Secondary frequency modulation multi-scale predictive control method for auxiliary thermal power generating unit of energy storage system

Abstract

The invention provides an energy storage system auxiliary thermal power generating unit secondary frequency modulation multi-scale prediction control method, which constructs a thermal power storage combined frequency modulation system, and comprises the following steps of: firstly, decomposing an AGC (Automatic Gain Control) instruction into multiple layers of instructions by utilizing a wavelet decomposition method, one is a main instruction with less nonlinearity and is responded by the output of a thermal power generating unit, and the other is a multi-layer nonlinear auxiliary instruction and is responded by the output of the thermal power generating unit; and the battery energy storage system outputs a response. And secondly, in order to improve AGC instruction response efficiency, an instruction prediction model is constructed by using a tree regression model, and a basis of a prediction control strategy is provided for a fire storage combined frequency modulation system. And finally, difference calculation is carried out on the superposition value of AGC response output of the thermal power generating unit and the energy storage system and the original AGC instruction, an error model is deduced, and an output value of the model serves as a reserved energy storage system capacity response instruction. The problems that when a traditional thermal power generating unit responds to an AGC instruction of a power grid dispatching center, response delay is high, the climbing speed is low, and the requirement for nonlinear frequency change of a power system caused by grid connection of a new energy power station at present cannot be met are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal energy storage combined with power grid secondary frequency regulation, and specifically relates to a multi-scale predictive control method for secondary frequency regulation of a thermal power unit assisted by an energy storage system. Background Art

[0002] As the proportion of renewable energy generation in the total grid-connected power generation continues to increase, the nonlinear grid-connected form it presents has greatly increased the difficulty of secondary frequency regulation of the power system. However, traditional thermal power units lack the ability to cope with this phenomenon. Figure 1 As shown in the figure, the Automatic Generation Control (AGC) command of the interconnected power grid in the Energy Management System (EMS) of the power system and the response curve of the thermal power unit not only have a huge error, but even reverse regulation occurs at some times. In order to improve the overall performance of thermal power units and respond quickly to frequency regulation commands, a large number of power plants have introduced battery energy storage systems (BESS). The introduction of energy storage systems composed of lithium-ion batteries can effectively solve the problems of high response delay and slow ramp rate of AGC commands of thermal power units. How to build a joint system of thermal storage is the key to realizing the technology of joint participation of thermal storage in the secondary frequency regulation of the power grid. Since the AGC command has a large number of sudden changes with high change rates, the AGC command response capability of the joint system of thermal storage cannot be effectively exerted without a fast and efficient control strategy. Summary of the invention

[0003] In response to the above problems, the present invention provides a multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by an energy storage system. The purpose is to solve the problem that traditional thermal power units have high response delay and slow climbing rate when responding to automatic generation control (AGC) instructions from the power grid dispatching center, and cannot meet the nonlinear frequency change requirements of the power system brought about by the connection of current new energy power stations to the grid.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A multi-scale predictive control method for secondary frequency regulation of a thermal power unit assisted by an energy storage system comprises the following steps:

[0006] Step 1: Decompose the AGC instructions issued by the power station control center into main instructions in the low frequency band and high frequency instructions in the high frequency band by using the basic Haar wavelet decomposition method;

[0007] Step 2: The AGC main command and high-frequency command are respectively predicted in advance using the prediction control system, wherein the prediction system uses the precision tree regression model as the prediction model; the main command is obtained by the precision tree regression model to obtain the main prediction command, and the high-frequency command is obtained by the precision tree regression model to obtain the high-frequency prediction command; the main prediction command is used as the response command of the thermal power unit, and the high-frequency prediction command is used as the response command of the energy storage system;

[0008] Step 3: Superimpose the AGC response output values ​​of the thermal power unit and the energy storage system, and the superimposed value is the AGC predictive control system response; perform difference calculation on the AGC predictive control system response value and the original AGC instruction, so as to obtain the AGC predictive control system response error; use polynomial fitting of the AGC predictive control system response error to obtain an error model, and use the output value of the model as the reserved energy storage system capacity response instruction; the reserved energy storage system capacity response instruction is a further guarantee for the predictive control instruction. When the error model input value is less than the predetermined threshold, the error model reserved energy storage system capacity response instruction output is not performed;

[0009] Step 4: After the above three steps, the thermal energy storage combined AGC command response is obtained, and the multi-scale predictive control process of the secondary frequency regulation of the energy storage system assisting the thermal power unit is completed.

[0010] Furthermore, in step 1, the AGC instruction is recorded as Y AGC , the main instruction after basic Haar wavelet decomposition is denoted as D AGC , high-frequency instructions are denoted as G AGC ;

[0011] The basic basis function group expression of wavelet decomposition is:

[0012]

[0013] Among them, f(n) is the combination of basic functions of wavelet decomposition, is the general form of the wavelet mother function, n is the number of decomposition levels, a n is the scale coefficient, b is the translation parameter, and j determines the frequency characteristics of the wavelet basis function. AGC D AGC and G AGC .

[0014] In step 2, the AGC main instruction and high-frequency instruction are predicted 15 seconds ahead using the predictive control system respectively; the basic steps of constructing the precision tree regression model are as follows:

[0015] a) Divide D AGC and G AGC The training data T(N,N) and verification data C(N,N) of AGC , D AGC and GAGC The data features are divided into two categories with clear features: DD1 and DD2. The selection of the partition threshold adopts a cross-validation strategy to verify the minimum Gini coefficient as the selected threshold;

[0016] b) Calculate the sum of square errors of DD1 and DD2 respectively, select the feature corresponding to the smallest square error, generate two child nodes: DD3 and DD4, and select the smallest Gini coefficient obtained by the cross-validation strategy as the selected threshold;

[0017] c) Repeat the above steps for DD3 and DD4, and the sub-node division cut-off condition is to generate sub-areas without obvious data features;

[0018] d) Add regularization measure in the process of sub-node division:

[0019]

[0020] where y d , is the true value of the node, Y d is the predicted value, d is the number of nodes, μ is the regularization coefficient, and ε is the feature division sub-area;

[0021] e) Substitute the validation data C(N,N) into the model. If the validation result accuracy is 98.5% or above, the precision tree model construction is terminated. Otherwise, repeat the above steps to build an accurate regression model.

[0022] The error model construction method in step three is as follows: the original AGC command given by the power generation control center to the unit is calculated with the AGC predictive control system response, and the error model is obtained by using the polynomial fitting method; if the difference between the two is less than 0.001MW, the predictive model command output is not performed, and it is directly used as the thermal storage combined AGC command response; the error model gives an output error response command to the supplementary energy storage system, and the AGC response output by the supplementary energy storage system is incorporated into the high-frequency response and is not output separately; the error model only outputs once at the same time and does not perform a cyclic operation.

[0023] The supplementary energy storage system is the capacity part reserved in the energy storage system for the response operation of the error model. This part is not shared with the energy storage part for high-frequency AGC response. At the same time, when multiple groups of energy storage systems assist thermal power units in secondary frequency modulation and multi-scale predictive control for joint response, only the capacity of the high-frequency AGC response part can be supplemented between different energy storage systems, while the supplementary energy storage systems are independent of each other.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention constructs a thermal energy storage combined frequency regulation system, and uses the wavelet decomposition method to decompose the AGC instruction into multiple layers of instructions, one of which is a main instruction with less nonlinearity, which is responded by the output of the thermal power unit, and the other is a multiple layer of nonlinear sub-instructions, which is responded by the output of the battery energy storage system. In order to improve the efficiency of AGC instruction response, a tree regression model is used to construct an instruction prediction model to provide a basis for the prediction control strategy for the thermal energy storage combined frequency regulation system. The present invention solves the problem that when traditional thermal power units respond to the automatic generation control (AGC) instructions of the power grid dispatching center, the response delay is high and the climbing rate is slow, which cannot meet the demand for nonlinear frequency changes of the power system brought about by the connection of new energy power stations to the grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a comparison chart of the AGC command and the response of the traditional thermal power unit in a power plant;

[0026] Figure 2 This is a schematic diagram of the secondary frequency regulation system of the energy storage system assisting the thermal power unit;

[0027] Figure 3 A multi-scale predictive control flow chart for secondary frequency regulation of thermal power units assisted by energy storage systems;

[0028] Figure 4 It is the wavelet decomposition result of AGC instructions (main instructions and high-frequency instructions);

[0029] Figure 5 It is the AGC main command (real) and the thermal power unit response command (prediction);

[0030] Figure 6 It is the error between the AGC main command (real) and the thermal power unit response command (prediction);

[0031] Figure 7 It is the AGC high-frequency command (real) and the energy storage system response command (prediction);

[0032] Figure 8 It is the error between the AGC high-frequency command (real) and the energy storage system response command (prediction);

[0033] Fig. 9 The error compensation model responds to commands (real) and outputs (predicted);

[0034] Fig.10 The error compensation model responds to the error between the command (true) and the output (prediction);

[0035] Fig.11 This is the frequency modulation performance test result. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is clearly and completely described below in conjunction with the accompanying drawings and specific embodiments.

[0037] Figure 1 The AGC instructions for a certain time period of a large coal-fired power plant in central and western my country given by the remote control center. This implementation method uses this time period data as the input data and frequency regulation performance evaluation standard of the multi-scale predictive control method for secondary frequency regulation of the energy storage system-assisted thermal power unit. The low-order model of the reheat steam turbine and generator is used in the simulation process of the present invention. This model is a mature application model and is not described in detail.

[0038] Figure 2 The schematic diagram of the hardware system composition of the multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by energy storage systems. The present invention uses three groups of thermal power units and three groups of energy storage systems as control objects. The AGC instructions come from the dispatching center and are sent to the coordination control center through the remote terminal unit (municipal power system). The coordination control center sends the AGC instruction response capability that can be achieved in this area to the power system. At the same time, it evaluates the capabilities of each power station in the area and gives the corresponding AGC instruction indicators to each power station. This implementation method uses three thermal power unit models and three groups of energy storage systems (the theoretical capacity value can achieve 1.2 times the capacity required for maximum frequency regulation).

[0039] Figure 3 This is a flow chart of multi-scale predictive control of secondary frequency regulation of thermal power units assisted by energy storage systems. The multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by energy storage systems described in this embodiment is used in the same way in each thermal power unit and energy storage system. Taking a group of thermal power units and a group of energy storage systems as an example, the method described in this embodiment includes the following steps:

[0040] Step 1: Decompose the AGC instructions issued by the power station control center into main instructions in the low frequency band and high frequency instructions in the high frequency band through the basic Haar wavelet decomposition method.

[0041] Step 2: The AGC main command and high-frequency command are predicted 15 seconds ahead using the prediction control system, where the prediction system uses the precision tree regression model as the prediction model. The main command is converted into the main prediction command through the precision tree regression model, and the high-frequency command is converted into the high-frequency prediction command through the precision tree regression model. The main prediction command is used as the response command of the thermal power unit, and the high-frequency prediction command is used as the response command of the energy storage system.

[0042] Step 3: Superimpose the AGC response output values ​​of the thermal power unit and the energy storage system. The superimposed value is the AGC predictive control system response. The difference between the AGC predictive control system response value and the original AGC instruction is calculated to obtain the AGC predictive control system response error. The error model is obtained by fitting the AGC predictive control system response error with a polynomial, and the output value of the model is used as the reserved energy storage system capacity response instruction. The reserved energy storage system capacity response instruction is a further guarantee for the predictive control instruction. When the error model input value is less than the predetermined threshold, the error model reserved energy storage system capacity response instruction output is not performed.

[0043] It should be noted that the error model here does not perform a threshold judgment cycle, that is, the error model outputs an output error response value at most once.

[0044] Step 4: After the above three steps, the thermal energy storage combined AGC command response is obtained, and the multi-scale predictive control process of the secondary frequency regulation of the energy storage system assisting the thermal power unit is completed.

[0045] Specific implementation method 2: This implementation method further illustrates the multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by energy storage systems described in specific implementation method 1. The specific implementation steps for constructing the predictive control system and the thermal storage joint response system described in step 1 are as follows:

[0046] Let AGC instruction be Y AGC , after hear wavelet decomposition, the low frequency band instruction (main instruction) is recorded as D AGC , high frequency instruction (high frequency instruction) is denoted as G AGC .

[0047] Step 1: Determine the basic basis function group expression of hear wavelet decomposition:

[0048]

[0049] Among them, f(n) is the combination of basic functions of wavelet decomposition, is the general form of the wavelet mother function, n is the number of decomposition levels, a n is the scale coefficient, b is the translation parameter, and j determines the frequency characteristics of the wavelet basis function. AGC D AGC and G AGC .

[0050] Step 2: Build a predictive control system (precision tree model):

[0051] The decision strategy selected in this implementation is to construct a decision tree based on the Gini coefficient for regression prediction. The basic steps for constructing a precision tree regression model are as follows:

[0052] a) Divide the training data T(N,N) and verification data C(N,N) of DAGC and GAGC, considering Y AGC , D AGC and G AGC The data features are divided into two categories of data DD1 and DD2 with clearer distinguishing characteristics, and the minimum Gini coefficient obtained by the cross-validation strategy is used as the threshold. In this implementation, the division threshold is 3.8, that is, less than 3.8 is DD1, and greater than 3.8 is DD2;

[0053] b) Calculate the sum of square errors of DD1 and DD2 respectively, select the feature corresponding to the smallest square error, generate two child nodes: DD3 and DD4, and use the smallest Gini coefficient obtained by cross-validation strategy as the threshold. The threshold for this step is 0.5;

[0054] c) Repeat the above steps for DD3 and DD4, and the sub-node division is terminated when a sub-area without obvious data features is generated. In this embodiment, in order to avoid overfitting, the maximum number of nodes is set to 12, that is, the division is stopped when more than 12 layers are exceeded.

[0055] d) Add regularization measure in the process of sub-node division:

[0056]

[0057] where y d , is the true value of the node, Y d is the predicted value, d is the number of nodes, μ is the regularization coefficient, and ε is the feature division sub-area;

[0058] e) Substitute the validation data C(N,N) into the model. If the validation result accuracy is 98.5% or above, the precision tree model construction is terminated. Otherwise, repeat the above steps to build a more accurate regression model.

[0059] Step 3: AGC predictive control system response operation process

[0060] The main command and high-frequency command are input into the prediction control system, that is, the constructed precision tree model, to obtain the main prediction command and high-frequency prediction command 15s ahead. The main prediction command is sent to the thermal power unit for command response, and the high-frequency prediction command is sent to the energy storage system for command response. The main response and high-frequency response are superimposed in the same time period to obtain the AGC prediction control system response.

[0061] Step 4: Error model construction

[0062] The difference between the original AGC command given by the power generation control center to the unit and the AGC predictive control system response is calculated, and the error model is obtained using the polynomial fitting method. If the difference between the two is less than 0.001MW, the predictive model command output is not performed, and it is directly used as the thermal storage combined AGC command response. The error model gives an output error response command to the energy storage system (supplementation), and the AGC response output by the supplementary energy storage system is incorporated into the high-frequency response and is not output separately. The error model only outputs once at the same time and does not perform a cyclic operation.

[0063] Step 5. After the above steps are completed, the fire-storage joint AGC command response value is obtained, and the construction of the prediction control system and the fire-storage joint response system is completed.

[0064] Example 1

[0065] Taking the AGC command of a power plant in a certain time period as input data, the performance of the multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by the energy storage system proposed in the present invention is verified.

[0066] The following preparations need to be made before performing multi-scale predictive control of secondary frequency regulation of thermal power units assisted by energy storage systems:

[0067] Step 1: 10 4 The time period length of s AGC instruction is the training data set, such as Figure 4 As shown in the AGC instructions.

[0068] Step 2: After the hear wavelet decomposition, the low frequency instruction (main instruction) a12 is recorded as D AGC , high frequency instruction b12 (high frequency instruction) is recorded as G AGC ,like Figure 4 As shown in a12 and b12.

[0069] Step 3: D AGC The main instruction precision tree model is trained by the predictive control system. AGC The high-frequency instruction precision tree model is trained by the predictive control system.

[0070] Step 4: Send the main prediction command to the thermal power unit for command response, and send the high-frequency prediction command to the energy storage system for command response. The main response and high-frequency response are superimposed in the same time period to obtain the AGC prediction control system response. The difference between the original AGC command given by the power generation control center to the unit and the AGC prediction control system response is calculated, and the error model is obtained using the polynomial fitting method.

[0071] Verification of the multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by energy storage systems proposed in this invention:

[0072] The AGC instructions of a thermal power plant for a period of 15000s are input into the prediction control system constructed by the present invention (hear wavelet decomposition is performed before input into the precision tree model). AGC Send it to the main instruction precision tree model and convert G AGC The high-frequency instruction precision tree model is sent to the main instruction precision tree model. The main instruction precision tree model outputs the main prediction instruction 15s ahead to the thermal power unit. The model prediction value and the actual value are as follows Figure 5 As shown in Figure 6 The high-frequency instruction precision tree model outputs 15s ahead high-frequency prediction instructions to the energy storage system. The model prediction value and the actual value are shown in Figure 7 As shown in Figure 8 After obtaining the AGC predictive control system response, it is input into the error model to obtain the output error response instruction, as shown in Fig. 9 The output error of the error model is shown as Fig.10 The frequency regulation performance of the multi-scale predictive control method for secondary frequency regulation of thermal power units assisted by energy storage systems proposed in the present invention is shown in Fig.11 As shown, it can be determined that the control method proposed in the present invention has excellent performance.

Claims

1. A multi-scale predictive control method for secondary frequency regulation of a thermal power unit assisted by an energy storage system, characterized in that: The following steps are involved: Step 1: Decompose the AGC instructions issued by the power station control center into main instructions in the low frequency band and high frequency instructions in the high frequency band by using the basic Haar wavelet decomposition method; Step 2: The AGC main command and high-frequency command are respectively predicted in advance using the prediction control system, wherein the prediction system uses the precision tree regression model as the prediction model; the main command is obtained by the precision tree regression model to obtain the main prediction command, and the high-frequency command is obtained by the precision tree regression model to obtain the high-frequency prediction command; the main prediction command is used as the response command of the thermal power unit, and the high-frequency prediction command is used as the response command of the energy storage system; Step 3: Superimpose the AGC response output values ​​of the thermal power unit and the energy storage system, and the superimposed value is the AGC predictive control system response; perform difference calculation on the AGC predictive control system response value and the original AGC instruction, so as to obtain the AGC predictive control system response error; use polynomial fitting of the AGC predictive control system response error to obtain an error model, and use the output value of the model as the reserved energy storage system capacity response instruction; the reserved energy storage system capacity response instruction is a further guarantee for the predictive control instruction. When the error model input value is less than the predetermined threshold, the error model reserved energy storage system capacity response instruction output is not performed; Step 4: After the above three steps, the thermal energy storage combined AGC command response is obtained, and the multi-scale predictive control process of the secondary frequency regulation of the energy storage system assisting the thermal power unit is completed.

2. A multi-scale predictive control method for secondary frequency modulation of a thermal power unit assisted by an energy storage system according to claim 1, characterized in that: In step 1, the AGC instruction is recorded as Y AGC , the main instruction after basic Haar wavelet decomposition is denoted as D AGC , high-frequency instructions are denoted as G AGC ; The expression of the basic basis function group of wavelet decomposition is: Among them, f(n) is the combination of basic functions of wavelet decomposition, is the general form of the wavelet mother function, n is the number of decomposition levels, a n is the scale coefficient, b is the translation parameter, and j determines the frequency characteristics of the wavelet basis function. AGC D AGC and G AGC .

3. A multi-scale predictive control method for secondary frequency modulation of a thermal power unit assisted by an energy storage system according to claim 2, characterized in that: In step 2, the AGC main instruction and high-frequency instruction are predicted 15 seconds ahead using the predictive control system respectively; the basic steps of constructing the precision tree regression model are as follows: a) Divide D AGC and G AGC The training data T(N,N) and verification data C(N,N) of AGC , D AGC and G AGC The data features are divided into two categories with clear features: DD1 and DD2. The selection of the partition threshold adopts a cross-validation strategy to verify the minimum Gini coefficient as the selected threshold; b) Calculate the sum of square errors of DD1 and DD2 respectively, select the feature corresponding to the smallest square error, generate two child nodes: DD3 and DD4, and select the smallest Gini coefficient obtained by the cross-validation strategy as the selected threshold; c) Repeat the above steps for DD3 and DD4, and the sub-node division cut-off condition is to generate sub-areas without obvious data features; d) Add regularization measure in the process of sub-node division: where y d , is the true value of the node, Y d is the predicted value, d is the number of nodes, μ is the regularization coefficient, and ε is the feature division sub-area; e) Substitute the validation data C(N,N) into the model. If the validation result accuracy is 98.5% or above, the precision tree model construction is terminated. Otherwise, repeat the above steps to build an accurate regression model.

4. A multi-scale predictive control method for secondary frequency modulation of a thermal power unit assisted by an energy storage system according to claim 2, characterized in that: The error model construction method in step three is as follows: the original AGC command given by the power generation control center to the unit is calculated with the AGC predictive control system response, and the error model is obtained by using the polynomial fitting method; if the difference between the two is less than 0.001MW, the predictive model command output is not performed, and it is directly used as the thermal storage combined AGC command response; the error model gives an output error response command to the supplementary energy storage system, and the AGC response output by the supplementary energy storage system is incorporated into the high-frequency response and is not output separately; the error model only outputs once at the same time and does not perform a cyclic operation.

5. A multi-scale predictive control method for secondary frequency modulation of a thermal power unit assisted by an energy storage system according to claim 4, characterized in that: The supplementary energy storage system is the capacity part reserved in the energy storage system for the response operation of the error model. This part is not shared with the energy storage part for high-frequency AGC response. At the same time, when multiple groups of energy storage systems assist thermal power units in secondary frequency modulation and multi-scale predictive control for joint response, only the capacity of the high-frequency AGC response part can be supplemented between different energy storage systems, while the supplementary energy storage systems are independent of each other.

Images