A bed pressure prediction method for circulating fluidized bed units based on time series information model
Through load condition division and sub-model construction based on the timing information model, the accuracy and parallel calculation problems of the bed pressure prediction of circulating fluidized bed machine units are solved, and the safe and stable operation of the unit is achieved.
Patent Information
- Application Number
- CN202311061915.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-22
- Publication Date
- 2025-08-12
- Estimated Expiration
- 2043-08-22
AI Technical Summary
The prior art is difficult to effectively predict the bed pressure of the circulating fluidized bed unit, resulting in insufficient combustion and unbalanced transverse material transportation, affecting the safe and stable operation of the unit.
Using a method based on the timing information model, the historical data is grouped through the load condition proportion division method, multiple timing information sub-models are constructed, and the bed pressure prediction is used to achieve parallel calculations using the memory modules of forget gates, input gates and candidate states.
It improves the accuracy and parallel computing capabilities of bed pressure prediction, ensures the safety and stability of unit operation, and optimizes operation control.
Smart Images
Figure CN117272093B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of intelligent power generation, and in particular relates to a bed pressure prediction method for a circulating fluidized bed unit based on a time series information model. Background Art
[0002] With the rapid development of new energy power technologies, to achieve peak-shaving and frequency-regulation strategies and overcome the adverse effects of the instability, intermittency, and uncertainty of new energy generators on the stable operation of the grid, existing technologies employ circulating fluidized bed boilers. These technologies address the poor load regulation capabilities of traditional thermal power generators, the need for oil injection to assist in furnace combustion during deep peak-shaving conditions, and the significant amount of pollutants generated by these units during deep peak-shaving conditions, which are detrimental to environmental protection and sustainable development. As a clean coal combustion technology, circulating fluidized bed boilers offer advantages such as high combustion efficiency, wide fuel adaptability, low pollutant emissions, good load regulation performance, and efficient and comprehensive resource utilization. They have been widely used in power generation and heating industries both domestically and internationally.
[0003] With the continuous development of large-capacity circulating fluidized bed units, the large-scale circulating fluidized bed has brought about the problem of incomplete combustion. It is difficult for traditional furnace types to ensure the penetration ability of secondary air during the combustion process, resulting in most of the furnace being in a negative oxygen combustion state, causing economic losses. While the existing technology uses a trouser-leg type boiler to solve the above-mentioned combustion problem, it also brings new problems of lateral transportation and fluctuation of materials on both sides of the furnace. When there is a large deviation in the materials on both sides of the furnace, the lateral transfer of materials is likely to form positive feedback, resulting in a sharp increase in the material on one side. In severe cases, it will cause the unit to shut down. Therefore, there is an urgent need for a bed pressure prediction method for a circulating fluidized bed unit. Through efficient and accurate prediction of the bed pressure, the concentration state of the material can be intuitively displayed, ensuring the balance of materials on both sides of the furnace to achieve smooth and safe operation of the unit. Summary of the Invention
[0004] The present invention aims to provide a method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model, which is characterized by comprising the following steps:
[0005] Step S1: collecting historical data from a distributed control system of a thermal power unit, normalizing the historical data, grouping the normalized historical data using a load condition ratio division method, and dividing the historical data into N training data sets based on the grouping results. The load condition ratio division method groups the data based on the load condition ratio, where the load condition ratio is the percentage of the operating load to the total unit capacity, and N is an integer greater than or equal to 2.
[0006] Step S2: constructing a time series information model, using the N training data sets constructed in step S1 to train the time series information model respectively, and constructing N time series information sub-models, the input of the time series information model includes: the unit load instruction, the primary air volume, the secondary air upper air volume, the secondary air lower air volume, the total fuel volume, the slag discharge volume, and the return valve opening data, and the output of the time series information model is the bed pressure measurement value after t seconds;
[0007] Step S3: collecting real-time data, normalizing the real-time data, grouping the normalized real-time data using a load condition proportion division method, and determining the data type of the real-time data based on the grouping result, wherein the data type includes N data subtypes corresponding to the N time series information sub-models respectively;
[0008] Step S4: Input the real-time data into the corresponding time series information sub-model according to the data sub-type of the real-time data determined in step S3, output the bed pressure measurement value after t seconds, and complete the bed pressure prediction of the circulating fluidized bed unit.
[0009] The value of N is 3, and the specific steps of the load condition proportion division method are:
[0010] When the load condition accounts for less than or equal to 50%, the input data is divided into group 1;
[0011] When the load condition accounts for more than 50% and less than or equal to 70%, the input data is divided into group 2;
[0012] When the load condition ratio is greater than 70% and less than or equal to 100%, the input data is divided into group 3.
[0013] The value of t in step S2 is 30.
[0014] The temporal information model includes a plurality of chain-connected memory modules, each of which includes a forget gate f, an input gate o, and a candidate state z;
[0015] The forget gate f is used to record the information that needs to be discarded in the cell state;
[0016] The input gate o is used to record the information that needs to be recorded in the cell state;
[0017] Candidate state z is the information that needs to be updated in the user's continued cell state;
[0018] The calculation steps of the time series information model are shown in formulas (1) to (5):
[0019] h1=cnn(concat(0,x)) (1)
[0020] f=sigmoid(Wf x+b f ) (2)
[0021] o=sigmoid(W o x+b o ) (3)
[0022] z=tanh(W z x+b z ) (4)
[0023] out=f×tanh(h1)+o×z (5)
[0024] Where: h1 is the time series information, f is the forget gate, o is the input gate, z is the candidate state, b is the batch size, time is the length of the time step, hidden is the number of hidden neurons in the time series information neural network model; x is the model input data, and out is the model output data; cnn represents the convolution operation, and concat represents the concatenation operation on the input data to ensure that the data has the same shape as the input after the convolution operation, so that the data is shifted in time series; W f is the weight of the forget gate, W o is the weight of the input gate, W z is the weight of the candidate state, b f is the bias value of the forget gate, b o is the bias value of the input gate, b z is the bias value of the candidate state;
[0025] The expression of the sigmoid activation function is:
[0026] The expression of the tanh activation function is:
[0027] The second object of the present invention is to disclose an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor, characterized in that when the processor executes the computer program, each step of the circulating fluidized bed unit bed pressure prediction method based on the timing information model of the present invention is implemented.
[0028] The third object of the present invention is to disclose a storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, each step of the circulating fluidized bed unit bed pressure prediction method based on the timing information model as described in the present invention is implemented.
[0029] The beneficial effects of the present invention are:
[0030] (1) Parameter prediction under multiple operating conditions is achieved by dividing the data segments, so that the model can take into account the data operating characteristics of the circulating fluidized bed unit under different operating conditions.
[0031] (2) The bed pressure parameters of the circulating fluidized bed unit are predicted through the time series information model. While fully considering the time series characteristics of the data, it overcomes the defect that the traditional LSTM model does not support parallel computing, can effectively improve the parallel computing capability of the model, and is more conducive to practical application on site.
[0032] (3) By predicting the bed pressure parameters of the circulating fluidized bed unit, the unit's operating safety and stability can be effectively improved, which is beneficial to the optimization and adjustment of operation control. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 Schematic diagram of a flow chart of a method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model according to the present invention;
[0034] Figure 2 Schematic diagram of the structure of the timing information model in the present invention. DETAILED DESCRIPTION
[0035] The present invention provides a method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model. The present invention is further described in detail below with reference to the accompanying drawings.
[0036] like Figure 1 As shown, an embodiment of the present invention discloses a method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model, comprising the following steps:
[0037] Step S1: collecting historical data from a distributed control system of a thermal power unit, normalizing the historical data, grouping the normalized historical data using a load condition ratio division method, and dividing the historical data into N training data sets based on the grouping results. The load condition ratio division method groups the data based on the load condition ratio, where the load condition ratio is the percentage of the operating load to the total unit capacity, and N is an integer greater than or equal to 2.
[0038] After comprehensively considering the balance between prediction accuracy and prediction performance, the N can be adjusted according to the specific object. In this embodiment, the value of N is 3, and the specific steps of the load condition proportion division method are:
[0039] When the load condition accounts for less than or equal to 50%, the input data is divided into group 1;
[0040] When the load condition accounts for more than 50% and less than or equal to 70%, the input data is divided into group 2;
[0041] When the load condition ratio is greater than 70% and less than or equal to 100%, the input data is divided into group 3.
[0042] In this embodiment, historical data is obtained from the distributed control system (i.e., DCS) of the on-site thermal power unit and normalized. The data is divided according to unit load. The load condition ratio is the percentage of operating load to total unit capacity. In this embodiment, the three grouping criteria are: less than 50%, 50%-70%, and 70%-100%.
[0043] Step S2: constructing a time series information model, using the N training data sets constructed in step S1 to train the time series information model respectively, and constructing N time series information sub-models, the input of the time series information model includes: the unit load instruction, the primary air volume, the secondary air upper air volume, the secondary air lower air volume, the total fuel volume, the slag discharge volume, and the return valve opening data, and the output of the time series information model is the bed pressure measurement value after t seconds;
[0044] In this embodiment, the value of t in step S2 is 30, which means that the output of the time series information model is the bed pressure measurement value 30 seconds later.
[0045] The timing information model has the same model structure as the N timing information sub-models. The hyperparameters in the N timing information sub-models are adjusted according to the data under various working conditions. In this embodiment, the value of N is 3, and three independent timing information sub-models are constructed. The three types of divided data are used to train these three timing information sub-models respectively.
[0046] Step S3: collecting real-time data, normalizing the real-time data, grouping the normalized real-time data using a load condition proportion division method, and determining the data type of the real-time data based on the grouping result, wherein the data type includes N data subtypes corresponding to the N time series information sub-models respectively;
[0047] Step S4: Input the real-time data into the corresponding time series information sub-model according to the data sub-type of the real-time data determined in step S3, output the bed pressure measurement value after t seconds, and complete the bed pressure prediction of the circulating fluidized bed unit.
[0048] In this embodiment, real-time data is divided according to the load condition proportion division method in S1, and the corresponding time series information sub-model is used to predict the real-time data based on the division results. In this embodiment, the value of N is 3, and the value of t in step S4 is 30, that is, the output of the time series information model is the bed pressure measurement value 30 seconds later.
[0049] like Figure 2As shown, in this embodiment, the temporal information model includes several chain-connected memory modules, and the memory modules include: a forget gate f, an input gate o, and a candidate state z;
[0050] The forget gate f is used to record the information that needs to be discarded in the cell state;
[0051] The input gate o is used to record the information that needs to be recorded in the cell state;
[0052] Candidate state z is the information that needs to be updated in the user's continued cell state;
[0053] The calculation steps of the time series information model are shown in formulas (1) to (5):
[0054] h1=cnn(concat(0,x)) (1)
[0055] f=sigmoid(W f x+b f ) (2)
[0056] o=sigmoid(W o x+b o ) (3)
[0057] z=tanh(W z x+b z ) (4)
[0058] out=f×tanh(h1)+o×z (5)
[0059] Where: h1 is the time series information, f is the forget gate, o is the input gate, z is the candidate state, b is the batch size, time is the length of the time step, hidden is the number of hidden neurons in the time series information neural network model; x is the model input data, and out is the model output data; cnn represents the convolution operation, and concat represents the concatenation operation on the input data to ensure that the data has the same shape as the input after the convolution operation, so that the data is shifted in time series; W f is the weight of the forget gate, W o is the weight of the input gate, W z is the weight of the candidate state, b f is the bias value of the forget gate, b o is the bias value of the input gate, b z is the bias value of the candidate state;
[0060] The expression of the sigmoid activation function is:
[0061] The expression of the tanh activation function is:
[0062] In this embodiment, the time series information model selects and calculates the previous time series data based on the current input. It extracts time series information through convolution, corrects the time series through concat, and concatenates the input data to ensure that the data has the same shape as the input after convolution, thus shifting the data in time. This overcomes the limitation of existing RNNs in achieving parallel computing and effectively improves the model's parallel computing capabilities.
[0063] This invention discloses a method for predicting bed pressure in circulating fluidized bed units based on a time series information model, applicable to all circulating fluidized bed units. Because the bed pressures on both sides of a circulating fluidized bed unit are strongly correlated, simply predicting the bed pressure on one side allows monitoring of both sides. This bed pressure prediction provides a clear visual indication of material concentration, ensuring material balance on both sides of the furnace and ensuring smooth and safe unit operation.
[0064] To verify the effectiveness of the disclosed method for predicting bed pressure in a circulating fluidized bed (CFB) unit based on a time series information model, the following experiment was conducted. A total of 60,000 sets of training data were used, with 20,000 sets of data for each of the three operating conditions. The computer used an i7-10850H processor and an RTX3070 GPU. This method achieved higher prediction accuracy than traditional LSTM models, improving accuracy by 5% to 10%.
[0065] To address the technical issues inherent in known technologies, this invention provides a method for predicting bed pressure in a circulating fluidized bed (CFB) unit based on a time series information model. By monitoring the unit's bed pressure based on this time series information model, on-site operators are provided with future bed pressure parameter information, effectively guiding their operations and improving the unit's operational safety and stability.
[0066] An optional embodiment of the present invention discloses an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model described in the present invention. The specific steps are the same as those of the first embodiment disclosed herein and are not further described here.
[0067] Another optional embodiment of the present invention discloses a storage medium storing a computer program. When executed by a processor, the computer program implements the steps of the method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model as described herein. The specific steps are the same as those of the first embodiment disclosed herein and are not further described here.
Claims
1. A method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model, characterized in that: The following steps are involved: Step S1: collecting historical data from a distributed control system of a thermal power unit, normalizing the historical data, grouping the normalized historical data using a load condition ratio division method, and dividing the historical data into N training data sets based on the grouping results. The load condition ratio division method groups the data based on the load condition ratio, where the load condition ratio is the percentage of the operating load to the total unit capacity, and N is an integer greater than or equal to 2. Step S2: constructing a time series information model, using the N training data sets constructed in step S1 to train the time series information model respectively, and constructing N time series information sub-models, the input of the time series information model includes: the unit load instruction, the primary air volume, the secondary air upper air volume, the secondary air lower air volume, the total fuel volume, the slag discharge volume, and the return valve opening data, and the output of the time series information model is the bed pressure measurement value after t seconds; Step S3: collecting real-time data, normalizing the real-time data, grouping the normalized real-time data using a load condition proportion division method, and determining the data type of the real-time data based on the grouping result, wherein the data type includes N data subtypes corresponding to the N time series information sub-models respectively; Step S4: Input the real-time data into the corresponding time series information sub-model according to the data sub-type of the real-time data determined in step S3, output the bed pressure measurement value after t seconds, and complete the bed pressure prediction of the circulating fluidized bed unit.
2. The method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model according to claim 1, characterized in that: The value of N is 3, and the specific steps of the load condition proportion division method are: When the load condition accounts for less than or equal to 50%, the input data is divided into group 1; When the load condition accounts for more than 50% and less than or equal to 70%, the input data is divided into group 2; When the load condition ratio is greater than 70% and less than or equal to 100%, the input data is divided into group 3.
3. The method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model according to claim 1, characterized in that: The value of t in step S2 is 30.
4. The method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model according to claim 1, characterized in that: The temporal information model includes a plurality of chain-connected memory modules, each of which includes a forget gate f, an input gate o, and a candidate state z; The forget gate f is used to record the information that needs to be discarded in the cell state; The input gate o is used to record the information that needs to be recorded in the cell state; Candidate state z is the information that needs to be updated in the user's continued cell state; The calculation steps of the time series information model are shown in formulas (1) to (5): h1=cnn(concat(0,x)) (1) f=sigmoid(W f x+b f ) (2) o=sigmoid(W o x+b o ) (3) z=tanh(W z x+b z ) (4) out=f×tanh(h1)+o×z (5) Where: h1 is the time series information, f is the forget gate, o is the input gate, z is the candidate state, b is the batch size, time is the length of the time step, hidden is the number of hidden neurons in the time series information neural network model; x is the model input data, and out is the model output data; cnn represents the convolution operation, and concat represents the concatenation operation on the input data to ensure that the data has the same shape as the input after the convolution operation, so that the data is shifted in time series; W f is the weight of the forget gate, W o is the weight of the input gate, W z is the weight of the candidate state, b f is the bias value of the forget gate, b o is the bias value of the input gate, b z is the bias value of the candidate state; The expression of the sigmoid activation function is: The expression of the tanh activation function is:
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, each step of the method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model as claimed in any one of claims 1 to 4 is implemented.
6. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method for predicting bed pressure of a circulating fluidized bed unit based on a time series information model as claimed in any one of claims 1 to 4 is implemented.