A user end application method and system of a production process adaptive recommendation algorithm

Abstract

The application discloses a user terminal application method and system of a production process self-adaptive recommendation algorithm, and belongs to the technical field of production management and process optimization. The method acquires operation data, historical feedback data and production context data containing device states and work order progress in real time through a collection agent deployed at a user terminal in a production site; based on an incremental learning algorithm, the data are used to update a user preference model and a production mode model in real time; according to the updated model and real-time context, personalized production recommendations are generated through a combination of a rule engine and a context perception algorithm and pushed to the user for execution; user execution feedback data are collected and directly used as historical feedback data for the next cycle, driving the model to continuously iterate and optimize, thus forming a complete self-adaptive closed loop. The application solves the problems of insufficient personalization and inability to optimize in real time of existing production recommendation systems, and significantly improves the accuracy of recommendations and user satisfaction.

Description

Technical Field

[0001] This invention belongs to the field of production management and process optimization technology, and specifically relates to a user-end application method and system for an adaptive recommendation algorithm for production processes based on artificial intelligence technology, which is particularly suitable for industrial production environments that require personalized and real-time production recommendation services. Background Technology

[0002] With the digitalization and intelligentization of industrial production processes, optimizing production decisions using production process data has become an important trend. Traditional production recommendation systems mostly build models based on historical statistical data of production equipment or systems, generating recommendation strategies such as equipment maintenance suggestions and process optimization schemes through offline calculations. However, such systems have the following problems: Low degree of personalization: Recommendation models are usually built based on overall production data, without taking into account the differences in operating habits, skill levels and preferences of different users, resulting in a low degree of matching between recommendation results and users' actual needs; Poor real-time performance: Model updates rely on offline batch training, which cannot respond to real-time user feedback during the production process and is difficult to adapt to dynamically changing production scenarios. Lack of closed-loop optimization mechanism: There is no effective data feedback loop between the recommendation system and users, the system cannot continuously optimize itself based on user behavior, and the improvement of recommendation effect is limited.

[0003] Currently, there are production support decision-making systems based on streaming technologies such as Spark Streaming that can collect equipment data in real time and generate suggestions. However, their models are still mainly trained offline and do not achieve personalized modeling and real-time iterative optimization on the user side. Therefore, there is an urgent need for a technical solution that can perceive user behavior in real time, dynamically update the model, and provide personalized production recommendations. Summary of the Invention

[0004] This invention provides a user-side application method for an adaptive recommendation algorithm for production processes. It aims to solve the problems of insufficient personalization and inability to optimize in real time in existing production recommendation systems by constructing a user feedback-driven closed-loop adaptive system.

[0005] According to one aspect of the present invention, a user-end application method for an adaptive recommendation algorithm for a production process is provided, forming a continuously looping adaptive closed-loop process, wherein a single loop cycle of the process includes the following steps: By deploying a data acquisition agent at the user's end in the production site, the system can acquire real-time user operation data during the production process, as well as user feedback data on historical recommendations, and obtain production context data including equipment status and work order progress. Based on the incremental learning algorithm, the pre-built user preference model and production mode model are updated in real time using the operation data, historical feedback data and production context data to obtain the optimized model for the current period. Based on the current cycle's optimization model, combined with real-time acquired production context data, personalized production recommendation content is generated through a combination of a rule engine and a context-aware algorithm, and then pushed to the user for execution. The system collects user feedback data on the recommended content generated in the current cycle and uses this feedback data directly as historical feedback data for the next cycle to drive continuous iterative optimization of the user preference model and production mode model.

[0006] As a further technical solution, the model is updated in real time based on an incremental learning algorithm, including: Calculate the feedback difference ΔR based on the user's acceptance or rejection of historical recommended content; Based on the feedback difference ΔR and the preset adjustment rate β, using the formula The model parameter θ is adjusted incrementally, where θ represents the parameter of the user preference model or production mode model.

[0007] As a further technical solution, the generation of personalized production recommendation content includes: Constructing context-weighted functions Where u represents user identity information, s represents the current working mode status, and e represents the real-time device status; Using the aforementioned context weighting function The alternative recommendation strategies in the rules engine are weighted and scored, and the strategy with the highest comprehensive score is selected to generate the final recommended content.

[0008] As a further technical solution, when updating the production mode model, the random forest ensemble learning method is adopted. The equipment status parameters of the preceding process, the output quality indicators and the process interval time are used as input features to integrate the analysis results of multiple decision trees in order to model the correlation patterns between complex production processes.

[0009] As a further technical solution, the raw data obtained through the acquisition agent is standardized and converted using a predefined JSON data template to unify the heterogeneous data format from different production devices.

[0010] According to one aspect of the present invention, a client application system for an adaptive recommendation algorithm for production processes is provided, comprising: The data acquisition module is configured to: acquire user operation data during the production process and user feedback data on historical recommendations in real time through the acquisition agent deployed on the user terminal at the production site, and acquire production context data including equipment status and work order progress. The model optimization module is configured to: based on the incremental learning algorithm, use the operation data, historical feedback data and production context data to update the pre-built user preference model and production mode model in real time to obtain the optimized model for the current period. The decision execution module is configured to: generate personalized production recommendation content based on the current cycle's optimization model and real-time acquired production context data, using a combination of a rule engine and a context-aware algorithm, and push it to the user for execution; The user feedback loop module is configured to: collect user feedback data on the recommended content generated in the current cycle, and use this feedback data directly as historical feedback data for the next cycle, so as to drive the continuous iterative optimization of the user preference model and the production mode model.

[0011] As a further technical solution, the model optimization module is also configured as follows: Calculate the feedback difference ΔR based on the user's acceptance or rejection of historical recommended content; Based on the feedback difference ΔR and the preset adjustment rate β, using the formula The model parameter θ is adjusted incrementally, where θ represents the parameter of the user preference model or production mode model.

[0012] As a further technical solution, the decision execution module is also configured as follows: Constructing context-weighted functions Where u represents user identity information, s represents the current working mode status, and e represents the real-time device status; Using the aforementioned context weighting function The alternative recommendation strategies in the rules engine are weighted and scored, and the strategy with the highest comprehensive score is selected to generate the final recommended content.

[0013] According to one aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a client application method for a production process adaptive recommendation algorithm.

[0014] According to one aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a client-side application method for a production process adaptive recommendation algorithm.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention collects personalized production information, including operational data, feedback data, and production context data, in real time on the user's end, providing a data foundation for building a model that accurately adapts to user preferences. By applying incremental learning algorithms, the system continuously incorporates the latest user feedback, dynamically updating the user preference model and production mode model, thereby achieving real-time optimization and improvement of recommendation performance. Combining a rule engine and context-aware algorithm decision-making mechanism, it can generate accurate recommendations that meet users' real-time needs based on complex and dynamic production environments, overcoming the limitations of traditional offline batch processing methods. The constructed complete user feedback closed-loop system, relying on an incremental update process, enables the recommendation service to continuously improve itself and enhance its quality. The synergistic effect of the above-mentioned technical aspects gives this solution good universality and adaptability, allowing for flexible deployment in various industrial production scenarios, ultimately effectively improving user work efficiency and satisfaction. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a user-side application method for an adaptive recommendation algorithm for production processes, provided in an embodiment of the present invention.

[0017] Figure 2 This is a flowchart illustrating the implementation of a client-side application system for a production process adaptive recommendation algorithm, as provided in an embodiment of the present invention. Detailed Implementation

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

[0019] It should be noted that, unless otherwise specified, the terms such as methods, modules, and algorithms used in this article have their usual meanings in related technical fields such as artificial intelligence, production management, and software engineering.

[0020] Example 1

[0021] This embodiment provides a client-side application method for an adaptive recommendation algorithm for production processes. For example... Figure 1 As shown, this method forms a continuously looping adaptive closed-loop process, and a single cycle of this process includes the following steps:

[0022] S101: Real-time data acquisition.

[0023] By pre-developing and deploying a lightweight data acquisition agent at the user's site in the production environment, three types of core data can be acquired in real time:

[0024] Operational data refers to the actual operation records performed by the user during the production process, such as starting / stopping equipment, adjusting process parameters, selecting tools, and filling out work orders.

[0025] Historical feedback data refers to user feedback behavior and results regarding the system's recommendations throughout historical periods, such as acceptance, rejection, and modification. For example, data on actual production efficiency improvement after a user adopted a previous process optimization suggestion.

[0026] Production context data refers to information describing the current state of the production environment, mainly including equipment status (such as running, standby, faulty, current load) and work order progress (such as current process, remaining time, delivery urgency). This data is typically obtained from Manufacturing Execution System (MES), Supervisory Control and Data Acquisition (SCADA) system, or equipment controllers.

[0027] The data acquisition agent adopts a modular connection interface design and enables dynamic access and data acquisition with production equipment and systems (such as MES and SCADA) of different brands and protocols through a configurable driver adapter.

[0028] To handle heterogeneous data from production equipment of different brands and models, the acquisition agent uses a predefined JSON data template to standardize and transform the acquired raw data. For example, the encoding of device status in different device messages (such as "1", "RUN", "running") is uniformly mapped to a standardized JSON field {"device_status": "running"}.

[0029] Standardized data is encrypted and transmitted to the backend model optimization module via secure network transmission protocols (such as HTTPS or MQTTS based on TLS) to ensure the security and integrity of the data transmission process.

[0030] S102: Incremental Model Optimization and Update.

[0031] This step utilizes data collected and transmitted by the S101 to continuously optimize and update the two pre-built core models in real time. Both models are based on incremental learning algorithms to dynamically adjust parameters, enabling them to quickly adapt to changes based on user feedback data without requiring complete retraining.

[0032] The feedback quantization and incremental learning mechanism in this embodiment is as follows:

[0033] First, the feedback difference ΔR is calculated based on user behavior, such as acceptance, rejection, or the effectiveness of implementation of historical recommendations. For example, a positive ΔR corresponds to a user accepting a recommendation, while a negative ΔR corresponds to a user rejecting it or experiencing poor results after implementation. The specific value of ΔR can also be quantified based on the actual changes in key performance indicators (KPIs) after implementation, such as the percentage reduction in working hours or changes in the defect rate.

[0034] Next, based on the feedback difference ΔR and the preset adjustment rate β (i.e., the learning rate) ), through formula The model parameters θ are incrementally adjusted. During this incremental adjustment, the feedback difference ΔR essentially indicates the direction and magnitude of the model's prediction error, acting similarly to the gradient signal of the loss function with respect to the parameters θ. This guides the model parameters to be updated in a direction that reduces future prediction errors. Here, θ represents the parameter vector of the user preference model or production mode model. The adjustment rate β controls the step size of the parameter updates, and its value can be determined through empirical settings, scheduling strategies (such as decay), or adaptive optimizers (such as Adam).

[0035] Specifically, the user preference model is updated as follows:

[0036] The user preference model is used to learn and predict a specific user's preference for different types of production recommendations (such as quick operations, spare parts selection, and process paths). This model can be built using a deep learning framework (such as TensorFlow). In the initial training phase, the model is trained and validated using historical production data. A loss function (such as mean squared error, MSE) is used to evaluate model performance, and the model is fine-tuned based on the evaluation results.

[0037] During the application phase, the model adjusts its parameters in real time based on user feedback through the aforementioned incremental learning mechanism. Furthermore, the system can implement an active learning strategy. For samples with high model prediction uncertainty (e.g., in fault classification tasks where the model's prediction probabilities for each fault category are similar), it filters them using the uncertainty metric U(x) and actively requests user annotations, thereby efficiently obtaining high-quality feedback data to further optimize the model.

[0038] Specifically, the production model update is as follows:

[0039] Production pattern models are used to learn and model the complex relationships and patterns among processes, equipment, and parameters in the production process. These models can be constructed using ensemble learning methods (such as random forests). Specifically, multiple decision trees are constructed for ensemble modeling, using equipment status parameters of preceding processes, output quality indicators, and process intervals as input features.

[0040] To achieve real-time updates, the ensemble model employs variant algorithms that support incremental learning. For example, it uses an online random forest method, allowing new branches to grow or the tree structure to be adjusted based on newly arriving samples (user feedback and production data) on top of the existing decision trees; or it achieves incremental adaptation by dynamically adjusting the weights of different decision trees in the ensemble. Simultaneously, the system analyzes time-series data such as equipment operating parameters to extract features such as trends, seasonality, and periodicity. These features serve as important inputs, enhancing the model's understanding and predictive ability regarding dynamic production patterns.

[0041] Through the above mechanism, both the user preference model and the production mode model can continuously absorb new user feedback and production data in an incremental manner, realizing collaborative optimization and self-evolution of the models, and providing a real-time and adaptive model foundation for subsequent accurate recommendations.

[0042] S103: Intelligent recommendation decision-making and execution.

[0043] This step generates and executes personalized production recommendations based on the updated optimized model (user preference model and production mode model) in S102 and combined with real-time acquired production context data.

[0044] The decision execution module, as the core of the system, has the following functions:

[0045] 1. Dynamic Acquisition and Analysis of Context: The module receives and integrates production context information provided by the data acquisition module in real time, mainly including equipment status (such as real-time operating parameters such as temperature, pressure, and vibration) and work order progress (such as current process, remaining time, and delivery urgency). The module cleans, extracts features, and fuses this information to form a comprehensive understanding of the current production environment.

[0046] 2. Fusion decision-making based on policy template library and context awareness: The decision-making mechanism adopts a combination of rule engine and context awareness algorithm. Its core is a pre-built recommendation policy template library and corresponding matching application mechanism.

[0047] Recommendation Strategy Template Library: The system maintains a recommendation strategy template library covering different typical production scenarios (such as high-speed precision machining, equipment preheating, fault early warning and handling, etc.). Each template defines the recommendation objective, applicable conditions, optional operation set, and initial weights for that scenario.

[0048] Context-aware algorithm: Designing a context-weighted function Where u represents user identity information, s represents the current operating mode state (such as stable operation mode or debugging mode) obtained through production mode model analysis, and e represents the real-time device state. This function quantifies the combined impact of user preferences and production constraints in the current specific context.

[0049] Policy matching and recommendation function: Implement a matching mechanism to dynamically select the most suitable policy template. This process can be abstracted as the calculation of a recommendation function R(h,C), where h represents the user's current state and is a context-weighted function. The system uses a comprehensive vectorized representation of u (user identity), s (current working mode state), and e (real-time device state); C represents the context information of historical patterns, which consists of pattern features extracted from historical data by the production mode model. The function R selects a set of candidate templates from the template library by calculating the correlation between the current state h and the historical pattern C, as well as the matching degree between h and the applicable conditions of each strategy template.

[0050] Rule Engine and Weighted Decision: The rule engine, embedding domain knowledge and hard constraints (such as security procedures), is responsible for compliance verification and filtering of candidate policy templates output by the recommendation function R. Subsequently, a context-weighted function is used... The validated candidate strategies are then given a refined weighted score. This function quantifies the combined impact of user preferences and production constraints in the current specific context (user u, working mode s, device e), thereby selecting the strategy with the highest overall score.

[0051] 3. Generate and push precise recommendations: Based on the selected optimal strategy and real-time data to populate specific parameters, the system generates final, executable, personalized production recommendations (e.g., suggesting reducing the spindle speed by 10% to gear X and changing to tool B). The system also determines the optimal timing for the recommendation (e.g., before the next tool change or when equipment parameters are about to exceed limits), and finally pushes it to the field user for execution through channels such as the user interface (HMI) and mobile terminals.

[0052] For example, suppose the system policy template library contains a high-temperature load reduction operation template, whose applicable conditions include "equipment temperature > threshold T1". Currently, a certain processing equipment has a high temperature (e: high temperature) and is in batch production mode (s: batch mode), and the current user (u: operator A) has a historical preference for conservative operation. The recommendation function R(h,C) matches the high-temperature load reduction operation template as a candidate based on the current state h (which combines u, s, and e) and the historical pattern C (such as the processing pattern that was effective under similar high-temperature conditions in the past). After the rule engine performs compliance checks, the context weighting function... Further calculations suggest that the sub-strategy of reducing engine speed might be given a higher weight due to user A's conservative preferences. Based on this, the system generates a specific recommendation to "immediately reduce the engine speed by 15%" and highlights it on the device control panel for user confirmation and execution.

[0053] Through the above process, the decision execution module tightly integrates the adaptive model, strategy template library and real-time production environment. Through a systematic matching and weighting mechanism, it outputs accurate and timely recommendations that conform to personalized preferences and historical patterns, as well as meet real-time constraints and domain rules.

[0054] S104: Feedback Acquisition and Closed-Loop Iterative Optimization.

[0055] This step forms the key feedback loop of the adaptive closed-loop process, aiming to collect users' actual responses to the recommendations and drive continuous system optimization. This module specifically performs the following operations:

[0056] 1. Feedback Behavior Collection and Analysis: The system monitors and records users' response behavior to the recommended content pushed by S103 in real time. This includes explicit feedback (such as explicitly clicking "Accept," "Reject," or "Ignore") and implicit feedback (such as performing a related action after the recommendation prompt, or failing to take action within the suggested time). The system calculates the user feedback function R(a,b), where a is the number of acceptances and b is the number of rejections, to quantify the user's immediate acceptance rate. Simultaneously, it records richer feedback content, such as users' manual modifications to the recommendations, evaluations filled in after execution, or effect notes.

[0057] 2. Feedback Data Integration: The collected execution feedback data (including behavior type, timestamp, relevant context, and any user input) is standardized and formatted. Subsequently, this data is integrated into the incremental learning sample dataset of the model optimization module (S102) in real time or near real time, serving as a key input for the next round of model iteration, ensuring the timeliness, accuracy, and closed-loop nature of the training data.

[0058] 3. Feedback-driven model iterative adjustment: Upon the arrival of new feedback data, an iterative adjustment process is immediately triggered. As described in S102, the system applies the incremental learning formula based on the feedback difference ΔR (which can be calculated from acceptance / rejection behavior, changes in KPIs after execution, etc.) and the adjustment rate β. The parameters θ of the user preference model and the production mode model are updated. This process enables the models to quickly adapt to the latest user preferences and actual production results, thereby achieving continuous improvement in recommendation accuracy and user satisfaction.

[0059] 4. Service Quality Assessment and System Optimization: The system periodically analyzes the macroscopic results of the feedback loop to assess overall service quality and identify optimization directions. This is quantified using the service quality optimization function Q=f(R,θ), where R represents comprehensive user feedback data over a period (e.g., average acceptance rate, satisfaction rating), and θ represents the current model parameters. By analyzing the trend of Q-value changes, the system can automatically or assist administrators in identifying problem areas (e.g., persistently low acceptance rates for a certain type of recommendation), thereby triggering broader optimization measures, such as adjusting the rules of the recommendation strategy template library, fine-tuning the parameters of the context weighting function, or initiating specialized incremental training of the model for specific scenarios. This ensures that the entire recommendation service can continuously improve itself and enhance its quality based on real feedback.

[0060] Specifically, the service quality optimization function Q can be specified as a weighted sum of the average acceptance rate of recommendations and the accuracy of the model on the validation set, i.e., Q = w1 × average acceptance rate + w2 × model accuracy. The overall health of the system is evaluated by monitoring the changes in the Q value.

[0061] Through S104, the system completes a closed loop from execution to feedback to learning, making the entire recommendation system an intelligent agent that can continuously evolve and become more and more adaptable along with the production process and user interaction.

[0062] Example 2

[0063] This embodiment provides a client-side application system for an adaptive recommendation algorithm for production processes, used to implement the aforementioned method embodiment. Please refer to... Figure 2 The system comprises the following modules implemented through software, hardware, or a combination of both. These modules work together to form a complete intelligent closed loop of perception, learning, decision-making, and feedback:

[0064] Data acquisition module 210: Configured to perform real-time data acquisition through a data acquisition agent deployed at the user's end in the production site. This module includes:

[0065] Modular access unit: It adopts modular connection interface and configurable driver to realize dynamic access with production equipment and systems (such as MES, SCADA, equipment controllers) of different brands and protocols.

[0066] Data standardization unit: Using a predefined JSON data template, raw operational data, feedback data, and production context data obtained from heterogeneous sources are parsed and mapped, and converted into a unified format.

[0067] Secure transmission unit: encrypts and transmits standardized data to the backend module through secure network transmission protocols (such as HTTPS, MQTTS).

[0068] Model optimization module 220: Connected to data acquisition module 210, configured to perform real-time collaborative optimization of the user preference model and production mode model based on incremental learning algorithms. This module includes:

[0069] Feedback processing unit: Receives feedback data from the feedback loop module and calculates the feedback difference ΔR.

[0070] Incremental learning engine: based on formula The engine incrementally adjusts the model parameters θ. For user preference models, it can incorporate active learning strategies to filter samples with high uncertainty. For production mode models built using random forests, the engine achieves incremental updates through mechanisms such as online random forests or integrated weight adjustments.

[0071] Feature engineering unit: Extracts features (such as trend and periodic features) from the input time series data for use by the model.

[0072] Decision execution module 230: Connected to model optimization module 220 and data acquisition module 210, configured to generate and push personalized recommendations. This module includes:

[0073] Context-aware computing unit: Receives and processes production context data in real time, and runs context-weighted functions. .

[0074] Recommendation Strategy Library and Rule Engine: This module maintains a library containing strategy templates for various production scenarios and implements domain rule and constraint validation through a rule engine. It implements the logic of the recommendation function R(s,C) to match the optimal strategy template.

[0075] Recommendation generation and scheduling unit: Based on the weighted scoring results, it generates specific recommended content and determines the best time to push the recommendations to the user through the interface.

[0076] User feedback loop module 240: Connected to decision execution module 230 and data acquisition module 210, configured to implement closed-loop iteration. This module includes:

[0077] Feedback monitoring unit: Monitors and records users' explicit and implicit feedback behaviors to recommendations, and calculates the feedback function R(a,b).

[0078] Data integration unit: Standardizes the feedback data and integrates it into the sample dataset of model optimization module 220 in real time.

[0079] Service evaluation unit: Periodically calculates the service quality optimization function Q=f(R,θ), analyzes system performance, and provides a basis for system-level parameter tuning.

[0080] like Figure 2As shown, after the system starts, the data acquisition module 210 continuously acquires data from the production site and transmits it in a standardized manner. The model optimization module 220 initializes the model using historical data and waits for feedback data. When the decision execution module 230 generates and pushes a recommendation based on the current model and real-time context, the user feedback loop module 240 immediately begins collecting user execution feedback. This feedback is immediately sent to the model optimization module 220, triggering incremental updates to the user preference model and the production mode model. The updated model will influence the next decision of the decision execution module 230, thus forming a continuously operating, self-optimizing adaptive closed loop. Through the above process, the modules work together to ensure that the system can dynamically adapt to changes in user preferences and the production environment.

[0081] Example 3

[0082] This embodiment provides an electronic device for implementing the aforementioned adaptive recommendation algorithm for production processes in a user-side application. This electronic device can be an edge computing server, industrial computer, high-performance gateway, or terminal device with sufficient computing power deployed on-site in a factory.

[0083] The electronic device includes: one or more processors; a memory for storing one or more computer programs; and a communication interface for data communication with a user-end data acquisition agent and a production system at the production site. The one or more computer programs are stored in the memory and configured to be executed by the one or more processors. When the computer programs are executed by the processors, the electronic device implements all steps of the user-end application method of the production process adaptive recommendation algorithm.

[0084] Specifically, when the processor executes the program, it controls the electronic device:

[0085] 1. Receive standardized data reported in real time from the data acquisition agent deployed at the user terminal in the production site through the communication interface. The data includes user operation data, historical feedback data, and production context data containing equipment status and work order progress.

[0086] 2. Run the model optimization algorithm, based on the incremental learning mechanism, and use the data to update the parameters of the pre-built user preference model and production mode model in real time.

[0087] 3. Run the decision execution algorithm, combine the updated model with the real-time production context, calculate personalized production recommendation content through the rule engine and context-aware function, and send the recommendation content to the user terminal for execution through the communication interface.

[0088] 4. Receive and process execution feedback data from the user end, integrate it into the model training set, drive the next round of model iteration optimization, and thus complete the closed-loop adaptive process.

[0089] This electronic device, through a combination of hardware and software, solidifies the methods described in this invention into an executable product, providing real-time, adaptive intelligent recommendation services for the production site.

[0090] Example 4

[0091] This embodiment provides a computer-readable storage medium on which a computer program (instructions) are stored. When the computer program is executed by the processor of an electronic device, it implements all the steps of the user-side application method of the production process adaptive recommendation algorithm.

[0092] The storage media include, but are not limited to: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, solid-state drives (SSDs), or various non-transitory computer-readable storage media capable of storing program code.

[0093] When the storage medium containing the program is loaded into an electronic device with processing capabilities (such as the device described in Embodiment 3), the device is configured to execute the method described in this invention, forming a complete product solution. Through the dissemination and application of this medium, the technical solution described in this invention can be deployed and implemented on different hardware platforms.

[0094] In summary, this invention has detailed the working mechanisms, data flows, and interaction logic of each component of the system through specific embodiments, fully disclosing the technical solution. Those skilled in the art can implement this invention without creative effort based on the description herein and common knowledge in the field. The solution provided by this invention solves the problems of insufficient personalization and real-time performance in traditional production recommendation systems, and has significant practical value.

[0095] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A client-side application method for an adaptive recommendation algorithm for production processes, characterized in that, A continuously looping, adaptive closed-loop process is formed, wherein a single cycle of the process includes the following steps: By deploying a data acquisition agent at the user's end in the production site, the system can acquire real-time user operation data during the production process, as well as user feedback data on historical recommendations, and obtain production context data including equipment status and work order progress. Based on the incremental learning algorithm, the pre-built user preference model and production mode model are updated in real time using the operation data, historical feedback data and production context data to obtain the optimized model for the current period. Based on the current cycle's optimization model, combined with real-time acquired production context data, personalized production recommendation content is generated through a combination of a rule engine and a context-aware algorithm, and then pushed to the user for execution. The system collects user feedback data on the recommended content generated in the current cycle and uses this feedback data directly as historical feedback data for the next cycle to drive continuous iterative optimization of the user preference model and production mode model.

2. The user-side application method of the adaptive recommendation algorithm for production process according to claim 1, characterized in that, The model is updated in real time based on incremental learning algorithms, including: Calculate the feedback difference ΔR based on the user's acceptance or rejection of historical recommended content; Based on the feedback difference ΔR and the preset adjustment rate β, using the formula The model parameter θ is adjusted incrementally, where θ represents the parameter of the user preference model or production mode model.

3. The user-side application method of the production process adaptive recommendation algorithm according to claim 1, characterized in that, The generation of personalized production recommendation content includes: Constructing context-weighted functions Where u represents user identity information, s represents the current working mode status, and e represents the real-time device status; Using the aforementioned context weighting function The alternative recommendation strategies in the rules engine are weighted and scored, and the strategy with the highest comprehensive score is selected to generate the final recommended content.

4. The user-side application method of the adaptive recommendation algorithm for production process according to claim 1, characterized in that, When updating the production model, the random forest ensemble learning method is used. The equipment status parameters, output quality indicators and process intervals of the preceding process are used as input features. The analysis results of multiple decision trees are integrated to model the correlation patterns between complex production processes.

5. The user-side application method of the adaptive recommendation algorithm for production process according to claim 1, characterized in that, The raw data obtained through the acquisition agent is standardized using a predefined JSON data template to unify the heterogeneous data format from different production devices.

6. A client-side application system for an adaptive recommendation algorithm for production processes, characterized in that, include: The data acquisition module is configured to: acquire user operation data during the production process and user feedback data on historical recommendations in real time through the acquisition agent deployed on the user terminal at the production site, and acquire production context data including equipment status and work order progress. The model optimization module is configured to: based on the incremental learning algorithm, use the operation data, historical feedback data and production context data to update the pre-built user preference model and production mode model in real time to obtain the optimized model for the current period. The decision execution module is configured to: generate personalized production recommendation content based on the current cycle's optimization model and real-time acquired production context data, using a combination of a rule engine and a context-aware algorithm, and push it to the user for execution; The user feedback loop module is configured to: collect user feedback data on the recommended content generated in the current cycle, and use this feedback data directly as historical feedback data for the next cycle, so as to drive the continuous iterative optimization of the user preference model and the production mode model.

7. The client application system for the production process adaptive recommendation algorithm according to claim 6, characterized in that, The model optimization module is further configured as follows: Calculate the feedback difference ΔR based on the user's acceptance or rejection of historical recommended content; Based on the feedback difference ΔR and the preset adjustment rate β, using the formula The model parameter θ is adjusted incrementally, where θ represents the parameter of the user preference model or production mode model.

8. The client application system for the production process adaptive recommendation algorithm according to claim 6, characterized in that, The decision execution module is further configured as follows: Constructing context-weighted functions Where u represents user identity information, s represents the current working mode status, and e represents the real-time device status; Using the aforementioned context weighting function The alternative recommendation strategies in the rules engine are weighted and scored, and the strategy with the highest comprehensive score is selected to generate the final recommended content.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements a client-side application method for a production process adaptive recommendation algorithm as described in any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a client-side application method for a production process adaptive recommendation algorithm as described in any one of claims 1 to 5.

Images