Central intelligent kitchen full life cycle operation optimization method based on digital twinning

Abstract

The invention provides a central intelligent kitchen full life cycle operation optimization method based on digital twinning, and relates to the technical field of intelligent kitchens, and the method comprises the steps: building a self-correction twinning model, and achieving the dynamic self-adaption of model parameters and structures; based on a multi-objective collaborative optimization algorithm, performing joint optimization on energy consumption, productivity and quality indexes; setting a trigger control logic in combination with the prediction deviation, the constraint conflict and the energy price change; a federal aggregation mechanism is adopted to realize unified updating of parameters and strategies among multiple sites, a global optimal strategy model is finally formed, and a collaborative optimal control scheme throughout construction, operation and maintenance stages is generated, so that dynamic optimal operation and full-life-cycle energy efficiency improvement of the central intelligent kitchen are realized. According to the method, the problems of staticizing of a digital twin model, simplification of an optimization target and insufficient multi-site collaboration in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent kitchen, in particular to a central intelligent kitchen full life cycle operation optimization method based on digital twinning. BACKGROUND

[0002] At present, central kitchen and cloud kitchen have realized a high degree of informatization management in food processing and supply chain management. Most enterprises have deployed enterprise resource planning systems, warehouse and production execution systems, and completed the basic process digitization from planning, material preparation, processing to distribution. At the same time, the automation level of kitchen equipment is continuously improved, and various sensors realize the collection of key process data such as temperature, humidity and energy consumption. However, existing intelligent control relies mainly on fixed thresholds or artificial experience rules, and digital twinning technology is mainly applied to static simulation of single equipment or local process, which is difficult to realize dynamic coordination across links and the whole process. In terms of energy consumption management and food safety, there is a common problem of information island, and the energy management system and the food safety monitoring system lack data connection, resulting in scattered decision-making and lagging optimization.

[0003] With the development trend of individualized consumer demand, time-efficient distribution and business scale, the operation mode of central kitchen is evolving towards network and flexibility. The interrelation between production links and energy systems is becoming closer, and time-of-use electricity price, energy storage scheduling and carbon emission constraints have become important factors affecting operating costs. At the same time, multi-site distributed central kitchen needs to realize the coordinated optimization of process, energy consumption and safety models under the premise of ensuring data security and privacy compliance. Various artificial intelligence and machine learning technologies are gradually applied to production prediction, equipment maintenance and energy consumption analysis, but most of them still remain in the stage of offline modeling or single-objective optimization, and have not yet formed an adaptive decision-making system under complex constraints. The real-time requirement of regulatory authorities for food safety traceability and process control is also increasing, which puts higher standards on the intelligent level of kitchen operation.

[0004] However, the existing technology still has many shortcomings. First, traditional digital twinning models are mostly static structures, which are difficult to dynamically correct with equipment aging, demand mutation or energy price changes, resulting in lagging strategy response. Second, most optimization systems use fixed weights or single-objective criteria, lack the ability to dynamically adjust the focus of the target based on risk and constraint conditions, and it is difficult to achieve overall balance between energy consumption, capacity, cost and food safety. Thirdly, in the face of multi-site distributed operation environment, data privacy and system heterogeneity make it impossible for each kitchen to effectively share decision-making knowledge, and the degree of collaborative optimization is limited. These problems make it difficult for central kitchen to achieve continuous multi-objective collaborative optimization in the full life cycle operation, and there is an urgent need for a new optimization method that combines digital twinning, self-evolution learning and federal collaborative decision-making to improve the intelligence, adaptability and overall operating efficiency of the system. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins. This invention solves the problems of static digital twin models, singular optimization objectives, and insufficient multi-site collaboration in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins, comprising: Acquire multi-source data from the central smart kitchen throughout the entire production and operation process and construct a digital twin basic model based on the multi-source data. The entire production and operation process includes: planning, raw material processing, warehousing management and food delivery. The digital twin basic model is calibrated online to obtain the digital twin dynamic model; The prediction error, confidence interval, and uncertainty index of the digital twin dynamic model are calculated to obtain the uncertainty assessment results; Based on the uncertainty assessment results, a virtual-real consistency analysis is performed on the digital twin dynamic model to obtain the deviation results between the simulation results and the actual operating state of the digital twin dynamic model. Determine whether the deviation result exceeds a preset threshold. If so, adaptively adjust the parameters and structure of the digital twin dynamic model to obtain a self-correcting twin model. The optimization objectives are determined and a multi-objective collaborative optimization scheme is obtained based on the self-calibrating twin model. The optimization objectives include: energy efficiency, production capacity, food safety, and operating costs. Based on the aforementioned multi-objective collaborative optimization scheme, the triggering conditions for prediction bias, constraint conflict, and energy price changes are set to obtain the operation control strategy; The operation control strategy is applied independently to the optimization process at multiple sites of the central smart kitchen, and the self-calibrating twin models and optimization results of each site are aggregated and uniformly updated through a federated collaboration mechanism to obtain the global optimal strategy model. Based on the global optimal strategy model, the optimal collaborative control scheme for the central smart kitchen throughout its entire lifecycle is determined.

[0007] The present invention discloses the following technical effects: The application provides a central intelligent kitchen full life cycle operation optimization method based on digital twinning, comprising: acquiring multi-source data of a central intelligent kitchen in a whole production and operation process and constructing a digital twinning basic model according to the multi-source data, wherein the whole production and operation process comprises planning, raw material processing, warehouse management and meal delivery; performing online parameter calibration on the digital twinning basic model to obtain a digital twinning dynamic model; calculating a prediction error, a confidence interval and an uncertainty index of the digital twinning dynamic model to obtain an uncertainty evaluation result; performing virtual-real consistency analysis on the digital twinning dynamic model according to the uncertainty evaluation result to obtain a deviation result between a simulation result of the digital twinning dynamic model and a real running state; judging whether the deviation result exceeds a preset threshold, and if yes, adaptively adjusting parameters and structures of the digital twinning dynamic model to obtain a self-correcting twinning model; determining an optimization target and obtaining a multi-objective collaborative optimization scheme according to the self-correcting twinning model, wherein the optimization target comprises energy consumption efficiency, production capacity, food safety and operation cost; based on the multi-objective collaborative optimization scheme, setting trigger conditions of prediction deviation, constraint conflict and energy price change to obtain a running control strategy; independently executing an optimization process by applying the running control strategy at sites of multiple central intelligent kitchens, and aggregating and uniformly updating self-correcting twinning models and optimization results of the sites through a federal collaborative mechanism to obtain a global optimal strategy model; and determining a collaborative optimal control scheme of the central intelligent kitchen in a full life cycle according to the global optimal strategy model. The application proposes a full-process dynamic optimization method based on digital twinning to solve problems such as high energy consumption, uneven production capacity distribution, scheduling response lag and virtual-real model deviation accumulation in the multi-site collaborative production of the central intelligent kitchen. By constructing a digital twinning basic model, a dynamic model, a self-correcting model and a global optimal strategy model, accurate mapping and real-time correction of production equipment state, energy consumption and task scheduling are realized. The method can continuously optimize virtual-real consistency without increasing manual parameter adjustment, so that the twinning system can adapt to changing environment and generate an optimal control strategy, thereby significantly improving the automatic decision-making level and response speed of the system, reducing energy consumption and manual intervention intensity, and realizing intelligent, efficient and stable operation of the central kitchen production process. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1A central intelligent kitchen full life cycle operation optimization method based on digital twinning is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0010] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0011] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0012] As shown in the drawings, Figure 1 The present application provides a central intelligent kitchen full life cycle operation optimization method based on digital twinning, which comprises: Step 100: acquiring multi-source data of a central intelligent kitchen in a whole production and operation process and constructing a digital twinning basic model according to the multi-source data, wherein the whole production and operation process comprises planning, raw material processing, warehouse management and meal delivery; Step 200: performing online parameter calibration on the digital twinning basic model to obtain a digital twinning dynamic model; Step 300: calculating the prediction error, confidence interval and uncertainty index of the digital twinning dynamic model to obtain an uncertainty evaluation result; Step 400: performing virtual-real consistency analysis on the digital twinning dynamic model according to the uncertainty evaluation result to obtain a deviation result between the simulation result of the digital twinning dynamic model and the real running state; Step 500: judging whether the deviation result exceeds a preset threshold, and if so, adaptively adjusting the parameters and structure of the digital twinning dynamic model to obtain a self-correcting twinning model; Step 600: determining an optimization target and obtaining a multi-objective collaborative optimization scheme according to the self-correcting twinning model, wherein the optimization target comprises energy efficiency, production capacity, food safety and operation cost; Step 700: setting trigger conditions of prediction deviation, constraint conflict and energy price change based on the multi-objective collaborative optimization scheme to obtain a running control strategy; Step 800: independently executing an optimization process by applying the running control strategy at multiple central intelligent kitchen sites, and aggregating and uniformly updating the self-correcting twinning model and optimization result of each site through a federal collaborative mechanism to obtain a global optimal strategy model; Step 900: determining a collaborative optimal control scheme of the central smart kitchen in the whole life cycle according to the global optimal strategy model.

[0013] Further, the specific implementation process of step 100 is as follows: The embodiment first acquires multi-source data in the whole process of production and operation of the central smart kitchen, takes the links of plan making, raw material processing, warehouse management and meal delivery as the collection objects, and forms a multi-dimensional data set in combination with equipment operation records, energy consumption collection terminals, raw material quality test results, environmental monitoring devices and task scheduling logs. The collection results are integrated through a unified time reference and data identification, the continuity and traceability of the data sources are ensured, and thus a complete original multi-source data set is obtained.

[0014] The embodiment performs fine processing on the multi-source data set, focuses on correcting multi-dimensional data noise, missing and abnormality, and extracts effective feature parameters that can represent the key state of the production process by using conventional data cleaning and feature extraction means. The extracted parameters cover core elements such as equipment operation characteristics, energy consumption fluctuation, raw material processing efficiency and task load relationship, form a cleaned feature data set that can comprehensively reflect the running state of the central smart kitchen, and realize the conversion of data from scattered sampling to structured expression.

[0015] The embodiment establishes a digital twin model structure representing the running state, energy consumption transmission and task allocation relationship based on the cleaned feature data set and in combination with the equipment operation mechanism and business logic of the central smart kitchen. The model structure is parameterized and fitted by calling historical running samples and field monitoring results, the calculation parameters are continuously adjusted to keep the virtual output consistent with the actual observation results, and finally a digital twin basic model is constructed, which can accurately map the actual kitchen production, reflect the dynamic process law and has verification reliability.

[0016] Specifically, after acquiring the multi-source data of the central smart kitchen and completing feature extraction, the embodiment realizes the construction of the digital twin basic model by establishing the correlation between production, energy consumption, equipment and logistics. The model takes real-time running data in the production process of the central kitchen as input, and comprehensively reflects the dynamic mapping relationship of equipment running state, energy consumption level, raw material flow characteristics and task load change law. The model construction process includes three main steps: first, time sequence matching of process state, equipment performance and energy consumption data is performed to determine the interaction influence relationship of each parameter in the production process; second, the change characteristics of raw material flow and warehouse capacity ratio in the logistics link are extracted to analyze their effect on production line stability; finally, the change rate of task load with time is introduced to represent the time-varying characteristics of the overall operation rhythm and resource allocation of the kitchen.

[0017] In the model, each parameter is derived from actual observation data or calculated physical quantities, ensuring the model has verifiability and physical meaning. Among them, the comprehensive response coefficient is used to measure the overall sensitivity of the production system to process, energy consumption and equipment performance, and its value is obtained by fitting historical operation data; the process coupling function reflects the nonlinear relationship between process state, energy input and equipment characteristics, which can be determined by establishing energy consumption-performance response curve and multiple regression; the comprehensive influence coefficient of logistics and inventory link is used to describe the influence of raw material inflow speed and inventory utilization rate on production plan execution, and its value is calibrated by logistics record data; the joint balance function expresses the matching degree of raw material supply and warehouse occupation, which is calculated by multi-time period monitoring data; the time dynamic factor represents the time sensitivity of task load to production scheduling, and its value range is determined by actual operation period; the system balance constant is used to quantify the offset caused by unobservable disturbances such as environmental changes and measurement deviations, to ensure that the virtual twin system is aligned with the actual running process.

[0018] Among them, the "process coupling function" refers to a mapping function reflecting the interaction between process state, energy input and equipment performance in the central kitchen production process, which is used to calculate the comprehensive energy efficiency; the "joint balance function" refers to a function describing the balance relationship between raw material flow and warehouse capacity, which is used to maintain the coordination of raw material supply and inventory management. Through the above design, the digital twin basic model established in this embodiment can quantify the dynamic relationship between production elements in a unified framework, realize accurate mapping and prediction of the real-time running state of the central intelligent kitchen, and provide reliable support for subsequent dynamic optimization and strategy decision-making.

[0019] In one specific embodiment, the comprehensive response coefficient of the production system is 0.82, which is obtained by fitting after analyzing the response delay and energy consumption fluctuation of different production processes, reflecting the comprehensive sensitivity of the overall process of the central kitchen to changes in input energy and equipment performance. The process coupling function is formed by process state, energy input and equipment performance parameters, among which the process state index is 0.75, the energy input is 105.6 kWh, and the equipment performance parameter is 0.93 after standardization, and the output is about 68.5 after nonlinear function calculation.

[0020] The comprehensive influence coefficient of logistics and inventory link is 0.65, representing the influence intensity of raw material circulation and storage state on production rhythm. The joint balance function output is calculated based on raw material flow and storage capacity proportion, wherein the raw material flow is 480 kg / h, the storage capacity proportion is 0.72, and the value after balance mapping is about 52.1. The time dynamic factor is set to 0.48, and the task load changes at a rate of 2.7 units / hour, which is used to depict the speed and dynamic response capability of task allocation adjustment. The system balance constant is 5.3, reflecting the average compensation deviation of measurement error, environmental fluctuation and unobservable disturbance. After substituting the above parameters into the model, the comprehensive representation output index of the digital twin basic model at this time is about 125.6, which can accurately reflect the running state of the central intelligent kitchen in the stable production stage.

[0021] Further, the specific implementation process of step 200 is: After establishing the digital twin basic model, in order to improve the adaptive ability of the model to real-time running changes, an online parameter calibration process is implemented to obtain a digital twin dynamic model. The specific implementation method is: during the operation of the central intelligent kitchen, real-time data such as equipment running state, environmental temperature and humidity, network communication delay and task load change are continuously collected, and the real-time data are compared and analyzed with the basic model output. The dynamic response difference of the model to time, environment and load disturbance is calculated by comparing the virtual and real deviations, so that the internal parameters of the model are progressively corrected, so that the model can reflect the time-evolving system behavior. In the online calibration process, an adaptive parameter updating strategy of sliding time window is adopted to dynamically capture the change trend of different time periods, ensuring that the model can keep consistent with the actual production state in real time.

[0022] In this process, the historical inheritance coefficient is derived from the output state of the base model, which describes the degree of retention of model parameters on historical behavior, and its value is set according to the system update frequency, between 0.7 and 0.9. The dynamic response coefficient is calculated by the time gradient change of the model output, which is used to adjust the response speed of the system to the sudden state, and its typical value is 0.3 to 0.5. The time sensitivity is generated by the change rate of the continuous output of the base model, and is determined according to the difference of real-time data sampling interval. The environmental interaction coefficient is obtained by regression fitting of environmental disturbance data on the stability of the production line, which is used to represent the comprehensive influence of external factors on the evolution of the virtual model, and its value is usually taken as 0.4 to 0.6. The external coupling function is composed of environmental conditions, network collaboration features and load disturbance intensity, among which environmental conditions include temperature, humidity, air pressure or air quality indicators, network collaboration features reflect data synchronization delay or communication bandwidth changes, and load disturbance intensity represents task density or parallel station variation degree. The virtual-real error compensation term is a correction amount established according to the deviation of model predicted value and field monitoring value, which is used to eliminate the instantaneous deviation caused by modeling approximation and data sampling error, and the value is generally dynamically adjusted over time.

[0023] wherein the "external coupling function" is a dynamic mapping function for establishing a nonlinear interaction relationship between the environment and the system. Its function is to quantify the influence of external fluctuations (such as temperature changes, communication delays or load peaks) as model correction inputs, so that the model can automatically adjust the prediction output according to the real-time changes of the environment. Through the synchronous execution of online parameter calibration and external coupling compensation, the digital twin dynamic model of the embodiment has real-time learning and rolling prediction capability, and can stably and accurately predict the energy consumption trend, equipment load change and production rhythm of the central intelligent kitchen in the future short period, ensuring that the virtual model and the actual operation process remain dynamically consistent.

[0024] Further, the specific implementation process of step 300 is: After obtaining the digital twin dynamic model, in order to evaluate its prediction reliability and model stability, the model output is analyzed for uncertainty. Specifically, the embodiment first obtains the simulation output data generated by the digital twin dynamic model within a certain time period, and at the same time, collects the measured data of the central intelligent kitchen in the same time period, including device running power, output, energy consumption level and task completion time, etc. Key production indicators. By establishing a unified time reference, the two types of data are accurately matched according to the time stamp to form a prediction-actual corresponding sample set. In the matching process, the missing data is linearly interpolated or completed by the median to ensure that the sample can accurately reflect the prediction performance of the model on the time sequence characteristics after alignment.

[0025] Based on the matched simulation output and the measured data sample, the model prediction error data is calculated. Specifically, by calculating the difference between the predicted value and the actual observation value at each time point or task cycle, an error vector is obtained, and the mean, variance and root mean square error of the error vector are calculated to measure the short-term and cumulative prediction deviation level of the model. Then, the error data is analyzed to determine whether it meets the normal or approximately normal distribution, thereby providing a statistical premise for confidence interval evaluation. For parts of the error distribution with significant skewness or abnormal fluctuations, the example uses quantile interval fitting method to correct the upper and lower limits of the error distribution, so that the calculation results have statistical robustness while ensuring accuracy.

[0026] The example further utilizes the method combining confidence interval estimation and variance decomposition to calculate the fluctuation range and sensitivity of each key variable, and then forms an uncertainty index set. The possible variation interval of the main process parameters in the model output is determined by the 95% confidence interval, and the variance decomposition technique is used to identify input factors that contribute more to the model error, such as environmental temperature, production intensity and equipment performance degradation. According to the uncertainty index set, the overall prediction stability score and credibility level of the model are calculated, and the prediction confidence is divided into three levels: stable, acceptable and need to be corrected. Finally, the uncertainty evaluation results output by the example can quantify the prediction accuracy and deviation range of the model under different operating conditions, providing a reliable basis for subsequent self-correction and global optimization strategy Further, the specific implementation process of step 400 is: After obtaining the uncertainty evaluation results, the example ensures that the digital twin dynamic model can accurately reflect the real operating state of the central intelligent kitchen by performing virtual-real consistency analysis on the model. Specifically, the example first extracts the simulation output of the digital twin dynamic model and the real operating data collected by the on-site monitoring system from the historical data, and aligns and normalizes the data according to a unified time scale. The normalization process converts data of different dimensions to the same numerical interval, thereby eliminating the influence of unit differences on subsequent comparisons. Then, combined with the uncertainty evaluation results obtained in the previous step, the observation samples with high confidence are selected as the input data for consistency analysis, ensuring that the analysis process is based on accurate and representative data.

[0027] On the basis of consistency analysis input data, the embodiment extracts multi-dimensional features of equipment running state, energy consumption change law and task scheduling execution, and constructs a feature comparison index set. Specifically, the index set includes equipment running efficiency deviation rate, unit time energy consumption deviation rate and task completion time difference value, etc., which are used to reflect the consistency degree of virtual simulation and actual operation from multiple angles. In the data processing process, the feature fluctuation sequence of different stages is calculated through time sliding window to identify the response deviation of the model in the running state conversion, task switching and energy consumption peak interval, so as to realize the extraction and quantitative description of dynamic consistency features.

[0028] According to the feature comparison index set, the embodiment adopts a difference vector calculation method to identify the difference pattern between the simulation output of the digital twin dynamic model and the actual running state. This method converts the deviation relationship between the virtual model output and the measured data into a quantifiable difference intensity and deviation trend by calculating the Euclidean distance and direction difference between the feature vectors, and then obtains the virtual-actual difference feature data. Finally, the deviation results of the model simulation results and the real running state are calculated according to the difference feature data, including overall deviation mean, peak deviation and trend deviation rate, etc. Through the above analysis steps, the embodiment can systematically identify the virtual-actual inconsistency source of the digital twin dynamic model, provide a direct basis for subsequent model adaptive optimization and parameter re-correction, and thus ensure the dynamic fidelity and decision reliability of the model in the long-term running process.

[0029] Further, the specific implementation process of step 500 is as follows: After completing the virtual-actual consistency analysis and obtaining the deviation results, the embodiment realizes the self-correction of the digital twin dynamic model through threshold determination and adaptive correction method, and obtains the self-corrected twin model. Specifically, the deviation results calculated from the model simulation results and the real running data are compared item by item, and when any key index (such as equipment running efficiency, energy consumption prediction value or task delay) deviation exceeds the preset threshold, the adaptive correction mechanism is started. This mechanism dynamically adjusts the weight parameters and structure mapping relationship in the model by comparing the real-time difference between the prediction output and the actual feedback signal, so that the model behavior converges to the real running state in a short time. The whole process is executed in a rolling time window mode to ensure the continuity of parameter correction and the real-time performance of model evolution.

[0030] In the self-correction process, the automatic correction rate is used to control the adjustment speed of the model when performing adaptive correction, and its value depends on the system response characteristics and the model stability requirements, generally between 0.2 and 0.6; the adaptive correction function is calculated according to the deviation and the nonlinear relationship of the model structure, which is used to automatically select the correction direction and amplitude, and can be realized by using multi-layer activation mapping or interpolation function, to correct the system error accumulated in complex environmental conditions; the dynamic matching coefficient reflects the influence of the change of the real running feedback signal on the model prediction update, and the value range is usually 0.4 to 0.8, which is set according to the sensitivity of device load fluctuation or energy consumption change; the time change rate of the real running feedback signal is derived from the data stream of the field sensor and the scheduling control system, which reflects the dynamic response speed and trend deviation direction in the actual production process, and is an important driving force for model self-correction.

[0031] The adaptive correction function is used to automatically select the optimal correction path according to the type and amplitude of the deviation characteristics, and its function is to map the complex error distribution to the model parameter adjustment amount that can be operated, so that the model can be self-recovered and optimized. The function generates a correction coefficient according to the size and direction of the input deviation signal during operation, and realizes the gradient update of the local structure weight. For example, when the device energy consumption prediction error continues to rise, the adaptive correction function will increase the correction weight of the energy consumption related parameters to reduce the cumulative error; when the load prediction error tends to be stable, its correction amplitude will automatically decay to maintain the system stability. Through the above multi-parameter linkage mechanism, the embodiment realizes the adaptive adjustment and self-correction of the digital twin dynamic model, so that the model maintains consistency and dynamic accuracy between virtual and real in long-term operation, and provides reliable model support for the continuous optimization of the central intelligent kitchen.

[0032] Further, the specific implementation process of step 600 is: After obtaining the self-correction twin model, in order to realize the efficiency and coordination of the central intelligent kitchen operation decision, the overall direction and target set of multi-objective optimization are first determined. Through comprehensive analysis of the operation requirements of the central kitchen, the embodiment clearly defines that the optimization targets not only include energy consumption reduction, but also cover production efficiency improvement and meal quality maintenance and other multi-dimensional requirements. Therefore, the embodiment defines three core optimization targets, i.e. minimizing unit energy consumption, maximizing device comprehensive utilization rate and optimizing meal processing quality stability. During the operation requirement analysis process, combined with historical production records and real-time operation data, the key factors affecting the above optimization targets are screened and quantified, thereby forming an optimization target set with actual constraint conditions. The target set can completely represent the balance relationship between production, energy consumption and quality of the central intelligent kitchen, and provide structured input for subsequent joint solution.

[0033] On the basis of determining the optimization target set, the embodiment extracts model variables and constraint conditions related to the target according to the dynamic output characteristics of the self-correcting twin model, thereby forming a model characteristic parameter set supporting the multi-target collaborative solution. The parameter set includes key variables such as production rate, equipment power load ratio, energy conversion efficiency, processing time fluctuation range, refrigeration and heating balance ratio, and task scheduling tightness. Each variable is obtained by jointly calculating the real-time simulation results of the twin model and the actual operation feedback to ensure the observability and operability of the model input. The constraint conditions are set according to the production equipment capacity of the central kitchen, the upper limit of energy consumption, the personnel operation capacity, and the food safety standards, forming an optimization boundary under the coupling of multiple constraints. Through this process, the embodiment establishes a parameterized optimization structure correlated with the self-correcting twin model, providing accurate data support for multi-target parallel computing.

[0034] Finally, based on the above model characteristic parameter set, the embodiment uses a multi-target collaborative optimization algorithm to perform parallel solution and weight balancing of energy consumption, production capacity, and quality indicators. Specifically, through a hierarchical solution strategy, the overall optimization process is divided into a local constraint optimization layer and a global trade-off decision layer: in the local optimization layer, the sub-optimal solutions are obtained for the minimum energy consumption, the maximum production capacity, and the quality stability indicators; in the global decision layer, the weight adjustment is performed according to the preset comprehensive benefit evaluation indicators (such as unit energy consumption cost, output time ratio, and quality deviation rate) to obtain a candidate set of collaborative solutions for multi-target optimization. Subsequently, the embodiment performs benefit sorting and sensitivity analysis on the candidate solution set to identify the optimal combination scheme for trade-off, and finally determines the multi-target collaborative optimization scheme for the operation decision of the central intelligent kitchen. Through this scheme, the central intelligent kitchen can achieve dynamic optimal operation under the conditions of controlled energy consumption, stable production capacity, and controllable quality.

[0035] Further, the specific implementation process of step 700 is as follows: After completing the solution of the multi-target collaborative optimization scheme, the embodiment constructs a multi-scene operation control strategy based on the scheme to ensure that the optimization results can be dynamically implemented under different operating conditions. Specifically, the embodiment first extracts key control variables and constraint coupling relationships from the optimal solution of the multi-target collaborative optimization scheme, performs sensitivity analysis on the three main indicators of energy consumption, production capacity, and quality, and identifies a set of key variables that are susceptible to fluctuations in the operating environment. Subsequently, combined with the prediction output variation law of the self-correcting twin model, the three types of potential disturbances, i.e., prediction deviation, constraint conflict, and energy price fluctuation, are identified, thereby forming a set of operation control trigger conditions. For example, when the energy consumption prediction value continuously deviates from the historical calibration curve, or when the task scheduling load exceeds the model set constraint range, it is identified as a trigger scene of the control trigger signal, and the subsequent control strategy adjustment logic is started.

[0036] Based on the identified set of operation control trigger conditions, the embodiment determines the threshold interval, response priority and corresponding weight parameters of each trigger condition to form a control trigger threshold and response weight configuration table. The threshold interval is determined according to the historical operation fluctuation range and the optimization target tolerance band, for example, the energy consumption deviation threshold is set to ±8%, the task delay threshold is set to ±5 minutes, and the energy price fluctuation threshold is set to ±10%. The determination of the response priority is divided into three levels according to the degree of influence on the operation state, that is, high, medium and low. The high-level event triggers immediate control action, the medium-level event starts the correction control instruction, and the low-level event is used as a prediction correction input to participate in subsequent scheduling optimization. The setting of the weight parameter is based on the principle of comprehensive balance of energy efficiency, production stability and economic benefit, and is obtained through empirical weight or machine learning model self-adjustment to ensure that the control response has hierarchy and stability under multiple factor trigger conditions.

[0037] The embodiment further establishes a mapping relationship from trigger condition to control action between the self-correcting twin model and real-time data stream based on the above-mentioned threshold and weight configuration table, and generates multi-scenario operation control logic in combination with rule decision and prediction control. The rule decision part realizes fast response according to the threshold and priority rules, and the prediction control part adjusts the upcoming operation state in advance through the future trend prediction output by the twin model. The combination of the two forms a self-learning and updating operation control strategy set. The embodiment verifies and evaluates the effect of the constructed control strategy in the digital twin simulation environment. The verification content includes response time, energy consumption improvement rate, operation stability and task completion accuracy. After evaluation, the optimal strategy set is selected to form the verified final operation control strategy, realizing dynamic collaborative control of energy consumption, production capacity and resources in the central intelligent kitchen under multiple scenarios.

[0038] Further, the specific implementation process of step 800 is as follows: After obtaining the verified operation control strategy, the embodiment deploys the same operation control logic at multiple kitchen sites and independently executes the optimization process to realize the overall optimization and strategy coordination of the cross-regional central intelligent kitchen. Specifically, the embodiment loads a self-correcting twin model and an operation control strategy module at each central kitchen site, and independently runs optimization calculation using local production data stream, environmental condition data and energy price information. The optimization objectives of each site remain the same, but the input data and operation constraints differ due to regional, climatic and task allocation differences. Through independent solution, the embodiment forms a local optimization result set at each site, including energy consumption efficiency, equipment utilization rate, process load balancing degree and quality maintenance indicators. The local optimization process is updated in a rolling period, and a feature output data package is automatically generated at the end of each period, which can be aggregated for federated coordination mechanism to provide input basis.

[0039] The embodiment further realizes data synchronization and knowledge sharing between sites through a federal collaborative mechanism without directly transmitting raw operation data, thereby protecting the independence and data security of each kitchen. When the federal collaborative mechanism is executed, firstly, the local optimization result set and the corresponding self-correcting twin model parameters uploaded by each site are received, and the synchronization aggregation of feature parameters and optimization indicators is completed through differential encryption or gradient compression technology. The synchronization content includes key features such as energy consumption model weight, device stability parameter, production rhythm adaptive coefficient and operation risk indicator. Through the aggregation analysis of the model features of each site, the embodiment forms a model parameter aggregation set required for global collaboration, which is used to represent the performance distribution and deviation trend in the overall network, and realizes effective sharing and dynamic fusion of parameters across sites.

[0040] Based on the formed model parameter aggregation set, the embodiment comprehensively calculates the parameter deviation and weight distribution of the self-correcting twin model of each site, and performs global aggregation update to obtain a global optimal strategy model. In the aggregation process, the embodiment adopts a double-layer fusion method of weighted average and deviation correction: in the first layer, the model parameters of different sites are integrated by weighted average, and the weight is determined according to the confidence score and data sample size of the optimization result of each site; in the second layer, the significantly deviated parameters in the aggregation result are updated smoothly to ensure the stability and consistency of the overall model. The finally formed global optimal strategy model can dynamically guide the operation scheduling of each central intelligent kitchen site, ensure energy economy and production balance, realize strategy unification, adaptive collaboration and continuous optimization in the whole network, and thus establish a digital twin decision-making system with global intelligent collaboration characteristics.

[0041] Specifically, after completing the optimization result and model parameter aggregation of each site under the federal collaborative mechanism, the embodiment forms a global optimal strategy model through multi-layer aggregation calculation to realize unified regulation and strategy decision of cross-regional central intelligent kitchens. Specifically, the embodiment takes the self-correcting twin model results output by each independent site at the same time step as the input basis, and combines and nonlinearly converts the model outputs of each site through a strategy mapping operator, thereby generating a comprehensive decision mapping result. The mapping process can automatically balance the production capacity difference between different sites, cooperatively adjust energy consumption allocation, and dynamically respond to global changes in the external environment. A global coordination coefficient is introduced in the aggregation process to adjust the influence proportion of local strategies in global decision-making, so that sites with higher weights have higher strategy power when the production capacity is concentrated or the energy consumption is abnormal, thereby realizing dynamic balance and optimal response of the whole network.

[0042] In this model, the aggregated global optimal strategy model is used to describe the unified control decision structure formed after multi-site collaborative optimization, and its core function is to achieve the overall optimization of cross-regional resource and task scheduling through dynamic strategy deduction. The number of central intelligent kitchen sites participating in federal collaboration is a variable parameter, which can be automatically updated according to the total number of connected nodes. The output of the self-correcting twin model of each site is derived from the predicted decision sequence after local optimization of the site, which usually includes equipment energy consumption distribution, yield prediction, and process state parameters. The value of the global coordination coefficient is calculated based on the performance of each site in the historical optimization period, which is used to control the fusion depth of different node information in the aggregation process.

[0043] The collaborative influence function is used to describe the comprehensive effect of external conditions on global strategy optimization, which consists of three elements: regional resource state, energy price fluctuation characteristics, and logistics scheduling load. The regional resource state reflects the comprehensive indicators of raw material supply, energy storage capacity, and human availability in the central kitchen location; the energy price fluctuation characteristics describe the change amplitude of electricity or gas cost in each region over time; and the logistics scheduling load represents the dynamic load level of order distribution density, material transportation pressure, and delivery time efficiency. Through the comprehensive modeling of the three types of influence factors, the collaborative influence function can dynamically correct the generation results of the global strategy model, achieving flexible resource allocation and energy optimal distribution across sites. The strategy mapping operator is used to map the above multi-dimensional input variables to the global optimal decision output, which is similar to a high-dimensional strategy space projection process, and can automatically identify the optimal collaborative path and output the strategy control solution with global adaptability. This embodiment realizes unified intelligent decision and continuous optimization of the digital twin network in a multi-site environment, enabling the system to maintain stable operation and efficient collaboration in complex environments.

[0044] More specifically, in the application scenario of the central intelligent kitchen group, this embodiment involves five collaborative sites located in different urban areas, each with differences in energy cost, supply stability, and production load. In a federal collaborative optimization process, this embodiment takes the data of the five sites as input and performs aggregation calculation by the global optimal strategy model. The number of sites participating in federal collaboration is 5. The output of the self-correcting twin model of each site reflects its comprehensive optimization level in the current period: the first site's model output is 92.3, corresponding to the highest running stability; the second site's is 88.5; the third site's is 90.1; the fourth site's is 85.7; and the fifth site's is 91.6. Based on the energy consumption and execution deviation statistics of the past two periods, the system automatically sets the global coordination coefficient to 0.55 to balance the strategy influence among multiple sites, so that the site with better energy consumption performance has slightly higher weight.

[0045] In the present polymerization process, the regional resource state parameters, energy price fluctuation characteristics and logistics scheduling load are taken as specific values for example illustration. The regional resource state is mainly based on capacity utilization and raw material accessibility, with the first station being 0.86, the second station being 0.78, the third station being 0.91, the fourth station being 0.74, and the fifth station being 0.88. The energy price fluctuation characteristics are represented by price variation rate, with the five stations being 0.12 yuan per degree, 0.15 yuan per degree, 0.10 yuan per degree, 0.18 yuan per degree, and 0.11 yuan per degree. The logistics scheduling load is represented by the average order density per unit time, which is 135 orders per hour, 142 orders per hour, 128 orders per hour, 150 orders per hour, and 139 orders per hour, respectively. The collaborative influence function calculates the external influence degree of each station according to these parameters, with the value ranging from 0.72 to 0.89, reflecting the differences in resource, price and scheduling pressure in different cities.

[0046] In the execution of the strategy mapping operator, the present embodiment inputs the output of the self-correcting twin model of each station and the result of the collaborative influence function into the aggregation calculation module, and generates a global optimal strategy model through weighted mapping calculation. Finally, the optimal output value of the global model is 90.7, corresponding to a global strategy evaluation level of A, indicating that the overall energy consumption control is good, the production capacity distribution is balanced, and the task response is timely. At this time, the weight distribution result is first station 0.23, second station 0.18, third station 0.21, fourth station 0.15, and fifth station 0.23, ensuring that the overall strategy tends to show stable and low energy consumption stations. Through this numerical example, the present embodiment demonstrates the value logic and interaction relationship of each parameter in collaborative optimization, effectively illustrating the actual implementation method and reproducibility of the calculation process of the global optimal strategy model under the condition of multi-station federal collaborative operation.

[0047] Further, the specific implementation process of step 900 is as follows: After obtaining the global optimal strategy model, the present embodiment further determines the collaborative optimal control scheme based on the model to realize the energy efficiency collaboration and strategy unification of the central intelligent kitchen in the whole life cycle stages of construction, operation and maintenance. Specifically, the present embodiment first combines the global optimal strategy model with the central intelligent kitchen life cycle management framework to divide the running characteristics of different stages, including the construction planning stage, the production operation stage, the equipment maintenance stage and the resource recycling stage. For each stage, the present embodiment predicts the trend of various core indicators such as energy intensity, equipment load rate and raw material utilization efficiency through the global model, and generates stage control targets according to the interaction between prediction deviation and constraint conditions. Taking the production operation stage as an example, the system automatically marks the process with high energy consumption in the model prediction, and automatically enables the energy redistribution module, thereby realizing stage energy optimization and production capacity balance, and providing quantitative basis for the formulation of whole life cycle control strategy.

[0048] On the basis of the phased target setting, the embodiment determines the collaborative control logic of each stage of the central intelligent kitchen through strategy layering and task mapping. The strategy layering is divided into three parts: a global coordination layer, a site optimization layer, and an execution feedback layer. The global coordination layer outputs unified constraints on overall energy consumption, production cycle, and resource allocation ratio based on a global optimal strategy model. The site optimization layer performs local adjustment according to the self-correcting twin model of each site to achieve fine control of device group start-stop, energy consumption ratio, and operation sequence. The execution feedback layer collects real-time operation data, task status, and energy efficiency change information, updates control parameters through a closed-loop feedback mechanism, and realizes dynamic correction and strategy iteration throughout the process. Through this layered mapping structure, the embodiment ensures the synchronization and coordination between the global scheduling target and the actions of each site.

[0049] After forming the collaborative control logic, the embodiment further generates a collaborative optimal control scheme for the central intelligent kitchen according to the whole life cycle operation law. The control scheme includes an energy consumption allocation optimization scheme, a task scheduling and device linkage scheme, and a whole life cycle performance balancing scheme. The energy consumption allocation optimization scheme realizes peak energy consumption reduction and valley compensation through time segmentation control algorithm; the task scheduling and device linkage scheme realizes task matching and energy consumption coupling optimization of multi-process series through multi-objective coordination rules; the whole life cycle performance balancing scheme periodically updates and re-optimizes the control strategy by continuously monitoring the equipment aging rate, energy efficiency attenuation trend, and maintenance cycle cost. Through the above steps, the embodiment realizes the global optimal collaboration of the central intelligent kitchen operation control in the whole life cycle, so that the system maintains energy consumption economy, production stability, and equipment sustainability in long-term operation, ensuring that the life cycle intelligent decision-making driven by digital twin has complete executable logic and accurate control closed loop.

[0050] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other.

[0051] The principles and implementation modes of the present application are described by applying specific examples in this paper. The above embodiment description is only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins, characterized in that, include: Acquire multi-source data from the central smart kitchen throughout the entire production and operation process and construct a digital twin basic model based on the multi-source data. The entire production and operation process includes: planning, raw material processing, warehousing management and food delivery. The digital twin basic model is calibrated online to obtain the digital twin dynamic model; The prediction error, confidence interval, and uncertainty index of the digital twin dynamic model are calculated to obtain the uncertainty assessment results; Based on the uncertainty assessment results, a virtual-real consistency analysis is performed on the digital twin dynamic model to obtain the deviation results between the simulation results and the actual operating state of the digital twin dynamic model. Determine whether the deviation result exceeds a preset threshold. If so, adaptively adjust the parameters and structure of the digital twin dynamic model to obtain a self-correcting twin model. The optimization objectives are determined and a multi-objective collaborative optimization scheme is obtained based on the self-calibrating twin model. The optimization objectives include: energy efficiency, production capacity, food safety, and operating costs. Based on the aforementioned multi-objective collaborative optimization scheme, the triggering conditions for prediction bias, constraint conflict, and energy price changes are set to obtain the operation control strategy; The operation control strategy is applied independently to the optimization process at multiple sites of the central smart kitchen, and the self-calibrating twin models and optimization results of each site are aggregated and uniformly updated through a federated collaboration mechanism to obtain the global optimal strategy model. Based on the global optimal strategy model, the optimal collaborative control scheme for the central smart kitchen throughout its entire lifecycle is determined.

2. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The multi-source data includes: Equipment operating parameter data, energy consumption record data, raw material attribute data, environmental information data, and task scheduling data.

3. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The process of acquiring multi-source data from the central smart kitchen throughout its production and operation, and constructing a digital twin model based on the multi-source data, includes: Operational data were collected from the planning, raw material processing, warehousing management and food delivery processes to obtain multi-source data. The multi-source data is preprocessed and key feature parameters of the preprocessed multi-source data are extracted to obtain a cleaned feature dataset. Based on the cleaning feature dataset, and according to the operating mechanism and business process of the central intelligent kitchen equipment, a digital twin model structure describing the operating status, energy consumption and task allocation relationship is constructed. The parameters of the digital twin model structure are calibrated and the performance is verified using historical operational data and on-site monitoring results to obtain the basic model of the digital twin. The expression for the basic model of the digital twin is: ； in, The overall response coefficient of the production system; For the process status Energy input With equipment performance parameters The process coupling function constitutes the process; This represents the combined impact coefficient of logistics and inventory processes. Indicates raw material flow rate Storage capacity ratio The joint equilibrium function; As a time-dynamic factor, Indicates kitchen task load rate of change over time; Here, is the system equilibrium constant, used to describe the residuals that handle unobservable disturbances or maintain virtual-real alignment; where, For the digital twin basic model at time The comprehensive representation output index.

4. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The expression for the digital twin dynamic model is: ； in, This is the historical inheritance coefficient; For dynamic response coefficients, This indicates the sensitivity of the underlying model to changes over time. For environmental interaction coefficients, Due to environmental conditions Network collaboration characteristics With load disturbance intensity The external coupling function constituted; This is the virtual / real error compensation term; among which, This is the predicted output of the digital twin dynamic model after time-series evolution.

5. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The prediction error, confidence interval, and uncertainty index of the calculated digital twin dynamic model are used to obtain uncertainty assessment results, including: Obtain simulation output data from the digital twin dynamic model and acquire measured data for the corresponding time period from actual production. Calculate the prediction error data based on the simulation output data and the measured data; Based on prediction error data, confidence interval estimation and variance decomposition methods are used to calculate the fluctuation range and sensitivity of key variables, and obtain a set of uncertainty indicators. Based on the uncertainty index set, the uncertainty assessment results of the digital twin dynamic model are determined.

6. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, Based on the uncertainty assessment results, a virtual-real consistency analysis is performed on the digital twin dynamic model to obtain the deviation results between the simulation results and the actual operating state of the digital twin dynamic model, including: The historical simulation output of the digital twin dynamic model is aligned and normalized with the corresponding real running data, and the input data for consistency analysis is determined based on the normalized data and the uncertainty assessment results. Based on the input data from consistency analysis, we extract equipment operation, energy consumption changes, and task scheduling to obtain a set of feature comparison indicators. Based on the aforementioned feature comparison index set, the difference vector calculation method is used to identify the difference patterns between the simulation output of the digital twin dynamic model and the actual operating state, thereby obtaining virtual-real difference feature data. The deviation between the simulation results of the digital twin dynamic model and the actual operating state is calculated based on the difference characteristics data between the virtual and real data.

7. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The expression for the self-calibrating twin model is: ； in, For automatic correction rate; It is an adaptive correction function; For dynamic matching coefficients; Indicates actual operational feedback signals The rate of change over time; This is the predicted output of the twin model after self-calibration calculation.

8. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, Determine the optimization objective and, based on the self-calibrating twin model, obtain a multi-objective collaborative optimization scheme, including: The operational requirements analysis of the central smart kitchen yielded the multi-objective optimization requirements and corresponding optimization objective set for the central smart kitchen. Based on the self-calibrating twin model and the set of optimization objectives, model variables and constraints related to each objective are extracted, resulting in a set of model feature parameters that support multi-objective solutions. Based on the model feature parameter set that supports multi-objective solution, a multi-objective collaborative optimization algorithm is adopted to solve energy consumption, production capacity and quality indicators in parallel and balance the weights, thus obtaining a candidate set of collaborative solutions for multi-objective optimization. The optimal solution of the candidate set of collaborative solutions for multi-objective optimization is calculated based on the comprehensive benefit evaluation index, and a multi-objective collaborative optimization scheme for the operation decision of the central intelligent kitchen is obtained.

9. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, Based on the multi-objective collaborative optimization scheme, the triggering conditions for prediction bias, constraint conflict, and energy price changes are set to obtain the operation control strategy, including: Based on the optimal results of the multi-objective collaborative optimization scheme, the sensitive variables and control constraints of the digital twin dynamic model under different operating scenarios are analyzed, and the set of operating control triggering conditions is obtained by classifying and identifying them according to the characteristics of prediction deviation, constraint conflict and energy price change. Based on the set of operation control triggering conditions, the threshold range, response priority, and corresponding weight parameters of each triggering condition are determined, resulting in a control triggering threshold and response weight configuration table. Based on the control trigger threshold and response weight configuration table, the self-calibrating twin model and real-time data stream establish a mapping relationship between trigger conditions and execution actions, and generate control logic by combining rule decision and predictive control, thus obtaining an initial operation control strategy set for multiple scenarios; Simulation verification and effect evaluation were conducted on the initial set of operation control strategies for multiple scenarios in a digital twin environment, and the verified operation control strategies were obtained.

10. The method for optimizing the entire lifecycle operation of a central smart kitchen based on digital twins according to claim 1, characterized in that, The optimization process is independently executed at multiple sites within the central smart kitchen by applying the aforementioned operation control strategy. A federated collaboration mechanism is then used to aggregate and uniformly update the self-calibrating twin models and optimization results of each site, resulting in a globally optimal strategy model, including: The operation control strategy was deployed and optimized independently at multiple central smart kitchen sites, resulting in a set of local optimization results for each site. The local optimization result set and the self-calibrating twin model are synchronized with the feature parameters and optimization indicators through a federated collaboration mechanism, resulting in the model parameter aggregation set required for global collaboration. Based on the aggregated set of model parameters, the parameter deviation and weight allocation of the self-calibrating twin models of each site are comprehensively calculated, and the parameters are uniformly updated through global aggregation to obtain the globally optimal strategy model. The expression for the globally optimal strategy model is: ； in, This is the globally optimal strategy model obtained through aggregation; The number of central smart kitchen sites participating in federal collaboration; Indicates the first The output of the self-calibrating twin model for each site; This is the global coordination coefficient; Based on the regional resource status Characteristics of energy price fluctuations With logistics scheduling load The resulting collaborative influence function; For policy mapping operators.

Images