Parameter determination method and device, equipment and storage medium

Abstract

The invention provides a parameter determination method and device, equipment and a storage medium, and relates to the technical field of data processing. The method comprises the steps that object attribute information of an object and product attribute information of a product associated with the object are obtained, the object is a production facility and a production tool in a production activity, and the object attribute information comprises geometric parameters of the object, geographic positions of the object and property characteristics of the object; the product attribute information comprises a category identifier of the product and production data of the product in a plurality of production stages, and the production data of each production stage comprises a current production stage identifier of the product, production state data of the product, a production record of the product, environment data corresponding to the product and an abnormal event of the product. And processing the attribute information of the object and the product through a preset parameter model, and obtaining quality parameters of the product. And determining a quality evaluation parameter of the user based on the evaluation rule and the product quality parameter. The method is used for improving the accuracy of performance quality parameters of to-be-returned resource quotas.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular, to a parameter determination method and device, equipment and a storage medium. BACKGROUND

[0002] In product collaborative production activities involving resource prepayment, in order to protect the interests of prepayment resources, the resource supporting party usually evaluates the prepayment resource return ability of the resource receiving party.

[0003] At present, in the process of evaluating the prepayment resource return ability, more attention is paid to the historical credibility of the resource receiving party. However, this scheme may ignore the influence of product output on the prepayment resource return ability of the resource receiving party, thereby leading to unreliable evaluation results of the prepayment resource return ability of the resource receiving party. SUMMARY

[0004] Embodiments of the present application provide a parameter determination method, device, equipment and storage medium, which can determine the quality of the product in the final stage by analyzing the time sequence characteristics of different element data in the product production process, and evaluate the product quality, thereby providing evaluation basis for the resource return ability of the resource receiving party.

[0005] To achieve the above purpose, embodiments of the present application adopt the following technical solutions: In a first aspect, a parameter determination method is provided, which comprises: First, the object attribute information of the target object and the product attribute information of the target product associated with the target object are obtained, wherein the target object is a production facility and a production tool in a production activity, and the object attribute information includes object geometric parameters, geographical position and property characteristics of the target object. The product attribute information includes the category identifier of the target product, the production data of the target product in multiple production stages, and the production data of each production stage includes the current production stage identifier of the target product, the production state data of the target product, the production record of the target product, the environmental data corresponding to the target product and the abnormal event of the target product. Then, the object attribute information and the product attribute information are processed by a preset parameter model to obtain the quality parameter of the target product; the preset parameter model has the function of obtaining the product quality parameter based on the object attribute information and the product attribute information; wherein the preset parameter model has the function of obtaining the product quality parameter based on the object attribute information and the product attribute information. Finally, the quality evaluation parameter is determined based on the preset evaluation rule and the product quality parameter.

[0006] The parameter determination method provided by the embodiment of the application can cover the analysis framework of the production process by introducing the attribute information of the target object and the dynamic timing data of the target product at different production stages (including the product at the current production stage and the production stage before it). In addition, the application can more accurately predict the quality parameters of the product at a future time point by deeply mining the correlation between the production factors (such as environmental data, state data and abnormal events) of the product at different production stages and the final yield of the product by using the parameter model. Compared with the related art, the application can provide a new evaluation direction for evaluating the resource return ability of the user by evaluating the product quality parameters to obtain product quality evaluation parameters, and can make the ability evaluation result more reliable.

[0007] In a possible implementation form of the first aspect, the target object includes a production facility in the agricultural field, and the production facility includes farmland, a greenhouse or a breeding pond.

[0008] It should be understood that when the scheme is applied to the agricultural field, the resource return ability of the agricultural product can be evaluated before the agricultural product is harvested by using the method. Meanwhile, the method can also decentralize the evaluation of the resource return ability to each stage of the agricultural production process, and provide a dynamic resource return ability evaluation means for the resource provider and the agricultural producer, so as to realize the refinement and foresight of the agricultural resource investment management.

[0009] In a possible implementation form of the first aspect, the production state data of the target product includes appearance form parameters of the target product, product weight parameters of the target product and production rate of the target product. The production record of the target product includes production material input record of the target product, production material addition record of the target product and production environment maintenance record of the target product. The environmental data corresponding to the target product includes temperature data, weather data, precipitation data and illumination data. The abnormal event of the target product includes an abnormal event type, an abnormal event processing mode and an abnormal event response time.

[0010] It should be understood that the scheme provides comprehensive and structured multi-dimensional input data for the parameter model by specifically defining the fine-grained parameters included in the production state data, the production record, the environmental data and the abnormal event, so that the product production process is more accurately and stereoscopically described, thereby laying a solid data foundation for accurately predicting the yield parameters.

[0011] In a possible implementation form of the first aspect, the quality parameter of the target product is determined by processing the object attribute information and the product attribute information by using a preset parameter model, and the quality parameter of the target product comprises: feature information obtained by performing feature construction based on the object attribute information of the target object and the product attribute information of the target product. The feature information comprises: a feature parameter for quantifying a deviation degree of product production state data from benchmark production state data, a feature parameter for quantifying a deviation degree of environment data from benchmark environment data, and a feature parameter for quantifying a frequency of abnormal events of the sample product.

[0012] It should be understood that, by performing feature extraction on the attribute information to obtain feature information with clear physical meaning and business orientation, the scheme can ensure that the accuracy and robustness of the prediction model parameter model are effectively improved, and the explainability of the model decision process is enhanced, in the case that the model structure is unchanged.

[0013] In a possible implementation form of the first aspect, the quality evaluation parameter of the target product is determined based on a preset evaluation rule and the quality parameter of the product, and the method comprises: determining a resource conversion amount of the target product based on a preset unit quality and unit resource amount conversion rule corresponding to the target product. The quality evaluation parameter of the target object for the resource quota to be returned is determined based on the preset evaluation rule, the resource conversion amount of the target product, and the resource amount to be returned of the target object. The resource amount to be returned is a resource amount that is pre-configured for the target user by a resource provision contract and is agreed to be returned by output of the target product.

[0014] It should be understood that, by establishing a mapping rule between unit quality and unit resource amount, the scheme converts the product quality parameter into a resource amount with a unified value benchmark, and then determines the resource conversion amount of the target product, thereby realizing objective and standardized measurement from product quality to resource value, and thus building a reliable bridge from product quality evaluation to user resource performance evaluation.

[0015] In a possible implementation form of the first aspect, the method further comprises: in a case where the quality evaluation parameter is lower than a preset quality evaluation parameter threshold, obtaining a plurality of quality parameters of the target product at different production stages. A quality change trend graph of the target product at different production stages is generated based on the plurality of quality parameters. Quality warning information of the target product is generated based on the quality change trend graph, the object attribute information, and the product attribute information.

[0016] It should be understood that the scheme realizes dynamic monitoring and risk early warning of the resource return ability by monitoring the quality evaluation parameters, tracing historical quality data, visualizing trend changes, and generating early warnings combined with multi-source information, thereby providing reliable timing basis for timely intervention and accurate decision-making.

[0017] In a possible implementation of the first aspect, the preset parameter model is obtained by: obtaining a training sample set. The training sample set includes a plurality of training samples. Each training sample includes object attribute information of a sample object, product attribute information of a sample product associated with the sample object, and a quality parameter label of the sample product. An initial model of the preset parameter model is trained based on the training sample set to obtain the preset parameter model. The initial model of the preset parameter model is a gradient boosting machine model.

[0018] It should be understood that the scheme provides a comprehensive and real data basis for model learning by constructing a training sample set including sample object attributes, sample product full-cycle production data, and real quality parameter labels. The gradient boosting machine model is used to learn features and perform nonlinear fitting, to deeply mine the mapping relationship between complex time sequence dynamic characteristics and quality parameters in the production process, so that the trained parameter model can capture the comprehensive influence mechanism of different production factors on product quality and accurately predict the final product quality level under specific production conditions.

[0019] In a possible implementation of the first aspect, the model training includes: obtaining a preset number of test sample sets from the training sample set. The initial model is trained based on the training sample set to obtain a parameter model to be evaluated. The parameter model to be evaluated is tested based on the test sample set to obtain a model evaluation index of the parameter model to be evaluated. The model evaluation index includes an area under the receiver operating characteristic curve (ROC-AUC). The model parameters in the parameter model to be evaluated are adjusted based on a preset model evaluation index-model parameter adjustment rule and the model evaluation index to obtain the preset parameter model.

[0020] It should be understood that by using the preset test sample set for model evaluation, and selecting the ROC-AUC index which is not sensitive to class imbalance as the evaluation standard in view of the characteristics of the production data, such as multiple feature dimensions and uneven sample distribution, and by iteratively optimizing the model combined with the parameter adjustment rule, the model evaluation deviation caused by uneven samples is effectively reduced, and the model has good generalization ability and robustness in real scenarios.

[0021] In a second aspect, a parameter determination apparatus is provided, and the parameter determination apparatus includes: The acquisition module is configured to acquire object attribute information of a target object and product attribute information of a target product associated with the target object. The target object is a production facility or a production tool in a production activity. The object attribute information includes geometric parameters of the target object, a geographic position of the target object, and property characteristics of the target object. The product attribute information includes a category identifier of the target product and production data of the target product at multiple production stages. The production data at each production stage includes a current production stage identifier of the target product, production state data of the target product, production records of the target product, environment data corresponding to the target product, and abnormal events of the target product.

[0022] The processing module is configured to process the object attribute information and the product attribute information by using a preset parameter model to obtain a quality parameter of the target product. The preset parameter model is configured to obtain the quality parameter of the target product based on the object attribute information and the product attribute information. The processing module is further configured to determine a quality evaluation parameter of the target user based on a preset evaluation rule and the quality parameter of the target product.

[0023] In a third aspect, an electronic device is provided. The method includes a memory and at least one processor. The memory is in communication with the processor. The memory is configured to store computer program code including computer instructions. When the processor executes the computer instructions, the electronic device performs the method of the first aspect and any possible implementation manner thereof.

[0024] In a fourth aspect, a computer readable storage medium is provided. The computer readable storage medium stores computer instructions. When the computer instructions are executed by a processor, the computer instructions are used to implement the method of the first aspect and any possible implementation manner thereof.

[0025] In a fifth aspect, a computer program product is provided. When the computer program product is run on a computer / executed by a processor of a computer, the computer program product implements the method of the first aspect and any possible implementation manner thereof. The computer can be the electronic device of the third aspect and any possible implementation manner thereof.

[0026] It can be understood that the parameter determination apparatus of the second aspect, the electronic device of the third aspect, the computer readable storage medium of the fourth aspect, and the computer program product of the fifth aspect can achieve the beneficial effects of the first aspect and any possible implementation manner thereof, which will not be described herein. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 A data processing flow diagram of a computing device is provided for the embodiments of the present application. Figure 2A structural schematic diagram of a data acquisition subsystem of a computing device provided by an embodiment of the present application is shown in FIG. 1. Figure 3 A flowchart of a parameter determination method provided by an embodiment of the present application is shown in FIG. 2. Figure 4 A flowchart of another parameter determination method provided by an embodiment of the present application is shown in FIG. 3. Figure 5 A flowchart of yet another parameter determination method provided by an embodiment of the present application is shown in FIG. 4. Figure 6 A flowchart of still another parameter determination method provided by an embodiment of the present application is shown in FIG. 5. Figure 7 A flowchart of a parameter model training method provided by an embodiment of the present application is shown in FIG. 6. Figure 8 A flowchart of another parameter model training method provided by an embodiment of the present application is shown in FIG. 7. Figure 9 A structural schematic diagram of a parameter determination apparatus provided by an embodiment of the present application is shown in FIG. 8. Figure 10 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 9. DETAILED DESCRIPTION

[0028] Hereinafter, the terms "first" and "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0029] The exemplary embodiments will be described in detail below with reference to the accompanying drawings. In the following description, the same numbers are used to designate the same elements in different drawings, unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present application. Rather, they are merely examples of devices and methods consistent with some aspects of the present application, as detailed in the appended claims.

[0030] In the technical solutions provided by the embodiments of the present application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data, etc. comply with the relevant legal regulations and do not violate public order and good customs.

[0031] It should be noted that in the embodiments of the present application, some software, components, models and the like of the prior art may be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the scheme.

[0032] In the product collaborative production activities involving resource prepayment, the resource supporting party needs to accurately assess the prepayment return ability of the resource receiving party (such as farmers or agricultural enterprises in agricultural production) to make reasonable decisions.

[0033] The related assessment method mainly depends on the historical credit status of the resource receiving party, which does not consider the product output in the assessment, resulting in a disconnection between the assessment result and the value of the product, and an inaccurate assessment of the resource return ability of the resource providing party, which is difficult to support effective resource recovery decisions and intervention decisions.

[0034] Therefore, how to consider the impact of the product on the resource return ability of the resource receiving party in the resource return ability assessment has become a key technical problem to be solved in the current field.

[0035] Based on this, the present application constructs an analysis framework that can deeply integrate the static attributes of the target object (such as farmland, production facilities) and the dynamic production data of the whole life cycle of the target product (such as crops). By collecting time-series data covering multiple production stages, and using machine learning models to mine the complex mapping relationship between these data and the final yield of the product, the quality of the product can be accurately estimated. Further, by evaluating the quality of the product, a more scientific, accurate and forward-looking assessment method is provided for the resource supporting party.

[0036] The parameter determination method and the parameter model training provided by the embodiments of the present application can be applied in a computing device. Specifically, the computing device can be a single server or a server cluster composed of multiple servers, or a computer, or a processor or processing chip in a server or computer. The embodiments of the present application do not limit the specific device form of the computing device. The embodiments of the present application can be applied to various scenarios that need to evaluate the quality of the product based on the production process.

[0037] For example, taking the product production scenario in the field of agriculture as an example. In this scenario, the target object is a plot of land to be planted with crops, and the target product is a specific crop (such as corn) planted on the plot. By obtaining attribute information such as soil properties and geographical location of the plot, and combining growth data such as plant height and leaf area index of corn at each growth stage, as well as abnormal events such as occurrence of diseases and pests, the expected yield of corn can be accurately predicted, providing support for production decisions.

[0038] For example, in the field of industrial production, the target object can be a specific production line or production facility, and the target product can be an industrial manufactured product (such as a specific model of mechanical equipment) produced by the line. By collecting state data such as operating parameters of production equipment and process completion progress, combined with condition data such as environmental temperature and humidity, dynamic evaluation of product qualification rate and production cycle can be realized, providing basis for production process optimization.

[0039] For another example, in the field of research and development projects, the target object is a research and development team and an experimental environment, and the target product is a technical achievement to be developed (such as a software). By tracking experimental data records, milestone completion, and technical indicator achievement, the technical feasibility of the project can be objectively evaluated, providing data support for project management.

[0040] For example, in the above agricultural field scenario, the data processing architecture of the computing device is described as follows: Figure 1 As shown in FIG. 1, it is a data processing flow diagram of a computing device provided by an embodiment of the present application. It includes the following: data collection and processing, model training, and quality evaluation.

[0041] Among them, data collection and processing includes data collection layer and data processing layer. Model training includes model training layer. Quality evaluation includes attribute information processing layer, early warning notification layer and user interface layer.

[0042] The data collection layer is used to obtain the attribute information of the sample plot and the attribute information of the sample crops on the sample plot. Specifically, it includes: obtaining the attributes such as the (geometric) area, geographical position and soil properties of the sample plot through the plot mapping module; obtaining the variety code and category identifier corresponding to the sample crops through the crop variety atlas module; obtaining the production state data (such as plant height, leaf area index, biomass) and production records (such as seeding amount, fertilization records, irrigation records) of the sample crops at different production stages through the production process monitoring module; obtaining the environmental data such as temperature, precipitation and light of the sample crops at different production stages through the weather monitoring module; obtaining the abnormal event types such as diseases and pests and their treatment records reported by the user through the inspection reporting module; obtaining the yield, quality grade and purchase price of the sample crops through the purchase monitoring module, and calculating the income parameter as the model label based thereon.

[0043] At the same time, the data collection layer associates and serializes the above-mentioned various data in time sequence according to different production stages (such as seeding period, seedling period, growth period and mature period) of the crops, forming a time series data set, and providing a structured data basis for subsequent model training.

[0044] The data processing layer is used to perform data cleaning and feature construction on the above-mentioned data to obtain a training data set.

[0045] a model training layer configured to train a parameter model according to a training data set.

[0046] an attribute information processing layer configured to input plot attribute information of a target plot and product attribute information of a target crop associated with the target plot into the trained parameter model to obtain a predicted product yield parameter of the target crop at a current growth stage. Further, the layer is configured to calculate a performance quality parameter based on the predicted product yield parameter, a preset unit yield-resource quantification mapping rule, and a target user's corresponding resource quota to be returned, and automatically generate a performance quality warning information when the parameter is identified to be lower than a preset threshold.

[0047] a warning notification layer configured to receive the performance quality warning information generated by the quality evaluation layer and push the performance quality warning information to a terminal device of a relevant management personnel of a resource support party through a preset message channel (such as an intranet message, a short message, or an email) so as to initiate necessary attention or a process.

[0048] a user interface layer configured to provide visual interactive interfaces for users in different roles. For example, the user interface layer provides a panoramic monitoring view including a performance quality parameter dynamic dashboard, a warning information list, and a trend analysis graph for a resource support party management user, and provides a production data reporting entry, a current yield prediction result, and a historical evaluation record query interface for a target product associated with a resource receiving party user. The layer supports data filtering and visual rendering based on user permissions, and realizes transparency and interactivity of the evaluation process.

[0049] In some embodiments, the data collection layer in the above Figure 1 may also be separately split from the computing device to form a separate data collection subsystem to complete real-time data collection and preliminary structured processing. As shown in Figure 2 , the data collection subsystem mainly includes the following functional modules and services: a plot surveying and mapping module. The plot surveying and mapping module is configured to call a map service to obtain plot geographic positioning information through GIS surveying and mapping technology and a drone, and the system supports field surveying and mapping or map drawing in a planting area, records a surveying and mapping track through a track service, generates a closed vector graph including latitude and longitude coordinates, and calculates plot geometric parameters and geographic position distribution based on the closed vector graph. The plot surveying and mapping module is further configured to generate a unique identification code for each plot to form standardized plot attribute information.

[0050] a crop variety atlas module. The crop variety atlas module is configured to establish a crop variety information database, assign a unique category identifier to each crop variety, and form a variety coding system. The crop variety atlas module provides basic data support for subsequent associated plot planting information.

[0051] Planting process monitoring module. The whole planting process is divided into multiple standard production stages, including pre-sowing plan, pre-sowing preparation, sowing, field management, yield measurement and field harvesting, etc. The data collection of each stage includes: Pre-sowing plan stage: record the association between land and variety, planting household, and obtain planting plan data.

[0052] Pre-sowing preparation stage: record the time, frequency, etc. of production records such as soil preparation, fertilization, etc.

[0053] Sowing stage: record sowing time, progress, etc. sowing operation data.

[0054] Field management stage: collect production status data such as emergence rate, growth index, etc.

[0055] Yield measurement stage: obtain yield estimation data through remote sensing or field measurement.

[0056] Field harvesting stage: record harvesting data such as harvesting progress, area, weight, etc.

[0057] Field monitoring module. Including: environmental monitoring module, abnormal event monitoring module and purchase monitoring module.

[0058] Among them, the abnormal event monitoring module records the field inspection path through the trajectory service, and collects abnormal event information including abnormal type, occurrence stage, reporting time, processing time, etc. Form a complete abnormal event account.

[0059] The environmental monitoring module accesses the meteorological monitoring equipment, continuously collects environmental data such as temperature, precipitation, and illumination, and records environmental abnormal information such as meteorological disasters.

[0060] The purchase monitoring module collects the actual harvesting weight through the weighing system after harvesting is completed, calculates the output value combined with the market situation, and provides quality parameter label data for model training.

[0061] Through the cooperative operation of the above modules, the data collection subsystem realizes the data collection system from land attribute collection to production process monitoring, thereby obtaining the collected data (that is, the data input into the data cleaning as Figure 1 the data collection layer).

[0062] As shown in Figure 3 , the parameter determination method provided by the embodiment of the application is applied in a computing device, and specifically includes the following content: S101, obtaining object attribute information of a target object and product attribute information of a target product associated with the target object.

[0063] Among them, the target object is a production facility and a production tool in a production activity.

[0064] The object attribute information includes object geometric parameters of the target object, geographical position of the target object, and property characteristics of the target object. The product attribute information includes category identification of the target product, and production data of the target product at multiple production stages. The production data of each production stage includes current production stage identification of the target product, production state data of the target product, production record of the target product, environment data corresponding to the target product, and abnormal event of the target product.

[0065] The target object refers to a facility or tool carrying out product production activities, and the target product refers to an entity produced by the target object.

[0066] In some embodiments, the target object includes a production facility in the agricultural field, including a farmland, a greenhouse, or a breeding pond. Alternatively, the target object includes a production facility and a production tool in the industrial field, including a factory building, a production line device, or a warehouse logistics facility.

[0067] The object attribute information corresponding to the target object is used to describe inherent characteristics of the target object. Specifically, the object attribute information can include the following: Geometric parameters are used to quantify the scale of the target object.

[0068] Geographical position is used to define the spatial position of the target object through latitude and longitude coordinates or relative position relationship, which is closely related to environmental factors, logistics costs, and the like. Property characteristics are used to describe the inherent attributes of the target object.

[0069] For example, the object geometric parameters refer to the acreage of a farmland or the floor area of a production line. The object property refers to the soil type, pH value of a farmland, or the device model, technical specification of a production line, and the like.

[0070] The product attribute information corresponding to the target product is used to describe the category of the target product and the dynamic production characteristics of the target product. Specifically, the product attribute information can include the following: The category identification is used to distinguish product categories (for example, crop variety codes or automobile model codes), and the core lies in obtaining product parameters of the product through type identification to associate the production data thereof.

[0071] The production data of multiple production stages includes the current production stage of the target product and the production data before the current production stage, and the production data of multiple production stages constitutes the dynamic production data of the target product.

[0072] In some embodiments, the production state data of the target product includes appearance form parameters of the target product, product weight parameters of the target product, and production rate of the target product. The production record of the target product includes production material input record of the target product, production material addition record of the target product, and production environment maintenance record of the target product. The environment data corresponding to the target product includes temperature data, weather data, precipitation data, and illumination data. The abnormal event of the target product includes abnormal event type, abnormal event processing mode, and abnormal event response time.

[0073] Specifically, the current production stage identifier reflects the production stage in which the current product is located.

[0074] For example, in the agricultural scenario, the current production stage identifier can include the sowing stage, the jointing stage, and the filling stage. Alternatively, in the industrial scenario, the current production stage identifier can include the stamping stage, the welding stage, and the painting stage.

[0075] The production state data reflects the real-time condition of the product during production, such as appearance form parameters (e.g., leaf area index, vehicle body paint thickness), product weight parameters (e.g., plant fresh weight, component weight), and production rate (e.g., daily growth, unit hourly output).

[0076] The production record includes specific production operations, including production material input record (e.g., sowing amount, raw material input), production material addition record (e.g., fertilizer, material supplement), and production environment maintenance record (e.g., irrigation, equipment maintenance).

[0077] The environment data captures the external conditions of the product production, including temperature, weather, precipitation, and illumination.

[0078] The abnormal event records the deviation from the normal production process, including the type (e.g., disease and insect pests, equipment failure), the processing mode (e.g., spraying pesticides, replacing parts), and the response time (the time interval from discovery to processing).

[0079] In one possible implementation, the computing device can obtain attribute information of the target object and the target product associated with the target object through a data collection subsystem as shown in Figure 2

[0080] S102, processing the object attribute information and the product attribute information through a preset parameter model to obtain a quality parameter of the target product.

[0081] The preset parameter model has the function of obtaining the product quality parameter based on the object attribute information and the product attribute information.

[0082] ​The quality parameter refers to the quality level that the target product can achieve after production. Essentially, the quality parameter converts various influencing factors in the product production process into the final output level indicator through model calculation. The specific meaning and form of the parameter can vary depending on different application fields.

[0083] For example, in an agricultural scenario, the quality parameter can be represented as a predicted total yield (such as crop weight), a predicted yield per unit area, or other physical output indicators. Alternatively, in an industrial scenario, it can be represented as a predicted number of qualified products, a predicted output efficiency, or other technical performance indicators.

[0084] In the actual reasoning process of the parameter model, the parameter model extracts key information such as time series change patterns and environmental response characteristics from the analysis of object attribute information (such as plot area and soil properties) and product attribute information (such as state data, environmental data, and abnormal events at each production stage). Based on the knowledge obtained from training, the model can comprehensively evaluate the development trend of the current production state, the cumulative impact of environmental factors, and the potential consequences of abnormal events, thereby determining the quality parameter of the product.

[0085] It should be understood that this reasoning process not only considers static attribute characteristics, but more importantly, incorporates dynamic change information of the production process, making the prediction results more timely and accurate.

[0086] In one possible implementation, for a preset parameter model, the computing device can directly obtain a trained parameter model through an external device and deploy it in the production environment of the computing device.

[0087] In another possible implementation, for a preset parameter model, the computing device can also train the parameter model itself to obtain the parameter model. In this case, the computing device can refer to the specific description of the parameter model training method in the following Figure 7 and the introduction content thereof, which will not be described in detail here.

[0088] S103, determining a quality evaluation parameter based on a preset evaluation rule and the product quality parameter.

[0089] From the foregoing S102, the product quality parameter reflects the product's output capacity, but the product's output capacity is not sufficient to be directly applied to evaluation decisions. Therefore, S103 needs to further analyze and interpret the product quality parameter through the preset evaluation rule, thereby determining a final quality evaluation parameter that is more comprehensive and has more decision-making guiding significance.

[0090] In some embodiments, the preset evaluation rule refers to a rule that compares the product quality parameter with a preset quality benchmark interval and outputs a standardized score based on the comparison result.

[0091] In some embodiments, the preset evaluation rule refers to an algorithm for calculating the ratio of the resource conversion amount to the amount of resources to be returned, and determining the quality evaluation parameter according to the ratio.

[0092] In this case, the specific implementation of S103 can refer to the following Figure 5 , which will not be described in detail here.

[0093] The parameter determination method provided by the embodiments of the present application can cover the analysis framework of the production process by introducing the attribute information of the target object and the dynamic timing data of the target product at different production stages (including the product at the current production stage and the production stage before it). Secondly, the present application uses the parameter model to deeply mine the correlation between the production factors such as environmental data, state data and abnormal events in different production stages and the final yield of the product, which can more accurately predict the quality parameters of the product at the future time point. Compared with related technologies, the present application can provide a new evaluation direction for evaluating the resource return ability of the user by evaluating the product quality parameters to obtain the product quality evaluation parameter, which can make the ability evaluation result more reliable.

[0094] In some embodiments, in the process of processing the attribute information by the model, in order to fully tap the data potential of the attribute information and enhance the perception ability of the model to the key influencing factors, the computing device can perform secondary feature construction based on domain knowledge on the basis of the attribute information. As shown in Figure 4 S102 can specifically include the following steps: S201, based on the object attribute information of the target object and the product attribute information of the target product, feature construction is performed to obtain feature information.

[0095] The feature information includes: a feature parameter for quantifying the deviation degree of the product production state data from the benchmark production state data (which can be summarized as a production state deviation degree feature parameter), a feature parameter for quantifying the deviation degree of the environmental data from the benchmark environmental data (which can be summarized as an environmental data deviation degree feature parameter), and a feature parameter for quantifying the frequency of abnormal events of the product (which can be summarized as an abnormal event influence degree feature parameter).

[0096] In some embodiments, for the production state deviation degree feature parameter, the computing device retrieves the corresponding benchmark production state data (such as the ideal growth model data or the historical statistical average value of the product category at the stage) from the preset knowledge base according to the category identifier of the product and the current production stage identifier, and compares and calculates the real-time collected production state data with the benchmark data.

[0097] For example, in the agricultural scenario, the feature can specifically include: leaf area index deviation: calculate the ratio or difference between the current leaf area index and the standard leaf area index of the same variety at this growth stage. Plant growth rate deviation: calculate the multiple of the standard deviation of the average daily growth of the plant height at the current stage from the historical average daily growth at the same stage. Fruit enlargement rate deviation: for fruit trees, calculate the deviation of the daily increase of fruit circumference / diameter from the standard value.

[0098] In some embodiments, the computing device obtains the historical benchmark environmental data of the same period from the associated environmental database (such as a meteorological database, a plant monitoring system) based on the geographical location of the object and the time information of the production stage, and calculates the deviation from the actual environmental data.

[0099] For example, in the agricultural scenario, the feature can specifically include: effective accumulated temperature deviation: calculate the difference between the actual accumulated effective accumulated temperature from sowing to the current stage and the average effective accumulated temperature of the same period in history. Precipitation anomaly percentage: calculate the percentage deviation of the total precipitation at the current stage from the average precipitation at the same period in history. Sunshine duration deviation: calculate the difference between the actual sunshine duration at the current stage and the average sunshine duration at the same period in history.

[0100] For example, in the industrial scenario, the feature can specifically include: clean room temperature and humidity deviation: calculate the cumulative length or area of temperature and humidity outside the standard control range in the production environment within a continuous time. Key process medium concentration deviation: such as the continuous root mean square error between the actual value and the standard value of the concentration of key media such as electroplating liquid and cleaning agent.

[0101] In some embodiments, the computing device calculates the type of abnormal event, the level of event impact, the duration and the processing efficiency according to the pre-defined quantification rules, and synthesizes a comprehensive impact index, rather than just the frequency of occurrence.

[0102] For example, in the agricultural scenario, the feature can specifically include: pest impact index: calculated by weighting the type of abnormal event (such as pest category), severity level (such as light, medium, and heavy), and duration. Meteorological disaster impact index: calculated by weighting the type of disaster (such as hail, drought), intensity (such as the degree of precipitation deficiency), and occurrence period (such as during the critical growth period).

[0103] For example, in the industrial scenario, the feature can specifically include: equipment failure impact index: calculated by weighting the failure level (such as downtime length), the criticality of the affected process (such as whether it is a critical device), and the repair response efficiency (such as average repair time). Incoming batch abnormality index: calculated by weighting the unqualified rate of the batch, the severity of the unqualified items (such as critical parameter out-of-tolerance).

[0104] It should be noted that from the above examples, it can be seen that the specific content of the feature information depends on the specific application scenario, and the embodiments of the present application do not make any limitation on the content of the feature information and the calculation method. Those skilled in the art can flexibly construct other feature parameters having strong correlation with the final yield according to the specific category of the target product, the production mode and the available data dimensions.

[0105] The embodiments of the present application can convert the experience and knowledge of domain experts into numerical indicators that can be directly learned by the model. This not only significantly enhances the interpretability of feature engineering, but more importantly, it provides a clearer and more solid bridge between the target and the result, enabling the model to more quickly and accurately capture key dynamic factors that affect the final yield, thereby comprehensively improving the stability and reliability of the prediction performance.

[0106] S202, processing the object attribute information, the product attribute information and the feature information through a preset parameter model, to obtain a quality parameter of the target product.

[0107] Among them, by introducing in the model reasoning, introducing the strongly related derived features, so that the model can more accurately establish the mapping relationship from the production process features to the final quality parameter in the reasoning process, thereby outputting more accurate, stable and business interpretable quality prediction results. The quality parameter comprehensively reflects the expected quality level that the target product can achieve under certain object attributes and dynamic production conditions.

[0108] It should be understood that this scheme can effectively improve the accuracy and robustness of the prediction model parameter model and enhance the interpretability of the model decision-making process by extracting features from the attribute information to obtain feature information with clear physical meaning and business directionality, while ensuring that the model structure remains unchanged.

[0109] In some embodiments, the preset evaluation rule can be understood as an algorithm for calculating the ratio of the resource conversion amount to the amount of resources to be returned and determining the quality evaluation parameter according to the ratio, according to the foregoing S103. In this case, the specific process of determining the quality evaluation parameter of the target product based on the preset evaluation rule is as shown in Figure 5 The specific process of determining the quality evaluation parameter of the target product based on the preset evaluation rule is as shown in S301, determining the resource conversion amount of the target product based on a preset unit mass and unit resource amount conversion rule corresponding to the target product.

[0110] Among them, the unit mass refers to a standardized measurement unit of the target product. The unit resource amount refers to the equivalent value of the quantified resources corresponding to the unit mass of the product.

[0111] The core of the preset unit mass and unit resource quantity conversion rule is to construct an internal measurement scale that maps product outputs of different quality levels to standard resource values. This rule converts the quality parameters representing the expected product quality and quantity into a unified resource conversion quantity through a predefined quality value coefficient (i.e., unit resource quantity). The quality value coefficient depends on the inherent value level of the target product corresponding to the given quality standard, thereby ensuring that high-quality outputs can be converted into higher resource values. Through this mechanism, the scheme provides a fair value data basis directly linked to product quality for subsequent objective quantitative evaluation of resource performance.

[0112] In one possible implementation, the computing device first acquires the category identifier of the target product and queries the preset unit mass and unit resource quantity conversion coefficient of the target product based on the identifier.

[0113] For example, in an industrial scenario, the target product is a specific model of engine, and the quality parameter is the predicted number of qualified products, which is 200 units. The preset unit resource quantity is 850 resource points per engine. The "resource points" take into account the standard man-hours, key material quotas, and technical complexity coefficients consumed by the product. The conversion rule is a direct mapping, so the resource conversion quantity = 200 units x 850 resource points / unit = 170,000 resource points.

[0114] S302, based on the preset evaluation rule, the resource conversion quantity of the target product and the quantity of resources to be returned by the target object, determines the quality evaluation parameter of the target user to which the target object belongs for the resource quota to be returned.

[0115] The target user to which the target object belongs refers to the resource receiver, i.e., the actual user of the pre-allocated resources.

[0116] For example, in an agricultural scenario, the target object is a specific plot of land, and the target user is the plot manager. In an industrial scenario, the target object is a production line, and the target user is the operation responsible party of the production line.

[0117] The quantity of resources to be returned is the quantity of resources pre-configured for the target user by the resource provision contract and agreed to be returned through the output of the target product.

[0118] The quantity of resources to be returned refers to the total quantity of various resources applied by the target user for product production before the production activity starts. The quality parameter alone is not enough to evaluate the return ability of the initial resource input, so the quality parameter needs to be converted into resource quantity and compared with the resource quota to be returned, thereby converting the product-level output prediction into an evaluation of the return ability of the resources to be returned.

[0119] In some embodiments, the preset evaluation rule includes the following processing logic: first, the ratio of the resource conversion amount to the amount of resources to be returned is calculated, and the ratio is mapped to a quality evaluation parameter according to a preset threshold interval, which is suitable for scenarios where the production process is stable and the resource input is clearly related to the product value. Second, for scenarios where the production cycle is long and each stage has a significant impact on the final output (such as tree cultivation and complex equipment manufacturing), the resource conversion amount can be dynamically corrected according to the completion degree of each production stage and the stage weight, and then the ratio is calculated based on the corrected resource conversion amount. Finally, for scenarios that are easily affected by abnormal events (such as open planting and precision component production), the resource conversion amount can be further corrected based on the frequency and impact of abnormal events, and then the resource conversion amount is determined to evaluate the quality of performance.

[0120] For example, in an agricultural scenario, assume that the amount of resources to be returned for a corn planting plot is 800 units, and the initial resource conversion amount is 1600 units after prediction and conversion. After considering the completion degree of each growth stage, the resource conversion amount is adjusted to 1521.6 units; further incorporating the influence of abnormal events, the resource conversion amount is finally determined to be 1460.7 units. By calculating the ratio of the resource conversion amount to the amount of resources to be returned (about 1.826), and comparing it with the preset threshold of “≥1.3 is excellent”, the quality evaluation parameter of the target user is determined to be “excellent”. In addition, the form of the quality evaluation parameter can be a score form (such as a percentage system or a percentage form) in addition to the grade form in the above example, and the specific form of the quality evaluation parameter is not limited in the embodiments of the present application.

[0121] It should be understood that the scheme converts the product quality parameter into a resource amount with a unified value benchmark by establishing a mapping rule between unit quality and unit resource amount, and then determines the resource conversion amount of the target product, thereby realizing the objective and standardized measurement from product quality to resource value, and thus building a reliable bridge from product quality evaluation to user resource performance evaluation.

[0122] In some embodiments, the computing device can perform resource performance analysis on the obtained quality evaluation parameter to discover potential resource return violation problems and urge the target user (i.e., the resource receiver) to optimize the production process of the product. In this case, the process is as shown in Figure 6 After S103, the method further includes: S401, if the quality evaluation parameter is lower than the preset quality evaluation parameter threshold, obtaining a plurality of quality parameters of the target product at different production stages.

[0123] The preset quality evaluation parameter threshold is a reference value for determining whether the product production state enters an early warning interval. When the computing device determines that the quality evaluation parameter is lower than the threshold, it indicates that the current production process deviates from the expected target.

[0124] The predicted quality parameters of different historical growth stages refer to the quality parameter data sequence recorded and stored by the computing device at each completed or ongoing production stage of the target product before the current time.

[0125] Specifically, the specific form of the quality evaluation parameter threshold depends on the specific form of the quality evaluation parameter.

[0126] For example, if the quality evaluation parameter is represented in the form of grades, for example, the grade sequence includes "excellent", "good", "medium", and "poor", the preset quality evaluation parameter threshold can be specifically set to "good". Alternatively, if the quality evaluation parameter is represented in the form of scores (such as percentage or percentage numerical form), the preset quality evaluation parameter threshold can be specifically set to a specific score, for example, 60 points or 60%. The above enumeration of specific values of the quality evaluation parameter threshold is only an example, and the specific setting mode and value of the preset quality evaluation parameter threshold are not limited in the embodiments of the present application. The actual business scene, management requirement strategy can be flexibly configured by the person skilled in the art.

[0127] In one possible implementation, the computing device retrieves snapshot data of the target product at the end of each production stage in the past in chronological order according to the identification information of the target user, the target plot, and the target product, to obtain multiple quality parameters of the target product at different production stages.

[0128] For example, in an agricultural planting scene, assuming that the obtained growth stage of the crop this time is the quality evaluation parameter of the grain filling stage, if the quality evaluation parameter of this stage is lower than the preset quality evaluation parameter threshold, the predicted quality parameters of the crop at the sowing stage and the jointing stage are obtained based on the crop identification, the target object identification, and the identification of the target user.

[0129] For another example, in an industrial manufacturing scene, the computing device retrieves the predicted quality parameter data saved at the end of the stamping stage, the end of the welding stage, and the end of the coating stage of the batch of products.

[0130] S402, based on the multiple quality parameters, generating a quality change trend chart of the target product at different production stages.

[0131] The quality parameter change trend chart is a kind of visual representation for intuitively reflecting the dynamic change of data. The horizontal axis is time or production stage sequence, and the vertical axis is the numerical value of the quality parameter. Through line charts and other forms, the trend of the quality parameter in the entire production cycle is clearly displayed.

[0132] Specifically, the computing device integrates the historical sequence data obtained in S501 with the quality parameter at the current time and arranges them in chronological order. According to the graph generation method, a trend chart of the parameter change over time is drawn. This chart can effectively reveal whether the quality parameter is steadily increasing, stably maintaining, or showing unexpected downward trend, etc.

[0133] In some embodiments, the graph generation method includes at least one of the following: The basic line chart generation method: the quality parameter values corresponding to each time point or production stage are taken as data points, which are connected in sequence according to time order to form a continuous trend line. This method is suitable for displaying the overall change trend and key turning points of the parameter.

[0134] The smooth curve fitting method: based on the data sequence, a preset curve fitting algorithm (such as polynomial fitting or spline interpolation algorithm) is used to generate a smooth trend curve, so as to more clearly display the macro trend of parameter change and reduce the visual interference caused by short-term fluctuations.

[0135] The multi-dimensional superimposed display method: on the basis of generating the basic trend chart, key environmental data (such as average temperature, cumulative precipitation) or important events (such as fertilization, irrigation, equipment maintenance) are superimposed on the trend chart in different forms of marks (such as color blocks, vertical line marks), establishing a visual association between parameter change and influencing factors.

[0136] It should be noted that the computing device selects and executes one or more of the above-mentioned graph generation methods according to the preset configuration or the specific application scene requirement, finally renders and generates the quality parameter change trend chart, and provides intuitive data support for the subsequent analysis steps.

[0137] S403, based on the quality change trend chart, the object attribute information and the product attribute information, generating the quality early warning information of the target product.

[0138] The performance quality early warning information is a structured text or signal output generated by the computing device after comprehensive analysis, aiming to describe the current situation, analyze the potential reasons and provide decision reference.

[0139] In some embodiments, the computing device can integrate and analyze the trend chart features, object attribute information and product attribute information through an integrated large language model to identify key influence patterns and generate prompt information in natural language description.

[0140] In one possible implementation, specifically, the computing device converts the multi-modal input data into a prompt that can be processed by a large language model. The prompt template includes: a trend feature description, a structured description extracted from a trend graph (e.g., a 15% drop in quality parameters during the irrigation period). Key object attributes: such as "soil type is sandy loam, water retention rating is medium". Related product attributes: such as "during the period when the drop occurs, the precipitation is recorded for 10 consecutive days below the historical same period average, and no irrigation operation is performed". Then, the computing device inputs the above structured prompt into a large language model pre-trained on professional corpus (data knowledge corresponding to the vertical field, such as agricultural knowledge, industrial production procedures). Based on its powerful context understanding and logical reasoning ability, the model performs fusion analysis on the input information, identifies the core factors causing the trend change, and generates an easy-to-understand natural language conclusion.

[0141] For example, based on the above input, the large language model may generate the following warning information on the quality of performance: "the system detects that the quality parameters of the target product in the irrigation period have a significant downward trend. Analysis shows that this trend is mainly related to insufficient precipitation during the same period and limited water retention capacity of the land, which cannot be supplemented by timely irrigation. It is recommended to immediately evaluate the soil moisture content and start the emergency irrigation plan to alleviate the potential impact of drought stress on the final yield."

[0142] In other embodiments, in addition to the quality change trend graph, the computing device can also generate an instruction response analysis graph based on a pre-set production management strategy to reflect the change in the degree of control of the production process of the target product. The pre-set production management strategy includes a set of standard operating procedures predefined according to the characteristics of the production stage, and the instruction response analysis graph quantitatively shows the standardization execution level of the production process by comparing the matching degree of the actual execution operation with the standard operating procedure.

[0143] It should be understood that this scheme realizes dynamic monitoring and risk warning of the resource return ability by monitoring the quality evaluation parameters, tracing the historical quality data, visualizing the trend change, and generating a warning combined with multi-source information, which provides a reliable time sequence basis for timely intervention and accurate decision-making.

[0144] The training method of the preset parameter model is described in detail below.

[0145] In some embodiments, the training process of the computing device for the parameter model includes the following steps as shown in Figure 7 S501, obtaining a training sample set.

[0146] ​The training sample set includes a plurality of training samples. Each training sample includes object attribute information of a sample object, product attribute information of a sample product associated with the sample object, and a quality parameter label of the sample product.

[0147] Specifically, the specific description of the object attribute information of the sample object and the sample attribute information of the sample product can refer to S101, which will not be described in detail here.

[0148] In some embodiments, the production data of the sample product in the training sample set at multiple production stages can be regarded as the time series data record of the product at all production stages in its complete production cycle, and these data collectively constitute a complete production history information set.

[0149] Specifically, for a sample product that has completed production and has been labeled with a final quality parameter label, the production data collected at the end of each specific production stage (such as the Kth stage, K = 1, 2,..., N) will form a training sub-sample with the same final quality parameter label. This means that from a single complete production cycle, multiple training sub-samples can be extracted, each sub-sample recording all historical and current production data up to the end of its corresponding production stage and being associated with the final yield result.

[0150] This way of constructing training samples ensures that the final trained parameter model can not only process data of all completed production stages, but more importantly, can make accurate predictions of the final yield based on incomplete, ongoing production process data, thereby meeting the demand for forward-looking assessment of mid-production projects in actual applications.

[0151] In some embodiments, referring to the method of feature construction of attribute information in S201, the computing device can also perform feature construction on the attribute information in the training sample during the model training process to obtain the feature information of each training sample. And the attribute information of the sample object and the sample product of each training sample and the feature information are taken as the training sample set. The acquisition of the feature information can refer to the above Figure 4 , which will not be described in detail here.

[0152] S502, model training is performed on the initial model of the parameter model based on the training sample set to obtain the parameter model.

[0153] The initial model of the parameter model is a gradient boosting machine model.

[0154] Specifically, the gradient boosting machine is a machine learning algorithm based on ensemble learning, and its core idea is to combine a plurality of weak parameter models (usually decision trees) to construct a strong parameter model.

[0155] Specifically, the gradient boosting machine model trains a series of weak decision trees in an iterative manner, where each subsequent tree aims to correct the prediction error of the previous tree. In each iteration, the algorithm calculates the negative gradient (i.e., residual) between the current model's predicted value and the true quality parameter label, and then trains a new decision tree to fit this negative gradient. The final model's prediction result is the weighted sum of all decision tree prediction results.

[0156] In one possible implementation, the process of model training an initial model of a parameter model based on a training sample set is as follows: First, the training sample set is preprocessed to construct the feature matrix and label vector required for model training. The computing device extracts and fuses the object attribute information and product attribute information in each training sample to form a corresponding feature vector; and takes the quality parameter label of the sample product as the target value of supervised learning to form a label vector.

[0157] Then, the gradient boosting machine model is iteratively optimized using a forward step-by-step algorithm and a gradient descent strategy. Specifically, it includes: Initialize the base predictor, usually set to a constant value that minimizes the loss function.

[0158] For each iteration (m = 1 to M): a. Calculate the negative gradient of the current model for all training samples, i.e., pseudo-residual.

[0159] b. Use the pseudo-residual calculated in the previous step as the new target value to fit a new regression decision tree.

[0160] c. Determine the optimal output value of the decision tree leaf node through linear search to minimize the overall loss function.

[0161] d. Update the trained decision tree to the current model with a certain learning rate (shrinkage rate). Finally, Finally, when the model training reaches the model training stopping condition, the training process is terminated. The final integrated strong predictor is output as the trained parameter model, which can output the quality parameter of the target product based on the input object attribute information and product attribute information.

[0162] In some embodiments, the model training stopping condition includes reaching a preset number of iterations, or the performance of the model on an independent validation set meets a preset convergence standard.

[0163] Specifically, the convergence criterion can be determined by at least one of the following: early stopping mechanism triggering: when the evaluation index (such as mean absolute error) of the model on the validation set no longer improves or the improvement is lower than the preset threshold Δ in consecutive P rounds (P≥1) of iterations, it is determined that the model has converged. Loss function convergence: when the decline amplitude of the training loss function value in consecutive Q rounds (Q≥1) of iterations is lower than the preset threshold ε, it is determined that the model training has reached a stable state. Performance meets the standard: when the evaluation index of the model on the validation set reaches the preset target performance threshold η, it is directly determined that the training is completed.

[0164] It should be understood that the scheme provides a comprehensive and real data basis for model learning by constructing a training sample set containing sample object attributes, sample product full-cycle production data, and real quality parameter labels. By using the feature learning and nonlinear fitting capability of the gradient boosting machine model, the mapping relationship between the complex time series dynamic characteristics in the production process and the quality parameters is deeply mined, so that the trained parameter model has the ability to capture the comprehensive influence mechanism of different production factors on product quality and accurately predict the final quality level of the product under specific production conditions.

[0165] In some embodiments, due to the high dimensionality of the feature parameters involved in the model training and the complex correlation between each feature and the revenue parameter, in order to comprehensively and stably evaluate the comprehensive classification performance of the model, as shown in Figure 8 S502 specifically includes the following steps: S601, obtaining a preset number of test sample sets from the training sample set.

[0166] The test sample set is independent of the training subset and the validation subset used for model training, and is used for the final unbiased evaluation of the trained model.

[0167] In some embodiments, the computing device can use stratified sampling or random sampling to reserve a portion of data (a preset number) from the original training sample set as a test sample set for subsequent model evaluation. The test sample set does not participate in parameter updating during model training to ensure the objectivity of the evaluation results.

[0168] For example, the preset number can be determined according to the total sample size by a preset proportion (such as 15%, 18%, or 20% of the total sample size). The embodiments of the present application do not limit the preset number.

[0169] S602, performing model training on the initial model based on the training sample set to obtain a parameter model to be evaluated.

[0170] This step can refer to the training method described in the foregoing S502, which will not be described in detail here.

[0171] S603, perform model testing on the parameter model to be evaluated based on the test sample set, to obtain a model evaluation index of the parameter model to be evaluated.

[0172] The model evaluation index includes an area under a receiver operating characteristic curve (ROC-AUC).

[0173] It should be noted that in the embodiments of the present application, the ROC-AUC index is used for a specific and non-traditional binary classification problem to evaluate the performance of the model in distinguishing between “high yield potential” and “low yield potential” production states. The specific construction and interpretation method is as follows: Specifically, since there are many feature parameters involved in the training (including original features and constructed derivative features), the model needs to learn a complex decision boundary in a high-dimensional feature space. In this case, the ROC-AUC index is not sensitive to the classification threshold, and can comprehensively reflect the overall performance of the model under different discrimination thresholds, that is, it can comprehensively measure the ability of the model to distinguish “high yield” samples from “low yield” samples. Therefore, it becomes the preferred index for evaluating the generalization performance of the model. This avoids the evaluation bias that may be caused by using a single threshold index (such as accuracy), and ensures the comprehensiveness and robustness of the evaluation results.

[0174] In some embodiments, the model evaluation index further includes at least one of the following: Precision: used to measure the proportion of truly “high yield potential” samples among all samples predicted by the model as “high yield potential”, focusing on evaluating the prediction accuracy of the model.

[0175] Recall: used to measure the proportion of samples successfully predicted by the model among all truly “high yield potential” samples, focusing on evaluating the discovery ability of the model for high-quality projects.

[0176] F1 score: is the harmonic mean of precision and recall, used to comprehensively evaluate the balance performance of the model between the two.

[0177] Mean squared error: directly measures the average squared difference between the predicted yield parameter value of the model and the true quality parameter label, used to evaluate the accuracy of regression prediction.

[0178] It should be understood that the embodiments can more comprehensively evaluate the performance of the model from different dimensions by combining multiple evaluation indexes, avoiding the limitations of a single index.

[0179] S604, adjust the model parameters in the parameter model to be evaluated based on the preset model evaluation index-model parameter adjustment rule and the model evaluation index, to obtain a preset parameter model.

[0180] The preset model evaluation index-model parameter adjustment rule refers to a pre-defined strategy set for guiding how to adjust a specific model parameter according to a value of a specific model evaluation index.

[0181] The preset model evaluation index-model parameter adjustment rule establishes a quantitative corresponding relationship between the evaluation result and the parameter optimization direction.

[0182] For example, the rule can be embodied as a series of "condition-action" mapping pairs. Specifically, the rule can be embodied as: Condition 1: If the ROC-AUC value of the model is less than 0.75.

[0183] Action 1: It is determined that the model has insufficient discrimination ability, and a structural enhancement strategy is triggered, for example, the number of decision trees is increased by 100 based on the original number, and the maximum depth of the decision tree is increased by 3 layers based on the original depth, so as to improve the ability of the model to capture complex patterns.

[0184] Condition 2: If the ROC-AUC value of the model is between 0.75 and 0.9.

[0185] Action 2: It is determined that the model has good foundation but needs to be optimized, and a fine tuning strategy is triggered, for example, the learning rate is reduced by half (for example, from 0.1 to 0.05), and the feature sampling ratio is fine tuned, so as to further improve the performance.

[0186] Condition 3: If the ROC-AUC value of the model is greater than 0.9.

[0187] Action 3: It is determined that the model has very excellent discrimination ability, and the adjustment target is shifted to generalization enhancement, for example, the minimum number of samples required for a leaf node is increased or the maximum depth of the tree is reduced, so as to further enhance the anti-overfitting ability of the model at the cost of a small amount of training set accuracy.

[0188] The model parameter refers to a configuration variable, i.e., a hyperparameter, which needs to be pre-set before the model training starts and will significantly affect the model structure, the training process or the final performance.

[0189] For example, the number of decision trees is used to control the complexity and fitting ability of the model. The maximum depth of the decision tree is used to control the complexity of a single tree and balance the fitting and generalization ability of the model. The learning rate is used to control the contribution of each tree to the final model and affects the stability and convergence speed of the training. The minimum number of samples required for a leaf node is used for regularization to prevent overfitting. The feature sampling ratio is used to introduce randomness and enhance the robustness and generalization ability of the model. The specific types of model parameters are not limited in the embodiments of the present application.

[0190] In some embodiments, the preset model evaluation index-model parameter adjustment rule can be determined according to the output result of an automatic hyperparameter tuning algorithm such as Bayesian optimization, grid search or random search. Specifically, the computing device takes the model evaluation index as the optimization target, and through multiple rounds of iterative sampling and evaluation in a preset model parameter value space, finds a parameter combination that optimizes the model evaluation index, thereby completing the final adjustment of the parameter model.

[0191] It should be understood that by using a preset test sample set for model evaluation, and in view of the characteristics of production data such as multiple dimensions and uneven sample distribution, selecting the ROC-AUC index which is not sensitive to class imbalance as the evaluation standard, and iteratively optimizing the model according to the parameter adjustment rule, the model evaluation bias caused by uneven samples is effectively reduced, and the model has good generalization ability and robustness in real scenarios.

[0192] Figure 9 A structural schematic diagram of a parameter determination apparatus provided by an embodiment of the present application is shown in FIG. 1. As shown in the figure, the parameter determination apparatus includes an acquisition module 901 and a processing module 902. Figure 9

[0193] The acquisition module 901 is configured to acquire object attribute information of a target object and product attribute information of a target product associated with the target object. The target object is a production facility and a production tool in a production activity. The object attribute information includes geometric parameters of the target object, a geographic position of the target object and property characteristics of the target object. The product attribute information includes a category identifier of the target product and production data of the target product in multiple production stages. The production data of each production stage includes a current production stage identifier of the target product, production state data of the target product, production records of the target product, environment data corresponding to the target product and an abnormal event of the target product.

[0194] The processing module 902 processes the object attribute information and the product attribute information by using a preset parameter model to obtain a quality parameter of the target product. The preset parameter model has a function of obtaining the quality parameter of the target product based on the object attribute information and the product attribute information. Based on a preset evaluation rule and the quality parameter of the target product, a quality evaluation parameter of a target user is determined.

[0195] In other embodiments, the target object includes a production facility in the field of agriculture, and the production facility includes farmland, a greenhouse or a breeding pond.

[0196] ​In some embodiments, the production state data of the target product comprises: an appearance form parameter of the target product, a product weight parameter of the target product, and a production rate of the target product. The production record of the target product comprises: a production material input record of the target product, a production material addition record of the target product, and a production environment maintenance record of the target product. The environment data corresponding to the target product comprises: temperature data, weather data, precipitation data, and illumination data. The abnormal event of the target product comprises: an abnormal event type, an abnormal event processing method, and an abnormal event response time.

[0197] In some embodiments, the processing module 902 is specifically configured to: perform feature construction based on the object attribute information of the target object and the product attribute information of the target product, to obtain feature information. The feature information comprises: a feature parameter for quantifying the deviation degree of the product production state data from the baseline production state data, a feature parameter for quantifying the deviation degree of the environment data from the baseline environment data, and a feature parameter for quantifying the abnormal event occurrence frequency of the sample product. The quality parameter of the target product is obtained by processing the object attribute information, the product attribute information, and the feature information through a preset parameter model.

[0198] In some embodiments, the processing module 902 is specifically configured to determine the resource conversion amount of the target product based on a preset unit mass and unit resource amount conversion rule corresponding to the target product. Based on a preset evaluation rule, the resource conversion amount of the target product and the to-be-returned resource amount of the target object, a quality evaluation parameter of a target user to which the target object belongs is determined for the to-be-returned resource quota. The to-be-returned resource amount is a resource amount that is pre-configured for the target user by a resource provision contract and is agreed to be returned through target product output.

[0199] In some embodiments, the processing module 902 is further configured to, in a case where the quality evaluation parameter is lower than a preset quality evaluation parameter threshold, obtain a plurality of quality parameters of the target product at different production stages; generate a quality change trend graph of the target product at different production stages based on the plurality of quality parameters; and generate quality warning information of the target product based on the quality change trend graph, the object attribute information, and the product attribute information.

[0200] In some embodiments, the processing module 902 is further configured to obtain a training sample set. The training sample set comprises a plurality of training samples. Each training sample comprises object attribute information of a sample object, product attribute information of a sample product associated with the sample object, and a quality parameter label of the sample product. An initial model of a preset parameter model is trained based on the training sample set to obtain the preset parameter model. The initial model of the preset parameter model is a gradient boosting machine model.

[0201] In some embodiments, the processing module 902 is further configured to obtain a preset number of test sample sets from the training sample set. The initial model is trained based on the training sample set to obtain a parameter model to be evaluated. The parameter model to be evaluated is tested based on the test sample set to obtain a model evaluation index of the parameter model to be evaluated. The model evaluation index includes a receiver operating characteristic curve area (ROC-AUC). The model parameters in the parameter model to be evaluated are adjusted based on a preset model evaluation index-model parameter adjustment rule and the model evaluation index to obtain a preset parameter model.

[0202] The parameter determination apparatus provided by the embodiments of the present application can perform the method shown in the parameter determination method embodiments, and the implementation principles and advantages thereof can refer to the related description in the method embodiments, which will not be repeated here.

[0203] Figure 10 A structural schematic diagram of an electronic device is provided in the embodiments of the present application. As shown in the figure, the electronic device includes a memory 1001, a transceiver 1002, and at least one processor 1003. Figure 10

[0204] The transceiver 1002 is used to interact with other devices to realize the transceiving of data. For example, in the embodiments of the present application, the transceiver 1002 can be specifically used to obtain object attribute information of a target object and product attribute information of a target product associated with the target object, send performance quality early warning information of a target user for the to-be-returned resource quota, and obtain a training sample set.

[0205] The memory 1001 is used to store computer program code, which includes computer instructions. The computer instructions run in the above-mentioned electronic device to implement the method shown in the above-mentioned method embodiments. For example, the memory can include a high-speed random access memory (RAM), and can also include a non-volatile memory (NVM), such as at least one disk memory, and can also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0206] ​The processor 1003 can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The processor 1003 can also be other general processors. The general processor can be a microprocessor or the processor can also be any conventional processor.

[0207] The memory 1001, the transceiver 1002 and the processor 1003 are communicatively connected. For example, the memory 1001, the transceiver 1002 can be connected with the processor 1003 through a system bus and complete mutual communication. The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0208] Optionally, the memory 1001 can be independent or integrated with the processor 1003. When the memory 1001 is independently arranged, the memory 1001 and the processor 1003 are connected through a system bus.

[0209] The embodiment of the present application further provides a chip for running instructions, which is used for executing the technical solution of the parameter determination method in the above embodiment.

[0210] The embodiment of the present application further provides a computer readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the technical solution of the parameter determination method in the above embodiment is implemented. Specifically, when the computer instructions are executed by the processor, the above electronic device can execute the technical solution of the parameter determination method in the above embodiment.

[0211] The embodiment of the present application further provides a computer program product, which comprises a computer program stored in a computer readable storage medium, at least one processor can read the computer program from the computer readable storage medium, and the at least one processor can implement the technical solution of the parameter determination method in the above embodiment when executing the computer program.

[0212] The computer readable storage medium can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The computer readable storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0213] An exemplary computer readable storage medium is coupled to the processor, so that the processor can read information from the storage medium and write information to the computer readable storage medium. Of course, the computer readable storage medium can also be an integral part of the processor. The processor and the computer readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the computer readable storage medium can also exist as discrete components in an electronic control unit or a host device, and the embodiment of the present application does not limit this.

[0214] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed ones can be indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0215] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to implement the embodiments of the present application.

[0216] In addition, the functional modules in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The unit formed by the above modules can be realized in the form of hardware or in the form of hardware plus software functional units.

[0217] The integrated modules realized in the form of software functional modules can be stored in a computer readable storage medium. The software functional modules stored in a storage medium include a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute part of the steps of the method of each embodiment of the present application.

[0218] It should be understood that the steps of the method disclosed in the embodiments of the present application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0219] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The aforementioned program can be stored in a computer readable storage medium. The program, when executed, executes steps including the above-mentioned method embodiments; and the aforementioned storage medium includes ROM, RAM, magnetic or optical disc, and various media that can store program codes.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or make equivalent replacements for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for determining parameters, characterized in that, include: Obtain the object attribute information of the target object and the product attribute information of the target product associated with the target object; The target objects are production facilities and tools used in production activities; The object attribute information includes: the geometric parameters of the target object, the geographical location of the target object, and the property characteristics of the target object; the product attribute information includes: the category identifier of the target product, and the production data of the target product in multiple production stages; the production data of each production stage includes: the current production stage identifier of the target product, the production status data of the target product, the production record of the target product, the environmental data corresponding to the target product, and the abnormal events of the target product; The object attribute information and the product attribute information are processed by a preset parameter model to obtain the quality parameters of the target product; the preset parameter model has the function of obtaining product quality parameters based on object attribute information and product attribute information. Based on the preset evaluation rules and the product quality parameters, the quality evaluation parameters are determined.

2. The method according to claim 1, characterized in that, The target objects include agricultural production facilities, including arable land, greenhouses, or aquaculture ponds.

3. The method according to claim 1, characterized in that, The production status data of the target product includes: the appearance parameters of the target product, the weight parameters of the target product, and the production rate of the target product. The production records of the target product include: the production material input records of the target product, the production material replenishment records of the target product, and the production environment maintenance records of the target product. The environmental data corresponding to the target product includes: temperature data, weather data, precipitation data, and light data; The abnormal events of the target product include: abnormal event type, abnormal event handling method, and abnormal event response time.

4. The method according to claim 1, characterized in that, The process of processing the object attribute information and the product attribute information through a preset parameter model, the quality parameters of the target product include: Feature information is obtained by constructing features based on the object attribute information of the target object and the product attribute information of the target product. The feature information includes: feature parameters for quantifying the degree of deviation between product production status data and benchmark production status data, feature parameters for quantifying the degree of deviation between environmental data and benchmark environmental data, and feature parameters for quantifying the frequency of abnormal events occurring in the sample product. The quality parameters of the target product are obtained by processing the object attribute information, the product attribute information, and the feature information through a preset parameter model.

5. The method according to claim 1, characterized in that, The process of determining the quality assessment parameters of the target product based on preset assessment rules and the product quality parameters includes: The resource conversion amount of the target product is determined based on the preset unit mass and unit resource quantity conversion rules corresponding to the target product. Based on preset evaluation rules, the resource conversion amount of the target product and the amount of resources to be returned to the target object, the quality evaluation parameters of the target user to which the target object belongs for the resource quota to be returned are determined; the amount of resources to be returned is the amount of resources pre-configured for the target user based on the resource provision contract and agreed to be returned through the output of the target product.

6. The method according to claim 1, characterized in that, The method further includes: If the quality assessment parameter is lower than the preset quality assessment parameter threshold, obtain multiple quality parameters of the target product at different production stages; Based on the multiple quality parameters, a quality change trend chart of the target product at different production stages is generated; Based on the quality change trend chart, the object attribute information, and the product attribute information, a quality early warning information for the target product is generated.

7. The method according to any one of claims 1-6, characterized in that, The preset parameter model is obtained in the following way: Obtain a training sample set; the training sample set includes multiple training samples; each training sample includes object attribute information of a sample object, product attribute information of a sample product associated with the sample object, and quality parameter labels of the sample product; The initial model of the preset parameter model is trained based on the training sample set to obtain the preset parameter model; the initial model of the preset parameter model is a gradient boosting machine model.

8. The method according to claim 7, characterized in that, The model training includes: Obtain a preset number of test samples from the training sample set; The initial model is trained based on the training sample set to obtain the parameter model to be evaluated; The parameter model to be evaluated is tested based on the test sample set to obtain the model evaluation index of the parameter model to be evaluated; the model evaluation index includes: area under the receiver operating characteristic curve (ROC-AUC); Based on the preset model evaluation index - model parameter adjustment rules and the model evaluation index, the model parameters in the parameter model to be evaluated are adjusted to obtain the preset parameter model.

9. A parameter determining device, characterized in that, include: The acquisition module is used to acquire object attribute information of the target object and product attribute information of the target product associated with the target object; The target objects are production facilities and tools used in production activities; The object attribute information includes: the geometric parameters of the target object, the geographical location of the target object, and the property characteristics of the target object; the product attribute information includes: the category identifier of the target product, and the production data of the target product in multiple production stages; the production data of each production stage includes: the current production stage identifier of the target product, the production status data of the target product, the production record of the target product, the environmental data corresponding to the target product, and the abnormal events of the target product; The processing module processes the object attribute information and the product attribute information, as well as the quality parameters of the target product, through a preset parameter model; the preset parameter model has the function of obtaining product quality parameters based on object attribute information and product attribute information; and the quality evaluation parameters of the target user are determined based on preset evaluation rules and the product quality parameters.

10. An electronic device, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the electronic device performs the method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-8.

12. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-8.

Images