Method and system for improving stability of power system based on coordination of internal and external resources of network

Abstract

The invention relates to a method and a system for improving stability of a power system based on coordination of internal and external resources of a network, and the method comprises the steps: considering voltage-reactive power characteristics, building a mathematical model which considers an inversion side extinction angle, a direct current and a commutation voltage of an internal resource VDCOL and an external resource SVG, and theoretically analyzing a coupling constraint relation of the coordination of the VDCOL and the SVG; establishing a DDPG basic model for collaborative optimization of VDCOL and SVG parameters, and determining a state space and an action space; considering a commutation failure recovery process, proposing a critical commutation voltage index of a multi-infeed subsequent commutation failure when the VDCOL and the SVG cooperate, and setting a reward function in combination with active power transmission and reactive power compensation strength of the system; and establishing a DDPG collaborative optimization model, completing model training and testing, and then issuing the model to a safety control master station so as to obtain an optimization parameter combination through reinforcement learning when a fault occurs. According to the invention, the commutation voltage can be stabilized, the commutation margin is increased, the suppression of multi-infeed continuous commutation failure is realized, and the stability of a power system is improved.

Description

Technical Field

[0001] The present invention relates to the field of high voltage direct current power transmission, and in particular to a method and system for improving the stability of an electric power system based on coordination of resources inside and outside the network. Background Art

[0002] In multi-infeed HVDC transmission systems, common commutation failure faults are particularly serious. When the fault causes the commutation voltage to drop sharply, commutation failure occurs, which may lead to serious consequences such as current surge and DC lockout. Reactive imbalance during the recovery process of the first commutation failure is the main reason for the continuous deterioration of the system and subsequent continuous commutation failures and even DC lockout. Therefore, it is a common idea to suppress commutation failure from the reactive characteristics.

[0003] Among them, the optimization of low voltage current limiting (VDCOL) characteristics is one of the methods to suppress continuous commutation failure from the network level. At present, existing studies have achieved staggered recovery of DC by increasing VDCOL delay units, considering the strength of the receiving AC power grid and proposing the system reactive operation short-circuit ratio index, considering the parameter sensitivity and current limiting effect and other constraints to establish a multi-DC transmission system optimization model, and reducing the reactive demand of the system at the same time.

[0004] Installing reactive power compensation equipment to provide additional support is the main method to suppress commutation failure from the outside of the grid. Among them, the static VAR generator (SVG) has a fast response speed, small harmonic current, and small loss, and is a relatively good reactive power compensation device. At present, there have been studies that configure SVG at the single-feed receiving end bus, which verifies that SVG can effectively regulate the bus voltage and suppress commutation failure. The literature optimizes the capacity and distribution of multi-feed SVG for better compensation effect.

[0005] The above two methods improve the commutation characteristics of the system from the aspects of reducing and compensating reactive power, but there are limitations when the two methods work. First, while VDCOL reduces the DC reactive power demand, it will also limit the DC transmission power. In severe cases, it will aggravate the power imbalance problem between the sending and receiving ends, and VDCOL does not match the system transient recovery process; secondly, for SVG, its reactive power compensation is directly related to the economic cost, and currently only a small number of SVGs with small capacity are configured in the power grid, resulting in poor suppression effect on commutation failure. At the same time, some studies have pointed out that the coupling effect between the SVG voltage outer loop control link and the DC inverter station may cause DC small signal instability, etc., and the SVG effect is not good. Further, for multi-input systems, multi-DC coupling will make VDCOL optimization difficult, and the reactive power required by multi-DC is much higher than the capacity of small SVG, and the limitations of VDCOL and SVG's independent role are more prominent.

[0006] Since both improve reactive power, it is not difficult to think that if the two are optimized and coordinated, it is expected to more effectively suppress the failure of multi-feed continuous commutation. However, for VDCOL and SVG, which both work on reactive power, there is a coupling relationship between the two. How to adjust the parameters to achieve a complementary state of coordination between the two is challenging, and there are few studies in this area. Secondly, there is mutual influence between different DCs, and the system transient recovery process and time scale are more complicated, and there are difficulties in the coordination of multiple VDCOLs. Thirdly, SVG can support the adjacent DC through local DC, and how much compensation is appropriate is also a problem that needs to be dealt with. Summary of the invention

[0007] The present invention provides a method and system for improving the stability of an electric power system based on coordination of resources inside and outside the grid, so as to solve at least one of the above-mentioned technical problems.

[0008] The technical solution of the present invention to solve the above technical problems is as follows: A method for improving the stability of a power system based on coordination of resources inside and outside the grid, comprising:

[0009] Step 1: Considering the voltage-reactive power characteristics of the multi-input HVDC transmission system, a mathematical model between the arc extinction angle, DC current and commutation voltage on the inverter side considering the in-grid resource VDCOL and the out-grid resource SVG is established, and based on the mathematical model, a coupling constraint relationship between the VDCOL and SVG collaboration is theoretically analyzed;

[0010] Step 2: Based on the coupling constraint relationship, a DDPG basic model for collaborative optimization of VDCOL and SVG parameters is established, and a state space and an action space of the DDPG basic model are determined;

[0011] Step 3: Considering the commutation failure recovery process, a critical commutation voltage index for subsequent commutation failure when VDCOL and SVG are coordinated is proposed, and the reward function of the DDPG basic model is set in combination with the system active transmission and reactive power compensation strength;

[0012] Step four, based on the state space, the action space and the reward function, a DDPG collaborative optimization model is established, the collected samples are used to train and test the DDPG collaborative optimization model, and the tested DDPG collaborative optimization model is sent to the security control main station, so that when a fault occurs, the optimized parameter combination can be obtained through reinforcement learning, and the optimized parameter combination can be called for control to suppress commutation failure.

[0013] On the basis of the above technical solution, the present invention has the following further solution.

[0014] Further, in the step 1, the mathematical model includes a single-feed mathematical model; the multi-feed high-voltage direct current transmission system includes at least three direct currents, and the at least three direct currents include a first direct current, a second direct current, and a third direct current; assuming that the third direct current including the SVG fails, the single-feed mathematical model is specifically:

[0015] Ua 4 ac3 +bU 3 ac3 +cU 2 ac3 +dU ac3 +e=0;

[0016] Among them, U ac3 is the commutation voltage of the third DC loop, a, b, c, d and e are all model parameters and are specifically expressed as follows:

[0017]

[0018] Specifically, B3 is the reactive compensation susceptance of the inverter side of the third DC loop, X3 is the reactance of the AC system at the receiving end of the third DC loop, E3 is the AC system voltage at the receiving end of the third DC loop, Q SVG is the reactive value output by SVG, I d3 is the DC current of the third DC circuit, A, B and C are all model parameters, and A=1.35k3N p3 , B = Acosγ3, C = 0.955X T3 N p3 ; Where k3 is the ratio of the converter transformer of the third DC circuit, N p3 is the number of pole pairs of the converter transformer of the third DC loop, γ3 is the arc extinction angle of the inverter side of the third DC loop, X T3 is the leakage reactance of the inverter-side converter transformer of the third DC loop.

[0019] Furthermore, in step 1, the mathematical model further includes a multi-input mathematical model; the multi-input mathematical model includes:

[0020] U 2 ac1 =E1U ac1 +(f Qc1 (U ac1 )-f Qd1 (U ac1 ,I d1 )-Q 13 -Q 12 -Q f )X3;

[0021] U 2 ac2 =E2Uac2 +(f Qc2 (U ac2 )-f Qd2 (U ac2 ,I d2 )+Q 12 -Q 23 )X3;

[0022] U 2 ac3 =E3U ac3 +(f Qc3 (U ac3 )-f Qd3 (U ac3 ,I d3 )+Q SVG +Q 13 +Q 23 )X3;

[0023] Among them, let i = 1, 2, 3, then: U aci is the commutation voltage of the ith DC circuit, E i is the AC system voltage at the receiving end of the i-th DC circuit, f Qci (U aci ) is the commutation voltage U of the i-th DC aci The reactive power compensation power Q of the inverter side of the i-th DC is represented by ci , f Qdi (U aci ,I di ) is the commutation voltage U of the i-th DC aci and the DC current I of the i-th DC circuit di The reactive power consumption of the ith DC inverter is Q di ;Q 12 is the reactive power transmission between the receiving busbars of the first and second DC loops, Q 13 is the reactive power transmission between the receiving busbars of the first and third DC loops, Q 23 It is the reactive power transmission between the receiving end busbars of the 2nd and 3rd DC loops.

[0024] Further, in the step 2, the action space is the VDCOL characteristic parameters of each DC loop in the multi-input high-voltage direct current transmission system and the reactive value output by SVG, and the action space is expressed as:

[0025] A=[A VDCOL1 A VDCOLi …A VDCOLn A SVG ];

[0026]

[0027] Where A is the action space, A VDCOLi is the VDCOL characteristic parameter of the i-th DC circuit, U dHi and U dLi are the starting point of recovery and current limiting of the VDCOL DC voltage of the i-th DC circuit respectively; A SVG Reactive value for SVG output.

[0028] Further, in the step 2, the state space is expressed as:

[0029] S=[S HVDC1 S HVDCi ...S HVDCn ];

[0030]

[0031] Where S is the state space, S HVDCi is the DC state of the i-th DC loop in the multi-infeed HVDC transmission system, U aci is the commutation voltage of the i-th DC circuit, I di is the DC current of the i-th DC circuit.

[0032] Furthermore, in step 3, a critical commutation voltage index for subsequent commutation failure due to multi-feed when VDCOL cooperates with SVG is proposed, specifically:

[0033] Combined with the coupling constraint relationship, taking into account the VDCOL control effect, taking the arc extinction angle as the critical arc extinction angle, the VDCOL control characteristics, the expressions of each bus voltage with respect to the DC current and SVG reactive compensation are combined to solve the critical commutation voltage index of each DC continuous commutation failure considering the recovery process under multiple infeed.

[0034] Further, in step 3, the reward function is described as:

[0035]

[0036] Where r is the reward function, U aci,min is the minimum commutation voltage of the i-th DC loop in the multi-infeed HVDC transmission system after a fault, U aci_th is the critical commutation voltage indicator, is the normalized active power of the DC line transmission of the i-th loop, is the reactive power output by the SVG after normalization, ω1, ω2 and ω3 are weighting coefficients.

[0037] Further, in step 4, the DDPG collaborative optimization model is trained using the collected samples, specifically:

[0038] Based on the state space, the action space and the reward function, four networks and the experience pool in the DDPG basic model are defined and constructed;

[0039] Based on the experience pool, the four networks are trained using the collected samples to update the network parameters of the four networks.

[0040] Further, the four networks include a Q network, a policy network, a target Q network and a target policy network, and the network parameters of the four networks include a Q network parameter, a policy network parameter, a target Q network parameter and a target policy network parameter;

[0041] The Q network parameters are updated by minimizing the loss function, the policy network parameters are updated by the policy gradient, the target Q network parameters are updated by the Q network parameters, and the target policy network parameters are updated by the policy network parameters.

[0042] On the basis of the above-mentioned method for improving the stability of an electric power system based on the coordination of resources inside and outside the grid, the present invention also provides a system for improving the stability of an electric power system based on the coordination of resources inside and outside the grid.

[0043] A system for improving the stability of an electric power system based on the coordination of resources inside and outside the grid comprises a processor, a memory and a computer program stored in the memory. When the computer program is executed by the processor, the method for improving the stability of an electric power system based on the coordination of resources inside and outside the grid as described above is implemented.

[0044] The beneficial effects of the present invention are as follows: the method and system for improving the stability of a power system based on the coordination of resources inside and outside the grid proposes a suppression idea of ​​using low-voltage current limiting control (VDCOL) of network resources to reduce the reactive power consumption of the inverter, and coordinating the reactive power compensation of a static VAR generator (SVG) with resources outside the grid for subsequent continuous commutation failure of multi-input DC caused by excessive reactive power shortage, and adaptively generates the optimal parameter combination based on the deep deterministic policy gradient algorithm (DDPG), which can stabilize the commutation voltage, increase the commutation margin, suppress the failure of continuous commutation of multiple inputs, and improve the stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of a method for improving the stability of a power system based on coordination of resources inside and outside the grid according to the present invention;

[0046] Figure 2 A schematic diagram of the topological structure of a three-infeed high-voltage direct current transmission system according to an embodiment of the present invention;

[0047] Figure 3 Schematic diagram of the coupling relationship between a single-circuit direct current VDCOL and SVG in an embodiment of the present invention;

[0048] Figure 4This is a schematic diagram of the convergence of the model training process in an embodiment of the present invention;

[0049] Figure 5 Schematic diagram of commutation failure in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0051] like Figure 1 As shown, the method for improving the stability of the power system based on the coordination of resources inside and outside the grid includes the following steps 1 to 4:

[0052] Step 1: Considering the voltage-reactive power characteristics of the multi-input HVDC transmission system, a mathematical model is established between the arc extinction angle, DC current and commutation voltage of the inverter side considering the in-grid resources VDCOL and the out-grid resources SVG, and based on the mathematical model, the coupling constraint relationship between VDCOL and SVG is theoretically analyzed. Specifically:

[0053] The multi-infeed HVDC transmission system includes at least three DC loops, including the first DC loop, the second DC loop and the third DC loop. In the receiving-end topology of the multi-infeed HVDC transmission system containing SVG, the SVG is set to work in reactive mode and connected to the AC bus of the HVDC transmission system through a transformer. The power loss generated by the transformer is ignored, and each DC inverter station has the following quasi-steady-state equation:

[0054]

[0055] P di =U di I di ; (4)

[0056]

[0057]

[0058]

[0059]

[0060] Where i is the DC number; U di ,I di , U di0 , U aci The DC voltage, DC current, no-load DC voltage, and commutation voltage of the i-th DC circuit are respectively di ,I di All are related to DC VDCOL control; is the power factor angle of the converter transformer of the i-th DC circuit; k i N is the ratio of the converter transformer of the i-th DC circuit; pi is the number of pole pairs of the converter transformer of the i-th DC circuit; γ i , U i are the arc extinction angle and commutation overlap angle of the inverter side of the i-th DC loop respectively; T i , X Ti are the inverter transformer ratio and leakage reactance of the i-th DC inverter; P di , Q di are respectively the active power transmitted by the i-th DC and the reactive power consumed by the inverter side; B ci is the equivalent susceptance of the reactive power compensation capacitor and filter of the inverter station of the i-th DC loop, Q ci is the reactive power generated by the ith DC circuit; Q aci is the reactive power of the i-th DC transmission to the AC system; Q ij represents the reactive power transmitted from the AC bus of the i-th DC circuit to the AC bus of the j-th DC circuit; Q SVG Reactive value for SVG output.

[0061] Formula (7) is the reactive dynamic equation of the AC bus, which is related to the reactive interaction between AC and DC, reactive compensation power, reactive consumption of the converter, and reactive transmission between DCs. In a multi-infeed HVDC transmission system, due to the existence of coupling impedance, the commutation failure of one DC due to a fault will quickly affect other DCs through reactive power. If it causes reactive imbalance in other DCs, it will cause commutation failure in other DCs.

[0062] The coupling relationship between VDCOL and SVG is analyzed by taking the i-th DC fault in a multi-infeed HVDC transmission system containing SVG as an example, where i=3 is assumed. When a three-phase short circuit fault occurs in the AC system of the third DC loop, equation (7) is simplified to:

[0063] Q ac3 =Q c3 -Q d3 +Q SVG -Q f ; (9)

[0064] Among them, Q c3 It can be directly calculated from the AC voltage through formula (6) and is calculated as Q c3 =f Qc3 (U ac3 );Q f Q is the reactive power loss due to fault, which is related to the ground impedance. d3 It can be simplified through equations (3) to (5) as follows:

[0065]

[0066] Substituting further into equations (1) to (5), we get Q d3 It can be written about U ac3 ,I d3 function, denoted as Q d3 =f Qd3 (U ac3 ,I d3 ).

[0067] After the fault, the commutation voltage U of the third DC circuit ac3 Being pulled low, its power frequency voltage can be expressed as:

[0068]

[0069] in, Z3 is the inverter side AC system voltage and impedance of the third DC loop, P ac3 +jQ ac3 is the impedance of the AC busbar at the receiving end of the third DC loop. Since the reactance in the high-voltage line is much larger than the resistance, the resistance is ignored. Substituting the above formula into formula (10), the above formula can be expressed as:

[0070]

[0071] Further, it can be simplified to get i ,I d3 , Q SVG Indicates, about U ac3 The univariate quartic equation (i.e. single-feed mathematical model) is:

[0072] Ua 4 ac3 +bU 3 ac3 +cU 2 ac3 +dU ac3 +e=0; (13)

[0073] Among them, U ac3 is the commutation voltage of the third DC loop, a, b, c, d and e are all model parameters and are specifically expressed as follows:

[0074]

[0075] Specifically, B3 is the reactive compensation susceptance of the inverter side of the third DC loop, X3 is the reactance of the AC system at the receiving end of the third DC loop, E3 is the AC system voltage at the receiving end of the third DC loop, Q SVG is the reactive value output by SVG, I d3 is the DC current of the third DC loop, A, B and C are all model parameters, k3 is the transformer ratio of the third DC loop, N p3is the number of pole pairs of the converter transformer of the third DC loop, γ3 is the arc extinction angle of the inverter side of the third DC loop, X T3 is the leakage reactance of the inverter-side converter transformer of the third DC loop.

[0076] The above analysis shows that during the DC commutation failure process, VDCOL and SVG act through reactive coupling, which is manifested in the effect on the commutation voltage.

[0077] Applying the above analysis process to the multi-input HVDC transmission system, we can obtain the calculation formula for the AC bus voltage at the receiving end of each DC loop under the action of multi-VDCOL and SVG when the AC bus at the receiving end of the jth (j≠i) DC loop fails.

[0078] For the convenience of analysis, this embodiment is described by taking a three-infeed HVDC transmission system as an example. The SVG is connected to the third DC loop. If a fault occurs in the AC busbar at the receiving end of the first DC loop, then for the first DC loop, according to equation (7) and combined with the above derivation, the following multi-infeed mathematical model can be obtained for each DC receiving end:

[0079] U 2 ac1 =E1U ac1 +(f Qc1 (U ac1 )-f Qd1 (U ac1 ,I d1 )-Q 13 -Q 12 -Q f )X3; (15)

[0080] U 2 ac2 =E2U ac2 +(f Qc2 (U ac2 )-f Qd2 (U ac2 ,I d2 )+Q 12 -Q 23 )X3; (16)

[0081] U 2 ac3 =E3U ac3 +(f Qc3 (U ac3 )-f Qd3 (U ac3 ,I d3 )+Q SVG +Q 13 +Q 23 )X3; (17)

[0082] Among them, let i = 1, 2, 3, then: U aci is the commutation voltage of the ith DC circuit, E i is the AC system voltage at the receiving end of the i-th DC circuit, f Qci (U aci ) is the commutation voltage U of the i-th DC aci The reactive power compensation power Q of the inverter side of the i-th DC is represented by ci , f Qdi (U aci ,I di ) is the commutation voltage U of the i-th DC aci and the DC current I of the i-th DC circuit di The reactive power consumption of the ith DC inverter is Q di ;Q 12 is the reactive power transmission between the receiving busbars of the first and second DC loops, Q 13 is the reactive power transmission between the receiving busbars of the first and third DC loops, Q 23 It is the reactive power transmission between the receiving end busbars of the 2nd and 3rd DC loops.

[0083] Combining the above equations, we can get U ac1 , U ac2 , U ac3 About Gamma i ,I d1 ,I d2 ,I d3 , Q SVG The expression of Q can be obtained, and then the law similar to that of single-feed can be obtained, that is, in order to maintain a certain commutation margin, Q SVG with I d3 There exists a feasible domain for the value of .

[0084] Step 2: Based on the coupling constraint relationship, a DDPG basic model for collaborative optimization of VDCOL and SVG parameters is established, and the state space and action space of the DDPG basic model are determined.

[0085] Specific:

[0086] DDPG is a deep reinforcement learning algorithm applied to continuous states and continuous actions, including Q network, policy network and corresponding target Q network and target policy network. Its principle is: the policy network outputs action a in state s, the Q network scores the action according to [s, a], the policy network adjusts parameters according to the score to try to get a higher score, and the Q network updates parameters according to the rewards of the environment, so that the actions of the policy network can finally get higher rewards from the environment. That is, the DDPG basic model needs to contain basic input parameters.

[0087] From step 1, we can see that VDCOL and SVG affect the commutation process through reactive power. As for SVG, it directly outputs reactive power, thus affecting the commutation process. After the commutation fails, VDCOL affects the reactive power by controlling the recovery process of the DC current. The VDCOL characteristic equation of each DC can be described as:

[0088]

[0089] Where U dHi , U dLi They are the starting point of recovery and current limiting of the VDCOL DC voltage of the i-th DC circuit.

[0090] From the above analysis, it can be seen that the SVG output reactive parameters and VDCOL characteristic parameters are both changeable parameters that can affect the commutation process. Therefore, the present invention sets the action space of the DDPG basic model as:

[0091] A=[A VDCOL1 A VDCOLi ...A VDCOLn A SVG ]; (19)

[0092]

[0093] Where A is the action space, A VDCOLi is the VDCOL characteristic parameter of the i-th DC circuit, U dHi and U dLi are the starting point of recovery and current limiting of the VDCOL DC voltage of the i-th DC circuit respectively; A SVG is the reactive value output by SVG. Each action space has the following constraints:

[0094]

[0095] Where U dLi,max , U dLi,min , U dHi,max , U dHi,min They are the two starting point parameters in the VDCOL curve (i.e., U dHi and U dLi ), Q N Rated capacity of SVG.

[0096] Secondly, in order to describe different commutation failure scenarios, the state space in the DDPG basic model is set to:

[0097] S=[S HVDC1 S HVDCi ...S HVDCn ]; (twenty two)

[0098]

[0099] Where S is the state space, S HVDCi is the DC state of the i-th DC in the multi-infeed HVDC power transmission system, obtained by collecting the receiving-end bus voltage amplitude and its change rate, and the DC current amplitude and its change rate; U aci is the commutation voltage of the i-th DC circuit, I di is the DC current of the i-th DC circuit.

[0100] Step 3: Considering the commutation failure recovery process, the critical commutation voltage index of the subsequent commutation failure when VDCOL and SVG are coordinated is proposed, and the reward function of the DDPG basic model is set in combination with the system active transmission and reactive power compensation strength.

[0101] For the DDPG basic model, after the state and action space are determined, the reward value after executing a specific action in each state needs to be given to evaluate the inhibitory effect of parameter synergy on commutation failure. To this end, this step is based on step 1 and proposes a critical commutation voltage index for subsequent commutation failure considering the recovery process to measure the commutation margin after the first commutation failure of the system under given control parameters.

[0102] First, the reason for proposing the critical commutation voltage index is explained. Generally speaking, the arc extinction angle is the most direct physical quantity to measure commutation failure, γ<γ th However, the arc extinction angle is difficult to measure in actual engineering, and if the arc extinction angle is monitored in real time and then the control strategy is calculated, it is too late for system stability and cannot be actually applied in the suppression of commutation failure. Therefore, in order to be able to judge the subsequent commutation failure as early as possible and adjust the control parameters in time to suppress the commutation failure, this paper converts the arc extinction angle criterion into the commutation voltage criterion U aci >U aci_th , U aci_th It is the critical commutation voltage index for the system to maintain subsequent normal commutation, corresponding to the γ min Equal to γ th critical situation.

[0103] Next, we will explain the critical commutation voltage indicator U aci_th The inverter fails to commutate for the first time, and the commutation voltage drops to the lowest point U ac_minAfter that, the DC recovery process is divided into two stages: voltage recovery and power recovery. There may be three situations for the commutation voltage. In the first case, due to the high strength of the receiving end, proper control strategy or minor fault, the commutation voltage quickly returns to the steady-state value and maintains at the rated value; in the second case, the voltage recovers quickly after reaching the lowest point. Compared with the first case, the rise speed changes from fast to slow. Due to the inappropriate control strategy at this time, such as the effect of VDCOL, the power recovery speed is greater than the voltage recovery, and the DC inverter station requires more reactive power, the commutation voltage is lowered, but the reactive power shortage is small, and the voltage can eventually return to a stable value; in the third case, because the reactive power shortage is not compensated, the commutation voltage is continuously lowered in the later stage of the second case, and the arc extinction angle continues to drop. When the arc extinction angle reaches the critical value γ th After that, it can be determined that the subsequent commutation failure occurs, that is, when the arc extinction angle is equal to the critical value again, the system is in a critical state of maintaining continuous normal commutation. At this time, the corresponding commutation voltage is the critical commutation voltage index U aci_th .

[0104] Finally, the results obtained by theoretical derivation are applied to calculate the critical commutation voltage index of the subsequent commutation failure in the recovery process under the coupling constraint relationship of the present invention. Combined with the coupling constraint relationship of step one, taking into account the VDCOL control effect, taking the arc extinction angle as the critical arc extinction angle, and combining the VDCOL control characteristics, the expressions of each bus voltage with respect to the DC current and SVG reactive compensation, the critical commutation voltage index of each DC continuous commutation failure in the recovery process under multiple feeds can be solved. That is: Combined with the coupling constraint relationship of step one, taking into account the VDCOL control effect, taking the arc extinction angle as the critical arc extinction angle γ th , the combined equations (15) to (18) can be solved to obtain the critical commutation voltage index U of each DC continuous commutation failure under multiple feeds and considering the recovery process. aci_th .

[0105] If U aci >U aci_th , indicating that the commutation voltage is always higher than the critical commutation voltage index during the entire recovery process, and continuous commutation failure will not occur. aci_min Considering that the fault point has a certain voltage support effect, the voltage at the starting point of commutation recovery is close to the instantaneous voltage of the fault, so U aci_min is the instantaneous fault voltage.

[0106] In addition, if the DC transmission power and SVG reactive capacity are also included in the reward function, the reward function can be described as:

[0107]

[0108] Where r is the reward function, U aci,minis the minimum commutation voltage of the i-th DC loop in the multi-infeed HVDC transmission system after a fault, U aci_th is the critical commutation voltage index, U aci,min -U aci_th Characterizes the commutation margin, U aci,min -U aci_th The larger the value of is, the higher the commutation margin is and the larger the reward value r is; is the normalized active power transmitted by the i-th DC line, obtained by multiplying the DC current and DC voltage of each DC line, calculated by equation (6) and equation (7) respectively. The more active power is transmitted, the greater the reward value r is; is the reactive power output of SVG after normalization. The larger the output, the higher the economic cost and the smaller the reward value r. ω1, ω2 and ω3 are all weighted coefficients that determine the importance of each item.

[0109] Step 4: Based on the state space, the action space and the reward function, a DDPG collaborative optimization model is established, the collected samples are used to train and test the DDPG collaborative optimization model, and the tested DDPG collaborative optimization model is sent to the security control master station to obtain an optimized parameter combination through reinforcement learning when a fault occurs, and the optimized parameter combination is called for control to suppress commutation failure. Specifically:

[0110] Based on the state space, the action space and the reward function, four networks and experience pools in the DDPG basic model are defined and constructed; using θ C ,θ μ ,θ C ′ and θ μ 'Initialize four networks and set related hyperparameters. The four networks include Q network, policy network, target Q network and target policy network. The network parameters of the four networks include Q network parameters, policy network parameters, target Q network parameters and target policy network parameters; θ C ,θ μ ,θ C ′ and θ μ ′ are the Q network parameters, strategy network parameters, target Q network parameters and target strategy network parameters respectively.

[0111] At the same time, fault scenarios with different fault sizes, fault locations, and system receiving end strengths are set in the simulation software and randomly combined to obtain samples for model training and testing.

[0112] For each round of training, we obtain the state S from the environment i , the policy network selects action a i After that, perform the action and observe the reward r iand the next state, and store them in the experience pool. If the sample in the experience pool is greater than the set minimum sampling number N, then randomly sample according to N, and calculate the target value y by combining the target Q network and the target strategy network i For the Q network, it updates the parameters θ by minimizing the loss function L C The loss function L represents the mean square error between the evaluation value and the target value, as shown in formula (25):

[0113]

[0114] Here, F() represents the evaluation function.

[0115] Update the policy network parameters θ using policy gradient μ , the policy gradient is calculated by formula (26):

[0116]

[0117] Then, the target Q network parameters are updated through the Q network parameters, and the target policy network parameters are updated through the policy network parameters. The update formula is shown in formula (27). Where τ is the network update rate.

[0118]

[0119] After training and testing, the DDPG collaborative optimization model is obtained and deployed in the security control master station for real-time parameter optimization. When the model is applied, the security control master station monitors the actual operating status. When a fault occurs, the training model is called online to generate the optimal parameter combination, and the optimal parameter combination is sent to the specific control link to suppress commutation failure.

[0120] The method of the present invention is described below with reference to specific examples.

[0121] This embodiment is based on the three-infeed HVDC transmission system built on the CIGRE HVDC transmission standard test model in PSCAD / EMTDC. Its topology is as follows: Figure 2 Among them, the specific parameters of the three-input high-voltage direct current transmission system are shown in Table 1, and the coupling impedance between the receiving ends is set to: Z 12 =1+j8.137, Z 13 =4+j32.547, Z 23 =2+j16.274. Secondly, a SVG model with a rated capacity of 300MVA was built in the PSCAD / EMTDC environment, and the parameters are shown in Table 2.

[0122] Table 1 High voltage DC transmission system parameters

[0123]

[0124]

[0125] Table 2 SVG model parameters

[0126]

[0127] Step 1: Analyze the coupling constraint relationship. Figure 1 The third DC circuit adopts the standard DC transmission system parameters shown in Table 1, and the other parameters are shown in Table 3. ref Under control, the commutation voltage offset ΔU ac3 Characterization I d3 , Q SVG The impact of commutation failure, such as Figure 3 shown.

[0128] Table 3 Other parameters of HVDC transmission system

[0129]

[0130] Figure 3 The z-axis of (a) corresponds to the deviation of the actual commutation voltage from the rated value. Since the sudden change of commutation voltage is the main cause of commutation failure, the deviation amplitude of commutation voltage can represent the DC commutation margin to a certain extent. Therefore, if the commutation voltage deviation is larger, it can be said that the voltage is more unstable and the system commutation margin is lower. Figure 3 The x-axis of (a) is the DC current I d3 The y-axis is the reactive value Q of SVG output SVG In the commutation process, the commutation voltage is different under the joint action of the two different parameters. Among them, the black line is the case when the commutation voltage offset is equal to 0. At this time, the commutation voltage remains stable, which means that the system commutation margin is the highest at this time, that is, it can be considered that I d3 , Q SVG Parameter coordination is the most ideal. Under other parameters, the commutation voltage is offset. It can be considered that in order to maintain the commutation voltage at a certain level, I d3 , Q SVG Mutually constrained, the two parameters exist in a feasible region. Figure 3 (b) Analysis, when Q SVG When constant, the commutation voltage increases first and then decreases with the DC current, and Q SVG The larger it is, the larger the DC current will be when the voltage offset is 0; Figure 3 (c) can also be obtained, when I d3 When the commutation voltage is constant, it decreases first and then increases with the increase of SVG reactive power, while I d3 The larger the value, the smaller the SVG reactive value corresponding to the voltage offset of 0. Figure 3 (b) Figure 3 (c) Available, Id3 , Q SVG When they exist at the same time, the two have different effects on the commutation voltage. SVG To maintain the commutation voltage, I d3 There is an optimal solution, which is not the smaller the better, that is, the VDCOL control command value cannot be too small. Similarly, if I d3 Definitely, Q SVG There is also an optimal solution, which is to configure the SVG output to not be as large as possible.

[0131] In summary, for a single-feed HVDC transmission system with SVG, when only VDCOL is acting, the smaller the DC current under VDCOL control, the smaller the required reactive power, which is more conducive to the stability of the commutation voltage. If only SVG is acting, the greater the SVG compensation reactive power, the more conducive to the recovery of the commutation voltage. When the two act at the same time, Figure 4 Analysis shows that in order to maintain the commutation voltage, VDCOL and SVG influence each other and have a coupling relationship. The most ideal situation of their coordination is no longer achieved at each boundary.

[0132] Step 2: Determine the state space and action space of the DDPG basic model.

[0133] for Figure 2 The three-infeed HVDC transmission system shown in the figure adopts the method of the present invention, and the state space and action space in the established DDPG basic model are respectively:

[0134] A=[U dL1 U dH1 U dL2 U dH2 U dL3 U dH3 A SVG ];

[0135]

[0136] Step 3: Determine the reward function of the DDPG base model.

[0137] for Figure 2 For the three-infeed HVDC transmission system shown in FIG. 1 , after adopting the method of the present invention, the reward function in the established DDPG basic model is:

[0138]

[0139] Step 4: Collect samples, train and test the model. Run the model under specific fault scenarios to obtain the optimal parameter combination.

[0140] For the DDPG basic model, the Q network is set to 4 layers, and the number of neurons in the input layer, hidden layer #1 and #2, and output layer are 6, 64, 64, and 2, respectively. Based on different random combinations of fault inductance, fault location, and system receiving end strength, a total of 40,000 training samples are obtained. The algorithm is run on the PyCharm platform based on Python language, using the Tensorflow2.0 framework.

[0141] The convergence of the DDPG collaborative optimization model training process represented by the loss function is as follows: Figure 4 As shown in the figure, it can be seen that the training loss of the DDPG collaborative optimization model decreases monotonically with the increase of the number of training times, close to zero, and the convergence curve is relatively smooth. In the early stage of training, the loss function takes a large value. When the number of training times reaches a certain volume, the loss drops sharply and basically reaches stability. This proves that the proposed DDPG collaborative optimization model has good convergence performance and can obtain stable collaborative parameters.

[0142] Set the fault scenario: The inverter-side AC bus of the first DC loop is set to a three-phase inductive grounding fault at t=1s. The grounding inductance L f =0.1H, fault duration 0.3s. Under conventional DC control only, the number of DC commutation failures is 3, 3, and 3. Running the DDPG collaborative optimization model, the optimized parameter combination is shown in Table 4. Under this optimized parameter combination, the three DC commutation failures are as follows Figure 5 shown.

[0143] Table 4 Optimized parameter combinations after running

[0144]

[0145] Depend on Figure 5 It can be seen that under the strategy proposed in the present invention, the number of DC commutation failures is 1, 1, and 1 respectively. Compared with conventional DC operation, the commutation failure can be effectively suppressed. Figure 5 (b1), (b2), and (b3) are the commutation voltage curves of each DC under the strategy. It can be seen intuitively that since the fault occurred in HVDC1, its commutation voltage drop is the largest, and HVDC3 is equipped with SVG for reactive power compensation, which makes the commutation voltage curve of HVDC3 higher than that of the other two DCs as a whole, and the system is the most stable. Figure 5 The dashed lines in (b1), (b2), and (b3) are the per-unit values ​​of the critical commutation voltage indicators of each DC circuit under the calculated optimization parameters. Among them, the per-unit values ​​of the critical commutation voltage indicators of the three DC circuits are 0.802pu, 0.885pu, and 0.901pu, respectively, which are lower than the minimum commutation voltage. This also shows that after the coordinated optimization, the system commutation margin has been enhanced.

[0146] On the basis of the above-mentioned method for improving the stability of an electric power system based on the coordination of resources inside and outside the grid, the present invention also provides a system for improving the stability of an electric power system based on the coordination of resources inside and outside the grid.

[0147] A system for improving the stability of an electric power system based on the coordination of resources inside and outside the grid comprises a processor, a memory and a computer program stored in the memory. When the computer program is executed by the processor, the method for improving the stability of an electric power system based on the coordination of resources inside and outside the grid as described above is implemented.

[0148] The present invention proposes a method and system for improving the stability of a power system based on the coordination of resources inside and outside the grid, and proposes a suppression idea of ​​using low-voltage current limiting control (VDCOL) of network resources to reduce the reactive power consumption of the inverter, and coordinating the reactive power compensation of a static VAR generator (SVG) with resources outside the grid, in order to solve the subsequent continuous commutation failure of multi-input DC caused by excessive reactive power shortage. The method and system adaptively generate the optimal parameter combination based on the deep deterministic policy gradient algorithm (DDPG), which can stabilize the commutation voltage, increase the commutation margin, suppress the continuous commutation failure of multi-input, and improve the stability of the power system.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for improving the stability of a power system based on coordination of resources inside and outside the grid, characterized in that: include: Step 1: Considering the voltage-reactive power characteristics of the multi-input HVDC transmission system, a mathematical model between the arc extinction angle, DC current and commutation voltage on the inverter side considering the in-grid resource VDCOL and the out-grid resource SVG is established, and based on the mathematical model, a coupling constraint relationship between the VDCOL and SVG collaboration is theoretically analyzed; Step 2: Based on the coupling constraint relationship, a DDPG basic model for collaborative optimization of VDCOL and SVG parameters is established, and a state space and an action space of the DDPG basic model are determined; Step 3: Considering the commutation failure recovery process, a critical commutation voltage index for subsequent commutation failure when VDCOL and SVG are coordinated is proposed, and the reward function of the DDPG basic model is set in combination with the system active transmission and reactive power compensation strength; Step four, based on the state space, the action space and the reward function, a DDPG collaborative optimization model is established, the collected samples are used to train and test the DDPG collaborative optimization model, and the tested DDPG collaborative optimization model is sent to the security control main station, so that when a fault occurs, the optimized parameter combination can be obtained through reinforcement learning, and the optimized parameter combination can be called for control to suppress commutation failure.

2. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In the step 1, the mathematical model includes a single-feed mathematical model; the multi-feed high-voltage direct current transmission system includes at least three direct currents, and the at least three direct currents include a first direct current, a second direct current, and a third direct current; assuming that the third direct current including the SVG fails, the single-feed mathematical model is specifically: aU 4 ac3 +bU 3 ac3 +cU 2 ac3 +dU ac3 +e=0; Among them, U ac3 is the commutation voltage of the third DC loop, a, b, c, d and e are all model parameters and are specifically expressed as follows: Specifically, B3 is the reactive compensation susceptance of the inverter side of the third DC loop, X3 is the reactance of the AC system at the receiving end of the third DC loop, E3 is the AC system voltage at the receiving end of the third DC loop, Q SVG is the reactive value output by SVG, I d3 is the DC current of the third DC circuit, A, B and C are all model parameters, and A=1.35k3N p3 , B = Acosγ3, C = 0.955X T3 N p3 ; Where k3 is the ratio of the converter transformer of the third DC circuit, N p3 is the number of pole pairs of the converter transformer of the third DC loop, γ3 is the arc extinction angle of the inverter side of the third DC loop, X T3 is the leakage reactance of the inverter-side converter transformer of the third DC loop.

3. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 2, characterized in that: In the step 1, the mathematical model further includes a multi-input mathematical model; the multi-input mathematical model includes: Among them, let i = 1, 2, 3, then: U aci is the commutation voltage of the ith DC circuit, E i is the AC system voltage at the receiving end of the i-th DC circuit, The commutation voltage U of the i-th DC circuit is aci The reactive power compensation power Q of the inverter side of the i-th DC is represented by ci , The commutation voltage U of the i-th DC circuit is aci and the DC current I of the i-th DC circuit di The reactive power consumption of the ith DC inverter is Q di ;Q 12 is the reactive power transmission between the receiving busbars of the first and second DC loops, Q 13 is the reactive power transmission between the receiving busbars of the first and third DC loops, Q 23 It is the reactive power transmission between the receiving end busbars of the 2nd and 3rd DC loops.

4. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In the step 2, the action space is the VDCOL characteristic parameters of each DC loop in the multi-input high-voltage direct current transmission system and the reactive value output by SVG, and the action space is expressed as: A=[A VDCOL1 A VDCOLi …A VDCOLn A SVG ]; Where A is the action space, A VDCOLi is the VDCOL characteristic parameter of the i-th DC circuit, U dHi and U dLi are the starting point of recovery and current limiting of the VDCOL DC voltage of the i-th DC circuit respectively; A SVG Reactive value for SVG output.

5. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In the step 2, the state space is expressed as: S=[S HVDC1 S HVDCi …S HVDCn ]; Where S is the state space, S HVDCi is the DC state of the i-th DC loop in the multi-infeed HVDC transmission system, U aci is the commutation voltage of the i-th DC circuit, I di is the DC current of the i-th DC circuit.

6. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In step 3, a critical commutation voltage index for subsequent commutation failure when VDCOL and SVG cooperate is proposed, specifically: Combined with the coupling constraint relationship, taking into account the VDCOL control effect, taking the arc extinction angle as the critical arc extinction angle, the VDCOL control characteristics, the expressions of each bus voltage with respect to the DC current and SVG reactive compensation are combined to solve the critical commutation voltage index of each DC continuous commutation failure considering the recovery process under multiple infeed.

7. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In step 3, the reward function is described as: Where r is the reward function, U aci,min is the minimum commutation voltage of the i-th DC loop in the multi-infeed HVDC transmission system after a fault, U aci_th is the critical commutation voltage indicator, is the normalized active power of the DC line transmission of the i-th loop, is the reactive power output by the SVG after normalization, ω1, ω2 and ω3 are weighting coefficients.

8. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 1, characterized in that: In step 4, the DDPG collaborative optimization model is trained using the collected samples, specifically: Based on the state space, the action space and the reward function, four networks and the experience pool in the DDPG basic model are defined and constructed; Based on the experience pool, the four networks are trained using the collected samples to update the network parameters of the four networks.

9. The method for improving the stability of a power system based on coordination of resources inside and outside the grid according to claim 8, characterized in that: The four networks include a Q network, a policy network, a target Q network and a target policy network, and the network parameters of the four networks include a Q network parameter, a policy network parameter, a target Q network parameter and a target policy network parameter; The Q network parameters are updated by minimizing the loss function, the policy network parameters are updated by the policy gradient, the target Q network parameters are updated by the Q network parameters, and the target policy network parameters are updated by the policy network parameters.

10. The power system stability improvement system based on the coordination of internal and external resources is characterized by: It includes a processor, a memory and a computer program stored in the memory. When the computer program is executed by the processor, it implements the method for improving the stability of the power system based on the coordination of internal and external resources as described in any one of claims 1 to 9.

Images