CN115717709B - Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model - Google Patents
Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model Download PDFInfo
- Publication number
- CN115717709B CN115717709B CN202211417405.6A CN202211417405A CN115717709B CN 115717709 B CN115717709 B CN 115717709B CN 202211417405 A CN202211417405 A CN 202211417405A CN 115717709 B CN115717709 B CN 115717709B
- Authority
- CN
- China
- Prior art keywords
- garbage
- model
- content
- furnace
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Incineration Of Waste (AREA)
Abstract
The invention relates to a heat value prediction technology of garbage in a furnace, and aims to provide a real-time prediction method of the heat value of garbage in the furnace based on an attention mechanism LSTM model. Comprising the following steps: extracting DCS control parameters from the historical data record, screening out input characteristic parameters for model training, extracting part of S parameters therefrom, calculating the calorific value of the garbage fed into the furnace, and using the garbage for model training; establishing an LSTM time sequence model based on a time attention mechanism as a training model; screening the model stored after multiple training, and selecting the model with the minimum root mean square error as a final prediction model; and extracting an input characteristic parameter from the DCS control parameter at the current moment, inputting a prediction model, and obtaining a predicted value of the heat value of the garbage entering the furnace in the next time step through calculation. The invention can effectively and accurately calculate the heat value of the garbage in the garbage incinerator, further uses the predicted result of the heat value of the garbage to adjust the control strategy, can reduce the emission of pollutants and improve the efficiency of incineration power generation.
Description
Technical Field
The invention relates to the technical field of heat value prediction of garbage in a furnace, in particular to a method for predicting the heat value of garbage in the furnace in real time based on an attention mechanism LSTM model.
Background
The garbage incineration technology is a main means for recycling and harmless utilization of household garbage, but the heat value of the garbage entering the furnace cannot be estimated in real time due to complex components and severe heat value fluctuation of the household garbage. Therefore, the real-time measurement and prediction of the heat value of the garbage in the garbage incinerator are needed, and the harmless, recycling and sustainable development of garbage treatment are realized.
Currently, techniques or main measurement methods adopted by technicians for garbage heat value measurement and estimation include: an oxygen bomb calorific value method, a formula estimation method (Dulong/Scheurer-Kestner and the like), a household garbage calorific value prediction method by utilizing grey correlation degree, and a method for establishing a garbage calorific value prediction model by using various neural networks. However, the methods have the characteristics of low precision, complex calculation model, difficult deployment, weak generalization and the like, and the real-time prediction of the garbage heat value cannot be realized. In the current research, an effective technical method capable of predicting the garbage calorific value in real time is lacking.
At present, only operation and maintenance personnel on the power plant site manually and roughly estimate the garbage heat value through the change trend of the steam load. Because the manual monitoring mode cannot accurately estimate the heat value of the garbage, hysteresis exists in control and regulation, and the combustion in the incinerator is easy to generate larger fluctuation. Therefore, it is highly desirable to provide a real-time measurement method for the heat value of the garbage in the furnace, which meets the actual requirements.
Disclosure of Invention
The invention aims to solve the technical problem of overcoming the defects in the prior art and providing a method for predicting the heat value of the garbage in the furnace in real time based on an attention mechanism LSTM model.
In order to solve the technical problems, the invention adopts the following solutions:
the method for predicting the heat value of the garbage in the furnace in real time based on the attention mechanism LSTM model comprises the following steps:
(1) Extracting DCS control parameters of different time points from historical data records of a DCS control system of the garbage incinerator; importing data into an industrial personal computer according to a fixed time interval to obtain a data set to be sorted;
(2) Based on the calculation of the correlation coefficient, SO is screened from the data set to be sorted 2 Concentration, flue gas temperature, upper temperature of furnace combustion grate, primary air temperature, smoke concentration, NOx content, HCl content, CO content and wet base CO 2 Content, fluoride content, dry basis O 2 The content and the flue gas humidity are 12 input characteristic parameters in total; performing data cleaningAfter washing, the data of the 12 input features are used for model training;
(3) Extracting DCS parameters for calculating C, H, O, S and moisture content parameters from the input characteristic parameters obtained in the step (2), calculating the heat value of the garbage entering the furnace at each moment, and adding a label for model training for the calculation result;
(4) Integrating the input characteristic parameters obtained in the step (2) with the heat value of the garbage fed into the furnace at each moment obtained in the step (3); then 80% of the total data was used as training set data and 20% was used as test set data;
(5) Establishing an LSTM time sequence model based on a time attention mechanism as a training model;
inputting training set data into a training model for training, and optimizing the state quantity h of a hidden layer by multiplying the state quantity h of the hidden layer of the LSTM model by a dot product of the attention moment matrix;
synchronously calculating the loss between the predicted value and the true value by using the root mean square error loss, and after each training round, reversely transmitting and updating the state quantity h of the predicted model at the next moment to gradually reduce the Euclidean distance between the predicted value of the heat value of the garbage of the furnace and the calculated result until the root mean square error loss is less than 1%, and storing the model;
screening the model stored after multiple training, and selecting the model with the minimum root mean square error as a final prediction model;
(6) And (3) extracting an input characteristic parameter from the DCS control parameter at the current time, inputting the prediction model obtained in the step (5), and obtaining the predicted value of the heat value of the garbage entering the furnace in the next time step through calculation.
The invention also provides a method for further improving the control strategy of the garbage incinerator by using the prediction result obtained by the prediction method, which comprises the following steps:
(a) Calculating the calorific value of the waste in the furnace at the current moment by referring to the contents of the steps (1) - (3) in the claim 1;
(b) Calculating the difference value between the heat value of the garbage entering the furnace at the current moment and the predicted value of the heat value of the garbage entering the furnace at the next time step, and the change condition of the amplitude of the difference value;
(c) And (c) according to the calculation result of the step (b), adjusting and optimizing the garbage feeding rate, ensuring that the total calorific value of the garbage fed into the incinerator in unit time tends to be stable, and keeping the garbage combustion state in the incinerator stable.
Description of the inventive principles:
the heat value of the garbage in the household garbage grate incinerator is determined by key factors such as the structure, the components and the like of the input materials, and the garbage heat value does not periodically change and fluctuates in distance at different continuous time points due to certain randomness of the components in the garbage. For the time sequence model, although the heat value of the garbage is interfered by uncertain factors such as the components of the garbage to be charged, the water content and the like, the future heat value change trend of the garbage is strongly related to the historical change trend. The invention utilizes the LSTM time sequence model to mine the change rule existing between the heat value trend of the future time period and the heat value trend of the historical time period, and the time attention mechanism is added to enable a computer to find the rule corresponding to the model in time areas with different scales.
The heat value of the garbage is difficult to measure in real time due to the complex components and the large difference of the garbage components at different moments. The current practice of operators is to use a sampling method to detect the calorific value of the garbage, and the period is usually one week or one month. Because the real-time garbage heat value cannot be obtained as the target value to participate in the neural network training, researchers can only train in the existing supervision neural network even through a data learning method, so that a prediction model meeting the actual application scene requirement is not reported until now, and no publication document records the research result of heat value prediction by using an LSTM time sequence model.
The invention creatively provides that the real-time CO content and CO in the DCS parameters are extracted in real time 2 The content, the oxygen content and the smoke water content are used for calculating the real-time numerical values of C, H, O, S and the water content, and the element content is used for calculating the real-time heat value so as to complete the LSTM time sequence model training process of supervised learning. In addition, compared with the traditional time sequence model, the LSTM time sequence model based on the time attention mechanism can enable the model to be enabled by adjusting weight to simulate the emphasis of human attention when information is processedFocusing on finding the relationship between each feature quantity and the target quantity on the local time series. Especially for the application scene, the requirement on the prediction time interval is not long, and the model modified by the method is more beneficial to searching for the characteristic variable trend in a short time interval.
According to the invention, the DCS data of each point location is finally extracted in real time and is input into the LSTM time sequence model based on the time attention mechanism, and the neural network is used for fitting and mining the relevance among the garbage heat value, each piece of DCS data and the historical working condition, so that the real-time prediction of the garbage heat value is realized.
Compared with the prior art, the invention has the beneficial effects that:
1. compared with the on-site operators which estimate the heat value of the garbage according to the data such as the load of the hearth or manual inaccuracy, the invention has more scientificalness and real-time performance, can efficiently and accurately calculate the heat value of the garbage in the garbage incinerator, and can realize the prediction of the heat value of the garbage in the incinerator and the diagnosis of stable combustion in the hearth.
2. Compared with a model obtained by training by using a supervision neural network in the prior study, the invention realizes real-time prediction of the garbage heat value and is used for adjusting a control strategy by extracting the DCS data of each point location in real time and inputting the DCS data into an LSTM time sequence model based on a time attention mechanism and fitting and excavating the relevance among the garbage heat value, each item of DCS data and the historical working condition by using the neural network, thereby reducing pollutant emission and improving the efficiency of incineration power generation. Therefore, the invention meets the requirements of practical application scenes.
Drawings
FIG. 1 is a schematic diagram of each designated point of a household garbage incinerator DCS;
FIG. 2 is a schematic diagram of a real-time prediction flow of the calorific value of the garbage entering the furnace;
FIG. 3 is a graph of MSE_loss variation trend after training in a specific application example;
fig. 4 is a graph showing comparison between predicted and actual values of the test heat collection value in a specific application example.
Detailed Description
Firstly, it should be noted that the present invention relates to a deep learning technology, which is an application of computer technology in the fields of industrial prediction and industrial control. In the implementation of the present invention, the application of multiple software functional modules may be involved. The applicant believes that the software programming skills of one skilled in the art would be fully available to practice the present invention in conjunction with the prior art, as the application document is read, with an accurate understanding of the principles and objects of the present invention. The foregoing software functional modules include, but are not limited to: the DCS parameter extraction module, the garbage heat value prediction module and the like belong to the category of the application files of the invention, and the applicant does not enumerate one by one.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described with reference to the accompanying drawings and specific embodiments, which are merely the most basic embodiments of the present invention, but not all embodiments. Other embodiments according to the invention are within the scope of the invention.
The invention discloses a method for predicting the heat value of garbage entering a furnace in real time based on an attention mechanism LSTM model, which comprises the following steps:
1. extracting DCS control parameters under different time points from historical data records of a DCS control system of the garbage incinerator; then importing the data into an industrial personal computer to obtain a data set to be sorted;
as shown in fig. 1, the extracted DCS control parameters include: in the garbage incinerator, the upper temperature of the fire grate at the point 1 and the primary air temperature at the point 2; in the chimney, flue gas SO at point 3 2 Concentration, flue gas temperature at point 4, smoke concentration at point 5, NO at point 6 x Concentration of HCl at point 7, CO at point 8, CO at point 9 2 Concentration, fluoride concentration at point 10, O at point 11 2 Concentration, flue gas humidity at point 12.
2. Based on the calculation of the correlation coefficient, SO is screened from the data set to be sorted 2 Concentration, flue gas temperature, upper temperature of furnace combustion grate, primary air temperature, smoke concentration, NOx content, HCl content, CO content and wet base CO 2 Content, fluorinationContent of substances, dry basis O 2 The content and the flue gas humidity are 12 input characteristic parameters in total, and the total of 12 input characteristic parameters are: after data cleansing, the data of these 12 input features were used for model training.
The screening refers to: calculating the pearson correlation coefficient of the input characteristic parameter and the real-time target heat value, and reserving the input characteristic parameter with the correlation coefficient larger than 0.3 for inputting a training model;
wherein, the calculation of pearson correlation coefficient pearson is shown as the following formula:
wherein: x is x i For a certain input parameter to be selected, y i The calculation result is the real-time target calorific value;for the mean value of the input parameter set, +.>The average value of the real-time target heat value group; i=1, 2..n.
The data cleaning refers to: and cleaning the extracted input characteristic parameter data according to the control parameter range in the operation process of the garbage incinerator, and removing the abnormal working condition value.
3. Extracting DCS parameters for calculating C, H, O, S, water content and other parameters from the input characteristic parameters obtained in the step 2, calculating the heat value of the garbage entering the furnace at each moment, and adding a label for a calculation result to be used for model training;
the DCS parameters extracted from the dataset include: CO content and CO for calculating carbon content 2 The content is as follows; the flue gas water content is used for calculating the hydrogen content; an oxygen content for calculating an oxygen content; the flue gas water content is used for calculating the moisture content, and the dry flue gas density and the flue gas water content are used for calculating the content of each element.
The heat value of the garbage entering the furnace at each moment is obtained through calculation according to the following formula; wherein formula (2) is used when the oxygen content exceeds 10%, and formula (3) is used when the oxygen content is below 10%:
wherein Q is the calorific value of the garbage in the furnace, and the unit is: kJ/kg; c is the carbon content in the garbage entering the furnace, and the unit is: the%; h is the hydrogen content in the garbage entering the furnace, and the unit is: the%; o is the oxygen content in the in-furnace garbage, in units of: the%; s is the sulfur content of the garbage entering the furnace, and the unit is: the%; w is the moisture content of the garbage entering the furnace, and the unit is: percent of the total weight of the composition.
Wherein the carbon content is as defined in (CO and CO 2 The sum of the contents of (1)/(the sum of the density of dry flue gas and the water content of flue gas); the hydrogen content is calculated according to (1/9 of the flue gas water content)/(the sum of the dry flue gas density and the flue gas water content); the oxygen content is calculated according to (the sum of the oxygen content/(the dry smoke density and the smoke water content), the sulfur content is smaller in the smoke components and calculated according to the fixed value of 0.1%, and the moisture content is calculated according to the smoke water content/(the sum of the dry smoke density and the smoke water content).
4. Integrating the input characteristic parameters obtained in the step 2 with the heat value of the garbage fed into the furnace at each moment obtained in the step 3; 80% of the total data was then used as training set data and 20% as test set data.
5. And establishing an LSTM time sequence model based on a time attention mechanism as a training model.
The training model is an LSTM time sequence model based on a time attention mechanism, wherein the input time sequence step length is 3, the output time sequence step length is 1, the input feature dimension is 12, the hidden layer of the LSTM model is 128 layers, the training round number is 60, and the learning rate is 0.48.
Inputting training set data into a training model for training, carrying out dot product multiplication on a hidden layer state quantity h of the LSTM model and an attention moment array through calculation of the following formulas (4) - (6), and optimizing the state quantity h of the hidden layer:
a·b＝a 1 b 1 +a 2 b 2 +...+a n b n (6)
wherein a and b are each a vector array containing n elements.
The loss between the predicted value and the true value is calculated using the root mean square error loss for synchronization:
wherein MSE_loss is root mean square error loss; n is the number of predicted instances, y i For a certain predicted heating value of the LSTM timing model,predicting the mean value of the heat value for the LSTM time sequence model; i=1, 2..n.
After each training round, the state quantity h of the prediction model at the next moment is required to be reversely transmitted and updated, so that the Euclidean distance between the predicted value of the heat value of the furnace garbage and the calculated result is gradually reduced until the root mean square error loss is less than 1%, and the model is saved;
the back propagation update specifically refers to: repeating the input feature vector x acting on different moments t using the same calculation unit t State vector h at this point t-1 Generating a new state vector h at the next moment t The method comprises the steps of carrying out a first treatment on the surface of the State quantity h to be used for updating t And (3) carrying out reverse gradient propagation, starting from an output layer of the model, solving the model gradient layer by layer from back to front by utilizing a chain rule of function derivation, thereby realizing optimal solution and omitting repeated derivation steps.
Screening the model stored after multiple training, and selecting the model with the minimum root mean square error as a final prediction model.
6. And (3) extracting an input characteristic parameter from the DCS control parameter at the current time, inputting the prediction model obtained in the step (5), and obtaining the predicted value of the heat value of the garbage entering the furnace in the next time step through calculation.
The invention further provides a method for further improving the control strategy of the garbage incinerator by utilizing the predicted result, which specifically comprises the following steps:
(a) Calculating the heat value of the garbage entering the furnace at the current moment by referring to the content of the step 1-3;
(b) Calculating the difference value between the heat value of the garbage entering the furnace at the current moment and the predicted value of the heat value of the garbage entering the furnace at the next time step, and the change condition of the amplitude of the difference value;
(c) And (c) according to the calculation result of the step (b), adjusting and optimizing the garbage feeding rate, ensuring that the total calorific value of the garbage fed into the incinerator in unit time tends to be stable, and keeping the garbage combustion state in the incinerator stable.
Specific application examples:
a typical garbage incineration power plant with a treatment capacity of 750 tons/day in China is selected for testing, the treated garbage is mainly urban household garbage, the data is normal operation data of DCS control parameters of various points of an incinerator for 30 days in a certain month, the data acquisition interval is 1 hour, and the total number of data is 533, and the total number of data is 28. C, H, O, S and water content in the same time period are synchronously extracted, specific data of a real-time target heat value are calculated according to an empirical formula, and pearson correlation coefficients of the target heat value and various sensor parameters are calculated.
The SO is obtained by extracting DCS control parameters of each point in 30-day data of a certain month 2 Concentration, flue gas temperature, upper temperature of furnace combustion grate, primary air temperature, smoke concentration, NOx content, HCl content, CO 2 Content (wet basis), fluoride content, O 2 The content (dry basis) of the real-time data of the humidity of the flue gas, and the pearson correlation coefficient of 12 parameters in total are all more than 0.3, belonging to the relevant parameters.
Industrial outliers in the above parameters were removed and fed into the LSTM time series neural network based on the time-attention mechanism and model trained using 80% of the dataset and model tested using 20% of the data. The training model is an LSTM time sequence model based on a time attention mechanism, wherein the input time sequence step length is 3, the output time sequence step length is 1, the input feature dimension is 12, the hidden layer of the LSTM model is 128 layers, the training round number is 60, and the learning rate is 0.48.
80% of training set is sent to a time sequence model for training, and FIG. 3 is a MSE_loss variation trend chart of training. And selecting the model with the lowest MSE_loss for storage as the optimal model for the reasoning of the subsequent models.
The 20% test set is sent to the LSTM time sequence model based on the time attention mechanism for reasoning, finally, the graph of fig. 4 is obtained, the trend change between the true value and the target value is similar, the time interval is 1 hour, the total data is 106 hours, and the average relative error in the final test set is 0.68%. The predicted value of the broken line can basically successfully predict the trend of the solid line target value.
Claims (9)
1. A method for predicting the heat value of garbage entering a furnace in real time based on an attention mechanism LSTM model is characterized by comprising the following steps:
(1) Extracting DCS control parameters of different time points from historical data records of a DCS control system of the garbage incinerator; importing data into an industrial personal computer according to a fixed time interval to obtain a data set to be sorted;
(2) Based on the calculation of the correlation coefficient, SO is screened from the data set to be sorted 2 Concentration, flue gas temperature, upper temperature of furnace combustion grate, primary air temperature, smoke concentration, NOx content, HCl content, CO content and wet base CO 2 Content, fluoride content, dry basis O 2 The content and the flue gas humidity are 12 input characteristic parameters in total; after data cleaning, the data of the 12 input features are used for model training;
(3) Extracting DCS parameters for calculating C, H, O, S and moisture content parameters from the input characteristic parameters obtained in the step (2), calculating the heat value of the garbage entering the furnace at each moment by using a formula (2) or a formula (3), and adding a label for a calculation result to be used for model training; in particular, the method comprises the steps of,
formula (2) is used when the oxygen content exceeds 10%, and formula (3) is used when the oxygen content is below 10%;
wherein Q is the calorific value of the garbage in the furnace, and the unit is: kJ/kg; c is the carbon content in the garbage entering the furnace, and the unit is: the%; h is the hydrogen content in the garbage entering the furnace, and the unit is: the%; o is the oxygen content in the garbage in the furnace, and the unit is: the%; s is the sulfur content of the garbage entering the furnace, and the unit is: the%; w is the moisture content of the garbage entering the furnace, and the unit is: the%;
wherein the carbon content is as defined in (CO and CO 2 The sum of the contents of (1)/(the sum of the density of dry flue gas and the water content of flue gas); the hydrogen content is calculated according to (1/9 of the flue gas water content)/(the sum of the dry flue gas density and the flue gas water content); the oxygen content is calculated according to (the sum of the oxygen content/(the dry smoke density and the smoke water content), the sulfur content is smaller in the smoke components and calculated according to the fixed value of 0.1 percent, and the moisture content is calculated according to the smoke water content/(the sum of the dry smoke density and the smoke water content);
(4) Integrating the input characteristic parameters obtained in the step (2) with the heat value of the garbage fed into the furnace at each moment obtained in the step (3); then 80% of the total data was used as training set data and 20% was used as test set data;
(5) Establishing an LSTM time sequence model based on a time attention mechanism as a training model;
inputting training set data into a training model for training, and optimizing the state quantity h of a hidden layer by multiplying the state quantity h of the hidden layer of the LSTM model by a dot product of the attention moment matrix;
synchronously calculating the loss between the predicted value and the true value by using the root mean square error loss, and after each training round, reversely transmitting and updating the state quantity h of the predicted model at the next moment to gradually reduce the Euclidean distance between the predicted value of the heat value of the garbage of the furnace and the calculated result until the root mean square error loss is less than 1%, and storing the model;
screening the model stored after multiple training, and selecting the model with the minimum root mean square error as a final prediction model;
(6) And (3) extracting an input characteristic parameter from the DCS control parameter at the current time, inputting the prediction model obtained in the step (5), and obtaining the predicted value of the heat value of the garbage entering the furnace in the next time step through calculation.
2. The method of claim 1, wherein the screening in the step (2) is to calculate pearson correlation coefficient of the input characteristic parameter and the real-time target heat value, and reserve the input characteristic parameter with the correlation coefficient greater than 0.3 for input of the training model;
wherein, the calculation of pearson correlation coefficient pearson is shown as the following formula:
wherein: x is x i For a certain input parameter to be selected, y i The calculation result is the real-time target calorific value;for the mean value of the input parameter set, +.>The average value of the real-time target heat value group; i=1, 2 … n.
3. The method of claim 1, wherein the data cleaning in the step (2) means cleaning the extracted input characteristic parameter data according to a control parameter range in the operation process of the garbage incinerator to remove abnormal working condition values.
4. The method according to claim 1, wherein in the step (3), the extracted DCS parameters include: CO content and CO for calculating carbon content 2 The content is as follows; the flue gas water content is used for calculating the hydrogen content; an oxygen content for calculating an oxygen content; the flue gas water content is used for calculating the moisture content, and the dry flue gas density and the flue gas water content are used for calculating the content of each element.
5. The method according to claim 1, wherein in the step (5), the training model has an input time sequence step length of 3, an output time sequence step length of 1, an input feature dimension of 12, a hidden layer of the lstm model of 128 layers, a training round number of 60, and a learning rate of 0.48.
6. The method according to claim 1, wherein in the step (5), the hidden layer state quantity h of the LSTM model is multiplied by the dot product of the attention moment matrix by the following formula:
a·b＝a 1 b 1 +a 2 b 2 +…+a n b n (6)
wherein a and b are each a vector array containing n elements.
7. The method according to claim 1, wherein in the step (5), the loss between the calculated predicted value and the true value is calculated according to the following formula:
wherein MSE_loss is root mean square error loss; n is the number of predicted instances, y i For a certain predicted heating value of the LSTM timing model,predicting the mean value of the heat value for the LSTM time sequence model; i=1, 2 … n.
8. The method according to claim 1, wherein in the step (5), the back propagation update means:
using the same meterThe computing unit repeatedly acts on the input feature vectors x at different moments t t State vector h at this point t-1 Generating a new state vector h at the next moment t The method comprises the steps of carrying out a first treatment on the surface of the State quantity h to be used for updating t And (3) carrying out reverse gradient propagation, starting from an output layer of the model, solving the model gradient layer by layer from back to front by utilizing a chain rule of function derivation, thereby realizing optimal solution and omitting repeated derivation steps.
9. A method for further improving a control strategy of a refuse incinerator by using a prediction result obtained by the method for predicting the calorific value of refuse entering a furnace in real time according to claim 1, comprising:
(a) Calculating the calorific value of the waste in the furnace at the current moment by referring to the contents of the steps (1) - (3) in the claim 1;
(b) Calculating the difference value between the heat value of the garbage entering the furnace at the current moment and the predicted value of the heat value of the garbage entering the furnace at the next time step, and the change condition of the amplitude of the difference value;
(c) And (c) according to the calculation result of the step (b), adjusting and optimizing the garbage feeding rate, ensuring that the total calorific value of the garbage fed into the incinerator in unit time tends to be stable, and keeping the garbage combustion state in the incinerator stable.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211417405.6A CN115717709B (en) | 2022-11-11 | 2022-11-11 | Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211417405.6A CN115717709B (en) | 2022-11-11 | 2022-11-11 | Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115717709A CN115717709A (en) | 2023-02-28 |
| CN115717709B true CN115717709B (en) | 2023-08-04 |
Family
ID=85255052
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211417405.6A Active CN115717709B (en) | 2022-11-11 | 2022-11-11 | Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115717709B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105864797A (en) * | 2016-04-01 | 2016-08-17 | 浙江大学 | Real-time prediction system and method for boiler entering heat value of circulating fluidized bed household garbage incineration boiler |
| CN106096763A (en) * | 2016-06-03 | 2016-11-09 | 河北省电力建设调整试验所 | A kind of method of the coal-fired calorific capacity predicting operating unit |
| CN109446187A (en) * | 2018-10-16 | 2019-03-08 | 浙江大学 | Complex equipment health status monitoring method based on attention mechanism and neural network |
| CN111144609A (en) * | 2019-11-13 | 2020-05-12 | 瀚蓝绿电固废处理(佛山)有限公司 | Boiler exhaust emission prediction model establishing method, prediction method and device |
| CN111855953A (en) * | 2019-09-23 | 2020-10-30 | 江苏方天电力技术有限公司 | Method and system for quickly feeding back calorific value of coal as fired in coal-fired power plant |
| CN111931346A (en) * | 2020-07-13 | 2020-11-13 | 西安热工研究院有限公司 | Real-time calculation method and application of low calorific value of coal as fired in coal-fired boiler |
| CN113076623A (en) * | 2021-03-02 | 2021-07-06 | 西安交通大学 | Dynamic estimation method and system for coal-fired calorific value of thermal power generating unit |
| CN114675607A (en) * | 2022-03-23 | 2022-06-28 | 浙江浙能兰溪发电有限责任公司 | Fired coal heat value prediction and correction control method combining flame temperature |
| CN115111594A (en) * | 2022-07-08 | 2022-09-27 | 浙江大学 | Intelligent regulation and control system and method for heat accumulating type thermal oxidation furnace |
| JP2022161065A (en) * | 2021-04-08 | 2022-10-21 | Ｊｆｅエンジニアリング株式会社 | Garbage property prediction device, combustion control device of incinerator, garbage property prediction method, learning method for garbage property prediction model, and garbage property predication model program |
-
2022
- 2022-11-11 CN CN202211417405.6A patent/CN115717709B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105864797A (en) * | 2016-04-01 | 2016-08-17 | 浙江大学 | Real-time prediction system and method for boiler entering heat value of circulating fluidized bed household garbage incineration boiler |
| CN106096763A (en) * | 2016-06-03 | 2016-11-09 | 河北省电力建设调整试验所 | A kind of method of the coal-fired calorific capacity predicting operating unit |
| CN109446187A (en) * | 2018-10-16 | 2019-03-08 | 浙江大学 | Complex equipment health status monitoring method based on attention mechanism and neural network |
| CN111855953A (en) * | 2019-09-23 | 2020-10-30 | 江苏方天电力技术有限公司 | Method and system for quickly feeding back calorific value of coal as fired in coal-fired power plant |
| CN111144609A (en) * | 2019-11-13 | 2020-05-12 | 瀚蓝绿电固废处理(佛山)有限公司 | Boiler exhaust emission prediction model establishing method, prediction method and device |
| CN111931346A (en) * | 2020-07-13 | 2020-11-13 | 西安热工研究院有限公司 | Real-time calculation method and application of low calorific value of coal as fired in coal-fired boiler |
| CN113076623A (en) * | 2021-03-02 | 2021-07-06 | 西安交通大学 | Dynamic estimation method and system for coal-fired calorific value of thermal power generating unit |
| JP2022161065A (en) * | 2021-04-08 | 2022-10-21 | Ｊｆｅエンジニアリング株式会社 | Garbage property prediction device, combustion control device of incinerator, garbage property prediction method, learning method for garbage property prediction model, and garbage property predication model program |
| CN114675607A (en) * | 2022-03-23 | 2022-06-28 | 浙江浙能兰溪发电有限责任公司 | Fired coal heat value prediction and correction control method combining flame temperature |
| CN115111594A (en) * | 2022-07-08 | 2022-09-27 | 浙江大学 | Intelligent regulation and control system and method for heat accumulating type thermal oxidation furnace |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115717709A (en) | 2023-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Song et al. | Hourly heat load prediction model based on temporal convolutional neural network | |
| CN111144609A (en) | Boiler exhaust emission prediction model establishing method, prediction method and device | |
| CN114742294A (en) | Neural network algorithm for carbon emission prediction | |
| Song et al. | Estimate the daily consumption of natural gas in district heating system based on a hybrid seasonal decomposition and temporal convolutional network model | |
| CN111260149A (en) | Method for predicting dioxin emission concentration | |
| Senthil Kumar et al. | Feature selection used for wind speed forecasting with data driven approaches | |
| Ding et al. | Estimating energy-related CO2 emissions using a novel multivariable fuzzy grey model with time-delay and interaction effect characteristics | |
| Zhang et al. | A novel grey Lotka–Volterra model driven by the mechanism of competition and cooperation for energy consumption forecasting | |
| Ye et al. | A nonlinear interactive grey multivariable model based on dynamic compensation for forecasting the economy-energy-environment system | |
| CN113947013A (en) | Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method | |
| CN115717709B (en) | Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model | |
| Karaçor et al. | Life performance prediction of natural gas combined cycle power plant with intelligent algorithms | |
| WO2023231667A1 (en) | Method for soft measurement of dioxin emission in mswi process based on integrated t-s fuzzy regression tree | |
| CN116401948A (en) | Online prediction method and system for generating amount of power station boiler ash based on LSTM | |
| Cui et al. | Multi-condition operational optimization with adaptive knowledge transfer for municipal solid waste incineration process | |
| CN116629553A (en) | Regional comprehensive energy scheduling method based on incomplete interval multi-objective fuzzy optimization | |
| CN115860232A (en) | Steam load prediction method, system, electronic device and medium | |
| Qiao et al. | A comprehensively improved interval type-2 fuzzy neural network for nox emissions prediction in mswi process | |
| CN110689156A (en) | Universal energy station optimization method and device | |
| Kang et al. | Research on forecasting method for effluent ammonia nitrogen concentration based on GRA-TCN | |
| CN113516269A (en) | Management method of multi-energy complementary energy hub equipment | |
| Li et al. | Prediction of typical flue gas pollutants from municipal solid waste incineration plants | |
| Tan | Carbon Emission Prediction with Macroeconomic Variables and Machine Learning | |
| CN113887130B (en) | Industrial boiler operation optimization method based on ensemble learning | |
| CN112669169B (en) | Short-term photovoltaic power prediction device and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |