An ethane-to-ethylene efficiency improvement method based on a refining scene

Abstract

The application discloses an ethane-to-ethylene efficiency improvement method based on a refining scene, which comprises the following steps: in multiple clients, based on the local running data of each production unit, a local model is trained by creating a hybrid loss function, a local model update amount and a set of contribution metadata are generated; a central server receives the local model update amount uploaded by each client and subjected to homomorphic encryption, determines the aggregation weight of each update amount according to the contribution metadata in plaintext form, and completes weighted aggregation in the encryption domain to update the global prediction model; in the client, the global model subjected to local fine-tuning is embedded into an online optimization problem as a prediction simulator, optimal operation parameters are solved and issued, and through performance monitoring and adaptive triggering steps, the continuous iteration of the model and the closed-loop improvement of the production efficiency are realized, the data privacy and safety are ensured, the reliability and performance of the model are improved, and the continuous adaptive optimization of the ethylene production efficiency is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and industrial process control technology, and in particular to an ethane-to-ethylene efficiency improvement method based on a refining scene. BACKGROUND

[0002] Ethane cracking to ethylene is one of the core processes in modern chemical industry, and its production efficiency directly affects the economic benefits of the entire petrochemical industry chain. In order to improve the ethylene yield, reduce the unit energy consumption and prolong the operation cycle of the cracking furnace, the industry is constantly exploring more accurate process modeling and optimization control technology. The current mainstream traditional method mainly relies on complex models established based on chemical reaction kinetics, fluid mechanics and heat transfer mechanisms. Although such models have clear physical meaning, their construction process is often too cumbersome, they are highly dependent on process parameters, and it is difficult to accurately capture dynamic nonlinear changes caused by factors such as raw material fluctuations, catalyst deactivation, and equipment aging, which limits the adaptability and prediction accuracy of the model.

[0003] With the development of artificial intelligence technology, data-driven methods such as deep neural networks are used to model the ethane cracking process, which has become a new research direction. However, direct application of existing data-driven techniques to complex refining scenarios often faces three major challenges. First, the data between multiple production units or plants under a large petrochemical group often forms a "data island" that cannot be physically concentrated due to commercial competition, data security regulations, and privacy protection requirements. Traditional machine learning models cannot train all the data centrally in this scenario, so the generalization ability of the model is severely limited, resulting in an upper limit to the performance of the model. Second, a purely data-driven model is a "black box" whose training process only relies on data correlation and lacks understanding of the underlying physical and chemical laws of the process. This not only requires a large amount of training data, but also its prediction results may violate basic chemical engineering laws such as mass conservation and energy conservation, resulting in poor robustness and insufficient reliability of the model when faced with unseen or abnormal conditions. Third, in the context of multi-party collaborative modeling, the process conditions, raw material batches, equipment status and data quality of each participating party (i.e., each production unit) often differ significantly, i.e., the data is Non-IID. If the contributions of each party are treated equally without distinction in distributed learning, low-quality or irrelevant data will seriously hinder the performance of the global prediction model, and even lead to model divergence. SUMMARY

[0004] This section is intended to introduce some aspects of embodiments of the present application, which are described below. This section is not intended to limit the scope of the application in any way.

[0005] In view of the above existing problems, the present application is proposed. Therefore, the present application provides an ethane-to-ethylene efficiency improvement method based on a refining scene to solve the problems raised in the background art.

[0006] To solve the above technical problems, the present application provides the following technical solutions: an ethane-to-ethylene efficiency improvement method based on a refining scene, which is based on a federated learning framework that integrates physical mechanism constraints to train and iterate a global prediction model for predicting an ethane cracking process. The specific steps include:

[0007] In a plurality of clients associated with a plurality of ethane cracking production unit data sources, based on local running data of each production unit, a local model is trained through a hybrid loss function including a data-driven loss component and a physical constraint loss component to generate a local model update;

[0008] Through a central server, the local model update from the plurality of clients is received, and an aggregation weight is determined for each local model update according to a preset contribution evaluation strategy, and then the plurality of local model updates received are weighted and aggregated to update the global prediction model;

[0009] Based on the updated global prediction model, an operating parameter setting value for guiding the ethane cracking process is generated.

[0010] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on a refining scene, wherein the local running data of each production unit is processed into a feature vector and a performance label vector, the feature vector includes at least one of raw material fingerprint features, process operation features, and equipment state features, and the performance label vector includes values related to ethylene yield, unit ethylene comprehensive energy consumption, or coking rate.

[0011] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on a refining scene, wherein the calculation method of the physical constraint loss component includes:

[0012] The partial derivative values of the prediction output of the local model with respect to one or more process input variables are obtained, and the size of the physical constraint loss component is determined according to the deviation between the partial derivative values and the target values determined by the ethane cracking chemical engineering mechanism.

[0013] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, wherein: the contribution degree evaluation strategy includes generating a set of contribution degree metadata on the client and uploading it together with the local model update, and the contribution degree metadata includes at least one of the following information:

[0014] The decline amplitude of the mixed loss function in the current local model training process;

[0015] A data quality indicator for characterizing the quality or distribution characteristics of the local running data;

[0016] And a physical compliance score quantifying the degree to which the local model satisfies the physical constraints after training.

[0017] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, wherein: the process of determining the aggregation weight follows the following rules:

[0018] The aggregation weight value assigned to any client is positively correlated with the decline amplitude, data quality, or physical compliance score contained in the contribution degree metadata uploaded by the client.

[0019] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, wherein: before updating the global prediction model, it further includes:

[0020] In the client, the local model update is encrypted using a homomorphic encryption algorithm;

[0021] In the central server, the weighted aggregation is performed directly on the encrypted local model update.

[0022] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, wherein: the method further includes an iterative loop step, after updating the global prediction model, the updated global prediction model is distributed to the multiple clients as the initial model for the next round of client model training step.

[0023] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, wherein: the process of generating the operating parameter setting value includes:

[0024] The updated global prediction model is embedded into an optimization problem with comprehensive energy consumption as the optimization objective function, and an optimization algorithm is used to solve the problem to obtain the optimal operating parameter sequence;

[0025] The first operation parameter in the sequence is set as the setting value of the current operation parameter.

[0026] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, before the global prediction model is used as a prediction simulator, a local fine-tuning step is further included.

[0027] At the client, the received global prediction model is trained using its associated local running data to generate a personalized prediction model adapted to the characteristics of the local data, and the personalized prediction model is used as the prediction simulator.

[0028] As a preferred scheme of the ethane-to-ethylene efficiency improvement method based on the refining scene, the method further includes a performance monitoring and adaptive triggering step.

[0029] After deploying the global prediction model at the client, the deviation between its prediction performance indicators and actual production data is continuously monitored.

[0030] When the statistical characteristics of the deviation exceed a preset threshold, the training and iteration steps of the global prediction model by the federated learning framework are automatically re-executed.

[0031] Compared with the prior art, the beneficial effects of the present scheme are as follows:

[0032] 1. The present scheme adopts a federated learning framework, effectively breaks the "data silos" between different production units by training locally at the client and uploading only model updates, ensures that the core running data of each party does not leave the local, fully safeguards the data privacy and commercial security, realizes the collaborative sharing of cross-unit implicit knowledge, builds a prediction model with a global perspective, and improves the generalization ability and prediction accuracy of the model.

[0033] 2. By introducing a hybrid loss function containing a physical constraint loss during client-side model training, the chemical engineering mechanism of the ethane cracking process is incorporated into the model as prior knowledge, overcoming the drawbacks of "black box" in pure data-driven models, ensuring that the prediction results of the model are physically reasonable and self-consistent, enhancing the robustness and reliability of the model when facing unseen or abnormal working conditions. At the same time, based on the contribution metadata weighting aggregation strategy, each party's contribution can be intelligently evaluated and differentiated, effectively suppressing the interference of low-quality data caused by the Non-IID characteristics of data, accelerating the model convergence, and obtaining a global model with better performance.

[0034] 3、The application also embeds the model after individual fine-tuning as a prediction simulator into an online optimization problem, reversely solves and issues optimal operation parameters, directly converts the model into actual production benefits, in addition, the application also contains performance monitoring and self-adaptive triggering steps, can autonomously perceive model performance decay and automatically start the relearning process, ensures that the ethane cracking process can be long-term and stably operated in the optimal state, and realizes the continuous improvement of ethylene efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0036] Figure 1 The overall flowchart of the ethane-to-ethylene efficiency improvement method based on the refining scene according to an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are 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 should be within the scope of protection of the present application.

[0038] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0039] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent of or selected from other embodiments.

[0040] The present application is described in detail in conjunction with the schematic diagram, and in the detailed description of the embodiments of the present application, the cross-sectional view of the device structure will be partially enlarged without general proportion for the convenience of description, and the schematic diagram is only an example, which should not limit the scope of protection of the present application herein. In addition, three-dimensional spatial dimensions including length, width and depth should be included in actual manufacture.

[0041] Meanwhile, in the description of the present application, it should be noted that the terms "upper, lower, inner and outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first, second or third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0042] Unless otherwise expressly specified and limited, the terms "mounting, connecting, connecting" in the present application should be understood broadly, for example: it can be fixed connection, detachable connection or integral connection; It can also be mechanical connection, electrical connection or direct connection, it can also be indirectly connected through intermediate medium, or it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0043] Embodiment 1

[0044] Reference Figure 1 For the first embodiment of the present application, the embodiment provides an ethane-to-ethylene efficiency improvement method based on a refining scene, comprising:

[0045] It should be noted that the present application scheme adopts a federal learning framework which integrates physical mechanism constraints to train and iterate a global prediction model for predicting the ethane cracking process, wherein the federal learning architecture adopts a centralized architecture, i.e. a client-server architecture;

[0046] Specifically, the present application scheme is executed in a star topology structure composed of a central server and multiple clients, wherein each client corresponds to an independent ethane cracking production unit and can access the local database and computing resources of the unit, the central server is responsible for task distribution, model aggregation and global prediction model management, but it itself does not contact any original production data from the clients, and the execution process of the whole scheme is iterative, each iteration aims to absorb the local knowledge of each client to optimize the performance of the global prediction model in an incremental manner;

[0047] S1, in the multiple clients associated with multiple ethane cracking production unit data sources, based on the local running data of each production unit, a local model is trained through a hybrid loss function comprising a data-driven loss component and a physical constraint loss component, to generate a local model update;

[0048] Further, the S1 step is independently and in parallel executed on each client selected to participate in the current (the round) federal learning training, and each client generates a local model update based on the local running data of the corresponding production unit. First, receive the latest global prediction model parameters from the central server. As the starting point for its local model training, the client then performs the following sub-steps:

[0049] S101. Local Data Preparation and Feature Processing Operations: Client The computing nodes access the historical operation database of their associated ethane cracking production units, stored locally, in real time via the OPC (OLE for Process Control) protocol or direct database connection. It is important to note that the raw time-series data stored in this database needs to undergo preprocessing operations such as data cleaning, alignment, missing value imputation, and outlier removal before it can be used. After the preprocessing operations are completed, the processed raw time-series data is used to construct structured training sample pairs. ,in, For sample index, For feature vectors, This is a performance label vector;

[0050] Specifically, this feature vector represents a digital description of the state of the ethane cracking process, and preferably, it contains information features in at least one of the following dimensions:

[0051] Raw material fingerprint characteristics: The concentrations of components such as ethane and propane, as well as the contents of key impurities such as trace sulfides, are obtained by directly using the measurement point values ​​after sensor alignment and analysis by near-infrared spectroscopy or gas chromatography. The concentrations of these components and the contents of these impurities are used as material property variables for entering the cracking furnace.

[0052] Process operating characteristics: These characteristics refer to variables that can be adjusted by the operator or the process control system (APC), such as the coil outlet temperature (COT), the steam to hydrocarbon mass ratio, and the average residence time of the material in the coil.

[0053] Equipment status characteristics: These characteristics are variables that reflect the current health status of the pyrolysis furnace, such as the tube skin temperature (TST) and the operating time of the pyrolysis furnace.

[0054] Specifically, the performance label vector is the actual performance indicator representing production efficiency at the time corresponding to the feature vector, and preferably, it includes the values ​​of one or more of the following key performance indicators:

[0055] Ethylene yield: expressed as the ratio of the mass of the target product ethylene to the mass of the feedstock ethane;

[0056] Unit ethylene comprehensive energy consumption: refers to the total energy consumed to produce unit mass of ethylene;

[0057] It should be noted that the ethylene yield and unit ethylene comprehensive energy consumption are not directly measured, but are calculated through the mechanism model of material balance and energy balance, and according to multiple measured values (such as raw material flow, product flow, fuel gas consumption, etc.);

[0058] Coking rate: refers to the coking speed of the inner wall of the furnace tube, which is indirectly calculated according to the pressure drop of the furnace tube, the change of the heat transfer coefficient, or estimated by a model, and directly affects the continuous operation cycle of the cracking furnace;

[0059] In addition, considering that the local model is very sensitive to the scale of the input data, after the feature vector and the performance label vector are determined, they also need to be standardized;

[0060] Specifically, the Z-score standardization is used in the present application scheme to ensure that all features are within a similar numerical range to accelerate the model convergence process, and the standardization processing formula of the feature vector is as follows:

[0061]

[0062] wherein, represents the standardized feature vector of the ith dimension , and are the mean and standard deviation of the ith dimension on the model training set, respectively;

[0063] It should be noted that the performance label vector and the feature vector are both standardized by Z-score, and the standardization processing formula is the same, so it will not be described again;

[0064] Specifically, the present application scheme uses a feedforward neural network (i.e., a multilayer perceptron (MLP)) as the backbone of the local model. In an implementable process of the present application scheme, the structure of the local model is: an input layer (the number of neurons is equal to the dimension of the feature vector ), 3-5 hidden layers (each layer contains 64-256 neurons, and ReLU is used as the activation function), and an output layer (the number of neurons is equal to the dimension of the performance label vector ), which is used to fit the complex nonlinear relationship of the cracking process;

[0065] S102, training by creating a local model of a hybrid loss function: the client uses its local data set to train its local model (with multiple rounds of training, with the goal of finding a set of updated model parameters The core of the training is to minimize the hybrid loss function;

[0066] It should be noted that the creation of the hybrid loss function aims to require the local model not only to learn, but also to learn in accordance with the chemical engineering mechanism;

[0067] Further, the hybrid loss function is composed of two parts, namely the data-driven loss component and the physical constraint loss component;

[0068] Further, for N samples in the training set, the mean square error between the local model prediction value and the performance label vector , which is to make the local model fit the historical data while learning, the purpose is to make the model prediction as close to the real production data as possible, the data-driven loss component is expressed as:

[0069]

[0070] wherein, represents the total number of samples of the local data set ;

[0071] Further, the purpose of the physical constraint loss component is to embed the known chemical engineering mechanism as prior knowledge into the local model, and to ensure that the prediction behavior of the local model conforms to the basic physical law, without using any performance label vector , but creating some input sample points to check whether the behavior of the local model on these input sample points is reasonable;

[0072] Specifically, in an implementable process of the present application scheme, the physical constraint term of the chemical engineering mechanism includes but is not limited to the following three forms:

[0073] Monotonicity constraint: assuming that within the normal operating range, the ethylene yield monotonically increases with the increase of the furnace tube outlet temperature ; by randomly generating a batch of virtual input sample points , the physical constraint loss component under the monotonicity constraint is obtained as:

[0074]

[0075] wherein, M represents the total number of virtual input sample points , represents the partial derivative of the ethylene yield output by the local model with respect to the input furnace tube outlet temperature, which is obtained by automatic differentiation;​ This is expressed as a predicted ethylene yield. It is a modified linear unit function. When the partial derivative is positive (in accordance with the chemical engineering mechanism), the output of ReLU is 0 and no loss is generated. When the partial derivative is negative (violating the chemical engineering mechanism), the output of ReLU is positive, which generates a quadratic penalty term (positive loss), thereby driving the model parameters to adjust in the direction that satisfies the monotonicity constraint.

[0076] Mass conservation constraint: Ideally, the mass of ethane input should equal the total mass of unreacted ethane, ethylene, hydrogen, methane, and other byproducts output. Therefore, this can be simplified to a single constraint: the sum of the mass fractions of all products (including unreacted substances) must be 1. (Assuming the model...) The output is a vector containing the mass fraction of each product. Let P be the total number of vectors representing the mass fractions of each product. Then, under this mass conservation constraint, the physical constraint loss component... It can be represented as:

[0077]

[0078] Boundary constraints: Since the yield of any product cannot be negative, the theoretical yield of ethylene should have an upper limit (approximately 55% yield from 80% conversion). The physical constraint loss component under this boundary constraint is then obtained. :

[0079]

[0080] in, This represents the theoretical upper limit of ethylene yield;

[0081] Furthermore, based on the physical constraints defined above and the data-driven loss component mentioned earlier, the created hybrid loss function can be expressed mathematically as follows:

[0082]

[0083] in, For a mixed loss function, , , These are represented as hyperparameters for each physical constraint term, primarily used to balance the importance of different physical constraint terms and satisfy... It can be set based on expert experience or through methods such as grid search, which will not be elaborated here;

[0084] S103, generating local model update: the client uses its local dataset and the mixed loss function created above to perform multiple rounds of gradient descent training through the Adam optimizer. Each round of training fine-tunes the local model parameters based on the total gradient calculated ;

[0085] Further, after the training is completed, in order to protect data privacy, the client does not directly upload its trained local model parameters, but instead calculates and generates an update of the local model parameters:

[0086]

[0087] It should be noted that this update essentially represents the change or gradient direction of the local model after absorbing local data knowledge (learning) and physical constraint loss relative to the initial global prediction model;

[0088] S2, through a central server, receive local model updates from multiple clients, and determine an aggregation weight for each local model update according to a pre-set contribution evaluation strategy, then perform weighted aggregation on the received multiple local model updates to update the global prediction model;

[0089] Further, the central server listens to and receives data packets uploaded from all participating clients (assuming a total of ) in the current round of training within a pre-set communication window, and in order to ensure the absolute security and privacy of the communication process, the client uploads not the plaintext model parameter update, but the ciphertext after encryption processing;

[0090] Specifically, after the client generates the local model parameter update , it immediately encrypts it using a homomorphic encryption algorithm (such as Paillier or CKKS) to generate an encrypted local model parameter update ;

[0091] It should be noted that the characteristic of homomorphic encryption is that it allows specific mathematical operations (such as addition and scalar multiplication) to be performed directly on ciphertext, and the decrypted result is consistent with the result of performing the same operation on plaintext, therefore, the data packet received by the central server from the client contains a binary tuple:

[0092] Where, is the contribution metadata associated with the update in plaintext form, which does not contain any sensitive production data and is only used by the server to evaluate the value of the client's contribution;

[0093] Furthermore, upon receiving all After receiving the data packet from each client, the central server updates the parameters for each encrypted local model. Determine a fair and efficient aggregation weight, denoted as . This process abandons the traditional federated learning approach of weighting by data volume, and instead relies on contribution metadata uploaded by the client. Conduct an assessment;

[0094] Specifically, the strategy employed in this assessment is as follows:

[0095] S201, Parsing Contribution Metadata The central server retrieves contribution metadata. Extract at least one of the following three dimensions of information:

[0096] The decrease in the mixed loss function ( This value reflects the client's... The "learning progress" during local model training is typically the difference in loss before and after training. ,in, and These represent the loss values ​​before and after training the local model, respectively.

[0097] Data quality indicators ( This metric quantifies the client The quality and uniqueness of the local dataset, in one feasible process of the present invention, is a comprehensive score. This comprehensive score combines the size of the local dataset, the entropy of the feature distribution (reflecting the diversity of the local dataset), and the KL divergence calculated by comparing the distribution of the local dataset with the global average distribution after aggregating all client data. The larger the divergence value, the more unique the client data is.

[0098] Furthermore, since the core of federated learning is that the original data does not leave the local machine, while the global average distribution directly aggregates all the data, the present invention adopts a global distribution approximation method to replace the global average distribution, so as to achieve the global average distribution without leaking the client's original data.

[0099] Specifically, each client Locally, each dimension of the data features is divided into equal intervals, and the number of samples falling into each interval is calculated to form a local histogram. Subsequently, each client, through Secure Multi-Party Computation (SMPC), jointly calculates the sum of all local histograms. This process does not require any party to know the specific histograms of the others; ultimately, the central server uses the aggregated global histogram. This can approximate the average distribution of global data;

[0100] Physical conformity score ( This rating quantifies the client's performance. The degree to which the trained local model adheres to the preset physical mechanisms, in one feasible process of the present invention, is defined as follows: ,in, It is the value of the physical constraint loss component after training. The higher the value, the more the local model's behavior conforms to physical laws;

[0101] S202, Calculate contribution metadata The extracted scores for each dimension: To eliminate the influence of dimensions, the central server first normalizes the raw values ​​of each dimension across all clients, obtaining a normalized score. Therefore, regarding the aforementioned decrease... Its normalization formula can be expressed as:

[0102]

[0103] in, This represents the decrease after normalization. Similarly, the normalized data quality score can be obtained using the same normalization formula. Physical conformity score This will not be elaborated upon here;

[0104] S203. Calculate the aggregate weight: The central server, based on the above strategy, weights and merges the scores of the three dimensions into the final aggregate weight. :

[0105]

[0106] in, Similar to the hyperparameters in the aforementioned hybrid loss function, they represent the importance of "learning efficiency," "data value," and "model reliability" in the aggregation decision, respectively, and are non-negative. In the early stages of model training, we can set them to relatively large values. Values ​​can be increased later in training to encourage dataset diversity. and The value of is used to focus on the convergence and physical consistency of the model;

[0107] Furthermore, for all clients Their respective aggregation weights were determined. After that, the central server performs an aggregation operation to update the global prediction model, it is noted that this process needs to be performed in the encrypted domain to ensure that even the central server cannot peek into any single client's model update details;

[0108] Further, the update rule of the global prediction model parameters is as follows:

[0109] Further, since the homomorphic encryption algorithm supports the multiplication operation of ciphertext and plaintext scalar (i.e. the aggregation weight ), and the addition operation between ciphertexts, the central server can directly calculate the encrypted global update amount, obtaining:

[0110]

[0111] wherein, represents the encrypted global update amount;

[0112] In addition, after calculating the encrypted global update amount, the central server updates the global prediction model in two ways:

[0113] First, server-side update: the central server sends the encrypted global update amount to a trusted third party or hardware security module (HSM) that has a private key for decryption, obtaining the plaintext global update amount , then the central server completes the update locally:

[0114]

[0115] Second, client-side collaborative decryption: a threshold encryption method based on secure multi-party computation (SMPC) is adopted, that is, multiple clients generate a public key through a secure interactive protocol and store private key shares in a decentralized manner. It should be noted that since each client only holds a private key share, it cannot reconstruct the complete private key alone. The central server broadcasts the encrypted global update to all clients. Each client that receives the request performs a decryption operation on the encrypted global update using its unique private key share. This decryption operation does not yield the final plaintext, but generates a partial decryption result. The client then sends the partial decryption result to the central server. It should be noted that, in order to prevent malicious clients from submitting fake partial decryptions, this process usually involves a zero-knowledge proof to prove that the partial decryption result is indeed generated from the correct private key share and the received encrypted global update, without exposing the private key share itself. When the central server collects at least F (F is a preset decryption threshold value, which is an integer greater than or equal to 2 and less than or equal to the total number of clients K) valid partial decryptions, it can combine these partial decryptions using a public aggregation algorithm to recover the final plaintext.

[0116] It should be noted that by updating the global prediction model, attackers cannot steal model information by compromising the central server. If they want to crack the encryption, they must compromise and control at least F different clients within the same decryption window period, thus increasing the attacker's attack cost and enhancing the security of the model.

[0117] S3, based on the updated global prediction model, generating operation parameter setting values for guiding the ethane cracking process;

[0118] It should be noted that after updating the global prediction model, the model is used to solve the production operation parameters of the ethane cracking process under optimal conditions, thereby achieving online optimization control of the ethane cracking process and improving ethylene production efficiency.

[0119] Further, this step is periodically executed (e.g., every 5-15 minutes) in the local control system of each client to adapt to real-time changes in production conditions.

[0120] In addition, in order to make the global prediction model better adapt to the unique equipment characteristics and operating condition drift of each production unit, a quick local fine-tuning step can be performed before it is used for optimization.

[0121] Specifically, the local fine-tuning step is as follows:

[0122] The client receives the latest global model parameters After that, it is not directly used, but is used as an initial parameter, and a small batch (for example, the past 24 hours) of high-quality local running data accumulated by itself is fine-tuned for several rounds (for example, 1-5 rounds) of fine-tuning training, and the process still uses the aforementioned created hybrid loss function for fine-tuning training, but the training rounds are relatively few, and the purpose is to quickly adapt the global prediction model to the latest dynamics of the local without "forgetting" the global knowledge;

[0123] After fine-tuning training, the client can obtain a personalized prediction model, which has both global vision and local characteristics, and therefore can be used as a prediction simulator to provide more accurate predictions.

[0124] Further, the actual production optimization target is formalized as a mathematical optimization problem, i.e., the ethylene comprehensive energy consumption minimization is preferred as the target of the scheme of the present application, and the ethylene production efficiency is improved, and for this purpose, the present application constructs a target function wherein, is the production operation parameter to be optimized;

[0125] Further, the process operation characteristics defined in the foregoing are used as decision variables of the target function, and the feedstock fingerprint characteristics and equipment state characteristics defined in the foregoing are used as state variables of the target function, and it should be noted that at the current time of optimization solving, the decision variables and state variables are regarded as known, fixed inputs, and their current values are obtained from the database;

[0126] It should be noted that since the unit of the target function is the ethylene comprehensive energy consumption, the prediction result of the personalized prediction model (or the global prediction model) is needed when calculating it, and based on this, we let the output of the personalized prediction model be the predicted performance label vector, which includes the performance indicators output by the prediction model, and its mathematical form is wherein, is a transpose operation, represents the predicted ethylene yield, represents the predicted unit ethylene comprehensive energy consumption, represents the predicted coking rate, is the personalized prediction model of the client k, and the preferred target of the scheme of the present application is to minimize the ethylene comprehensive energy consumption and improve the ethylene production efficiency, and therefore, the target function is the predicted unit ethylene comprehensive energy consumption:

[0127]

[0128] It is worth noting that, based on the definition of the above objective function, maximizing the ethylene yield or minimizing the coking rate can also improve the efficiency of ethylene production, so the definition of the objective function depends on the performance indicators output by the prediction model;

[0129] Further, in order to ensure that the operation parameters solved are physically feasible and meet the safety production regulations, a series of constraint conditions must be imposed;

[0130] Specifically, taking the furnace tube outlet temperature and the steam to hydrocarbon mass ratio (steam to hydrocarbon ratio) as an example, the operation variable boundary constraint can be set, that is:

[0131]

[0132]

[0133] wherein, and respectively represent the lower limit temperature and the upper limit temperature of the furnace tube outlet temperature COT, and correspondingly, and respectively represent the minimum ratio and the maximum ratio of the steam to hydrocarbon ratio;

[0134] Specifically, taking the predicted ethylene yield and the coking rate as an example, the predicted ethylene yield cannot be lower than the expected target value, and the predicted coking rate cannot be higher than the expected warning value, so the process performance constraint is as follows:

[0135]

[0136] wherein, is the expected target value (i.e. the expected ethylene yield), and correspondingly, is the expected warning value (i.e. the expected coking rate);

[0137] Specifically, taking the furnace tube outer wall temperature as an example, the furnace tube outer wall temperature cannot exceed the material tolerance limit, which is the equipment safety constraint:

[0138]

[0139] wherein, is the upper limit furnace tube outer wall temperature;

[0140] In summary, the complete optimization problem can be represented as:

[0141] wherein, is the inequality form of the above constraint, represents the constraint condition of the production operation parameter to be optimized, and the range of the production operation parameter cannot exceed the lower limit of its own range and upper limit ,;

[0142] Further, but after the optimization problem is determined, a nonlinear programming (NLP) optimization algorithm such as sequential quadratic programming (SQP) or interior-point methods is called by the computing nodes of the clients to solve it; at the same time, the algorithm takes the personalized prediction model as the core, and finds the optimal production operation parameter combination sequence that satisfies all the constraint conditions by iteratively calculating the gradient of the objective function and the constraint conditions on the decision variables;

[0143] Specifically, the first operation parameter in the sequence is set as the operation parameter setting value at the current time, and then the setting value is sent to the distributed control system or process control system of each client (such as a factory A), to overwrite the original setting value, and the actual actuator (such as a valve, a heater, etc.) is driven by calling the underlying PID controller to make the actual process variable track the setting value;

[0144] In addition, the present application also includes a performance monitoring and adaptive triggering step to cope with the performance degradation of the model due to unmodeled operating condition changes (such as catalyst deep deactivation, major changes in raw material composition);

[0145] Specifically, for the personalized prediction model of each client, the predicted performance indicators (such as , , , etc.) are continuously compared with the actual performance indicators obtained by mechanism calculation, the prediction deviation is calculated, the statistical characteristics of the deviation are analyzed, the mean and standard deviation within the sliding time window are calculated, and the corresponding mean threshold and standard deviation threshold are preset based on the mean and standard deviation. When one of the following conditions occurs, it indicates that the model performance has significantly declined:

[0146] Case one: when the absolute value of the mean exceeds the mean threshold, it means that there is a persistent systematic prediction deviation;

[0147] Case two: when the standard deviation exceeds the standard deviation threshold, it means that the prediction stability of the model has sharply declined;

[0148] It should be noted that the mean threshold and the standard deviation threshold can be determined based on statistical analysis of the prediction deviation sequence generated in a stable running period at the beginning of model deployment, for example, the mean threshold can be set to 2 times the mean deviation in the period, and the standard deviation threshold can be set to 3 times the standard deviation of the deviation in the period (following the 3 sigma principle) to ensure that regular fluctuations will not cause false triggering;

[0149] Specifically, once one of the above situations is triggered, the central server will receive a "model retraining" request from the client. When the central server receives enough (at least 2 or more based on the star topology of the present solution) requests from the clients, it will automatically restart a new round of federated learning process, i.e. from step S1, using the latest data accumulated by each party to retrain a new global prediction model that can better adapt to the current working condition;

[0150] It should be noted that by adaptively updating the global prediction model, it is ensured that the production operation parameters in the process of ethane-to-ethylene can be continuously operated in the optimal state, thereby systematically improving the efficiency of ethylene production.

[0151] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as object-oriented programming languages Java and interpreted scripting language JavaScript, etc.

[0152] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks.

[0153] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0154] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0155] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the preferred embodiments by those of skill in the art once they have the benefit of the present disclosure. Therefore, the appended claims are intended to encompass within their scope all possible variations and modifications of the preferred embodiments. 1

[0156] It is apparent that a person skilled in the art can make various changes and modifications to the application without departing from the spirit and scope thereof. Thus, if these modifications and changes fall within the scope of the claims and their equivalents, it is intended to include them in the application.

Claims

1. An ethane-to-ethylene efficiency improvement method based on a refining scene, characterized in that, The method is based on a federated learning framework incorporating physical mechanism constraints, and trains and iterates a global prediction model for predicting an ethane cracking process, including the following steps: In a plurality of clients associated with a plurality of ethane cracking production unit data sources, based on the local running data of each production unit, a local model is trained through a hybrid loss function comprising a data-driven loss component and a physical constraint loss component, to generate a local model update; The physical constraint loss component is calculated in the following way: Obtain the partial derivative value of the prediction output of the local model with respect to one or more process input variables, and determine the size of the physical constraint loss component according to the deviation between the partial derivative value and the target value determined by the ethane cracking chemical engineering mechanism; Through a central server, receive the local model update from the plurality of clients, and determine an aggregation weight for each local model update according to a pre-set contribution evaluation strategy, then perform weighted aggregation on the received plurality of local model updates to update the global prediction model; Based on the updated global prediction model, generate an operating parameter setting value for guiding the ethane cracking process.

2. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, The local running data of each production unit is processed into a feature vector and a performance label vector, the feature vector comprising at least one of raw material fingerprint features, process operation features and equipment state features, and the performance label vector comprising values related to ethylene yield, unit ethylene comprehensive energy consumption or coking rate.

3. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, The contribution evaluation strategy includes generating a set of contribution metadata at the client and uploading it together with the local model update, the contribution metadata including at least one of the following information: The descent amplitude in the current local model training process; A data quality indicator for characterizing the quality or distribution characteristics of the local running data; And a physical compliance score quantifying the extent to which the local model satisfies the physical constraints after training.

4. The method according to claim 3, wherein the method is based on the scenario of an ethane-to-ethylene plant. The process of determining the aggregation weight follows the following rules: The aggregation weight value assigned to any client is positively correlated with the descent amplitude, data quality or physical compliance score contained in the contribution metadata uploaded by the client.

5. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, Before updating the global prediction model, it further includes: At the client, encrypt the local model update using a homomorphic encryption algorithm; At the central server, weighted aggregation is performed directly on the encrypted local model update.

6. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, The method further includes an iteration step, after updating the global prediction model, the updated global prediction model is distributed to the plurality of clients as the initial model for the next round of client model training.

7. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, The process of generating the operating parameter setting value includes: Embed the updated global prediction model as a prediction simulator into an optimization problem with comprehensive energy consumption as the optimization objective function, and solve the problem using an optimization algorithm to obtain an optimal operating parameter sequence; Take the first operating parameter in the sequence as the setting value of the current operating parameter.

8. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 7, wherein, Before using the global prediction model as a prediction simulator, a local fine-tuning step is included: At the client side, the received global prediction model is trained using its associated local running data to generate a personalized prediction model that fits the characteristics of the local data, and the personalized prediction model is used as the prediction simulator.

9. The method for improving the efficiency of ethane-to-ethylene based on the refining scene according to claim 1, wherein, The method further includes a performance monitoring and adaptive triggering step: After deploying the global prediction model at the client side, the deviation between its prediction performance indicators and the actual production data is continuously monitored; When the statistical characteristics of the deviation exceed the preset threshold, the training and iteration steps of the global prediction model by the federated learning framework are automatically re-executed.

Images