A method for constructing a data model of an intelligent power plant based on group plant collaboration

Abstract

The present invention relates to a method for constructing a data model of an intelligent power plant based on group-plant collaboration, comprising the steps of: deploying a comprehensive decision-making module on the group side for realizing comprehensive display and comprehensive decision-making of each plant; extracting business data, constructing and training the data model of the intelligent power plant on the group side. The beneficial effects of the present invention are as follows: the present invention adopts a cloud-edge-end distributed elastic architecture mode, which includes the edge end of the first zone on the plant side, the fourth zone on the plant side, and the cloud data aggregation application and the Internet of Things platform on the group side (based on the DCS system of the power plant, building a data acquisition platform for each device), the first zone on the plant side is the actual production area of ​​the power plant, and the fourth zone on the plant side is the production management area. The elastic architecture can provide component support for business, models, and data according to demand for each part; the present invention solves the problems of low data interconnection rate, insufficient information integration, and difficulty in knowledge accumulation caused by data islands.

Description

Technical Field

[0001] The present invention belongs to the field of electric power information systems, and in particular relates to a method for constructing a data model of an intelligent power plant based on group plant collaboration. Background Art

[0002] In the intelligent information transformation of power production enterprises, the existing distributed process control system (DCS), monitoring information system (SIS), and enterprise resource management system (ERP) can uniformly store and manage the real-time data generated in the energy production process. However, due to the differences in manufacturers, equipment performance, and equipment parameter suppliers, the system integration mode, design standards, and development technology used are different, and the systems cannot achieve effective information interconnection in data transmission, resulting in data island problems between systems, which further leads to the inability to share information. In order to solve the problems of low data interconnection rate, insufficient information integration, and difficulty in knowledge accumulation caused by data islands, it is necessary to establish a method for building a data model of an intelligent power plant based on group plant collaboration, realize data collaboration and intelligent production applications, and provide effective decision-making for remote early warning, diagnosis, and intelligent maintenance of equipment. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for constructing a data model of an intelligent power plant based on group plant collaboration.

[0004] This smart power plant data model based on group-plant collaboration has constructed a business layer, a model layer and a data layer; the business layer adopts a cloud-edge distributed elastic architecture, which includes area one on the plant side, area four on the plant side and the group side, among which the group side and area one on the plant side constitute two dual registration centers; a comprehensive decision-making module and an Internet of Things platform are deployed on the group side, an intelligent application module is deployed in area one on the plant side, and an enabling service module is deployed in area four on the plant side. The enabling service module adopts a microservice architecture to achieve front-end and back-end separation and independent deployment of services; the Internet of Things platform is a data collection platform for each device built by the power plant DCS system; the cloud-edge distributed elastic architecture provides component support for the business layer, model layer and data layer as needed; the model layer includes a model module, a model management module and a model deployment module; the data layers of area one on the plant side, area four on the plant side and the group side are all equipped with in-memory databases and relational databases; the data layer corresponding to area four on the plant side is also equipped with a time series database.

[0005] Preferably, the business layer, model layer, and data layer are constructed in a top-down manner.

[0006] Preferably, area one on the plant side is the actual production area of ​​the power plant, and area four on the plant side is the production management area.

[0007] Preferably, each enabling service module is a software module, and uses RESTful API (Representational State Transfer Interface) to achieve communication.

[0008] The method for constructing the data model of the intelligent power plant based on group-plant collaboration includes the following steps:

[0009] Step 1: Deploy a comprehensive decision-making module on the group side to realize comprehensive display and comprehensive decision-making of each plant; extract business data, build and train smart power plant data models on the group side;

[0010] Step 2: The group side exports the trained smart power plant data model to the designated server through the model export operator; transmits it to the server in the four zones on the plant side through the dedicated network of the power Internet of Things, and optimizes and adjusts the smart power plant data model in the four zones on the plant side;

[0011] Step 3: The smart power plant data model optimized and adjusted in step 2 in the four zones on the plant side is sent and deployed to the first zone on the plant side through the network isolation device, and the smart power plant data model is used; the smart application module for realizing the business is deployed in the first zone on the plant side; the training set is used to fine-tune the smart power plant data model to obtain the mathematical model and the mechanism model, and the mathematical model and the mechanism model are used for online early warning and monitoring (the mathematical model is obtained by training with historical production data and selecting a suitable algorithm; the mechanism model is obtained by adjusting the experimental data according to the classical mechanism formula);

[0012] The mathematical model constructed is:

[0013]

[0014]

[0015] In the above formula, x is the historical data from the production system; is the estimated value; y is the true value; δ is the upper limit of the deviation value; f(x) is the parameter function, and the constraint condition of the parameter function is that the parameters in the parameter function satisfy that the difference between the estimated value and the true value does not exceed the upper limit of the deviation value; D is the historical data, which comes from the PI database;

[0016] The constructed mechanism model is:

[0017]

[0018] In the above formula, x is the historical data from the production system; ε is the correction value of the mechanism model; is the estimated value; g(x) is the mechanism calculation formula;

[0019] Step 4. After the optimized and adjusted smart power plant data model is deployed in the first area of ​​the plant side, the operation personnel shall regularly evaluate the smart power plant data models of the first area of ​​the plant side, the fourth area of ​​the plant side and the group side; if the evaluation score of only one of the smart power plant data models in the first area of ​​the plant side, the fourth area of ​​the plant side and the group side is lower than the standard, the configuration parameters of the smart power plant data model in the fourth area of ​​the plant side are adjusted, and the smart power plant data model is redeployed; if the evaluation scores of two or more smart power plant data models in the first area of ​​the plant side, the fourth area of ​​the plant side and the group side are lower than the standard, the smart power plant data model is rebuilt and trained on the group side, and the configuration parameters of the smart power plant data model are sent to the fourth area of ​​the plant side, and the newly trained smart power plant data model is redeployed to the first area of ​​the plant side to realize the coordination and update of the group plant data models.

[0020] Preferably, step 1 specifically comprises the following steps:

[0021] Step 1.1, business data extraction: Use unified coding to sample data and map equipment measurement points to a unified KKS coding library; the equipment digital mirror operator quickly determines the physical meaning of business data according to the unified coding rules, extracts business data from the plant side area 1 through the plant side area 4 to the group side, and obtains a historical data set;

[0022] Step 1.2: Build a smart power plant data model and configure the smart power plant data model by dragging and dropping graphical operators;

[0023] y out =h(x in )

[0024] =h m (y m-1 )

[0025] =h m (h m-1 (y m-2 ))

[0026] =h m (h m-1 (···(h1(x in ))))

[0027] In the above formula, h() is the main smart power plant data model; x in The input of the main smart power plant data model; out As output; according to the function points, the main smart power plant data model is divided into multiple nested sub-smart power plant data models, where m is the number of sub-smart power plant data models, and the model input of the first layer in the multiple nested sub-smart power plant data models is x in , the model input of the mth layer is the output of the m-1th layer sub-model;

[0028] The graphical operator splits the algorithm into the device digital mirror operator, data screening operator, validity analysis operator, correlation analysis operator, normalization operator, model training operator and model export operator, and each operator is connected in series;

[0029] Step 1.3: The model training operator completes the training of the smart power plant data model based on the correlation coefficient and the normalized valid measurement point data:

[0030]

[0031] In the above formula, Model is the smart power plant data model, and the input is the correlation coefficient matrix cor and the normalized effective measurement point data norm. is the effective data, θ a Output the evaluation value of the measurement point threshold for the smart power plant data model.

[0032] Preferably, step 1.2 specifically comprises the following steps:

[0033] Step 1.2.1: The data screening operator will filter the business data through the rule method and trend method, remove the data that affects the model training from the historical data set, and obtain the filtered data;

[0034]

[0035]

[0036] In the above formula, x is the actual production data; the rule range is R; SR is the business data filtered by the rule method; the rule method obtains the normal operating range of the measuring point according to the operating procedures. When the measuring point value is not within the rule range, the data is filtered by the SR value, and only the data within the rule range is retained; the trend range is A, and SA is the business data filtered by the trend method; the trend method obtains the operating range of the measuring point according to the common sense of the operating personnel. When the measuring point value is not within the rule range, the data is filtered by the SA value, and only the data within the rule range is retained;

[0037] Step 1.2.2, the validity analysis operator performs numerical validity analysis on the data in the data set processed in step 1.2.1, divides valid measurement points into invalid measurement points according to the proportion of null values ​​and invalid values ​​of the sparse statistical measurement points, and selects valid measurement points for model training;

[0038]

[0039]

[0040]

[0041] In the above formula, r null is the null value ratio; n null The number of null values; n total is the total amount of values; r error is the proportion of invalid values; n error is the number of invalid values; θ null is the threshold of the null value ratio; if the measured point data is greater than θ null , then the data of the measuring point is judged to be invalid data; θ error is the threshold of invalid value proportion. If the measured point data is greater than θ error , then the data of the measuring point is judged to be invalid data;

[0042] Step 1.2.3: Use the correlation analysis operator to analyze the effective measuring points, and obtain the correlation coefficient by calculating the covariance matrix of the effective measuring points:

[0043]

[0044] In the above formula, the person method is used to calculate the correlation coefficient. is valid data;

[0045] Step 1.2.4: The normalization operator uses the maximum and minimum value method to remove the multi-dimensional dimensions of the effective measurement point data itself, and obtains the normalized effective measurement point data:

[0046]

[0047] In the above formula, Min represents the minimum value, Max represents the maximum value, x is the current data, and norm is the normalized valid measurement point data.

[0048] Preferably, in step 2, the four zones on the factory side serve as data hubs for transmitting and storing data; the data are transmitted to the one zone on the factory side and the group side respectively.

[0049] Preferably, the criterion for model evaluation in step 4 is: within a period of time, the operating personnel compare the deviation between the actual situation in the actual operation process and the model results. If the deviation is small, the performance of the smart power plant data model is good; if the deviation is large, the performance of the smart power plant data model is poor.

[0050] Preferably, a forward network isolation device and a reverse network isolation device are used for data communication between the plant side zone 1 and the plant side zone 4; the forward isolation device ensures that the plant side zone 1 performs one-way encrypted data transmission to the specific port of the plant side zone 4 through the network layer, and the plant side zone 1 transmits the production data in full to the time series database of the plant side zone 4 through the forward isolation device; a reverse isolation device is used to ensure that the plant side zone 4 performs one-way encrypted data transmission to the specific port of the plant side zone 1 through the network layer, and the plant side zone 4 transmits the model parameters to the plant side zone 1 through the reverse isolation device; the plant side zone 1 provides the collection, storage, and transmission of real-time production data, and the collection of real-time production data is performed in accordance with the standard industrial automation control hardware interface protocol; the storage of real-time production data is divided into full storage, operation storage, and system storage, and the full storage transmits the real-time data to the production time series database through the deployed measurement point interface for storage; the operation storage of real-time production data transmits the measurement point data frequently read in the production time series database to the memory database for storage, and the storage of the power plant DCS system is to store the system parameters in the relational database to support system operation.

[0051] As a preferred option, the group side and the four areas on the plant side use a dedicated power Internet of Things network for data encryption communication; the memory database corresponding to the four areas on the plant side accesses the measurement point data in its own time series database, and the relational database of the four areas on the plant side stores the data necessary for the training and normal operation of the smart power plant data model.

[0052] The beneficial effects of the present invention are:

[0053] The present invention adopts a cloud-edge distributed elastic architecture model, which includes the "edge" belonging to the first and fourth zones on the plant side and the "cloud" data aggregation application and the Internet of Things platform (mainly based on the DCS system of the power plant, building a data collection platform for each device) on the group side. The first zone on the plant side is the actual production area of ​​the power plant, and the fourth zone on the plant side is the production management area. The elastic architecture enables each part to provide component support for business, models, and data as needed.

[0054] The business layer adopts a service architecture with two locations and dual registration centers. The two locations and dual registration centers are the group side and the power plant side. The group side deploys comprehensive decision-making, and the power plant deploys intelligent applications and empowerment services. The service adopts a microservice architecture to achieve front-end and back-end separation and independent deployment of services. Services communicate through representational state transfer interfaces. According to the bottom-up design principle, the first area of ​​the plant side realizes the intelligent application of business, and uses the mathematical model and mechanism model fine-tuned by the training set for online early warning and monitoring. The fourth area of ​​the plant side realizes the comprehensive display and management of the plant side equipment, and the group side realizes the comprehensive display and comprehensive decision-making of each plant.

[0055] The present invention solves the problems of low data interconnection rate, insufficient information integration, difficulty in knowledge accumulation, etc. caused by data islands. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a diagram of the data model architecture of a smart power plant based on group-plant collaboration of the present invention;

[0057] Figure 2 A flow chart for constructing an induced draft fan warning model in Example 2;

[0058] Figure 3 It is a processing flow chart of each operator of the induced draft fan early warning model in the second embodiment;

[0059] Figure 4 This is a result diagram of the early warning model of the induced draft fan of a coastal power plant A in Example 2;

[0060] Figure 5 This is a result diagram of the induced draft fan early warning model of a coastal power plant B in Example 2. DETAILED DESCRIPTION

[0061] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0062] Embodiment 1

[0063] Embodiment 1 of the present application provides a Figure 1 The data model of a smart power plant based on group plant collaboration is shown below:

[0064] The business layer, model layer and data layer were constructed in a top-down manner; the business layer adopts a cloud-edge distributed elastic architecture, which includes area 1 on the plant side (the actual production area of ​​the power plant), area 4 on the plant side (production management area) and the group side, among which the group side and area 1 on the plant side constitute two dual registration centers; a comprehensive decision-making module and an Internet of Things platform are deployed on the group side, an intelligent application module is deployed in area 1 on the plant side, and an enabling service module is deployed in area 4 on the plant side. The enabling service module adopts a microservice architecture to achieve front-end and back-end separation and independent deployment of services; the Internet of Things platform is a data collection platform for each device built by the DCS system of the power plant; the cloud-edge distributed elastic architecture provides component support for the business layer, model layer and data layer as needed; the model layer includes a model module, a model management module and a model deployment module; the data layers of area 1 on the plant side, area 4 on the plant side and the group side are all equipped with in-memory databases and relational databases; the data layer corresponding to area 4 on the plant side is also equipped with a time series database.

[0065] Embodiment 2

[0066] Based on the first embodiment, taking the training and deployment of the induced draft fan warning model as an example, the second embodiment of the present application provides a method such as Figure 2 The construction method of the induced draft fan early warning model is shown in the figure:

[0067] Step 1: Deploy a comprehensive decision-making module on the group side to realize comprehensive display and comprehensive decision-making of each plant; extract business data, build and train smart power plant data models on the group side;

[0068] Step 1.1, business data extraction: Use unified coding to sample data, and map each equipment measurement point in Table 1 to a unified KKS coding library; the equipment digital mirror operator quickly determines the physical meaning of the business data according to the unified coding rules, and extracts the business data from the plant side area 1 through the plant side area 4 to the group side to obtain a historical data set;

[0069] Table 1 Measurement points of induced draft fan warning model

[0070] Serial number Measurement point description Measuring point type Measuring point 1 Air preheater flue gas side differential pressure Calculation Point Measuring point 2 Inlet flue gas temperature of induced draft fan Calculation Point Measuring point 3 Number of operating absorption tower circulation pumps Calculation Point Measuring point 4 Differential pressure before and after demister Calculation Point Measuring point 5 A induced draft fan static pressure head Calculation Point Measuring point 6 A. Temperature rise of induced draft fan Calculation Point Measuring point 7 A Deviation between the command and feedback of the induced draft fan blades Calculation Point Measuring point 8 A. Change rate of static pressure head of induced draft fan (5 seconds) Calculation Point Measuring point 9 A. Temperature rise rate of induced draft fan (1 minute) Calculation Point Measuring point 10 A Draft fan current change rate (5 seconds) Calculation Point Measuring point 11 Current deviation Calculation Point Measuring point 12 Furnace negative pressure PI original measurement point Measuring point 13 Unit load PI original measurement point Measuring point 14 Total coal PI original measurement point Measuring point 15 Total air volume PI original measurement point Measuring point 16 Ambient temperature Calculation Point Measuring point 17 A induced draft fan bearing vibration X direction PI original measurement point Measuring point 18 A induced draft fan bearing vibration Y direction PI original measurement point Measuring point 19 A induced draft fan current PI original measurement point Measuring point 20 A. Inlet flue gas pressure of induced draft fan PI original measurement point Measuring point 21 A draught fan outlet flue gas pressure PI original measurement point Measuring point 22 A. Exhaust gas temperature of induced draft fan PI original measurement point Measuring point 23 Induced draft fan blade command offset PI original measurement point Measuring point 24 A Control instructions for the induced draft blade adjustment device PI original measurement point Measuring point 25 A. Valve position of the induced draft blade adjustment device PI original measurement point Measuring point 26 B induced draft fan current PI original measurement point

[0071] Step 1.2: Build a smart power plant data model and configure the smart power plant data model by dragging and dropping graphical operators;

[0072] y out =h(x in )

[0073] =h m (y m-1 )

[0074] =h m (h m-1 (y m-2 ))

[0075] =h m (h m-1 (···(h1(x in ))))

[0076] In the above formula, h() is the main smart power plant data model; x in The input of the main smart power plant data model; out As output; according to the function points, the main smart power plant data model is divided into multiple nested sub-smart power plant data models, where m is the number of sub-smart power plant data models, and the model input of the first layer in the multiple nested sub-smart power plant data models is x in , the model input of the mth layer is the output of the m-1th layer sub-model;

[0077] like Figure 3As shown, the graphical operator splits the algorithm into the device digital mirror operator, data screening operator, validity analysis operator, correlation analysis operator, normalization operator, model training operator and model export operator, and each operator is connected in series;

[0078] Step 1.2.1: The data screening operator will filter the business data through the rule method and trend method, remove the data that affects the model training from the historical data set, and obtain the filtered data;

[0079]

[0080]

[0081] In the above formula, x is the actual production data; the rule range is R; SR is the business data filtered by the rule method; the rule method obtains the normal operating range of the measuring point according to the operating procedures. When the measuring point value is not within the rule range, the data is filtered by the SR value, and only the data within the rule range is retained; the trend range is A, and SA is the business data filtered by the trend method; the trend method obtains the operating range of the measuring point according to the common sense of the operating personnel. When the measuring point value is not within the rule range, the data is filtered by the SA value, and only the data within the rule range is retained;

[0082] Step 1.2.2, the validity analysis operator performs numerical validity analysis on the data in the data set processed in step 1.2.1, divides valid measurement points into invalid measurement points according to the proportion of null values ​​and invalid values ​​of the sparse statistical measurement points, and selects valid measurement points for model training;

[0083]

[0084]

[0085]

[0086] In the above formula, r null is the null value ratio; n null The number of null values; n total is the total amount of values; r error is the proportion of invalid values; n error is the number of invalid values; θ null is the threshold of the null value ratio; if the measured point data is greater than θ null , then the data of the measuring point is judged to be invalid data; θ error is the threshold of invalid value proportion. If the measured point data is greater than θ error , then the data of the measuring point is judged to be invalid data;

[0087] Step 1.2.3: Use the correlation analysis operator to analyze the effective measuring points, and obtain the correlation coefficient by calculating the covariance matrix of the effective measuring points:

[0088]

[0089] In the above formula, the person method is used to calculate the correlation coefficient. is valid data;

[0090] Step 1.2.4: The normalization operator uses the maximum and minimum value method to remove the multi-dimensional dimensions of the effective measurement point data itself, and obtains the normalized effective measurement point data:

[0091]

[0092] In the above formula, Min represents the minimum value, Max represents the maximum value, x is the current data, and norm is the normalized valid measurement point data;

[0093] Step 1.3: The model training operator completes the training of the smart power plant data model based on the correlation coefficient and the normalized valid measurement point data:

[0094]

[0095] In the above formula, Model is the smart power plant data model, and the input is the correlation coefficient matrix cor and the normalized effective measurement point data norm. is the effective data, θ a Output the evaluation value of the measurement point threshold for the smart power plant data model;

[0096] Step 2: The group side exports the trained smart power plant data model to the designated server through the model export operator; transmits it to the server in the four zones on the plant side through the dedicated network of the power Internet of Things, and optimizes and adjusts the smart power plant data model in the four zones on the plant side;

[0097] Step 3: The smart power plant data model optimized and adjusted in step 2 in the four zones on the plant side is sent and deployed to the first zone on the plant side through the network isolation device, and the smart power plant data model is used; the smart application module for realizing the business is deployed in the first zone on the plant side; the training set is used to fine-tune the smart power plant data model to obtain the mathematical model and the mechanism model, and the mathematical model and the mechanism model are used for online early warning and monitoring (the mathematical model is obtained by training with historical production data and selecting a suitable algorithm; the mechanism model is obtained by adjusting the experimental data according to the classical mechanism formula);

[0098] The mathematical model constructed is:

[0099]

[0100]

[0101] In the above formula, x is the historical data from the production system; is the estimated value; y is the true value; δ is the upper limit of the deviation value; f(x) is the parameter function, and the constraint condition of the parameter function is that the parameters in the parameter function satisfy that the difference between the estimated value and the true value does not exceed the upper limit of the deviation value;

[0102] The constructed mechanism model is:

[0103]

[0104] In the above formula, x is the historical data from the production system; ε is the correction value of the mechanism model; is the estimated value; g(x) is the mechanism calculation formula;

[0105] Step 4. After the optimized and adjusted smart power plant data model is deployed in the first area of ​​the plant side, the operation personnel shall regularly evaluate the smart power plant data models of the first area of ​​the plant side, the fourth area of ​​the plant side and the group side; if the evaluation score of only one of the smart power plant data models in the first area of ​​the plant side, the fourth area of ​​the plant side and the group side is lower than the standard, the configuration parameters of the smart power plant data model in the fourth area of ​​the plant side are adjusted, and the smart power plant data model is redeployed; if the evaluation scores of two or more smart power plant data models in the first area of ​​the plant side, the fourth area of ​​the plant side and the group side are lower than the standard, the smart power plant data model is rebuilt and trained on the group side, and the configuration parameters of the smart power plant data model are sent to the fourth area of ​​the plant side, and the newly trained smart power plant data model is redeployed to the first area of ​​the plant side to realize the coordination and update of the group plant data models.

[0106] In this embodiment, the induced draft fan early warning model is deployed to a coastal power plant a and a coastal power plant b respectively, and the model is tested through historical data to obtain the following results: Figure 4 and Figure 5 The current deviation diagram reflecting the stall of the induced draft fan is shown as the model warning result. Analysis shows that the model can send the alarm information of the stall of the induced draft fan about 10 minutes in advance. The model has a strong warning effect on the equipment.

[0107] For the above models, personnel will conduct regular model evaluations on the factory side. If the model evaluation score of a single factory is lower than the standard, the model will be adjusted and redeployed in the four areas of the factory. If the model evaluation scores of multiple factories are lower than the standard, the model needs to be retrained as a common model in the group and completed and deployed to achieve coordination and updating of data models of group factories.

Claims

1. A method for constructing a data model of an intelligent power plant based on group-plant collaboration, characterized in that: The business layer, model layer and data layer were constructed; the business layer adopts a cloud-edge-end distributed elastic architecture, which includes the plant-side zone 1, the plant-side zone 4 and the group side, where the group side and the plant-side zone 1 form two dual registration centers; the group side deploys a comprehensive decision-making module and an IoT platform, the plant-side zone 1 deploys an intelligent application module, and the plant-side zone 4 deploys an enabling service module; the IoT platform is a data collection platform for various devices built by the power plant DCS system; the cloud-edge-end distributed elastic architecture provides component support for the business layer, model layer and data layer; The model layer includes a model module, a model management module and a model deployment module; the data layers of the plant side area 1, the plant side area 4 and the group side are all equipped with memory databases and relational databases; the data layer corresponding to the plant side area 4 is also equipped with a time series database; the construction method of the smart power plant data model based on group plant collaboration includes the following steps: Step 1: Deploy the comprehensive decision-making module on the group side; extract business data, build and train the smart power plant data model on the group side; Step 2: The group side exports the trained smart power plant data model to the designated server through the model export operator; transmits it to the server in the four zones on the plant side through the dedicated network of the power Internet of Things, and optimizes and adjusts the smart power plant data model in the four zones on the plant side; Step 3: The smart power plant data model optimized and adjusted in step 2 in the four zones on the plant side is sent and deployed to the first zone on the plant side through the network isolation device, and the smart power plant data model is used; the smart application module for realizing the business is deployed in the first zone on the plant side; the training set is used to fine-tune the smart power plant data model to obtain the mathematical model and the mechanism model, and online early warning and monitoring are performed through the mathematical model and the mechanism model; The mathematical model constructed is: In the above formula, x is the historical data from the production system; is the estimated value; y is the true value; δ is the upper limit of the deviation value; f(x) is the parameter function, and the constraint condition of the parameter function is that the parameters in the parameter function satisfy that the difference between the estimated value and the true value does not exceed the upper limit of the deviation value; D is the historical data, which comes from the PI database; The constructed mechanism model is: In the above formula, x is the historical data from the production system; ε is the correction value of the mechanism model; is the estimated value; g(x) is the mechanism calculation formula; Step 4. After the optimized and adjusted smart power plant data model is deployed in the plant side area 1, the operation personnel shall regularly evaluate the smart power plant data models in the plant side area 1, the plant side area 4 and the group side; if the evaluation score of only one of the smart power plant data models in the plant side area 1, the plant side area 4 and the group side is lower than the standard, the configuration parameters of the smart power plant data model in the plant side area 4 are adjusted, and the smart power plant data model is redeployed; if the evaluation scores of two or more smart power plant data models in the plant side area 1, the plant side area 4 and the group side are lower than the standard, the smart power plant data model is rebuilt and trained on the group side, and the configuration parameters of the smart power plant data model are sent to the plant side area 4, and the newly trained smart power plant data model is redeployed to the plant side area 1.

2. The method for constructing a smart power plant data model based on group-plant collaboration according to claim 1 is characterized by: Step 1 specifically includes the following steps: Step 1.1, business data extraction: Use unified coding to sample data and map equipment measurement points to a unified KKS coding library; the equipment digital mirror operator quickly determines the physical meaning of business data according to the unified coding rules, extracts business data from the plant side area 1 through the plant side area 4 to the group side, and obtains a historical data set; Step 1.2: Build a smart power plant data model and configure the smart power plant data model by dragging and dropping graphical operators; y out =h(x in ) =h m (y m-1 ) =h m (h m-1 (y m-2 )) =h m (h m-1 (···(h1(x in )))) In the above formula, h() is the main smart power plant data model; x in The input of the main smart power plant data model; out As output; according to the function points, the main smart power plant data model is divided into multiple nested sub-smart power plant data models, where m is the number of sub-smart power plant data models, and the model input of the first layer in the multiple nested sub-smart power plant data models is x in , the model input of the mth layer is the output of the m-1th layer sub-model; The graphical operator splits the algorithm into the device digital mirror operator, data screening operator, validity analysis operator, correlation analysis operator, normalization operator, model training operator and model export operator, and each operator is connected in series; Step 1.3: The model training operator completes the training of the smart power plant data model based on the correlation coefficient and the normalized valid measurement point data: In the above formula, Model is the smart power plant data model, and the input is the correlation coefficient matrix cor and the normalized effective measurement point data norm. is the effective data, θ a Output the evaluation value of the measurement point threshold for the smart power plant data model.

3. The method for constructing a smart power plant data model based on group-plant collaboration according to claim 2 is characterized in that: Step 1.2 specifically includes the following steps: Step 1.2.1: The data screening operator will filter the business data through the rule method and trend method, remove the data that affects the model training from the historical data set, and obtain the filtered data; In the above formula, x is the historical data from the production system; the rule range is R; SR is the business data filtered by the rule method; the rule method obtains the normal operating range of the measuring point according to the operating procedures. When the measuring point value is not within the rule range, the data is filtered by the SR value, and only the data within the rule range is retained; the trend range is A, and SA is the business data filtered by the trend method; the trend method obtains the operating range of the measuring point according to the common sense of the operating personnel. When the measuring point value is not within the rule range, the data is filtered by the SA value, and only the data within the rule range is retained; Step 1.2.2, the validity analysis operator performs numerical validity analysis on the data in the data set processed in step 1.2.1, divides valid measurement points into invalid measurement points according to the proportion of null values ​​and invalid values ​​of the sparse statistical measurement points, and selects valid measurement points for model training; In the above formula, r null is the null value ratio; n null The number of null values; n total is the total amount of values; r error is the proportion of invalid values; n error is the number of invalid values; θ null is the threshold of the null value ratio; if the measured point data is greater than θ null , then the data of the measuring point is judged to be invalid data; θ error is the threshold of invalid value proportion. If the measured point data is greater than θ error , then the data of the measuring point is judged to be invalid data; Step 1.2.3: Use the correlation analysis operator to analyze the effective measuring points, and obtain the correlation coefficient by calculating the covariance matrix of the effective measuring points: In the above formula, the person method is used to calculate the correlation coefficient. is valid data; Step 1.2.4: The normalization operator uses the maximum and minimum value method to remove the multidimensional dimensions of the effective measurement point data itself, and obtain the normalized effective measurement point data: In the above formula, Min represents the minimum value, Max represents the maximum value, x is the historical data from the production system, and norm is the normalized valid measurement point data.

4. The method for constructing a smart power plant data model based on group-plant collaboration according to claim 1 is characterized in that: In step 2, the four zones on the factory side serve as data hubs for transmitting and storing data; The data is transmitted to the first area of ​​the factory side and the group side respectively.

5. The method for constructing a smart power plant data model based on group-plant collaboration according to claim 1 is characterized in that: The standard for model evaluation in step 4 is: within a period of time, the operator compares the deviation between the actual situation in the actual operation process and the model results. If the deviation is small, the performance of the smart power plant data model is good; if the deviation is large, the performance of the smart power plant data model is poor.

Images