Fire storage double-layer cooperative automatic power generation control and adjustment method

Through the coordinated control of lithium battery and flywheel energy storage hybrid system, the prediction model and dynamic filter coefficient method are used to solve the problems of low adjustment accuracy of thermal power units and low economics of energy storage systems, and efficient and fast automatic power generation control is achieved. Improves system performance and security.

CN118970916APending Publication Date: 2024-11-15STATE GRID HENAN ELECTRIC POWER ELECTRIC POWER SCI RES INST +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411034754.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In automatic power generation control, thermal power units have low adjustment accuracy and slow response. Frequent load changes lead to equipment losses, affecting safety and economic benefits. The existing energy storage system is not economically high when used alone or its life is affected by the state of charge.

Method used

Adopting a hybrid system for lithium batteries and flywheel energy storage, the power demand is predicted by adding a long and short-term memory network with attention mechanism, combined with time domain rolling optimization and dynamic adjustment filter coefficient method based on charge state of the process to coordinate the real-time energy storage of lithium batteries and flywheel energy storage. Power distribution enables efficient and fast automatic power generation control and regulation.

Benefits of technology

It improves automatic power generation control performance, achieves efficient and fast power adjustment, extends the service life of lithium batteries, and reduces the equipment loss of thermal power units and the economic cost of energy storage systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118970916A_ABST
    Figure CN118970916A_ABST
Patent Text Reader

Abstract

The invention discloses a fire storage double-layer cooperative automatic power generation control and adjustment method, and relates to the field of automatic power generation control. The method comprises the following steps: inputting historical power data of a thermal power plant unit into a power demand prediction model to obtain an energy storage power demand prediction result at a prediction moment; the power demand prediction model is a long short-term memory network added with an attention mechanism; according to the energy storage power demand prediction result, outputting the total power of the hybrid energy storage system in the prediction time domain by using the model prediction control model; optimizing the total power increment of the hybrid energy storage system in the prediction time domain by adopting a time domain rolling optimization method, and further determining a planned power target instruction of the hybrid energy storage system at each moment in the prediction time domain; and according to the planned power target instruction, performing coordinated distribution on real-time power of energy storage of a lithium battery and a flywheel in the hybrid energy storage system by adopting a dynamic adjustment filter coefficient method based on the state of charge. According to the invention, efficient and rapid automatic power generation control adjustment can be realized, and the automatic power generation control performance is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of automatic power generation control, and in particular to a method for controlling and regulating double-layer coordinated automatic power generation of thermal storage. Background Art

[0002] The Automatic Generation Control (AGC) system can detect the frequency deviation in the power network in real time, convert it into the corresponding active power shortage, and reasonably allocate it to the frequency-regulated power source under its control, so as to adjust its active output accordingly. However, due to the low regulation accuracy and slow ramp rate of thermal power units, their real-time power response is weak, and they perform poorly in the process of tracking AGC instructions. At the same time, frequent variable load operation will cause serious equipment loss of the generator set, which will have a serious impact on the safety and economic benefits of thermal power units.

[0003] Energy storage systems can greatly enhance the AGC regulation performance of thermal power units due to their characteristics of energy time shifting, rapid response, and flexible deployment. Lithium batteries are one of the most commonly used energy storage systems, with the characteristics of large throughput and low cost, but their frequent and irregular charging and discharging will reduce their service life. Flywheel energy storage, as a new type of energy storage system, has the characteristics of instantaneous response and life that is not affected by the state of charge, but its energy capacity is low and it is not economical when used alone for AGC regulation. Therefore, the technical characteristics of the two can be fully combined to assist thermal power units in achieving efficient and rapid AGC regulation by configuring a hybrid system of lithium batteries and flywheel energy storage. Summary of the invention

[0004] The purpose of this application is to provide a thermal storage double-layer collaborative automatic power generation control and regulation method, which can realize efficient and rapid automatic power generation control and regulation and improve the automatic power generation control performance.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] The present application provides a thermal-storage dual-layer collaborative automatic power generation control and regulation method, comprising: obtaining historical power data of thermal power plant units; inputting the historical power data of the thermal power plant units into a power demand prediction model of a hybrid energy storage system to obtain a prediction result of the energy storage power demand at the prediction moment; the power demand prediction model is a long short-term memory network with an attention mechanism added; the hybrid energy storage system includes a lithium battery and a flywheel energy storage; according to the prediction result of the energy storage power demand at the prediction moment, using a model predictive control model to output the total power of the hybrid energy storage system in the prediction time domain; based on the total power of the hybrid energy storage system in the prediction time domain and the prediction result of the energy storage power demand at the prediction moment, using a time domain rolling optimization method to optimize the total power increment of the hybrid energy storage system in the prediction time domain; according to the total power increment of the hybrid energy storage system in the prediction time domain, determining the planned power target instruction of the hybrid energy storage system at each moment in the prediction time domain; according to the planned power target instruction of the hybrid energy storage system at each moment in the prediction time domain, using a dynamic adjustment filter coefficient method based on the state of charge to coordinate the real-time power of the lithium battery and the flywheel energy storage at each moment in the prediction time domain.

[0007] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0008] The present application provides a thermal power storage dual-layer coordinated automatic power generation control and regulation method. The upper layer designs a power demand prediction model to predict the total power of the hybrid energy storage system, and uses a time domain rolling optimization method to optimize the total power increment of the hybrid energy storage system in the predicted time domain, providing the lower layer with a reference trajectory for the actual power generation of the hybrid energy storage; in view of the different characteristics of lithium batteries and flywheel energy storage in the hybrid energy storage system, the lower layer control adopts a dynamic adjustment filter coefficient method based on the state of charge to coordinate the real-time power distribution of lithium batteries and flywheel energy storage. By considering the thermal power-hybrid energy storage dual-layer coordination of power demand estimation, efficient and rapid automatic power generation control and regulation are achieved, and the automatic power generation control performance is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0010] Figure 1 It is a flow chart of a method for controlling and regulating double-layer coordinated automatic power generation of thermal storage in one embodiment of the present application.

[0011] Figure 2 A brief flowchart of a method for controlling and regulating double-layer coordinated automatic power generation of thermal storage provided in another embodiment of the present application. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0013] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0014] In an exemplary embodiment, Figure 1 As shown, a method for controlling and regulating double-layer thermal storage coordinated automatic power generation is provided, comprising the following steps 101 to 106. Among them:

[0015] Step 101: Obtain historical power data of thermal power plant units.

[0016] Step 102: Input the historical power data of the thermal power plant units into the power demand prediction model of the hybrid energy storage system to obtain the energy storage power demand prediction result at the prediction time; the power demand prediction model is a long short-term memory network with an attention mechanism added; the hybrid energy storage system includes lithium batteries and flywheel energy storage.

[0017] Step 103: Based on the energy storage power demand prediction result at the prediction time, the model predictive control model is used to output the total power of the hybrid energy storage system in the prediction time domain.

[0018] Step 104: Based on the total power of the hybrid energy storage system in the forecast time domain and the energy storage power demand forecast result at the forecast time, a time domain rolling optimization method is used to optimize the total power increment of the hybrid energy storage system in the forecast time domain.

[0019] Step 105: Determine the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain according to the total power increment of the hybrid energy storage system in the forecast time domain.

[0020] Step 106: Based on the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain, a dynamic adjustment filter coefficient method based on the state of charge is used to coordinate and allocate the real-time power of the lithium battery and the flywheel energy storage at each moment in the forecast time domain.

[0021] Implement the above steps 101 to 106. The upper layer designs a power demand prediction model to predict the total power of the hybrid energy storage system, and uses the time domain rolling optimization method to optimize the total power increment of the hybrid energy storage system in the prediction time domain, providing the lower layer with a reference trajectory of the actual power of the hybrid energy storage; in view of the different characteristics of lithium batteries and flywheel energy storage in the hybrid energy storage system, the lower layer control adopts a dynamic adjustment filter coefficient method based on the state of charge to coordinate the real-time power distribution of lithium batteries and flywheel energy storage. By considering the dual-layer coordination of thermal power and hybrid energy storage with power demand estimation, efficient and rapid automatic power generation control adjustment is achieved, and the automatic power generation control performance is improved.

[0022] In another exemplary embodiment of the present application, real data of a thermal power plant is collected, and the data is preprocessed to identify and correct abnormal values ​​and missing values ​​in the data. Then the above step 101 is replaced by the following steps 201 to 206:

[0023] Step 201: Select one month in each quarter of the year to collect historical operation data of the thermal power plant units from the distributed control system of the generator sets; the historical operation data of the units include automatic power generation control instructions and the actual power generation of the thermal power units.

[0024] Step 202: Use the mean square method to eliminate abnormal data in the unit operation history data, and use the linear fitting method to replace the abnormal data.

[0025] Step 203: normalize the unit operation history data that replaces the abnormal data to obtain normalized unit operation history data.

[0026] Step 204: Calculate respectively the difference between the automatic power generation control instructions at adjacent historical moments in the normalized unit operation history data, and the difference between the actual power generation of the thermal power units at adjacent historical moments.

[0027] Step 205: According to the normalized unit operation history data, the automatic generation control command and the actual power of the thermal power unit at the same historical moment are calculated according to the formula P d (t) = P AGC (t)-P g (t), calculate the historical power demand of the hybrid energy storage system at each historical moment; where P d (t) is the historical power demand of the hybrid energy storage system at historical time t, P AGC (t) is the automatic power generation control instruction at historical time t, P g (t) is the actual power generated by the thermal power unit at historical moment t.

[0028] Step 206: The normalized historical unit operation data, the difference of automatic power generation control instructions at adjacent historical moments, the difference of actual power generation of thermal power units at adjacent historical moments, and the historical power demand of the hybrid energy storage system at each historical moment constitute the historical power data of the thermal power plant units.

[0029] Taking a 330MW thermal power plant in North China as an example, real data of a 330MW thermal power plant in North China is collected, and these data are preprocessed to identify and correct abnormal values ​​and missing values ​​in the data. The specific process of the above data preprocessing is as follows:

[0030] Step 1-1 Data collection. First, collect historical data from the distributed control system (DCS) system of the generator set, and use it as the historical data of the unit operation in four months of the year (one month in each quarter) to avoid errors caused by different power loads caused by seasons or natural reasons. The data sampling interval is 3s. The data collection range should include no less than 40 power command fluctuation levels, and there should be no less than 15 large-scale load increase and decrease data per month. It is also necessary to ensure that the thermal power unit operates without shutdown and failure within the data collection range; the collected data includes AGC command P AGC and actual power P g The collected data can also include the system's primary air volume Q1, coal supply B and water supply D.

[0031] Step 1-2: Abnormal value processing. The abnormal data collected in step 1-1 is processed by the mean square method. The calculation formula of the mean square method is shown below.

[0032]

[0033] In the formula, N is the number of samples, x i is the i-th sample value. If |x i -μ>3δ, then x i The outliers are judged to be replaced by linear fitting method.

[0034] Step 1-3 is normalized. To speed up the network training, the data obtained in step 1-2 is normalized. The values ​​are converted to the range of [0,1]. The normalization process is shown in the following formula.

[0035]

[0036] In the formula, x min is the minimum value of the processed feature data, x max is the maximum value among the processed feature data.

[0037] In another exemplary embodiment of the present application, the establishment of the power demand prediction model in the above step 102 can be implemented by the following steps 301 to 302:

[0038] Step 301: Determine the input variable of the power demand prediction model as x(t)=[P g (t),P AGC (t),ΔP g (t),ΔP AGC (t),P d (t)]; where x(t) is the input variable of the power demand prediction model at historical time t, P g (t) is the actual power generated by the thermal power unit at the historical moment t, P AGC (t) is the automatic power generation control command at historical time t, ΔP g (t) is the difference in actual power generation of the thermal power unit at the historical time t and the historical time t-1, ΔP AGC (t) is the command difference between the automatic power generation control at the historical time t and the historical time t-1, P d (t) is the historical power demand of the hybrid energy storage system at historical time t.

[0039] Step 302: Add an attention mechanism between the hidden layer and the fully connected layer of the long short-term memory network to form a power demand prediction model for the hybrid energy storage system.

[0040] Since the attention mechanism is added to the long short-term memory network, the power demand prediction model is an ultra-short-term HESS (Hybrid Energy Storage System) power demand prediction model based on LSTM (Long Short-Term Memory)-AM (Attention Mechanism). The following is a more detailed introduction to the establishment process of the power demand prediction model:

[0041] Step 2-1 Determination of input variables. Consider the power demand sequence P of the HESS at prediction time k d =[P d (1),P d (2),...,P d (k)] (assuming that HESS discharge is positive and charge is negative), then:

[0042] P d (k) = P AGC (k)-P g (k)

[0043] Where P d is the ultra-short-term forecast data of HESS power demand value, P AGCAGC instruction, P g It is the actual power of thermal power unit.

[0044] From the above formula, we can see that P d (k) is not only related to the AGC instruction, but also affected by P g The following sequences are selected as input variables of the power demand forecasting model:

[0045] x(t)=[P g (t),P AGC (t),ΔP g (t),ΔP AGC (t),P d (t)]

[0046] Step 2-2 adds AM. When building the LSTM model, AM can be used to make different neurons hide vectors h at different times. t Assign the corresponding weight β tj , merge the hidden vectors into a new hidden vector h according to their importance t ' and input to the decoder to extract important LSTM neurons and timing information and improve weights to improve prediction accuracy. The calculation formula of AM is:

[0047] e tj =a(h t-1 ',h t )

[0048]

[0049] In the formula, e tj is the attention score, h t-1 ' is the hidden layer state of another LSTM network in the decoder at the previous moment, h t is the state of the encoder at each moment; j represents the jth element in the sequence; the attention score is calculated by the function a(·); β tj Then it corresponds to e tj The attention weight of C t is the output of the AM layer at time t; T represents the length of the sequence.

[0050] Step 2-3 designs a prediction model based on LSTM-AM. The LSTM layer input vector x(t),...,x(t-H+1) is the H multidimensional feature vectors before the time to be predicted. The AM layer is used as the interface of the LSTM network. The hidden layer state of all neurons of the LSTM model at time H is used as the input of the AM layer, where H is the time window length of the input sequence. First, calculate the score e tj , represents the degree of fit between the input sequence component and the current output at position t. The higher the value, the greater the correlation. Then according to the above formula, according to the attention score etj Get the weight β tj Finally, the obtained attention weights are assigned to different intermediate layer states, and the sum is obtained to obtain the generated vector C input to the decoder t . Vector C t Through the fully connected layer, the energy storage power demand prediction result is calculated.

[0051] In another exemplary embodiment of the present application, the establishment of the model predictive control model in the above step 103 can be implemented by the following steps 401 to 403:

[0052] Step 401: Establish the power balance equation of the automatic power generation control system as P dif (k) = P d (k)-P h (k) = P d (k)-(P b (k)+P f (k)); where P dif (k) is the power difference after compensation at prediction time k, P d (k) is the energy storage power demand forecast result at the k forecast time, P h (k) is the actual power of the hybrid energy storage system at prediction time k, P b is the power of lithium battery, P f The power of the flywheel energy storage.

[0053] Step 402: Establish the charge state calculation equation of the hybrid energy storage system: In the formula, S h (k) is the charge state of the hybrid energy storage system at prediction time k, S b (k) and S f (k) are the charge states of the lithium battery and flywheel energy storage at prediction time k, E b and E f They are the capacities of lithium battery and flywheel energy storage respectively.

[0054] Step 403: Based on the power balance equation and the state of charge calculation equation, the state space equation of the model predictive control model is constructed as follows: Where x(k+1) and x(k) are the input vectors at the k+1 prediction time and the k prediction time, respectively. dif (k) P h (k) S h (k)] T ; △u(k) is the control vector, △u(k)=[△P h (k)], ΔP h (k) is the total power increment of the hybrid energy storage system at the k prediction time; △v(k) is the disturbance vector, Δv(k)=[ΔPd (k)], ΔP d (k) is the predicted increment of energy storage power demand at prediction time k; y(k) is the output vector, y(k) = [P dif (k)S h (k)] T ; A, B, C, D are all coefficient matrices, D =

[100] T , E h is the capacity of the hybrid energy storage system, and △T is the sampling time.

[0055] The following is a more detailed introduction to the process of establishing the Model Predictive Control (MPC) model:

[0056] Establish an MPC model. The MPC designed in this application is used for real-time rolling optimization control to track AGC instructions. It takes the real-time data collected at time k and the ultra-short-term HESS power demand forecast data obtained by prediction as input, and rolls to optimize the planned power generation of HESS in the prediction time domain. The above process is repeated at time k+m, where m is the length of the prediction time domain and the control time domain.

[0057] After obtaining the power demand at the future moment, the HESS system will adjust its power output to track the power demand. Due to the state of charge (SOC) and capacity limitations of the energy storage system, it is impossible to fully compensate for the power demand. The difference can be calculated by the following formula:

[0058] P dif (k) = P d (k)-P h (k) = P d (k)-(P b (k)+P f (k))

[0059] The above formula is the power balance equation of the automatic power generation control system.

[0060] State of charge S of hybrid energy storage system h It is calculated based on the real-time state of charge of the lithium battery and flywheel energy storage, as shown in the following formula.

[0061]

[0062] The state space equation of the MPC model is constructed based on the above two equations, as shown below:

[0063]

[0064] In the formula, the coefficient matrices A, B, C, and D are introduced, and the specific expressions are as follows. According to the power balance equation of the automatic power generation control system and the real-time charge state calculation formula, select P dif , P h and S h As the state vector, P dif and S h As the output vector, the total power P of the HESS h As the control vector, the ultra-short-term predicted power increment ΔP of the HESS power demand at prediction time k d (k) as the disturbance vector, ΔP h (k) is the total power increment of HESS at prediction time k.

[0065] x(k)=[P dif (k)P h (k)S h (k)] T

[0066] y(k)=[P dif (k)S h (k)] T

[0067] Δu(k)=[ΔP h (k)]

[0068] Δv(k)=[ΔP d (k)]

[0069]

[0070] As the MPC prediction model iterates at each time point, the output vector Y(k) and state vector X(k) of the prediction model within the set prediction period can be obtained, as shown in the following formula.

[0071] Y(k)=[y(k+1|k),y(k+2|k),...,y(k+m|k)] T

[0072] X(k)=[x(k+1|k),x(k+2|k),...,x(k+m|k)] T

[0073] In the formula, y(k+n|k) and x(k+n|k) are defined as the output vector and state vector at time k+n in the prediction domain at prediction time k. The value of n ranges from 1 to m, and m is the length of the prediction domain and the control domain.

[0074] In another exemplary embodiment of the present application, rolling optimization is adopted. The upper layer MPC rolling optimization utilizes the real-time data P of the system dif , Sh , P h , and combined with the ultra-short-term power forecast results of HESS power demand ΔP d (k) By integrating the above data and feedback correction, ΔP h The optimization objectives considered by the upper control model include the compensated power difference P dif As small as possible, S h Keep it near the expected value and minimize the power change of HESS. Use the time domain rolling optimization method to optimize ΔP h (k) to optimize the tracking effect. Then the objective function of the time domain rolling optimization method in the above step 104 is:

[0075]

[0076] y ref (k) = [P dif,plan (k+1|k),S h,ref (k+1|k),

[0077] P dif,plan (k+2|k),S h,ref (k+2|k),...,

[0078] P dif,plan (k+m|k),S h,ref (k+m|k)] T

[0079] Δu i (k)=[Δu(k+1|k),Δu(k+2|k),...,Δu(k+m|k)] T

[0080] In the formula, y i is the output vector of the i-th upper layer rolling optimization cycle, y ref (k) is the target tracking vector in the prediction time domain at the k prediction time, Δu i (k) is the control vector of the i-th upper layer rolling optimization cycle, P dif,plan (k+w|k) is the target deviation at each moment in the prediction time domain at the k prediction moment, S h,ref (k+w|k) is the target state of charge value of the hybrid energy storage system in the prediction time domain at the k prediction time, Δu(k+w|k) is the control vector at each moment in the prediction time domain at the k prediction time, w=1,2,…,m, m is the length of the prediction time domain, and the time length of the upper rolling optimization cycle is equal to the time length of the prediction time domain; Q and R are both weight coefficient matrices.

[0081] On this basis, in order to ensure the feasibility and effectiveness of the optimization results, the following constraints are set:

[0082] The state of charge constraints are:

[0083] S h,min ≤S h (k+n|k)≤S h,max

[0084] In the formula, S h (k+n|k) is the charge state of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k; S h,min and S h,max They are the minimum state of charge and maximum state of charge of the hybrid energy storage system respectively.

[0085] The power constraint is:

[0086] P h,min (k)≤P h (k+n|k)≤P h,max (k)

[0087] P h,min (k)-P h (k+n|k)≤ΔP h (k+n|k)≤P h,max (k)-P h (k+n|k)

[0088] Where P h (k+n|k) is the total power of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k, P h,min (k), P h,max (k) are the minimum and maximum total power of the hybrid energy storage system in the prediction time domain at the k prediction time determined according to the real-time charge state of the lithium battery and the flywheel energy storage; ΔP h (k+n|k) is the total power increment of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k.

[0089] In order to avoid power cycling within the HESS and reduce unnecessary charging or discharging, P h (k) and P dif (k) should have the same sign. However, when P d When it is close to 0, the predicted value may be positive or negative. The logical constraints of energy storage charging and discharging are:

[0090]

[0091] Where, ε is a constant, ε<0, and can be adjusted according to the results; P d(k+n|k) is the energy storage power demand forecast result at time k+n in the forecast time domain at forecast time k, P dif (k+n|k) is the power difference after compensation at time k+n in the prediction time domain at prediction time k.

[0092] The above steps 103 and 104 constitute an upper-level control strategy based on MPC rolling.

[0093] In another exemplary embodiment of the present application, the above step 105 can be replaced by the following steps 501 to 503:

[0094] Step 501: extracting the control vector at each moment in the total power increment of the hybrid energy storage system in the prediction time domain.

[0095] Step 502: Based on the control vector at each moment, the state space equation of the model predictive control model is used Calculate the state vector at each moment in the next rolling optimization cycle; where x(k+1) and x(k) are the input vectors at the k+1 prediction moment and the k prediction moment, respectively, and x(k) = [P dif (k)P h (k)S h (k)] T ; △u(k) is the control vector, △u(k)=[△P h (k)], ΔP h (k) is the total power increment of the hybrid energy storage system at the k prediction time; Δv(k) is the disturbance vector, Δv(k) = [ΔP d (k)], ΔP d (k) is the predicted increment of energy storage power demand at prediction time k; y(k) is the output vector, y(k) = [P dif (k)S h (k)] T ; A, B, C, D are all coefficient matrices, D =

[100] T , E h is the capacity of the hybrid energy storage system, and ΔT is the sampling time.

[0096] Step 503: The total power of the hybrid energy storage system in the state vector at each moment in the next rolling optimization cycle is used as the planned power target instruction of the hybrid energy storage system at each moment in the measurement time domain.

[0097] In another exemplary embodiment of the present application, in order to realize the coordinated distribution of real-time power of lithium batteries and flywheel energy storage, the above step 106 can be implemented by the following steps 601 to 603:

[0098] Step 601: According to the real-time state of charge of the flywheel energy storage, a method of dynamically adjusting the filter coefficient based on the state of charge is adopted to determine the real-time filter time constant.

[0099] Step 602: Using limit control, determine the real-time power constraints of the lithium battery and flywheel energy storage as Where Δt is the lower layer control period, P z,min , P z,max are the lower and upper limits of the physical power of energy storage respectively; P z,min (k+n),P z,max (k+n) are the lower and upper limits of the power of the energy storage at time k+n in the prediction time domain at prediction time k, respectively. z,min , S z,max are the lower and upper limits of the state of charge of the energy storage, S z (k+n-1) is the state of charge of the energy storage at time k+n-1 in the prediction time domain at prediction time k, E z is the energy storage capacity, z=b or f.

[0100] Step 603: Based on the planned power target instruction and real-time filtering time constant of the hybrid energy storage system at each moment in the forecast time domain, combined with the real-time power constraints of the lithium battery and flywheel energy storage, according to the formula Obtain the real-time power command of the lithium battery and flywheel energy storage at each moment in the prediction time domain; where τ is the real-time filter time constant, T is the sampling period, P b_ref (k+n) is the power instruction of the lithium battery at time k+n in the prediction time domain at prediction time k, P b_ref (k+n-1) is the power instruction of the lithium battery at time k+n-1 in the prediction time domain at prediction time k, P f_ref (k+n) is the power command of the flywheel energy storage at time k+n in the prediction time domain at prediction time k, is the planned power target instruction of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k.

[0101] The specific implementation of the above step 601 is as follows:

[0102] 1) If the real-time charge state of the flywheel energy storage is [S f,low , S f,high ] interval, the flywheel energy storage (FES) is in normal working state, then the filter time constant is set to be equal to the initial value of the filter time constant; where S f,low is the lower limit threshold of the charge state of the flywheel energy storage, S f,high It is the upper limit threshold of the charge state of the flywheel energy storage.

[0103] 2) If the real-time charge state of the flywheel energy storage is [Sf,min , S f,low ], the discharge capacity of FES is poor, but the charging capacity is good. According to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; where S f,min is the minimum state of charge of the flywheel energy storage, S f is the real-time charge state of the flywheel energy storage, τ d It is the maximum variation range of the filter time constant increment Δτ.

[0104] 3) If the real-time charge state of the flywheel energy storage is [S f,high , S f,max ], the discharge capacity of FES is better, but the charging capacity is worse. According to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; where S f,max It is the maximum state of charge of the flywheel energy storage.

[0105] 4) If the real-time charge state of the flywheel energy storage is in [0, S f,min ] interval (FES is in the discharge forbidden area), and According to the formula The filter time constant increment Δτ is determined, and the filter time constant is obtained by combining the filter time constant increment Δτ and the filter time constant initial value, and the FES bears more high-frequency components.

[0106] 5) If the real-time charge state of the flywheel energy storage is in [0, S f,min ] interval, and Then set the filter time constant equal to zero; prohibit FES discharge, and let LiB bear all the

[0107] 6) If the real-time charge state of the flywheel energy storage is [S f,max , 1] interval (FES is in the charging forbidden area), and According to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; FES is responsible for more High frequency components.

[0108] 7) If the real-time charge state of the flywheel energy storage is [S f,max , 1], and Then set the filter time constant equal to zero. FES charging is prohibited, and LiB bears all the

[0109] The above steps 105 and 106 constitute the real-time control strategy of the lower-level HESS. The time scale of the lower-level control strategy is further refined below. For example, each cycle of the upper layer is subdivided into 5 control cycles. Each control cycle of the lower layer is carried out simultaneously. The control strategy of each control cycle is:

[0110] Step 4-1 Input of the lower control model. According to the above steps, the control vector ΔU(k) ​​in the prediction time domain at prediction time k is obtained. Next, according to the state space equation of the MPC model, the first control vector Δu(k+1|k) in ΔU(k) ​​is extracted, and combined with the prediction model at prediction time k, the state vector x(k+1|k) of the next rolling control cycle is calculated. The element P in this state vector x(k+1|k) is h (k+1|k), which is taken as the HESS planned power target instruction P of the lower control model h,ref (k), which is then sent to the lower-level control system. When a new control cycle arrives, the entire process is repeated to continuously optimize and control. Through such cyclic iterations, effective regulation of the entire system is achieved.

[0111] Step 4-2 Dynamic filter coefficient method design. The time scale of the lower-level control strategy is further refined, and each cycle of the upper layer is subdivided into 5 control cycles. Equation (24) is the input of the lower-level model, which is the power command of the HESS in the rolling control cycle obtained by the upper-level optimization. The real-time fluctuation of HESS power demand is also included.

[0112]

[0113] Where i′ represents the lower-layer scheduling time, which is different from the upper-layer scheduling time k.

[0114] The HESS power is filtered through a first-order digital low-pass filter. The high frequency component is allocated to FES, and the low frequency component is allocated to LiB. According to the working principle of the first-order digital low-pass filter, the power required by FES and LiB is as shown in the following formula.

[0115]

[0116] In the formula, τ is the time constant of the low-pass filter, and T is the sampling period. This application proposes a strategy to dynamically adjust the filter time constant according to the change of SOC, thereby changing the power distribution of FES and LiB. Assume that the initial value of the filter time constant is τ0, the increment is Δτ, τ d is the maximum variation range of Δτ.

[0117] The hybrid energy storage control strategy may refer to the specific implementation of step 601 above.

[0118] Limiting control is adopted to ensure that the final power command of FES and LiB meets the power limit constraint updated in real time by SOC. The following formula is the final power command received by FES and LiB.

[0119]

[0120] Among them, Δt is the lower layer control period, P b / f,min , P b / f,max are the lower and upper limits of the physical power of energy storage respectively; P b / f,min (i′), P b / f,max (i′) are the lower and upper limits of the energy storage power at time i′, respectively, which are determined by the real-time SOC.

[0121] When a new control cycle comes, the above process is repeated to continuously optimize and control. Through such cyclic iteration, effective regulation of the entire system is achieved.

[0122] In another exemplary embodiment of the present application, in order to ensure the accuracy of power distribution, at the end of the 5 control cycles of the lower layer, the upper model will perform feedback correction processing on the system state vector x(k+1|k) at the time k+1 predicted by the k prediction time. After the above step 106, the method may also include: after coordinating the real-time power distribution of the lithium battery and the flywheel energy storage at all times in the prediction time domain, according to the formula Correct the total power of the hybrid energy storage system at the next prediction moment; where p h,fdk (k+m+1|k) is the total power of the hybrid energy storage system corrected at the k+m+1 prediction time, P f (k+n) is the power command of the flywheel energy storage obtained at time k+n in the prediction time domain at prediction time k, P b (k+n) is the power instruction of the lithium battery obtained at time k+n in the prediction time domain of prediction time k, and m is the length of the prediction time domain.

[0123] The brief process of this application method is as follows Figure 2 As shown. This application aims at the energy storage system when the hybrid energy storage system assists the automatic power generation control of the thermal power unit to improve the AGC performance, and proposes a thermal power-hybrid energy storage dual-layer collaborative AGC regulation control strategy based on model predictive control. The upper layer designs a model predictive control rolling optimization model, combined with the ultra-short-term prediction results of the hybrid energy storage power demand obtained by the long short-term memory network with the attention mechanism, to achieve comprehensive consideration of tracking deviations, state of charge and other targets in the entire regulation process, while providing the lower layer with a reference trajectory of the actual power generation of the hybrid energy storage; in view of the different characteristics of the hybrid energy storage, the lower-layer control adopts a dynamic adjustment filter coefficient method based on the state of charge to coordinate the real-time power distribution of the lithium battery and the flywheel.

[0124] In order to make the decision forward-looking, reasonably decide whether to reserve margin for HESS, optimize the utilization of HESS capacity, enable thermal power plants to improve their AGC performance, reduce HESS costs, and improve AGC compensation, a thermal power-hybrid energy storage double-layer collaborative AGC regulation control strategy considering power demand estimation was established. This strategy has little impact on LiB, and the algorithm takes a short time. In terms of economy, after deducting the HESS cost, the net benefit of this strategy is very high.

[0125] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The present application uses specific examples to illustrate the principles and implementation methods of the present application. The above examples are only used to help understand the method and core ideas of the present application. At the same time, for those skilled in the art, according to the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A method for controlling and regulating double-layer thermal storage coordinated automatic power generation. It is characterized in that include: Obtain historical power data of thermal power plant units; Inputting the historical power data of the thermal power plant units into the power demand prediction model of the hybrid energy storage system to obtain the energy storage power demand prediction result at the prediction time; the power demand prediction model is a long short-term memory network with an attention mechanism added; the hybrid energy storage system includes lithium batteries and flywheel energy storage; According to the energy storage power demand forecast results at the forecast time, the model predictive control model is used to output the total power of the hybrid energy storage system in the forecast time domain; Based on the total power of the hybrid energy storage system in the forecast time domain and the energy storage power demand forecast results at the forecast time, the time domain rolling optimization method is used to optimize the total power increment of the hybrid energy storage system in the forecast time domain; According to the total power increment of the hybrid energy storage system in the forecast time domain, determining the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain; According to the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain, the dynamic adjustment filter coefficient method based on the state of charge is adopted to coordinate the real-time power of the lithium battery and flywheel energy storage at each moment in the forecast time domain.

2. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that Obtain historical power data of thermal power plant units, including: One month is selected in each quarter of the year to collect historical operation data of the thermal power plant units from the distributed control system of the generator sets; the historical operation data of the units include automatic power generation control instructions and the actual power generation of the thermal power units; The mean square method is used to eliminate abnormal data in the historical operation data of the unit, and the linear fitting method is used to replace the abnormal data; Normalizing the unit operation history data that replaces the abnormal data to obtain normalized unit operation history data; Calculate respectively the difference of automatic power generation control instructions at adjacent historical moments in the normalized unit operation history data, and the difference of actual power generation of thermal power units at adjacent historical moments; According to the normalized unit operation history data, the automatic generation control command and the actual power of the thermal power unit at the same historical moment, according to the formula P d (t) = P AGC (t)-P g (t), calculate the historical power demand of the hybrid energy storage system at each historical moment; where P d (t) is the historical power demand of the hybrid energy storage system at historical time t, P AGC (t) is the automatic power generation control instruction at historical time t, P g (t) is the actual power generated by the thermal power unit at historical moment t; The normalized historical operation data of the units, the difference of automatic power generation control instructions at adjacent historical moments, the difference of actual power generation of the thermal power units at adjacent historical moments and the historical power demand of the hybrid energy storage system at each historical moment constitute the historical power data of the thermal power plant units.

3. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that The establishment of the power demand prediction model specifically includes: Determine the input variable of the power demand prediction model as x(t)=[P g (t),P AGC (t),ΔP g (t),ΔP AGC (t),P d (t)]; where x(t) is the input variable of the power demand prediction model at historical time t, P g (t) is the actual power generated by the thermal power unit at the historical moment t, P AGC (t) is the automatic power generation control command at historical time t, ΔP g (t) is the difference in actual power generation of the thermal power unit at the historical time t and the historical time t-1, ΔP AGC (t) is the command difference between the automatic power generation control at the historical time t and the historical time t-1, P d (t) is the historical power demand of the hybrid energy storage system at historical time t; An attention mechanism is added between the hidden layer and the fully connected layer of the long short-term memory network to form a power demand prediction model for the hybrid energy storage system.

4. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that The establishment of the model predictive control model specifically includes: The power balance equation for the automatic power generation control system is established as P dif (k) = P d (k)-P h (k) = P d (k)-(P b (k)+P f (k)); where P dif (k) is the power difference after compensation at prediction time k, P d (k) is the energy storage power demand forecast result at the k forecast time, P h (k) is the actual power of the hybrid energy storage system at prediction time k, P b is the power of lithium battery, P f The power of energy stored in the flywheel; The charge state calculation equation of the hybrid energy storage system is established as follows: In the formula, S h (k) is the charge state of the hybrid energy storage system at prediction time k, S b (k) and S f (k) are the charge states of the lithium battery and flywheel energy storage at prediction time k, E b and E f are the capacities of lithium battery and flywheel energy storage respectively; Based on the power balance equation and the state of charge calculation equation, the state space equation of the model predictive control model is constructed as follows: Where x(k+1) and x(k) are the input vectors at the k+1 prediction time and the k prediction time, respectively. dif (k) P h (k) S h (k)] T ; Δu(k) is the control vector, Δu(k)=[ΔP h (k)], ΔP h (k) is the total power increment of the hybrid energy storage system at the k prediction time; Δv(k) is the disturbance vector, Δv(k) = [ΔP d (k)], ΔP d (k) is the predicted increment of energy storage power demand at prediction time k; y(k) is the output vector, y(k) = [P dif (k) S h (k)] T ; A, B, C, D are all coefficient matrices, D=[1 0 0] T , E h is the capacity of the hybrid energy storage system, and ΔT is the sampling time.

5. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that The objective function of the time domain rolling optimization method is: Δu i (k)=[Δu(k+1|k),Δu(k+2|k),...,Δu(k+m|k)] T ; In the formula, y i is the output vector of the i-th upper layer rolling optimization cycle, y ref (k) is the target tracking vector in the prediction time domain at the k prediction time, Δu i (k) is the control vector of the i-th upper layer rolling optimization cycle, P dif,plan (k+w|k) is the target deviation at each moment in the prediction time domain at the k prediction moment, S h,ref (k+w|k) is the target state of charge value of the hybrid energy storage system in the prediction time domain at the k prediction time, Δu(k+w|k) is the control vector at each moment in the prediction time domain at the k prediction time, w=1,2,…,m, m is the length of the prediction time domain, and the time length of the upper rolling optimization cycle is equal to the time length of the prediction time domain; Q and R are both weight coefficient matrices.

6. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 5, It is characterized in that The constraints of the time domain rolling optimization method include: state of charge constraints, power constraints and energy storage charging and discharging logic constraints; The state of charge constraint is: S h,min ≤S h (k+n|k)≤S h,max Where S h (k+n|k) is the charge state of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k; S h,min and S h,max are the minimum state of charge and maximum state of charge of the hybrid energy storage system respectively; The power constraint is: P h,min (k)≤P h (k+n|k)≤P h,max (k); P h,min (k)-P h (k+n|k)≤ΔP h (k+n|k)≤P h,max (k)-P h (k+n|k); Where P h (k+n|k) is the total power of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k, P h,min (k), P h,max (k) are the minimum and maximum total power of the hybrid energy storage system in the prediction time domain at the k prediction time determined according to the real-time charge state of the lithium battery and the flywheel energy storage; ΔP h (k+n|k) is the total power increment of the hybrid energy storage system at time k+n in the prediction time domain at prediction time k; The energy storage charging and discharging logic constraints are: Where, ε is a constant, ε<0; P d (k+n|k) is the energy storage power demand forecast result at time k+n in the forecast time domain at forecast time k, P dif (k+n|k) is the power difference after compensation at time k+n in the prediction time domain at prediction time k.

7. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that According to the total power increment of the hybrid energy storage system in the forecast time domain, the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain is determined, specifically including: Extract the control vector at each moment in the total power increment of the hybrid energy storage system in the prediction time domain; According to the control vector at each moment, the state space equation of the model predictive control model is used Calculate the state vector at each moment in the next rolling optimization cycle; where x(k+1) and x(k) are the input vectors at the k+1 prediction moment and the k prediction moment, respectively, and x(k) = [P dif (k) P h (k) S h (k)] T ; Δu(k) is the control vector, Δu(k)=[ΔP h (k)], ΔP h (k) is the total power increment of the hybrid energy storage system at the k prediction time; Δv(k) is the disturbance vector, Δv(k) = [ΔP d (k)], ΔP d (k) is the predicted increment of energy storage power demand at prediction time k; y(k) is the output vector, y(k) = [P dif (k) S h (k)] T ; A, B, C, D are all coefficient matrices, D=[1 0 0] T , E h is the capacity of the hybrid energy storage system, ΔT is the sampling time; The total power of the hybrid energy storage system in the state vector at each moment in the next rolling optimization cycle is used as the planned power target instruction of the hybrid energy storage system at each moment in the measurement time domain.

8. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that According to the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain, the real-time power of the lithium battery and flywheel energy storage at each moment in the forecast time domain is coordinated and allocated by using the dynamic adjustment filter coefficient method based on the state of charge, including: According to the real-time charge state of the flywheel energy storage, a dynamic adjustment filter coefficient method based on the charge state is adopted to determine the real-time filter time constant; Using limiting control, the real-time power constraints of lithium batteries and flywheel energy storage are determined as Where Δt is the lower layer control period, P z,min , P z,max are the lower and upper limits of the physical power of energy storage respectively; P z,min (k+n),P z,max (k+n) are the lower and upper limits of the power of the energy storage at time k+n in the prediction time domain at prediction time k, respectively. z,min , S z,max are the lower and upper limits of the state of charge of the energy storage, S z (k+n-1) is the state of charge of the energy storage at time k+n-1 in the prediction time domain at prediction time k, E z is the capacity of energy storage, z = b or f; According to the planned power target instruction and real-time filtering time constant of the hybrid energy storage system at each moment in the forecast time domain, combined with the real-time power constraints of lithium batteries and flywheel energy storage, according to the formula Obtain the real-time power command of the lithium battery and flywheel energy storage at each moment in the prediction time domain; where τ is the real-time filter time constant, T is the sampling period, P b_ref (k+n) is the power instruction of the lithium battery at time k+n in the prediction time domain at prediction time k, P b_ref (k+n-1) is the power instruction of the lithium battery at time k+n-1 in the prediction time domain at prediction time k, P f_ref (k+n) is the power command of the flywheel energy storage at time k+n in the prediction time domain at prediction time k, is the planned power target instruction of the hybrid energy storage system at time k+n in the prediction domain at prediction time k.

9. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 8, It is characterized in that According to the real-time charge state of the flywheel energy storage, the real-time filter time constant is determined by using the dynamic adjustment filter coefficient method based on the charge state, which specifically includes: If the real-time charge state of the flywheel energy storage is [S f,low , S f,high ] interval, let the filter time constant be equal to the initial value of the filter time constant; where S f,low is the lower limit threshold of the charge state of the flywheel energy storage, S f,high is the upper limit threshold of the state of charge of the flywheel energy storage; If the real-time charge state of the flywheel energy storage is [S f,min , S f,low ], then according to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; where S f,min is the minimum state of charge of the flywheel energy storage, S f is the real-time charge state of the flywheel energy storage, τ d is the maximum variation range of the filter time constant increment Δτ; If the real-time charge state of the flywheel energy storage is [S f,high , S f,max ], then according to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; where S f,max is the maximum state of charge of the flywheel energy storage; If the real-time charge state of the flywheel energy storage is in [0, S f,min ] interval, and According to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; If the real-time charge state of the flywheel energy storage is in [0, S f,min ] interval, and Then let the filter time constant equal to zero; If the real-time charge state of the flywheel energy storage is [S f,max , 1], and According to the formula Determine the filter time constant increment Δτ, and combine the filter time constant increment Δτ and the filter time constant initial value to obtain the filter time constant; If the real-time charge state of the flywheel energy storage is [S f,max , 1], and Then set the filter time constant equal to zero.

10. The thermal storage double-layer coordinated automatic power generation control and regulation method according to claim 1, It is characterized in that According to the planned power target instruction of the hybrid energy storage system at each moment in the forecast time domain, the dynamic adjustment filter coefficient method based on the state of charge is adopted to coordinate the real-time power of the lithium battery and flywheel energy storage at each moment in the forecast time domain, and then it also includes: After coordinating and allocating the real-time power of lithium batteries and flywheel energy storage at all times in the prediction time domain, according to the formula Correct the total power of the hybrid energy storage system at the next prediction moment; where p h,fdk (k+m+1|k) is the total power of the hybrid energy storage system corrected at the k+m+1 prediction time, P f (k+n) is the power command of the flywheel energy storage obtained at time k+n in the prediction time domain at prediction time k, P b (k+n) is the power instruction of the lithium battery obtained at time k+n in the prediction time domain of prediction time k, and m is the length of the prediction time domain.

Citation Information

Cited By

  • Intelligent control method for flexibility and deep peak regulation of thermal power generating unit

    CN121192839A

  • Marine crane group dynamic power prediction and cooperative control system and method

    CN121918397A