sensor interface modules and user feedback interfaces. The collected data includes real-time readings from relevant sensors, such as the status of the equipment being repaired, environmental conditions, and available resources, as well as input information provided by human operators, such as indications of changes in task urgency or feedback on the success of the last repair operation. All data is timestamped and preprocessed according to the needs of decision-making calculations, such as filtering sensor noise and standardizing the data, to provide effective input for subsequent decision analysis.

[0037] S2. Dynamic Weight Adjustment: Based on machine learning algorithms, the weights of multiple repair decision objectives are adjusted according to the data and feedback information. After receiving new data, the system uses machine learning algorithms to drive the weighting module to update the weights of each decision objective. The core algorithm includes: Dynamic Weighted Average (DWA) algorithm: The system dynamically adjusts weight allocation based on operator feedback. For example, when an operator emphasizes speed requirements, the system gradually increases the weight of the repair time target. At the same time, the weight of the repair quality target is reduced accordingly. Or resource consumption target weight Maintain total weight normalization.

[0038] Uncertainty-based weighted (UW) algorithm: The system evaluates the uncertainty of the prediction results for each target and automatically reduces the weight of targets with high uncertainty. This uncertainty includes the difficulty in accurately estimating repair time due to unknown equipment damage.

[0039] After the above processing, a set of weights that are updated in real time is generated. The weight vector reflects both the current operation priority and the reliability of available information for each objective. The weights can be dynamically changed according to the actual situation during different decision cycles, achieving adaptive adjustment to the real-time scenario.

[0040] S3. Solution Optimization: Based on the adjusted weights, the multi-objective problem is transformed into a single-objective optimization through weighted and scalarization methods, generating a unique Pareto optimal solution.

[0041] Given adjusted weights, optimization module 34 evaluates various possible repair actions to determine the optimal solution. In an embedded repair support environment, possible actions include different repair strategies, such as patching versus complete component replacement, using rapid repair materials and standard parts, or resource allocation schemes such as assigning one expert and multiple technicians to perform the task. Each option produces specific results in terms of time, quality, and resources. The optimization module assigns a composite score to each option using a weighted scalar function:

[0042] in, For repair time, For the sake of repair quality, To repair resource consumption, As time weight, For quality weight, For resource weights. The index is appropriately normalized or scaled and signed to ensure that all metrics are consistently maximized or minimized; for example, repair time is minimized when used as a cost metric, and quality is maximized when used as a utility metric.

[0043] The system then searches for the optimal score. The options are as follows. Since the weighting process has already incorporated the current decision-maker's preferences and contextual factors such as task urgency and resource availability into the weights, optimizing this single objective... It can effectively generate Pareto optimal solutions adapted to the current scenario. The search process is designed to run efficiently to adapt to embedded environments: if the decision space is small, it can be solved directly through simple deterministic calculations; if more complex scenarios need to be handled, a lightweight optimization algorithm is used. In addition, the system supports maintaining a small library of heuristic rules or pre-computed options for typical scenarios, and quickly selecting the optimal solution through weighted scoring to avoid complex calculations and ensure real-time decision-making on resource-constrained embedded hardware.

[0044] S4. Solution Output: Output the solution in real time to guide remediation decisions.

[0045] Once step S3 determines the optimal repair solution, the system immediately outputs a decision recommendation via its output module. The recommended solution is presented in a user-understandable format, possibly including key principles if necessary; for example, the system might display: "Recommended Action: Perform a rapid repair on the hydraulic lines. Rationale: Given the current combat situation, this is the fastest option, using the fewest critical resources, and is expected to restore 80% functionality." The operator can then choose to follow this guidance, or, in an autonomous system, the system itself will initiate the repair action under the control of the automatic repair mechanism. By providing a single, clear solution, the system significantly reduces the user's cognitive burden and accelerates the decision-making process.

[0046] S5, Feedback Learning: Used to collect repair result data and user feedback, and update the machine learning model to optimize subsequent weight adjustment strategies.

[0047] Whether performed manually or automatically, the system evaluates the results after the recommended repair actions are executed. Sensors may provide data on the actual time spent on the repair, the performance of the repair equipment, and the resources consumed. Users may also provide feedback on whether the solution is satisfactory or whether any objectives have not been adequately met. This information is fed back into the system to update its internal models. Machine learning components can use this data to adjust future behavior, including updating any outcome prediction models or refining how weighted algorithms interpret similar feedback in the future. Over time, the learning process makes the system more accurate and consistent with the specific preferences of its users or the typical conditions of its deployment environment. The next time step 100 is triggered, the system incorporates this learned knowledge, thus closing a loop of continuous improvement.

[0048] In summary, the decision support system of this invention achieves a balance between multiple objectives in a constantly changing environment. Importantly, it locks in a single optimal solution in each iteration, rather than generating a series of choices. This single-solution approach, guided by dynamic weighting, ensures rapid convergence of decisions. Since the system does not spend time presenting or analyzing dominated solutions, it can meet the stringent time constraints of battlefield operations. Furthermore, the simplicity of weighting and computation means the algorithm can run rapidly on embedded hardware. Real-time adaptation is evident in steps 100, 200, and 500: the system incorporates information promptly as new data arrives or results are observed. This contrasts with static decision support tools, which may require manual reconfiguration or cannot be adjusted in a timely manner.

[0049] Therefore, the present invention has the following advantages: 1. Through dynamic weight adjustment, the system can respond to environmental changes in milliseconds and continuously output the optimal solution, thereby improving real-time adaptability.

[0050] 2. By using a single solution generation mechanism, information overload is avoided, and decision-making efficiency can be guaranteed even in high-pressure scenarios.

[0051] 3. Lightweight algorithm design supports embedded deployment, eliminating the need for high-performance processors and broadening application scenarios.

[0052] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent multi-objective decision support system for embedded environments, characterized in that, include: Input interface module: Used to receive data from sensing devices and feedback from operators; Weight adjustment module: used to adjust the weights of multiple decision objectives based on the data information and the feedback information using machine learning algorithms; Solution optimization module: used to evaluate potential remediation actions with weights adjusted by the weight adjustment module, transform the multi-objective problem into a single composite objective optimization problem through weighted and scalar functions based on the evaluation results, and output a single solution as the Pareto optimal trade-off; Output module: Used to convert the single solution into executable instructions or a visual form and then output it.

2. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The weight adjustment module uses a dynamic weighted average algorithm to dynamically adjust the weights based on changes in sensor data or operator instructions, thereby gradually adjusting the target priority.

3. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The weight adjustment module reduces the weight of targets with high uncertainty in the prediction results through an uncertainty-weighted algorithm.

4. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The scheme optimization module assigns a composite score to each option using a weighted and scalar quantization function. The formula for calculating the assigned composite score is as follows: in, For repair time, For the sake of repair quality, To repair resource consumption, As time weight, For quality weight, This refers to resource weights.

5. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The input interface module includes: Sensor interface unit: used for real-time acquisition of equipment diagnostic data, environmental data, and resource availability data; User feedback interface unit: used to receive operator evaluations of the repair results and optimize weight adjustment strategies.

6. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The scheme optimization module and the weighting module work together in real time. When the weighting module adjusts the weights based on sensor data or user feedback, the scheme optimization module immediately recalculates the optimal solution.

7. The multi-objective decision support system for embedded environments according to claim 1, characterized in that, The output module supports human-computer interaction interfaces or machine-executable instructions.

8. An intelligent multi-objective decision support method for embedded environments, characterized in that, include: S1. Data Collection: Receive data information from sensing devices and feedback information from operators; S2. Dynamic weight adjustment: Based on machine learning algorithms, the weights of multiple repair decision objectives are adjusted according to the data information and feedback information. S3. Solution Optimization: Based on the adjusted weights, the multi-objective problem is transformed into a single-objective optimization through weighted and scalarization methods, generating a unique Pareto optimal solution. S4. Solution Output: Output the solution in real time to guide remediation decisions.

9. The intelligent multi-objective decision support method for embedded environments according to claim 8, characterized in that, The dynamic weight adjustment step includes: A dynamic weighted average algorithm can be used to gradually adjust the priority of targets based on real-time feedback; alternatively, an uncertainty-based weighted algorithm can be used to reduce the weight of uncertain targets.

10. The intelligent multi-objective decision support method for embedded environments according to claim 8, characterized in that, It also includes feedback learning steps: Collect repair results data and user feedback, update the machine learning model, and optimize subsequent weight adjustment strategies.

Images