WO2023082788A1

WO2023082788A1 - Method and apparatus for predicting oxygen content in flue gas and load, method and apparatus for selecting prediction model, and method and apparatus for predicting flue gas emission

Info

Publication number: WO2023082788A1
Application number: PCT/CN2022/116583
Authority: WO
Inventors: 刘胜伟; 杨杰; 余真鹏
Original assignee: 新智我来网络科技有限公司
Priority date: 2021-11-11
Filing date: 2022-09-01
Publication date: 2023-05-19

Abstract

Provided are a method and apparatus for predicting oxygen content in flue gas and a load, a method and apparatus for selecting a prediction model, and a method and apparatus for predicting flue gas emission. The method for predicting oxygen content in flue gas and a load comprises: by means of a participant, respectively determining data sets of a plurality of groups of local devices and a sample weight which corresponds to a data set of a target device; on the basis of the data sets of the plurality of groups of local devices and corresponding sample weights, obtaining, by means of training, prediction neural network models of the plurality of groups of devices; uploading the prediction neural network models of the plurality of groups of local devices to a central node for model aggregation, so as to obtain an aggregated prediction neural network model; training the aggregated prediction neural network model on the basis of a preset training condition, so as to obtain a joint prediction model; and predicting the value of oxygen content in flue gas of the target device on the basis of the joint prediction model and the sample weight corresponding to the data set of the target device.

Description

Flue gas oxygen content load prediction method, prediction model selection method, flue gas emission prediction method and device

technical field

The present disclosure relates to the field of comprehensive energy technology, and in particular to a flue gas oxygen content load prediction method, a prediction model selection method, a flue gas emission prediction method and a device.

Background technique

With the wide application of comprehensive energy, thermal efficiency is an important indicator to measure gas-fired boilers. Generally, by controlling the oxygen content of the boiler flue gas to the optimal design value, the highest thermal efficiency under different energy equipment can be achieved. The oxygen content of flue gas is often measured and maintained by zirconia measuring instrument, but the cost is high.

For example, in the field of distributed energy, small gas-fired boilers generally abandon the installation of zirconia measuring instruments in order to save costs, resulting in the inability to achieve closed-loop control and optimal operation of thermal efficiency, especially when the calorific value of gas is unstable, thermal efficiency will be mostly affected Sacrifice. That is, the data distribution of different boilers is not the same, which greatly affects the prediction accuracy of the data, so that the prediction of the oxygen content load of the boiler flue gas will not be accurate, which brings a lot of trouble to the enterprises or factories that use energy equipment. Big economic loss, so urgently need to solve this problem at present.

Contents of the invention

In view of this, the embodiments of the present disclosure provide a method, device, computer equipment, and computer-readable storage medium for predicting the oxygen content load of flue gas based on joint learning, so as to solve the problem of the inability to improve the flue gas load of energy equipment in the prior art. The accuracy of the oxygen load prediction results in a waste of resources.

The first aspect of the embodiments of the present disclosure provides a flue gas oxygen content load prediction method, including:

The participants respectively determine the sample weights corresponding to the data sets of the local multiple groups of devices and the target device data sets;

According to the data sets of multiple sets of local equipment and the corresponding sample weights, train the prediction neural network model of multiple sets of equipment;

Upload the predictive neural network models of multiple groups of local devices to the central node for model aggregation to obtain the aggregated predictive neural network models;

Train the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model;

According to the joint prediction model and the sample weight corresponding to the target equipment data set, the oxygen content value of the flue gas of the target equipment is predicted.

The second aspect of the embodiments of the present disclosure provides a flue gas oxygen content load prediction device, including:

A determination module, used for the participants to respectively determine the sample weights corresponding to the data sets of multiple groups of local devices and the data sets of the target device;

The first training module is used to train the prediction neural network models of multiple sets of equipment according to the data sets of multiple sets of local equipment and the corresponding sample weights;

The aggregation module is used to upload the predictive neural network models of multiple groups of local devices to the central node for model aggregation, so as to obtain the aggregated predictive neural network models;

The second training module trains the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model;

The prediction module is used to predict the oxygen content value of the flue gas of the target device according to the joint prediction model and the sample weight corresponding to the target device data set.

The third aspect of the embodiments of the present disclosure provides a flue gas oxygen content load prediction method, which is applied in a joint learning framework, including:

Obtain the equipment data of the first participant and the equipment data of the second participant under the federated learning architecture; among them, the first participant is the participant who proposes the prediction demand, and the second participant is other participants except the first participant square;

using the device data of the first party and the device data of the second party to train a predictive classifier;

determining weight data of the device data of the first party with respect to the device data of the second party based on the predictive classifier;

Train a predictive gradient boosting model based on the device data and weight data of the second participant;

Using a predictive gradient boosting model to predict flue gas oxygen loads for first-party equipment.

The fourth aspect of the embodiments of the present disclosure provides a flue gas oxygen content load prediction device, which is applied in a joint learning framework, including:

The acquisition module is used to acquire the equipment data of the first participant and the equipment data of the second participant under the joint learning architecture; wherein, the first participant is the participant who proposes the prediction demand, and the second participant is the participant other than the first participant parties other than the Party;

The first training module uses the equipment data of the first participant and the equipment data of the second participant to train the predictive classifier;

A calculation module, configured to determine weight data of the equipment data of the first participant with respect to the equipment data of the second participant according to the predictive classifier;

The second training module is used to train the predictive gradient boosting model based on the equipment data and weight data of the second participant;

The prediction module is used to predict the flue gas oxygen content load of the equipment of the first participant by using the predicted gradient boosting model.

According to the fifth aspect of the embodiments of the present disclosure, a method for selecting a flue gas oxygen content load prediction model is provided, including:

Based on the federated learning architecture, receive the training data set and test data set from the prediction equipment of the participating parties;

Preprocess the data in the training data set and the test data set of the prediction device, and obtain the preprocessed device data set;

Calculate the evaluation index value of each piece of data in the preprocessed equipment data set according to the established prediction model group;

According to the minimum evaluation index value, the flue gas oxygen content load prediction model suitable for the prediction equipment is determined.

The sixth aspect of the embodiments of the present disclosure provides a flue gas oxygen content load prediction model selection device, including:

The receiving module is used to receive training data sets and test data sets from prediction devices of participating parties based on the federated learning architecture;

The preprocessing module is used to preprocess the data in the training data set and the test data set of the prediction device, and obtain the preprocessed device data set;

A calculation module, configured to calculate the evaluation index value of each piece of data in the preprocessed device data set according to the established prediction model group;

The prediction module is used to determine the flue gas oxygen content load prediction model suitable for the prediction equipment according to the minimum evaluation index value.

The seventh aspect of the embodiments of the present disclosure provides a flue gas emission prediction method, including:

According to the local energy data, train the local energy data measurement model;

Based on the joint learning framework, the local energy data measurement model is trained according to the test data and the energy data of the target energy equipment, and the test data prediction model and the target energy data prediction model are respectively obtained;

Based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, calculate the first sample migration weight and the second sample migration weight, wherein the first sample migration weight is the local energy data for the target energy equipment The sample migration weight of the energy data, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device;

Using the local energy data and the first sample migration weight, the test data and the second sample migration weight, respectively train the local energy data network model and the test data network model;

Receive the joint learning prediction model from the central node after the aggregation and training of the local energy data network model and the test data network model;

According to the joint learning prediction model, the flue gas emission of the target energy equipment is predicted.

The eighth aspect of the embodiments of the present disclosure provides a flue gas emission prediction device, including:

The first training module is used to train the local energy data measurement model according to the local energy data;

The second training module is used to train the local energy data measurement model based on the joint learning framework according to the test data and the energy data of the target energy equipment, and respectively obtain the test data prediction model and the target energy data prediction model;

A calculation module, configured to calculate the first sample migration weight and the second sample migration weight based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, wherein the first sample migration weight is the local energy The sample migration weight of the data for the energy data of the target energy device, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device;

The third training module is used to train the local energy data network model and the test data network model respectively by using the local energy data and the first sample transfer weight, the test data and the second sample transfer weight;

Establishing a module for receiving a joint learning prediction model from the central node after the aggregation and training of the local energy data network model and the test data network model;

The prediction module is used to predict the flue gas emission of the target energy equipment according to the joint learning prediction model.

A ninth aspect of the embodiments of the present disclosure provides a computer device, including a memory, a processor, and a computer program stored in the memory and operable on the processor, and the processor implements the steps of the above method when executing the computer program.

A tenth aspect of the embodiments of the present disclosure provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above method are implemented.

Compared with the prior art, the beneficial effects of the embodiments of the present disclosure at least include: determining the sample weights corresponding to the datasets of multiple groups of local devices and the datasets of the target device through the participating parties; Sample weight, training to obtain the prediction neural network model of multiple groups of devices; upload the prediction neural network model of multiple groups of local devices to the central node for model aggregation to obtain the aggregated prediction neural network model; train the aggregated model according to preset training conditions predictive neural network model to obtain a joint forecasting model; according to the joint forecasting model and the sample weight corresponding to the target equipment data set, predict the oxygen content value of the flue gas of the target equipment. The embodiments of the present disclosure solve the problem of waste of resources caused by the inability to improve the accuracy of load prediction of flue gas oxygen content of energy equipment in the prior art.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following will briefly introduce the drawings that need to be used in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only of the present disclosure For some embodiments, those skilled in the art can also obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic diagram of a joint learning architecture according to an embodiment of the present disclosure;

Fig. 2 is a flowchart of a flue gas oxygen content load prediction method provided by an embodiment of the present disclosure;

Fig. 3 is a block diagram of a flue gas oxygen content load prediction device provided by an embodiment of the present disclosure;

Fig. 4 is a flow chart of another flue gas oxygen content load prediction method provided by an embodiment of the present disclosure;

Fig. 5 is a block diagram of another flue gas oxygen content load prediction device provided by an embodiment of the present disclosure;

Fig. 6 is a flow chart of a method for selecting a flue gas oxygen content load prediction model provided by an embodiment of the present disclosure;

Fig. 7 is a block diagram of a flue gas oxygen content load prediction model selection device provided by an embodiment of the present disclosure;

Fig. 8 is a flow chart of a flue gas emission prediction method provided by an embodiment of the present disclosure;

Fig. 9 is a block diagram of a flue gas emission prediction device provided by an embodiment of the present disclosure;

Fig. 10 is a schematic diagram of a computer device provided by an embodiment of the present disclosure.

Detailed ways

In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and techniques are presented for a thorough understanding of the embodiments of the present disclosure. It will be apparent, however, to one skilled in the art that the present disclosure may be practiced in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present disclosure with unnecessary detail.

Joint learning refers to the comprehensive utilization of various AI (Artificial Intelligence, artificial intelligence) technologies on the premise of ensuring data security and user privacy, and joint multi-party cooperation to jointly mine the value of data, and to promote new intelligent business forms and models based on joint modeling. Federated learning has at least the following characteristics:

(1) Participating nodes control the weakly centralized joint training mode of their own data to ensure data privacy and security in the process of co-creating intelligence.

(2) In different application scenarios, use screening and/or combining AI algorithms and privacy-preserving calculations to establish multiple model aggregation optimization strategies to obtain high-level, high-quality models.

(3) On the premise of ensuring data security and user privacy, based on a variety of model aggregation optimization strategies, obtain an efficiency method to improve joint learning, where the efficiency method can be solved by solving information including computing architecture parallelism and large-scale cross-domain network Interaction, intelligent perception, exception handling mechanism, etc., to improve the overall efficiency of joint learning.

(4) Obtain the needs of multi-party users in each scenario, determine and reasonably evaluate the true contribution of each joint participant through the mutual trust mechanism, and distribute incentives.

Based on the above methods, it is possible to establish an AI technology ecology based on joint learning, give full play to the value of industry data, and promote the implementation of scenarios in vertical fields.

A method and device for predicting flue gas emission based on joint learning according to an embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

FIG. 1 is a schematic diagram of a joint learning architecture according to an embodiment of the present disclosure. As shown in FIG. 1 , the architecture of joint learning may include a server (central node) 101 , a participant 102 , a participant 104 and a participant 104 . A participant can be composed of one or more clients.

In the joint learning process, the basic model can be established by the server 101, and the server 101 sends the model to the participant 102, the participant 104 and the participant 104 with which a communication connection is established. The basic model can also be uploaded to the server 101 after being created by any participant, and the server 101 sends the model to other participants that have established communication connections with it. Participant 102, participant 104, and participant 104 build a model according to the downloaded basic structure and model parameters, use local data for model training, obtain updated model parameters, and encrypt and upload the updated model parameters to the server 101. The server 101 aggregates the model parameters sent by the participant 102 , the participant 104 and the participant 104 to obtain global model parameters, and returns the global model parameters to the participant 102 , the participant 104 and the participant 104 . The participant 102, the participant 104 and the participant 104 iterate their respective models according to the received global model parameters until the models finally converge, thereby realizing the training of the models. In the joint learning process, the data uploaded by participant 102, participant 104, and participant 104 are model parameters, local data will not be uploaded to the server 101, and all participants can share the final model parameters, so data can be guaranteed Co-modeling is achieved on the basis of privacy.

It should be noted that the number of participants is not limited to the above three, but can be set according to needs, which is not limited in this embodiment of the present disclosure.

Fig. 2 is a flow chart of a method for predicting the load of oxygen content in flue gas provided by an embodiment of the present disclosure. Wherein, in Fig. 2, the execution subject is set as a participant, and the participant can be a client or an independent server, which is collectively referred to as a participant here; the central node can be a cloud or an integrated server. As shown in Figure 2, the flue gas oxygen content load prediction method includes:

S201, the participant respectively determines sample weights corresponding to data sets of multiple groups of local devices and target device data sets.

Specifically, the participants respectively determine the sample weights corresponding to the data sets of multiple groups of local devices and the target device data sets in the following ways:

Step 1. The participant selects the data sets of multiple sets of local devices and the data sets of the target device;

Among them, the data set can be steam boiler flue gas temperature, economizer outlet temperature, instantaneous value of flue gas flow, steam boiler gas temperature, steam boiler flue gas standard flow rate, steam boiler natural gas inlet pressure, steam boiler flue flow rate, steam boiler Condenser inlet smoke temperature, steam boiler exhaust gas temperature, steam boiler flue gas pressure, steam boiler condenser inlet pressure, steam boiler main steam instantaneous flow, steam boiler operating status, steam boiler natural gas inlet instantaneous flow, etc. The limit is a steam boiler, but it can also be a gas furnace or other energy equipment.

Step 2, merging the data sets of the local multiple groups of devices and the data sets of the target device to obtain merged data;

Step 3, using the merged data to train a kernel density estimation model;

Step 4. According to the kernel density estimation model, respectively determine the sample weights corresponding to the data sets of the local multiple groups of devices and the data sets of the target device.

S202. According to the data sets of multiple sets of local equipment and the corresponding sample weights, train the prediction neural network models of multiple sets of equipment.

S203. Upload the predictive neural network models of multiple groups of local devices to the central node for model aggregation, so as to obtain the aggregated predictive neural network models.

Specifically, uploading the prediction neural network models of multiple groups of local devices to the central node for model aggregation to obtain the aggregated prediction neural network model can be achieved in the following ways:

Step 1. Upload the prediction neural network models of multiple groups of local devices to the central node;

Step 2. Responding to the information fed back by the central node;

Step 3, receiving the aggregated prediction neural network model delivered by the central node.

S204. Train the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model.

Wherein, the preset training condition may include a preset number of training times or a predicted convergence state value of model training, and the like.

Specifically, training the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model can be achieved in the following manner:

Step 1. Responding to the aggregated prediction neural network model issued by the central node;

Step 2. Determine the preset training conditions;

Step 3: Train the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model.

S205. Predict the oxygen content value of the flue gas of the target device according to the joint prediction model and the sample weight corresponding to the target device data set.

Specifically, according to the joint prediction model and the sample weight corresponding to the target equipment data set, the prediction of the oxygen content value of the flue gas of the target equipment can be achieved in the following ways:

Step 1. Participants upload the joint prediction model to the central node for joint learning and training;

Step 2: Joint learning and training of the joint prediction model in response to feedback from the receiving engine;

Step 3. Send the joint prediction model trained by joint learning to the target device;

Step 4. According to the joint prediction model and the sample weight corresponding to the target equipment data set, predict the oxygen content value of the flue gas of the target equipment.

Further, the present disclosure also provides relevant embodiments for the optimization of the predictive neural network model:

Step 1. Participants use the data sets of multiple sets of local devices to establish training samples for predictive neural network models;

Step 2, using the data set of the target device to establish a predictive neural network model test sample;

Step 3. According to the prediction neural network model training sample and the model test sample, the sample prediction value is obtained;

Step 4, according to the norm of the error matrix of the sample predicted value and the sample expected value, the fitness value of the predicted neural network model is obtained;

Step 5: Update the particles in the population in the predictive neural network model according to the fitness value of the predictive neural network model to obtain an optimized predictive neural network model.

Specifically, it can be further illustrated by the following examples:

The optimization prediction neural network model is as follows:

(a) Coding the learning parameters of the predictive neural network model to obtain the initial particle population, coding rules: each parameter is represented by 13bit binary code, and these parameters are spliced into a particle;

(b) decoding to obtain the parameters of the predictive neural network model, and assigning the parameters to the predictive neural network model;

(c) training a predictive neural network model using training samples;

(d) using the test sample to test the predictive neural network model to obtain the sample predicted value;

(e) Select the norm of the error matrix of the predicted value of the predicted sample and the expected value as the fitness value;

(f) Update the particles in the population.

The parameters involved in the update of the particle algorithm are: speed, position, individual extremum, and group extremum of the population. The speed and position update methods are shown in the following formulas (1) and (2). At the same time, in order to prevent the blind search of particles, it is generally recommended to limit their position and speed to the interval [-Xmax, Xmax], [-Vmax, Vmax] .

Variable description:

X _i ＝(x _i1 , x _i2 ,..... x _iD ) represents a population particle with D dimension, and also represents a solution to the problem

V _i =(v _i1 ,v _i2 ,.....v _iD ) means the velocity of a population particle with D dimension

P _i ＝(p _i1 ,p _i2 ,.......p _iD ) means the extremum value of individual particles with D dimension

P _g ＝(p _g1 ,p _g2 ,...p _gD ) means the population extremum with D dimension

w is the inertia weight, d=1, 2,...D, i=1,2,...n, k is the current iteration number, Vid is the speed of the particle, c1, c2 are non Negative constants, called acceleration factors, r1, r2 are random numbers distributed in [0,1].

(g) Perform optimal crossover on individuals in the population. Individual particles are updated by crossing with individual extremum particles. The crossover method uses an integer crossover method. First, two crossover positions are selected, and then the individual and the individual extremum are crossed. The obtained new individuals adopt the strategy of retaining excellent individuals, and update the particles only when the fitness value of the new particle is greater than the fitness value of the old particle.

(h) Perform optimal crossover on the population. This step is similar to the eighth operation, except that the individual extremum is replaced by the group extremum.

(i) The mutation operation is performed on the particle operation in the population. The mutation operation adopts the two-bit exchange method within the individual. First, the mutation positions pos1 and pos2 are randomly selected, and then the two mutation positions are exchanged. For the obtained new individuals, the strategy of retaining excellent individuals is adopted, and the particles are updated only when the fitness value of the new particles is better than that of the old particles.

(j) Get a new population.

(k) Judging whether the termination condition is satisfied, or the maximum number of iterations is reached, or an error smaller than the limit is satisfied.

(l) If the condition is not satisfied, go to the third step, otherwise, decode the particle swarm to obtain the initial parameters of the optimal prediction neural network model network.

A further example of a flue gas oxygen content load prediction method based on joint learning provided in this disclosure is as follows: There are three boiler data sets of boiler 1, boiler 2, and boiler 3, where boiler 1 and boiler 2 are local equipment, boiler 3 is the target equipment.

First, select the data of boiler 1, boiler 2, and boiler 3 to obtain local equipment data set A and local equipment data set B (hereinafter referred to as data set A and data set B respectively), and target equipment data set C (hereinafter referred to as data set C).

Then, merge the data at dataset A, dataset B, and dataset C, use the merged data to train a KDE (kernel density estimation model) model, input the data of dataset A into the KDE model, and get The sample weight of boiler 1 data set A, input the data of boiler 2 data set B into the KDE model, and obtain the sample weight of B.

Specifically, in a multi-energy station, there will be many boiler data of different models and different processes. At this time, the data of multiple boilers can be used to improve the prediction accuracy and reduce the installation of boiler sensors, thereby reducing costs and resource waste.

Thirdly, at boiler 1, the predictive neural network model can be trained using data from dataset A and sample weights; at the same time, at boiler 2, the predictive neural network model can be trained using data from dataset B and sample weights. Then upload the predictive neural network models trained by dataset A and dataset B to the central node for model aggregation. When the central node sends the aggregated model to Boiler 1 and Boiler 2, Boiler 1 and Boiler 2 use the aggregated model to train respectively to obtain a joint prediction model, and repeat this many times until the model is trained to Until it converges, upload the joint prediction model to the central node.

Fourth, the central node sends the joint prediction model trained in the third step to boiler 3, and uses the joint prediction model to predict the oxygen content value of the flue gas at boiler 3.

According to the technical solution provided by the embodiments of the present disclosure, the participants respectively determine the sample weights corresponding to the data sets of the local multiple groups of devices and the target device data sets; according to the data sets of the local multiple groups of devices and the corresponding sample weights, multiple groups The predictive neural network model of the device; upload the predictive neural network model of multiple groups of local devices to the central node for model aggregation to obtain the aggregated predictive neural network model; train the aggregated predictive neural network model according to preset training conditions to A joint prediction model is obtained; according to the joint prediction model and the sample weight corresponding to the target equipment data set, the oxygen content value of the flue gas of the target equipment is predicted. In order to solve the resource waste problem caused by the inability to improve the accuracy of load prediction of flue gas oxygen content of energy equipment in the prior art.

All the above optional technical solutions may be combined in any way to form optional embodiments of the present application, which will not be repeated here.

Fig. 3 is a schematic diagram of a flue gas oxygen content load prediction device provided by an embodiment of the present disclosure. As shown in Figure 3, the flue gas oxygen content load prediction device includes:

Determining module 301, used for participants to respectively determine the sample weights corresponding to the data sets of multiple groups of local devices and the data sets of the target device;

The first training module 302 is used to train the prediction neural network models of multiple sets of equipment according to the data sets of local multiple sets of equipment and the corresponding sample weights;

Aggregation module 303, for uploading the predictive neural network models of multiple groups of local devices to the central node for model aggregation, so as to obtain the aggregated predictive neural network models;

The second training module 304 trains the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model;

The prediction module 305 is configured to predict the oxygen content value of the flue gas of the target device according to the joint prediction model and the sample weight corresponding to the target device data set.

It should be understood that the sequence numbers of the steps in the above embodiments do not mean the order of execution, and the execution order of each process should be determined by its functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

Fig. 4 is a flow chart of another method for predicting the load of oxygen content in flue gas provided by an embodiment of the present disclosure. The method for predicting the load of flue gas oxygen content based on sample migration in FIG. 4 can be executed by the server in FIG. 1 . As shown in Figure 4, the flue gas oxygen load prediction method includes:

S401. Acquire device data of the first participant and device data of the second participant under the joint learning architecture.

Among them, the first participant is the participant who proposes the forecast demand, and the second participant is other participants except the first participant.

Specifically, by receiving the equipment data set from the first participant and the equipment data set from the second participant; and then filtering the equipment data set of the first participant and the equipment data set of the second participant according to the preset screening features , to obtain the sample size of the equipment data of the first participant and the volume of equipment data of the second participant respectively; Device data for one party and device data for a second party.

S402. Using the device data of the first participant and the device data of the second participant, train a prediction classifier.

Specifically, the device data of the first participant and the device data of the second participant can be processed through tagging to obtain the tag data of the device data of the first participant and the tag data of the device data of the second participant; The label data of the equipment data of one participant and the label data of the equipment data of the second participant to obtain the combined label data; finally, according to the combined label data, a predictive classifier is trained.

S403. Determine, according to the prediction classifier, weight data of the equipment data of the first participant with respect to the equipment data of the second participant.

Specifically, by using the predictive classifier, the equipment failure probability value corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment data of the second participant can be respectively obtained; and then according to the equipment data of the first participant The corresponding equipment failure probability value and the equipment failure probability value corresponding to the equipment data of the second participant determine the weight data of the equipment data of the first participant with respect to the equipment data of the second participant.

Further, for realizing using the predictive classifier to obtain the equipment failure probability value corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment data of the second participant, respectively, by using the predictive classifier, respectively Classifying the equipment data of the first participant and the equipment data of the second participant to obtain equipment failure data corresponding to the equipment data of the first participant and equipment failure data corresponding to the equipment data of the second participant; then, respectively Calculate the equipment failure probability value corresponding to the equipment failure data corresponding to the equipment data of the first participant and the equipment failure data corresponding to the equipment data of the second participant.

S404. Train a predictive gradient boosting model based on the device data of the second participant and the weight data.

Specifically, the predictive gradient boosting model can be trained based on the equipment data of the second participant and the weight data of the equipment data of the first participant with respect to the equipment data of the second participant; Data training predictive gradient boosting model to obtain the predicted value of the equipment of the first participant; then according to the norm of the error matrix between the predicted value of the equipment of the first participant and the expected value of the equipment, the fitness value of the predictive gradient boosting model is obtained; Finally, according to the fitness value of the predictive gradient boosting model, the particles in the population in the predictive gradient boosting model are updated to obtain an optimized predictive gradient boosting model.

Further, optimizing the predictive gradient boosting model can be achieved in the following ways:

First, determine the population in the predictive gradient boosting model and the particles in the population;

Then, it is judged whether the fitness value corresponding to the current particle in the population is greater than the fitness value of the previous old particle; if it is smaller, it is necessary to update the population and the particle in the population in the predictive gradient boosting model.

S405. Use the predictive gradient boosting model to predict the oxygen content load of the flue gas of the equipment of the first participant.

According to the technical solution provided by the embodiments of the present disclosure, by obtaining the device data of the first participant and the device data of the second participant under the joint learning architecture; wherein, the first participant is the participant who proposes the prediction demand, and the second participant The party is other parties except the first party; use the equipment data of the first party and the equipment data of the second party to train the predictive classifier; according to the predictive classifier, determine that the equipment data of the first party The weight data of the equipment data of the second participant; based on the equipment data and weight data of the second participant, train the predictive gradient boosting model; use the predictive gradient boosting model to predict the flue gas oxygen content load of the equipment of the first participant. In order to solve the problem of inaccuracy in flue gas oxygen content load prediction due to the difference in data distribution of energy equipment generated under different processes in the prior art, and save the cost of energy equipment sensors.

Fig. 5 is a schematic diagram of a flue gas oxygen content load prediction device provided by an embodiment of the present disclosure, which is applied in a joint learning framework. As shown in Figure 5, the flue gas oxygen content load prediction device based on sample migration includes:

The acquisition module 501 is used to acquire the equipment data of the first participant and the equipment data of the second participant under the joint learning framework; wherein, the first participant is the participant who proposes the prediction demand, and the second participant is the a party other than the party;

The first training module 502 uses the device data of the first participant and the device data of the second participant to train a predictive classifier;

A calculation module 503, configured to determine the weight data of the equipment data of the first participant with respect to the equipment data of the second participant according to the predictive classifier;

The second training module 504 is configured to train a predictive gradient boosting model based on the equipment data and weight data of the second participant;

The prediction module 505 is configured to use the predicted gradient boosting model to predict the flue gas oxygen content load of the equipment of the first participant.

According to the technical solution provided by the embodiments of the present disclosure, by obtaining the device data of the first participant and the device data of the second participant under the joint learning architecture; wherein, the first participant is the participant who proposes the prediction demand, and the second participant The party is other parties except the first party; use the equipment data of the first party and the equipment data of the second party to train the predictive classifier; according to the predictive classifier, determine that the equipment data of the first party The weight data of the equipment data of the second participant; based on the equipment data and weight data of the second participant, train the predictive gradient boosting model; use the predictive gradient boosting model to predict the flue gas oxygen content load of the equipment of the first participant. In order to solve the problem of inaccurate prediction of the oxygen content load in the flue gas caused by the difference in the data distribution of the energy equipment generated under different processes in the prior art, and save the cost of the energy equipment sensor.

Fig. 6 is a schematic flowchart of a method for selecting a flue gas oxygen content load prediction model provided by an embodiment of the present disclosure, and the method may be executed by the server in Fig. 1 . As shown in Figure 6, the selection method of the flue gas oxygen content load prediction model includes:

S601. Based on the federated learning architecture, receive a training data set and a testing data set from a prediction device of a participant.

Wherein, the training data set may be different energy equipment models (for example, different boiler models), and the test set data may be flue gas oxygen content data and corresponding characteristic data of energy equipment under different processes.

Specifically, according to the attributes of the prediction equipment, it can be determined that the attributes of the prediction equipment correspond to the oxygen content data of the flue gas of the prediction equipment; then, the features of the oxygen content data of the flue gas of the prediction equipment are extracted; The characteristics of the oxygen content data are used to construct the training data set and test data set of the prediction equipment respectively.

S602. Perform preprocessing on the data in the training data set and the data in the testing data set for predicting the device, and obtain a preprocessed device data set.

Specifically, it may be determined whether the data in the training data set and the test data set of the prediction device are abnormal; if there is an exception, the data in the training data set and the test data set of the prediction device are abnormally processed. Then, data normalization processing is performed on the data in the training data set and the data in the testing data set of the prediction device after the exception processing.

S603. Calculate the evaluation index value of each piece of data in the preprocessed device data set according to the established prediction model group.

Among them, the prediction model group can be composed of xgboost algorithm, SVR algorithm, neural network algorithm, belief network algorithm, decision tree algorithm, random forest regression algorithm, gradient boosting tree regression algorithm, linear regression algorithm, deep learning algorithm and other algorithms. The present invention is not limited.

Specifically, a prediction model group can be established according to the properties of the predicted device and each piece of data in the preprocessed device data set; then the root mean square error of each piece of data in the training set and the test set can be calculated using the prediction model group ; Furthermore, the root mean square error of each piece of data in the obtained training set and test set can be used as the evaluation index value of each piece of data in the preprocessed device data set.

Further, for the implementation of using the prediction model group to calculate the root mean square error of each piece of data in the training set and the test set, preferably, the training data set of the prediction device can be used to train the algorithm in the prediction model group to obtain Prediction result; the algorithm in the prediction model group can be trained by using the test data set of the prediction device to obtain the test result; then, according to the prediction result and the test result, the root mean square of each piece of data in the training set and the test set can be obtained error.

For example, calculate each piece of data in the training set and test set for prediction, get the root mean square error (set to rmse) of each piece of data, and use the root mean square error as the value of each piece of data in the preprocessed device data set Evaluate the index value, and then select the algorithm with the smallest rmse value as the corresponding algorithm label of the training set and test set (label each algorithm in the algorithm group, and the label value is 1, 2, 3, etc.)

About the process of obtaining rmse in the steps: use the training set to train the algorithm given by the algorithm group, use the test set to test the prediction results obtained by the training algorithm, and use the prediction results and the test set to obtain the rmse index. The calculation formula of rmse index is as follows:

Among them, n≥1 is the label value; y _i is the training set,

is the test set; i≥1 is the corresponding data set number.

S604, according to the smallest evaluation index value, determine a flue gas oxygen content load prediction model suitable for the prediction equipment.

Specifically, the evaluation index value of each piece of data in the preprocessed equipment data set can be sorted from small to large; and according to the sorting result, the smallest evaluation index value is selected; and then the corresponding prediction model in the prediction model group is called tag value;

When the minimum evaluation index value matches the label value corresponding to the prediction model in the prediction model group, it is determined that the prediction model corresponding to the label value corresponding to the prediction model is a flue gas oxygen content load prediction model suitable for the prediction equipment.

Among them, for the method of calling the label value corresponding to the prediction model in the prediction model group, the classification algorithm can be used to cluster the training data set of the prediction device to obtain at least two types of training cluster data; then, by calling the classification Classify the data of at least two types of training clusters; then according to the classified data of at least two types of training clusters, train at least two classifiers corresponding to at least two types of training clusters; Cluster prediction to obtain at least one class from at least two training clusters. According to at least one category in the at least two training clusters, determine the classifier corresponding to at least one category in the at least two training clusters; finally, determine the corresponding prediction model in the prediction model group according to the category label value corresponding to the classifier tag value.

Further example: the binary algorithm can be used to cluster the training set data to obtain the number of clusters K categories (K is a constant), and then use the classification algorithm to cluster the K categories to obtain the corresponding data in the K categories, and select a category For example, the gradient boosting regression tree classifies K types of data respectively, and the corresponding label of the training data (given by step S203) has K types of data, so K classifiers are trained. Then, the clustering prediction operation is performed on the test set data. The prediction result is a certain category in the 1-K clustering, and then the corresponding classifier is used to perform the classification operation to obtain an output result, which corresponds to the algorithm group. A predictive model, and then use this algorithm to predict the data, and then judge the correctness of the predictive model.

According to the technical solution provided by the embodiments of the present disclosure, based on the federated learning architecture, the training data set and the test data set of the prediction device from the participant are received; the data in the training data set and the test data set of the prediction device are preprocessed , and obtain the preprocessed equipment data set; according to the establishment of the prediction model group, calculate the evaluation index value of each piece of data in the preprocessed equipment data set; according to the minimum evaluation index value, determine the flue gas oxygen content suitable for predicting the equipment load forecasting model. In order to improve the prediction of the oxygen content of the flue gas of the energy equipment, and reduce the measurement cost of the existing technology.

Fig. 7 is a schematic diagram of a flue gas oxygen content load prediction model selection device provided by an embodiment of the present disclosure. As shown in Figure 7, the device includes:

The receiving module 701 is configured to receive training data sets and test data sets from prediction devices of participating parties based on the joint learning architecture;

A preprocessing module 702, configured to preprocess the data in the training data set and the test data set of the prediction device, and obtain a preprocessed device data set;

The calculation module 703 is used to calculate the evaluation index value of each piece of data in the preprocessed device data set according to the established prediction model group;

The prediction module 704 is configured to determine a flue gas oxygen content load prediction model suitable for the prediction device according to the smallest value of the evaluation index.

In the field of distributed energy, in order to save costs, small gas-fired boilers generally abandon the installation of zirconia measuring instruments, resulting in the inability to achieve closed-loop control and optimal thermal efficiency operation, especially when the calorific value of gas is unstable, and the oxygen content in flue gas cannot be accurately measured quantity. Fig. 8 is a schematic flowchart of a smoke emission prediction method provided by an embodiment of the present disclosure, and the smoke emission prediction method based on joint learning can be executed by the participants in Fig. 1 . As shown in Figure 8, the method includes:

S801. Train a local energy data measurement model according to the local energy data.

Specifically, local energy data can refer to the flue gas temperature, flue gas flow rate, and equipment inlet pressure of local equipment, such as steam boiler flue gas temperature, economizer outlet temperature, instantaneous value of flue gas flow, steam boiler gas temperature, steam boiler Flue gas standard flow rate, steam boiler natural gas inlet pressure, steam boiler flue gas flow rate, steam boiler condenser inlet flue temperature, steam boiler exhaust gas temperature, steam boiler flue gas pressure, steam boiler condenser inlet pressure, steam boiler main steam instantaneous Flow rate, operating status of steam boiler, instantaneous flow rate of natural gas inlet of steam boiler, etc.

Furthermore, before training the local energy data measurement model according to the local energy data, the following methods can also be used to organize or filter the local energy data, test data and energy data of the target energy equipment: first, select the sample data set; Labeling the data in the data set, and obtaining the label data corresponding to the data in the sample data set; respectively determining the label data corresponding to the local energy data, test data, and energy data of the target energy equipment.

S802. Based on the joint learning framework, train the local energy data measurement model according to the test data and the energy data of the target energy equipment, and respectively obtain the test data prediction model and the target energy data prediction model.

Wherein, the test data may be selected from local energy data, or may be energy data extracted from other related equipment; the energy data of the target energy equipment may be the energy data of the equipment to be predicted.

Specifically, it can be achieved in the following ways: Based on the federated learning framework, the participants send the local energy data prediction model to the central node; in response to the feedback information of the central node, train the local energy data according to the test data and the energy data of the target energy equipment The test model is used to obtain the test data prediction model and the target energy data prediction model respectively.

S803. Calculate the first sample migration weight and the second sample migration weight based on the local energy data prediction model, the test data prediction model, and the target energy data prediction model; wherein, the first sample migration weight is the local energy data for the target energy equipment The sample migration weight of the energy data, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device.

Specifically, in the framework based on federated learning, first, the participants send the local energy data prediction model, test data prediction model and target energy data prediction model to the central node; the central node can After the local energy data prediction model, test data prediction model and target energy data prediction model are sorted or adjusted, they are sent to relevant participants; then, in response to the feedback information from the central node, based on the local energy data prediction model, test data prediction The model and the target energy data prediction model perform target classification on the local energy data, test data and energy data of the target energy equipment respectively; finally, calculate the first sample migration weight and the second sample migration weight according to the target classification.

S804. Using the local energy data and the first sample transfer weight, the test data and the second sample transfer weight, respectively train the local energy data network model and the test data network model.

Specifically, it is preferable to establish an application data set by using the local energy data and the first sample migration weight; and then according to the established application data set, train the local energy data network model; and then use the test data and the second sample migration weight, Establish the expected data set; then, the test data network model can be trained according to the expected data set. The local energy data network model and the test data network model can be trained in parallel, or one of the two models can be trained first, which is not limited in this disclosure.

S805. Receive a joint learning prediction model from the central node after aggregate training of the local energy data network model and the test data network model.

Specifically, the local energy data network model and the test data network model can be uploaded to the central node; the central node performs aggregation training on the local energy data network model and the test data network model. Then, after receiving the joint learning prediction model from the central node, the local energy data network model and the test data network model are aggregated and trained.

Furthermore, the joint learning prediction model can be optimized according to the prediction conditions; wherein the prediction conditions include: the prediction value of the model parameters and the judgment of the fitness of the model parameters.

S806. Predict the flue gas emission of the target energy equipment according to the joint learning prediction model.

According to the technical solution provided by the embodiments of the present disclosure, the local energy data measurement model is trained according to the local energy data; based on the joint learning framework, the local energy data measurement model is trained according to the test data and the energy data of the target energy equipment, and the test data predictions are respectively obtained model and target energy data prediction model; based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, the first sample migration weight and the second sample migration weight are calculated, wherein the first sample migration weight is the local The energy data is aimed at the sample migration weight of the energy data of the target energy device, and the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device; using the local energy data and the first sample migration weight, the test data and the second sample migration weight Two-sample migration weights to train the local energy data network model and the test data network model respectively; receive the joint learning prediction model after the aggregation training of the local energy data network model and the test data network model from the central node; according to the joint learning prediction model, the target energy Equipment for flue gas emission predictions. In order to solve the problem of inaccurate measurement of flue gas emission due to the different distribution of equipment in the prior art. This in turn saves resource costs for the actual sensor installation.

Fig. 9 is a schematic diagram of a flue gas emission prediction device provided by an embodiment of the present disclosure. As shown in Figure 9, the flue gas emission prediction device includes:

The first training module 901 is used to train the local energy data measurement model according to the local energy data;

The second training module 902 is used to train the local energy data measurement model based on the joint learning framework according to the test data and the energy data of the target energy equipment, and respectively obtain the test data prediction model and the target energy data prediction model;

Calculation module 903, configured to calculate the first sample migration weight and the second sample migration weight based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, wherein the first sample migration weight is the local energy data For the sample migration weight of the energy data of the target energy device, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device;

The third training module 904 is used to train the local energy data network model and the test data network model respectively by using the local energy data and the first sample transfer weight, test data and the second sample transfer weight;

Establishment module 905, used to receive the joint learning prediction model from the central node after the aggregation and training of the local energy data network model and the test data network model;

The prediction module 906 is used to predict the smoke emission of the target energy equipment according to the joint learning prediction model.

FIG. 10 is a schematic diagram of a computer device 10 provided by an embodiment of the present disclosure. As shown in FIG. 10 , the computer device 10 of this embodiment includes: a processor 1001 , a memory 1002 , and a computer program 1003 stored in the memory 1002 and capable of running on the processor 1001 . When the processor 1001 executes the computer program 1003, the steps in the foregoing method embodiments are implemented. Alternatively, when the processor 1001 executes the computer program 1003, the functions of the modules/units in the foregoing device embodiments are implemented.

Exemplarily, the computer program 1003 can be divided into one or more modules/units, and one or more modules/units are stored in the memory 1002 and executed by the processor 1001 to complete the present disclosure. One or more modules/units may be a series of computer program instruction segments capable of accomplishing specific functions, and the instruction segments are used to describe the execution process of the computer program 1003 in the computer device 10 .

The computer device 10 may be a computer device such as a desktop computer, a notebook, a palmtop computer, or a cloud server. The computer device 10 may include, but is not limited to, a processor 1001 and a memory 1002 . Those skilled in the art can understand that FIG. 4 is only an example of the computer device 10, and does not constitute a limitation to the computer device 10. It may include more or less components than those shown in the illustration, or combine certain components, or different components. , for example, computer equipment may also include input and output equipment, network access equipment, bus, and so on.

The processor 1001 may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), on-site Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

The memory 1002 may be an internal storage unit of the computer device 10 , for example, a hard disk or a memory of the computer device 10 . The memory 1002 can also be an external storage device of the computer device 10, for example, a plug-in hard disk equipped on the computer device 10, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card ( Flash Card), etc. Further, the storage 1002 may also include both an internal storage unit of the computer device 10 and an external storage device. The memory 1002 is used to store computer programs and other programs and data required by the computer equipment. The memory 1002 can also be used to temporarily store data that has been output or will be output.

Those skilled in the art can clearly understand that for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used for illustration. In practical applications, the above-mentioned functions can be assigned to different functional units, Completion of modules means that the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment may be integrated into one processing unit, or each unit may exist separately physically, or two or more units may be integrated into one unit, and the above-mentioned integrated units may adopt hardware It can also be implemented in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the present application. For the specific working processes of the units and modules in the above system, reference may be made to the corresponding processes in the aforementioned method embodiments, and details will not be repeated here.

In the above-mentioned embodiments, the descriptions of each embodiment have their own emphases, and for parts that are not detailed or recorded in a certain embodiment, refer to the relevant descriptions of other embodiments.

Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementation should not be considered beyond the scope of the present disclosure.

In the embodiments provided in the present disclosure, it should be understood that the disclosed apparatus/computer equipment and methods may be implemented in other ways. For example, the device/computer device embodiments described above are only illustrative, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, and multiple units or components can be Incorporation may either be integrated into another system, or some features may be omitted, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

A unit described as a separate component may or may not be physically separated, and a component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If an integrated module/unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure realizes all or part of the processes in the methods of the above embodiments, and can also be completed by instructing related hardware through computer programs. The computer programs can be stored in computer-readable storage media, and the computer programs can be processed. When executed by the controller, the steps in the above-mentioned method embodiments can be realized. A computer program may include computer program code, which may be in source code form, object code form, executable file, or some intermediate form or the like. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory (Read-Only Memory, ROM), random access Memory (Random Access Memory, RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer readable media may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, computer readable media may not Including electrical carrier signals and telecommunication signals.

The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be described in the foregoing embodiments Modifications to the technical solutions recorded, or equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in this disclosure. within the scope of protection.

Claims

A flue gas oxygen content load prediction method, characterized in that, comprising:

The participants respectively determine the sample weights corresponding to the data sets of the local multiple groups of devices and the target device data sets;

According to the data sets of multiple sets of local equipment and the corresponding sample weights, train the prediction neural network model of multiple sets of equipment;

Upload the predictive neural network models of multiple groups of local devices to the central node for model aggregation to obtain the aggregated predictive neural network models;

Train the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model;

According to the joint prediction model and the sample weight corresponding to the target equipment data set, the oxygen content value of the flue gas of the target equipment is predicted.
The method according to claim 1, wherein the participants respectively determine the sample weights corresponding to the data sets of the local multiple groups of devices and the target device data sets include:

The participant selects the data sets of multiple sets of local devices and the data sets of the target device;

Merge the data sets of the local multiple groups of devices with the data sets of the target device to obtain the merged data;

training a kernel density estimation model using the merged data;

According to the kernel density estimation model, the sample weights corresponding to the data sets of the local multiple groups of devices and the data sets of the target device are respectively determined.
The method according to claim 1, wherein, according to the joint prediction model and the sample weight corresponding to the target equipment data set, predicting the oxygen content value of the flue gas of the target equipment includes:

Participants upload the joint prediction model to the central node for joint learning and training;

Joint learning and training of the joint prediction model in response to feedback from the receiving engine;

Send the joint prediction model trained by joint learning to the target device;

According to the joint prediction model and the sample weight corresponding to the target equipment data set, the oxygen content value of the flue gas of the target equipment is predicted.
The method according to claim 1, further comprising:

Participants use the data sets of multiple sets of local devices to establish training samples for predictive neural network models;

Use the data set of the target device to establish a predictive neural network model test sample;

Obtaining a sample prediction value according to the model training sample and the model testing sample;

According to the norm of the error matrix between the sample predicted value and the sample expected value, the fitness value of the predicted neural network model is obtained;

The particles in the population in the prediction neural network model are updated according to the fitness value to obtain an optimized prediction neural network model.
A flue gas oxygen content load prediction device, characterized in that it comprises:

A determination module, used for the participants to respectively determine the sample weights corresponding to the data sets of multiple groups of local devices and the data sets of the target device;

The first training module is used to train the prediction neural network models of multiple sets of equipment according to the data sets of multiple sets of local equipment and the corresponding sample weights;

The aggregation module is used to upload the predictive neural network models of multiple groups of local devices to the central node for model aggregation, so as to obtain the aggregated predictive neural network models;

The second training module trains the aggregated prediction neural network model according to preset training conditions to obtain a joint prediction model;

The prediction module is used to predict the oxygen content value of the flue gas of the target device according to the joint prediction model and the sample weight corresponding to the target device data set.
A flue gas oxygen content load prediction method, characterized in that the method is applied in a joint learning framework, including:

Obtain the equipment data of the first participant and the equipment data of the second participant under the federated learning architecture; among them, the first participant is the participant who proposes the prediction demand, and the second participant is other participants except the first participant square;

using the device data of the first party and the device data of the second party to train a predictive classifier;

determining weight data of the device data of the first party with respect to the device data of the second party based on the predictive classifier;

training a predictive gradient boosting model based on the device data of the second participant and the weight data;

Using the predictive gradient boosting model to predict the flue gas oxygen content load of the equipment of the first party.
The method according to claim 6, wherein acquiring the equipment data of the first participant and the equipment data of the second participant under the joint learning framework comprises: receiving the equipment data set and the second participant's equipment data from the first participant According to the preset screening features, the equipment data set of the first party and the equipment data set of the second party are screened to obtain the sample size of the equipment data of the first party and the data set of the second party respectively. Amount of equipment data; determining the sample volume of equipment data of the first participant and the amount of equipment data of the second participant as the equipment data of the first participant and the equipment data of the second participant, respectively;

Alternatively, using the equipment data of the first participant and the equipment data of the second participant, training the predictive classifier includes: labeling and processing the equipment data of the first participant and the equipment data of the second participant to obtain the first participant The tag data of the device data of the first party and the tag data of the device data of the second party; the tag data of the device data of the first party and the tag data of the device data of the second party are merged to obtain the combined tag data; Train a predictive classifier on the combined labeled data.
The method according to claim 6, wherein, according to the predictive classifier, determining the weight data of the equipment data of the first participant with respect to the equipment data of the second participant comprises:

Using the predictive classifier, respectively obtain the equipment failure probability value corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment data of the second participant;

According to the equipment failure probability value corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment data of the second participant, determine the weight data of the equipment data of the first participant with respect to the equipment data of the second participant .
The method according to claim 8, characterized in that, using the predictive classifier, respectively obtaining the equipment failure probability value corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment data of the second participant include :

Using the predictive classifier, respectively classify the equipment data of the first participant and the equipment data of the second participant, so as to obtain the equipment fault data corresponding to the equipment data of the first participant and the equipment data corresponding to the second participant equipment failure data;

The equipment failure data corresponding to the equipment data of the first participant and the equipment failure probability value corresponding to the equipment failure data corresponding to the equipment data of the second participant are respectively calculated.
The method according to claim 6, wherein, based on the equipment data of the second participant and the weight data, training the predictive gradient boosting model further comprises:

training a predictive gradient boosting model based on the device data of the second participant and the weight data;

Obtain the test data of the first participant to train the predictive gradient boosting model to obtain the device prediction value of the first participant;

Obtain the fitness value of the predictive gradient boosting model according to the norm of the error matrix between the predicted value of the equipment of the first participant and the expected value of the equipment;

According to the fitness value, the particles in the population in the predictive gradient boosting model are updated to obtain an optimized predictive gradient boosting model.
The method according to claim 10, wherein the particles in the population in the predictive gradient boosting model are updated according to the fitness value, so as to obtain an optimized predictive gradient boosting model comprising:

determining a population in the predictive gradient boosting model and particles in the population;

Judging whether the fitness value corresponding to the particle in the current population is greater than the fitness value of the previous old particle;

If it is less than, the population in the predictive gradient boosting model and the particles in the population need to be updated.
A flue gas oxygen content load forecasting device, characterized in that the application of the device in the joint learning framework includes:

The acquisition module is used to acquire the equipment data of the first participant and the equipment data of the second participant under the joint learning architecture; wherein, the first participant is the participant who proposes the prediction demand, and the second participant is the participant other than the first participant parties other than the Party;

The first training module uses the device data of the first participant and the device data of the second participant to train a predictive classifier;

A calculation module, configured to determine weight data of the equipment data of the first participant with respect to the equipment data of the second participant according to the predictive classifier;

A second training module, configured to train a predictive gradient boosting model based on the equipment data of the second participant and the weight data set;

A prediction module, configured to use the predictive gradient boosting model to predict the flue gas oxygen content load of the equipment of the first participant.
A method for selecting a flue gas oxygen content load prediction model, characterized in that it includes:

Based on the federated learning architecture, receive the training data set and test data set from the prediction equipment of the participating parties;

Preprocess the data in the training data set and the test data set of the prediction device, and obtain the preprocessed device data set;

Calculate the evaluation index value of each piece of data in the preprocessed equipment data set according to the established prediction model group;

According to the minimum evaluation index value, the flue gas oxygen content load prediction model suitable for the prediction equipment is determined.
The method according to claim 13, wherein, according to the established prediction model group, calculating the evaluation index value of each piece of data in the preprocessed equipment data set includes: according to the properties of the predicted equipment and the predicted preprocessed equipment For each piece of data in the data set, a prediction model group is established; using the prediction model group, the root mean square error of each piece of data in the training set and test set is calculated respectively; the root mean square error of each piece of data in the training set and test set is obtained Error, as the evaluation index value of each piece of data in the preprocessed device data set;

Alternatively, according to the smallest evaluation index value, determining a flue gas oxygen content load prediction model suitable for predicting equipment includes: sorting the evaluation index values of each piece of data in the preprocessed equipment data set from small to large; according to the sorting As a result, select the minimum evaluation index value; call the label value corresponding to the prediction model in the prediction model group; when the minimum evaluation index value matches the label value corresponding to the prediction model in the prediction model group, determine the The prediction model corresponding to the above label value is the flue gas oxygen content load prediction model suitable for the prediction equipment.
A flue gas oxygen content load prediction model selection device, characterized in that it includes:

The receiving module is used to receive training data sets and test data sets from prediction devices of participating parties based on the federated learning architecture;

The preprocessing module is used to preprocess the data in the training data set and the test data set of the prediction device, and obtain the preprocessed device data set;

A calculation module, configured to calculate the evaluation index value of each piece of data in the preprocessed device data set according to the established prediction model group;

The prediction module is used to determine a flue gas oxygen content load prediction model suitable for the prediction equipment according to the minimum value of the evaluation index.
A flue gas emission prediction method, characterized in that it comprises:

According to the local energy data, train the local energy data measurement model;

Based on the joint learning framework, the local energy data measurement model is trained according to the test data and the energy data of the target energy equipment, and the test data prediction model and the target energy data prediction model are respectively obtained;

Based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, calculate the first sample migration weight and the second sample migration weight, wherein the first sample migration weight is the local energy data for the target energy equipment The sample migration weight of the energy data, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device;

Using the local energy data and the first sample migration weight, the test data and the second sample migration weight, respectively train the local energy data network model and the test data network model;

Receive the joint learning prediction model from the central node after the aggregation and training of the local energy data network model and the test data network model;

According to the joint learning prediction model, the flue gas emission of the target energy equipment is predicted.
The method according to claim 16, wherein, based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, calculating the first sample transfer weight and the second sample transfer weight comprises: based on a joint learning framework , the participant sends the local energy data prediction model, test data prediction model and target energy data prediction model to the central node; in response to the feedback information of the central node, based on the local energy data prediction model, test data prediction model and target energy data A predictive model, performing target classification on the local energy data, test data and energy data of the target energy equipment respectively; calculating the first sample migration weight and the second sample migration weight according to the target classification;

Alternatively, using the local energy data and the first sample migration weights, the test data and the second sample migration weights, respectively training the local energy data network model and the test data network model includes: using the local energy data and the first sample migration weights, establishing Applying the data set; according to the application data set, training the local energy data network model; using the test data and the second sample transfer weight to establish the expected data set;

According to the expected data set, the test data network model is trained.
A flue gas emission prediction device, characterized in that it comprises:

The first training module is used to train the local energy data measurement model according to the local energy data;

The second training module is used to train the local energy data measurement model based on the joint learning framework according to the test data and the energy data of the target energy equipment, and respectively obtain the test data prediction model and the target energy data prediction model;

A calculation module, configured to calculate the first sample migration weight and the second sample migration weight based on the local energy data prediction model, the test data prediction model and the target energy data prediction model, wherein the first sample migration weight is the local energy The sample migration weight of the data for the energy data of the target energy device, the second sample migration weight is the sample migration weight of the test data for the energy data of the target energy device;

The third training module is used to train the local energy data network model and the test data network model respectively by using the local energy data and the first sample transfer weight, the test data and the second sample transfer weight;

Establishing a module for receiving a joint learning prediction model from the central node after the aggregation and training of the local energy data network model and the test data network model;

The prediction module is used to predict the flue gas emission of the target energy equipment according to the joint learning prediction model.
A computer device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, when the processor executes the computer program, the computer program according to claim 1 is implemented. steps of the method described above.
A computer-readable storage medium storing a computer program, wherein the computer program implements the steps of the method according to claim 1 when the computer program is executed by a processor.