Low-load coordination control method and system for thermal power generating unit

Abstract

The invention discloses a thermal power generating unit low-load coordination control method and system, and the method comprises the steps: collecting the data of a thermal power generating unit, building an auxiliary machine trip prediction model, and predicting the fault of an auxiliary machine; the D-S evidence theory is used for judging the tripping risk of the auxiliary machine, and if the fault does not trigger the fault load reduction RB function of the auxiliary machine, fault classification is carried out; and automatically controlling the fault auxiliary machine according to the fault classification result, and entering low-load stable transition after the automatic control is completed. According to the method, the structure is simple, design is reasonable, the fan blade opening degree and the speed of the primary air fan with the tripping risk are attenuated to 0 according to a decline function, the fan blade opening degree and the speed of the primary air fan without the tripping risk are increased to target values according to an incremental function, the decline function and the total incremental function are symmetrical, complementation is achieved, and the tripping risk is reduced. An auxiliary machine tripping prediction model is constructed, door filter optimization is adopted for predicting the auxiliary machine fault probability, and evidence combination rules are adopted for carrying out evidence fusion on fault evidences of different auxiliary machines.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems and equipment, and in particular to a low-load coordinated control method and system for a thermal power unit. Background Art

[0002] With the rapid development of renewable energy power generation in my country and the overcapacity of coal-fired power generation, the flexibility transformation of coal-fired power generation is imperative. As the proportion of renewable energy in the power grid gradually increases, the demand for peak-shaving power sources also gradually increases. Compared with renewable energy and other power sources, coal-fired power generation has better peak-shaving performance. Although the current conventional control logic can meet the needs of deep peak-shaving of 300MW, important auxiliary machines trip under this load, lacking effective accident control logic, and being completely operated by operating personnel, due to human uncertainty, improper and untimely handling can easily lead to boiler fire extinguishing and even damage to equipment.

[0003] Therefore, a system is needed that can provide early warning of tripping faults of important auxiliary machines, take effective measures in advance based on the predicted results of tripping faults, and perform smooth switching. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing control method lacks effective accident control logic, is completely operated by operating personnel, and fails to handle faults in a timely manner.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a low-load coordinated control method for a thermal power unit, comprising: collecting data of the thermal power unit, establishing an auxiliary machine tripping prediction model, and predicting auxiliary machine failures; using the DS evidence theory to judge the auxiliary machine tripping risk, and if the fault does not trigger the auxiliary machine fault load reduction RB function, then performing fault classification; performing automatic control of the faulty auxiliary machine according to the fault classification result, and entering a low-load smooth transition after the automatic control is completed.

[0007] As a preferred solution of the low-load coordinated control method for thermal power units described in the present invention, the auxiliary machine tripping prediction model includes collecting the operating parameters of the auxiliary machine in the time period [Tt, T] before the tripping occurs to construct a data set X, the auxiliary machine is divided into three types: primary fan, water supply pump and coal mill, and the data set X is divided into a training set and a test set; a convolutional neural network is used to construct the auxiliary machine tripping prediction model, and a gate filter φ is added after each convolution layer. P , door filter φ P is [1, C P , 1, 1], the output feature map of the convolution layer is multiplied by the gate filter in the channel direction, thereby changing the standard neural network model module, combined with C PThe model is optimized by random assignment; the tripping risk is accurately predicted in the unbalanced data set through the loss function L. The regularization term L2 is added on the basis of the convolutional neural network loss term L1, and the model parameters are adjusted, which is expressed as:

[0008] L=(1-α)1+αL2

[0009]

[0010] Among them, α represents the weight of L2; C P represents the number of channels in the gate filter added after the Pth convolutional layer; X represents the input data; Y represents the label; θ represents the parameter of the auxiliary machine trip prediction model; δ represents the gradient; β represents the coefficient; Indicates the modulation coefficient.

[0011] As a preferred solution of the low-load coordinated control method for thermal power units described in the present invention, the steps of the DS evidence theory are:

[0012] Based on the DS evidence theory, the identification framework Θ={F} is set, and the risk W ij Assign the value of m to ij {F};

[0013] The weighted average of multi-time evidence is expressed as:

[0014]

[0015] Evidence combination rules are used for evidence fusion. The evidence combination rules for any two groups are:

[0016]

[0017] The fused evidence is converted into the auxiliary machine trip probability BetP(A); where m p (A) represents the weighted average evidence information; ω t Represents evidence m at time T-t+1 (T-t+1)ij (A) weight; l represents the total time; m ⊕ (C) represents the fused evidence information; m p represents the evidence to be integrated in the identification framework; A represents the power set 2 Θ A subset of; B represents the power set 2 Θ A subset of .

[0018] As a preferred scheme of the low-load coordinated control method for thermal power units described in the present invention, the fault classification includes the risk of primary fan tripping, the risk of water supply pump tripping and the risk of coal mill tripping. The faulty auxiliary machine is determined according to the DS evidence theory output BetP(A) and enters automatic control.

[0019] As a preferred solution of the low-load coordinated control method for a thermal power unit according to the present invention, when the primary fan has a tripping risk, the primary fan enters the automatic tripping control, and the fan blade opening y1 of the primary fan with a tripping risk decays to zero within a Y time, which is expressed as:

[0020] y1=f1(Y) y1∈[0,μ1]

[0021] The speed y2 of the primary fan with tripping risk decays to zero within Y time, expressed as:

[0022] y2=f2(Y) y2∈[0,v1]

[0023] The fan blade opening y of the primary fan without tripping risk 3(ε-r) In Y time, it increases from μ2 to f(s T ), the host computer calculates the opening of the primary fan that did not trip when the primary fan tripped, and converts the opening f(s T ) is fed back to the automatic control system of the primary fan that has not tripped, expressed as:

[0024] y 3(ε-r) =f 3(ε-r) (Y) y 3(ε-r) ∈[μ2,f(s T )]

[0025]

[0026] The upper computer calculates the output rate v3 of the primary fan without tripping risk based on the output of the primary fan without tripping risk and feeds it back to the automatic control system of the primary fan without tripping risk. The rate y of the primary fan without tripping risk is 4(ε-r) The increase from v2 to v3 in Y time is expressed as:

[0027] y 4(ε-r) =f 4(ε-r) (Y) y 4(ε-r) ∈[v2,v3]

[0028]

[0029] Wherein, μ1 represents the current blade opening of the primary fan with a tripping risk; v1 represents the current blade speed of the primary fan with a tripping risk; μ2 represents the current blade opening of the primary fan without a tripping risk; f(s T ) represents the output blade opening of the primary fan without tripping risk; r represents the number of primary fans that tripped; η 1T represents the influencing factor of ammonia escape; η 2T represents the impact factor of sulfur dioxide; C1T Indicates the concentration of ammonia in the flue at time T; C set Indicates the set concentration value of ammonia in the flue; C 2T Indicates the concentration of sulfur dioxide in the flue at time T; b 2T Represents the fitting coefficient; v2 represents the current speed of the primary fan without tripping risk; v3 represents the output speed of the primary fan without tripping risk; Q represents the expected wind pressure of the primary fan without tripping.

[0030] As a preferred scheme of the low-load coordinated control method for thermal power units described in the present invention, wherein: when the feed water pump is at risk of tripping, it enters the feed water pump tripping automatic control, and the feed water pump that has not tripped follows the feed water flow value corresponding to the fuel amount M corresponding to the unit load target value UL, and the multi-order inertia time of the change is switched from the conventional 20 to 30 seconds to 10 seconds; the recirculation valve of the feed water pump that has not tripped is fully closed at time Y; during the low-load smooth transition control action, the manual cutting conditions for the large deviation between the control quantity and the set value in the automatic control of the cold and hot air doors of the coal mill and the manual cutting conditions for the large deviation between the valve and the opening instruction feedback and the set value are shielded.

[0031] As a preferred solution of the low-load coordinated control method for thermal power units described in the present invention, when the coal mill has a tripping risk, the coal mill tripping automatic control is entered, and the upper computer calculates the target fuel quantity M g , the running coal feeder automatically puts in the fuel quantity automatic control, which is expressed as:

[0032]

[0033] Among them, M w Indicates the amount of fuel for stable combustion; M max Indicates the actual maximum acceptable fuel quantity of the pulverizing system; indicates the current actual fuel quantity; M P Indicates the maximum fuel quantity of a single coal feeder; UL indicates the current load value of the unit; UL w Indicates the steady-state combustion load value of the unit; M 1max Indicates the maximum fuel quantity of a single coal feeder; M 2max It indicates the maximum fuel quantity of a single coal mill; z indicates the number of coal mills remaining in operation.

[0034] In the second aspect, the present invention also provides a low-load coordinated control system for thermal power units, including a fault prediction module, which collects data of thermal power units, establishes and optimizes an auxiliary machine tripping prediction model based on a convolutional neural network, and uses the trained auxiliary machine tripping prediction model to predict the tripping risk of the auxiliary machine; a fault classification module, which obtains new evidence based on the DS evidence theory, judges and classifies the auxiliary machine tripping risk, and judges whether the fault triggers the auxiliary machine fault load reduction RB function. If the RB function is triggered, the process ends, and if the RB function is not triggered, automatic control is entered; an automatic control module, which performs automatic control of a fan, a water supply pump or a coal mill based on the judgment result of the DS evidence theory, and enters a low-load smooth transition after the automatic control is completed.

[0035] In a third aspect, the present invention further provides a computing device, comprising: a memory and a processor;

[0036] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the low-load coordinated control method of the thermal power unit are implemented.

[0037] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for low-load coordinated control of thermal power units.

[0038] The beneficial effects of the present invention are as follows: the method of the present invention has a simple structure and a reasonable design. The fan blade opening and speed of a primary fan with a risk of tripping are respectively decayed to 0 according to a decreasing function, and the fan blade opening and speed of a primary fan without a risk of tripping are increased to a target value according to an increasing function. The decreasing function and the total increasing function are symmetrical, thereby achieving complementarity. An auxiliary machine tripping prediction model is constructed and gate filter optimization is used to predict the auxiliary machine failure probability. The fault evidence of different auxiliary machines is fused using evidence combination rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0040] Figure 1 An overall flow chart of a low-load coordinated control method for a thermal power unit provided by an embodiment of the present invention;

[0041] Figure 2A waveform diagram of an increasing function and a decreasing function of a low-load coordinated control method for a thermal power unit provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0043] Example 1

[0044] Reference Figure 1 and Figure 2 , is an embodiment of the present invention, and provides a low-load coordinated control method for a thermal power unit, comprising:

[0045] S1: Collect data of thermal power units, establish auxiliary unit tripping prediction model, and predict auxiliary unit failure.

[0046] Furthermore, the thermal power unit includes at least three coal mills in operation and a primary fan, a primary fan, an induced draft fan and a water supply pump in double row operation. A variety of sensors are used to collect the operating parameters of the auxiliary machine in the time period [Tt, T] before the trip occurs as positive samples, and the operating parameters of the auxiliary machine in normal operation are collected as negative samples to construct a data set X, X = {X1, ..., X h}, X i is the operating parameter subset of the i-th auxiliary machine, X i ={X i1 , ..., X in}, X ij are the operating parameters of the jth tripping fault of the i-th auxiliary machine, where 1≤i≤h, 1≤j≤n, n is not less than 2, h=5, and the auxiliary machines include coal mill, primary fan and feed water pump. The data set X is divided into training set and test set in a ratio of 7:3, and the ratio of positive and negative samples in the training set is 1:1.

[0047] A convolutional neural network is used to build an auxiliary machine trip prediction model, and a gate filter φ is added after each convolution layer. P , door filter φ P is [1, C P ,1,1], the auxiliary machine tripping prediction model is trained using the training set, the loss value of the iterative training is back-propagated and the gate filter of the auxiliary machine tripping prediction model is updated, and the trained network model is obtained after completing the iterative training.

[0048] For one or more of the C PRandomly assign 0, and the rest of the C P The number of image channels of the output feature map after the Pth convolution layer is the same. The classification results of the auxiliary machine trip prediction model after random assignment of 0 are verified using the validation set. If the classification results meet the requirements, the C assigned 0 is deleted from the auxiliary machine trip prediction model. P The corresponding convolutional layer.

[0049] Add a gate filter φ after each convolutional layer P , the output feature map of the convolutional layer can be multiplied by the gate filter in the channel direction, thereby changing the standard neural network model module. P When 0 is randomly assigned, it is like removing the corresponding convolutional layer, thereby optimizing the auxiliary machine tripping prediction model.

[0050] The loss function L is used to accurately predict the tripping risk in an unbalanced data set. The regularization term L2 is added to the convolutional neural network loss term L1, and the model parameters are adjusted, which is expressed as:

[0051] L=(1-α)L1+αL2

[0052]

[0053] Among them, α represents the weight of L2; C P represents the number of channels in the gate filter added after the Pth convolutional layer; X represents the input data; Y represents the label; θ represents the parameter of the auxiliary machine tripping prediction model; δ represents the gradient; β represents the coefficient; Indicates the modulation coefficient.

[0054] It should be noted that the coefficients β and It is expressed as:

[0055]

[0056]

[0057] Among them, β′ represents the weight of the image feature belonging to the positive sample; Indicates the probability that the image feature belongs to a positive sample; β′=0.25, γ=2.

[0058] In machine learning, we hope that the predicted data distribution learned by the model on the training data is as close to the real data distribution as possible. Since cross entropy describes the measure of the amount of information required to eliminate the uncertainty of the real data distribution, cross entropy is used in machine learning to eliminate the uncertainty of the system at the lowest cost.

[0059] At the same time, during the training phase, it is necessary to consider the gate filter φ P Therefore, the first loss function L1 is based on the gate filter φP .

[0060] Furthermore, a sensor group is used to collect auxiliary machine operating parameters, and the host computer receives the auxiliary machine operating parameters in real time. The auxiliary machine tripping prediction model is used to predict the tripping risk of the auxiliary machine, and the tripping risk prediction result of the auxiliary machine is obtained. The risk here refers to the confidence level.

[0061] S2: Use DS evidence theory to judge the risk of auxiliary machine tripping. If the fault does not trigger the auxiliary machine fault load reduction RB function, perform fault classification.

[0062] Furthermore, the steps of DS evidence theory are:

[0063] Based on the DS evidence theory, the identification framework Θ={F} is set, and the risk W ij Assign the value of m to ij {F};

[0064] The weighted average of multi-time evidence is expressed as:

[0065]

[0066] Evidence combination rules are used for evidence fusion. The evidence combination rules for any two groups are:

[0067]

[0068] The fused evidence is converted into the auxiliary machine trip probability BetP(A); where m p (A) represents the weighted average evidence information; ω t Represents evidence m at time T-t+1 (T-t+1)ij (A) weight; l represents the total time; m ⊕ (C) represents the fused evidence information; m p represents the evidence to be integrated in the identification framework; A represents the power set 2 Θ A subset of; B represents the power set 2 Θ A subset of .

[0069] Furthermore, it is determined whether the fault triggers the auxiliary machine fault load reduction RB function, and the low load smooth transition is triggered when the unit load is 30% to 50% MCR; the auxiliary machine fault load reduction RB function is triggered when the unit load is less than 30% MCR or the unit load is greater than 50% MCR. MCR stands for maximum continuous power.

[0070] Fault classification includes the risk of primary fan tripping, feed water pump tripping and coal mill tripping. The faulty auxiliary machine is determined based on the output BetP(A) of DS evidence theory and enters automatic control.

[0071] S3: Automatically control the faulty auxiliary machine according to the fault classification result, and enter a low-load smooth transition after the automatic control is completed.

[0072] Furthermore, when a primary fan has a tripping risk, the primary fan tripping automatic control is entered, and the fan blade opening y1 of the primary fan with a tripping risk decays to zero within Y time, which is expressed as:

[0073] y1=f1(Y) y1∈[0,μ1]

[0074] The speed y2 of the primary fan with tripping risk decays to zero within Y time, expressed as:

[0075] y2=f2(Y) y2∈[0,v1]

[0076] These two steps enable the rapid, smooth and automatic exit of primary fans that are at risk of tripping.

[0077] The fan blade opening y of the primary fan without tripping risk 3(ε-r) In Y time, it increases from μ2 to f(s T ), the host computer calculates the opening of the primary fan that did not trip when the primary fan tripped, and converts the opening f(s T ) is fed back to the automatic control system of the primary fan that has not tripped, expressed as:

[0078] y 3(ε-r) =f3( ε-r) (Y) y 3(ε-r) ∈[μ2,f(s T )]

[0079]

[0080] It should be noted that f(r) represents the opening function of the non-tripped primary fan related to r, and the mapping relationship between f(r) and r is made in advance through experiments and actual data.

[0081] When the unit is running at low load, the furnace flue gas temperature drops, resulting in a decrease in the desulfurization reaction efficiency, which in turn causes ammonia escape, resulting in poor flue flow uniformity or blockage. Therefore, when calculating the opening of the primary fan that has not tripped, the effect of ammonia escape on flue flow is taken into account. When the ammonia concentration is large, that is, greater than the set ammonia concentration in the flue, C 1T / C set The ratio of is greater than 1, so ln(C1T / C set ) is positive, so η 1t The value is greater than 1, so when the ammonia concentration is large, the primary fan is opened at an excessive value, which ensures the efficiency of the desulfurization reaction, the activity of the catalyst, the smoothness of the flue, and good applicability.

[0082] One indicator of flue flue smoothness is the concentration of SO2 in the flue, so adding η 2T Used to characterize the flue gas flow and the influence factor of primary fan opening. Coal-fired boiler: The emission concentration limit of SO2 in the flue gas from the chimney or flue is 200~300㎎ / m 3 . η 2T It indicates that the concentration value of SO2 is positively correlated with the influencing factor. The larger the concentration value of SO2, the larger the influencing factor of SO2, which is consistent with the actual situation.

[0083] y1=f1(Y) is a decreasing function that intersects both the positive half axis of the X axis and the positive half axis of the Y axis. 3(ε-r) =f 3(ε-r) (Y) is an increasing function passing through the origin of the coordinate axis. For example

[0084] y1=f1(Y)=Y 2 ,y 3(ε-r) =-Y 2 +b2

[0085] The parabolic function is used to increase or decrease. When the adjustment time Y is short, the coefficient of the parabolic function can be changed. The curvature and slope of the parabola can be adjusted to change the speed of increase of the rate and opening.

[0086] like Figure 2 As shown, it should be noted that y1 and y 3(ε-r) Symmetrical about y=y0, the decreasing function and the total increasing function are symmetrical, thereby achieving complementarity and achieving the functions of fast, smooth and automatic exit of primary fans with tripping risks and fast, smooth and automatic output replacement of primary fans without tripping risks.

[0087] The upper computer calculates the output rate v3 of the primary fan without tripping risk based on the output of the primary fan without tripping risk and feeds it back to the automatic control system of the primary fan without tripping risk. The rate y of the primary fan without tripping risk is 4(ε-r) The increase from v2 to v3 in Y time is expressed as:

[0088] y 4(ε-r) =f 4(ε-r) (Y) y 4(ε-r) ∈[v2,v3]

[0089]

[0090] The host computer is based on UL=min{UL p , UL w}Calculate the unit load target value UL when the primary fan trips, and then calculate the expected wind pressure of the primary fan that has not tripped, expressed as:

[0091]

[0092] Wherein, μ1 represents the current blade opening of the primary fan with a tripping risk; v1 represents the current blade speed of the primary fan with a tripping risk; μ2 represents the current blade opening of the primary fan without a tripping risk; f(s T ) represents the output blade opening of the primary fan without tripping risk; r represents the number of primary fans that tripped; η 1T represents the influencing factor of ammonia escape; η 2T represents the impact factor of sulfur dioxide; C 1T Indicates the concentration of ammonia in the flue at time T; C set Indicates the set concentration value of ammonia in the flue; C 2T Indicates the concentration of sulfur dioxide in the flue at time T; b 2T represents the fitting coefficient; v2 represents the current speed of the primary fan without tripping risk; v3 represents the output speed of the primary fan without tripping risk; Q represents the expected wind pressure of the primary fan without tripping; Q p represents the current wind pressure of the primary fan that has not tripped; ΔQ represents the wind pressure increment of the primary fan that has not tripped; M represents the fuel quantity corresponding to the unit load target value UL; K M represents the proportional factor of the change in furnace heat load caused by the increase in fuel quantity to M; τ represents the lag time when the fuel quantity increases to M; P b Indicates the primary fan outlet pressure; P indicates atmospheric pressure.

[0093] It should be noted that the unit of v3 is m / s, the unit of Q is cubic, and the stable combustion load value of the unit is generally determined by the stable combustion test without the combustion-supporting system, which is generally 35% to 50% of the rated load of the unit. When the unit is a 1000MW supercritical unit, the stable combustion load is 350MW to 500MW.

[0094] y2=f2(Y) is a decreasing function that intersects both the positive half axis of the X axis and the positive half axis of the Y axis. 4(ε-r) =f 4(ε-r) (Y) is an increasing function passing through the origin of the coordinate axis. Similarly, y2 and y 4(ε-r) Refer to y1 and y 3(ε-r) The function can be selected from polynomial function and trigonometric function.

[0095] At this time, the feed water pump follows the changes in the feed water flow value corresponding to the fuel amount M corresponding to the unit load target value UL, and the multi-order inertia time of the change is switched from the conventional 20 to 30 seconds to 5 seconds; during the low-load smooth transition control action, the manual cut-off conditions for the large deviation between the control quantity and the set value in the primary fan automatic control and the manual cut-off conditions for the large deviation between the valve and the opening command feedback and the set value are shielded.

[0096] Furthermore, when the feed water pump is at risk of tripping, the feed water pump enters automatic tripping control. The feed water pump that has not tripped follows the change of the feed water flow value corresponding to the fuel amount M corresponding to the unit load target value UL, and the multi-order inertia time of the change is switched from the conventional 20 to 30 seconds to 10 seconds.

[0097] The recirculation valve of the water supply pump that has not tripped is fully closed at time Y. The closing rate here obeys a decreasing function, which is a decreasing function that intersects both the positive half axis of the X axis and the positive half axis of the Y axis. The independent variable of the decreasing function is Y.

[0098] During the low-load smooth transition control action, the manual cut-off conditions when the control quantity in the automatic control of the cold and hot air doors of the coal mill deviates greatly from the set value, as well as the manual cut-off conditions when the valve and opening command feedback deviates greatly from the set value are shielded.

[0099] Furthermore, when the coal mill trips, the coal mill tripping automatic control is activated, and the upper computer calculates the target fuel quantity M g , the running coal feeder automatically puts in the fuel quantity automatic control, which is expressed as:

[0100]

[0101] Among them, M w Indicates the amount of fuel for stable combustion; M max Indicates the actual maximum acceptable fuel quantity of the pulverizing system; indicates the current actual fuel quantity; M P Indicates the maximum fuel quantity of a single coal feeder; UL indicates the current load value of the unit; UL w Indicates the steady-state combustion load value of the unit; M 1max Indicates the maximum fuel quantity of a single coal feeder; M 2max It indicates the maximum fuel quantity of a single coal mill; z indicates the number of coal mills remaining in operation.

[0102] It should be noted that before the low-load smooth transition is triggered, the control mode of the unit is the closed-loop regulation of the unit power on the turbine side, the closed-loop regulation of the machine front pressure on the boiler side + ULD feedforward mode. After the low-load smooth transition is triggered, the control mode of the unit is switched to the closed-loop regulation of the machine front pressure on the turbine side, and the boiler side tracking of the low-load smooth transition target fuel quantity instruction mode.

[0103] Furthermore, it is determined whether the automatic control is completed. If so, the system enters a low-load smooth transition, otherwise it returns to automatic control.

[0104] The conditions for the completion of automatic control include: the deviation between the actual load value of the unit and the unit load target value UL is less than 15MW; the automatic control adjustment of the primary fan is stable, and the furnace pressure fluctuation range is less than ±200Pa; the oxygen content of the furnace flue gas flue gas fluctuates within ±0.5%; the automatic control adjustment of the feed water pump is stable, and the main steam temperature fluctuation is less than ±3℃.

[0105] It should be noted that in the determination of the conditions for the completion of a smooth low-load transition, the specific conditions for the stable adjustment of the automatic feed water control are: the superheat fluctuation at the separator outlet is less than ±5°C, and the drum water level fluctuation is less than ±25mm.

[0106] The present embodiment also provides a low-load coordinated control system for a thermal power unit, including a fault prediction module, which collects data of the thermal power unit, establishes and optimizes an auxiliary machine tripping prediction model based on a convolutional neural network, and uses the trained auxiliary machine tripping prediction model to predict the tripping risk of the auxiliary machine; a fault classification module, which obtains new evidence based on the DS evidence theory, judges and classifies the auxiliary machine tripping risk, and judges whether the fault triggers the auxiliary machine fault load reduction RB function. If the RB function is triggered, the process ends, and if the RB function is not triggered, automatic control is entered; an automatic control module, which performs automatic control of a fan, a water supply pump or a coal mill based on the judgment result of the DS evidence theory, and enters a low-load smooth transition after the automatic control is completed.

[0107] This embodiment also provides a computing device, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the low-load coordinated control method of thermal power units proposed in the above embodiment.

[0108] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the low-load coordinated control method for a thermal power unit proposed in the above embodiment is implemented.

[0109] The storage medium proposed in this embodiment and the low-load coordinated control method for thermal power units proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0110] Through the above description of the implementation method, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccessMemory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0111] Example 2

[0112] The following is an embodiment of the present invention, which provides a low-load coordinated control method for a thermal power unit. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0113] A simulation experiment was conducted using historical data from three coal mills, two primary fans and a water supply pump. By manually modifying some parameters in the historical data, controllable fault data was generated to verify the beneficial effects of the present invention. The data collected at a certain moment was selected for analysis and display. The analysis results are shown in the following table.

[0114] Table 1 Experimental data table

[0115]

[0116]

[0117] It can be seen from the data that the present invention has very high accuracy in the fault judgment stage, and can realize accurate analysis of fault probability based on the collectible data of the auxiliary machine. The auxiliary machine trip prediction model is constructed and optimized by gate filter, and the cross entropy loss function and gate filter are added to the loss function to eliminate the uncertainty of the system at the lowest cost and improve the accuracy of the auxiliary machine trip prediction model.

[0118] The evidence combination rule is used to fuse the fault evidence of different auxiliary machines to predict the auxiliary machine failure probability, and the fused evidence is obtained. The fused evidence is converted into the auxiliary machine tripping probability, and the need for a smooth transition to low load is determined based on the auxiliary machine tripping probability.

[0119] The blade opening and speed of the primary fan with tripping risk decay to 0 according to the decreasing function, and the blade opening and speed of the primary fan with no tripping risk increase to the target value according to the increasing function. The decreasing function and the total increasing function are symmetrical, thereby achieving complementarity and achieving the functions of fast, smooth and automatic exit of the primary fan with tripping risk and fast, smooth and automatic output replacement of the primary fan with no tripping risk.

[0120] The automatic control of the coal mill and water feed pump also achieves timeliness and accuracy that cannot be achieved by manual operation. It can prevent failures in advance and achieve a smooth transition of the system.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A low-load coordinated control method for a thermal power unit, characterized in that: include: Collect data from thermal power units, establish auxiliary machine trip prediction models, and predict auxiliary machine failures; Use DS evidence theory to judge the risk of auxiliary machine tripping. If the fault does not trigger the auxiliary machine fault load reduction RB function, then perform fault classification; The faulty auxiliary machine is automatically controlled according to the fault classification results, and a smooth low-load transition is entered after the automatic control is completed.

2. The low load coordinated control method for a thermal power unit according to claim 1, characterized in that: The auxiliary machine trip prediction model includes collecting the operating parameters of the auxiliary machine in the time period [Tt, T] before the trip occurs to construct a data set X, the auxiliary machine is divided into three types: primary fan, feed water pump and coal mill, and the data set X is divided into a training set and a test set; A convolutional neural network is used to build an auxiliary machine trip prediction model, and a gate filter φ is added after each convolution layer. P , door filter φ P is [1, C P , 1, 1], the output feature map of the convolution layer is multiplied by the gate filter in the channel direction, thereby changing the standard neural network model module, combined with C P The random assignment of achieves model optimization; The loss function L is used to accurately predict the tripping risk in an unbalanced data set. The regularization term L2 is added to the convolutional neural network loss term L1, and the model parameters are adjusted, which is expressed as: L=(1-α)L1+αL2 Among them, α represents the weight of L2; C P represents the number of channels in the gate filter added after the Pth convolutional layer; X represents the input data; Y represents the label; θ represents the parameter of the auxiliary machine tripping prediction model; δ represents the gradient; β represents the coefficient; Indicates the modulation coefficient.

3. The low load coordinated control method for a thermal power unit according to claim 2, characterized in that: The steps of the DS evidence theory are: Based on the DS evidence theory, the identification framework Θ={F} is set, and the risk W ij Assign the value of m to ij {F}; The weighted average of multi-time evidence is expressed as: Evidence combination rules are used for evidence fusion. The evidence combination rules for any two groups are: Convert the fused evidence into auxiliary machine trip probability BetP(A); Among them, m p (A) represents the weighted average evidence information; ω t Represents evidence m at time T-t+1 (T-t+1)ij (A) weight; l represents the total time; Represents the fused evidence information; m p represents the evidence to be integrated in the identification framework; A represents the power set 2 Θ A subset of; B represents the power set 2 Θ A subset of .

4. The low load coordinated control method for a thermal power unit according to claim 3, characterized in that: The fault classification includes the tripping risk of the primary fan, the tripping risk of the feed water pump and the tripping risk of the coal mill. The auxiliary machine with the fault is determined according to the output BetP(A) of the DS evidence theory and enters automatic control.

5. The low-load coordinated control method for a thermal power unit according to claim 4, characterized in that: When the primary fan has a tripping risk, the primary fan enters the primary fan tripping automatic control, and the fan blade opening y1 of the primary fan with a tripping risk decays to zero within the Y time, which is expressed as: y1=f1(Y)y1∈[0,μ1] The speed y2 of the primary fan with tripping risk decays to zero within Y time, expressed as: y2=f2(Y)y2∈[0,v1] The fan blade opening y of the primary fan without tripping risk 3(ε-r) In Y time, it increases from μ2 to f(s T ), the host computer calculates the opening of the primary fan that did not trip when the primary fan tripped, and converts the opening f(s T ) is fed back to the automatic control system of the primary fan that has not tripped, expressed as: y 3(ε-r) =f 3(ε-r) (Y) y 3(ε-r) ∈[μ2,f(s T )] The upper computer calculates the output rate v3 of the primary fan without tripping risk based on the output of the primary fan without tripping risk and feeds it back to the automatic control system of the primary fan without tripping risk. The rate y of the primary fan without tripping risk is 4(ε-r) The increase from v2 to v3 in Y time is expressed as: <h2 style=";text-align:left;direction:ltr">y<h2 style=";text-align:left;direction:ltr"> 4(ε-r) <h2 style=";text-align:left;direction:ltr"> =f<h2 style=";text-align:left;direction:ltr"> 4(ε-r) <h2 style=";text-align:left;direction:ltr"> (Y) y<h2 style=";text-align:left;direction:ltr"> 4(ε-r) <h2 style=";text-align:left;direction:ltr"> ∈[v2, v3] Wherein, μ1 represents the current blade opening of the primary fan with a tripping risk; v1 represents the current blade speed of the primary fan with a tripping risk; μ2 represents the current blade opening of the primary fan without a tripping risk; f(s T ) represents the output blade opening of the primary fan without tripping risk; r represents the number of primary fans that tripped; η 1T represents the influencing factor of ammonia escape; η 2T represents the impact factor of sulfur dioxide; C 1T Indicates the concentration of ammonia in the flue at time T; C set Indicates the set concentration value of ammonia in the flue; C 2T Indicates the concentration of sulfur dioxide in the flue at time T; b 2T Represents the fitting coefficient; v2 represents the current speed of the primary fan without tripping risk; v3 represents the output speed of the primary fan without tripping risk; Q represents the expected wind pressure of the primary fan without tripping.

6. The low-load coordinated control method for a thermal power unit according to claim 5, characterized in that: When the feedwater pump has a tripping risk, it enters the feedwater pump tripping automatic control, and the feedwater pump that has not tripped follows the feedwater flow value corresponding to the fuel amount M corresponding to the unit load target value UL, and the multi-order inertia time of the change is switched from the conventional 20 to 30 seconds to 10 seconds; The recirculation valve of the feedwater pump that has not tripped is fully closed at time Y; During the low-load smooth transition control action, the manual cut-off conditions when the control quantity in the automatic control of the cold and hot air doors of the coal mill deviates greatly from the set value, as well as the manual cut-off conditions when the valve and opening command feedback deviates greatly from the set value are shielded.

7. The low-load coordinated control method for a thermal power unit according to claim 6, characterized in that: When the coal mill has a tripping risk, it enters the coal mill tripping automatic control, and the host computer calculates the target fuel quantity M g , the running coal feeder automatically puts in the fuel quantity automatic control, which is expressed as: Among them, M w Indicates the amount of fuel for stable combustion; M max Indicates the actual maximum acceptable fuel quantity of the pulverizing system; indicates the current actual fuel quantity; Mp indicates the maximum fuel quantity of a single coal feeder; UL indicates the current load value of the unit; UL w Indicates the steady-state combustion load value of the unit; M 1max Indicates the maximum fuel quantity of a single coal feeder; M 2max It indicates the maximum fuel quantity of a single coal mill; z indicates the number of coal mills remaining in operation.

8. A low-load coordinated control system for a thermal power unit using the method according to any one of claims 1 to 7, characterized in that: include, The fault prediction module collects data from thermal power units, establishes and optimizes an auxiliary machine trip prediction model based on a convolutional neural network, and uses the trained auxiliary machine trip prediction model to predict the trip risk of the auxiliary machine; The fault classification module obtains new evidence based on the DS evidence theory, judges and classifies the risk of auxiliary machine tripping, and determines whether the fault triggers the auxiliary machine fault load reduction RB function. If the RB function is triggered, the process ends; if the RB function is not triggered, automatic control is entered; The automatic control module performs automatic control of the fan, water supply pump or coal mill based on the judgment results of the DS evidence theory. After the automatic control is completed, it enters a low-load smooth transition.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images