CN113964885B

CN113964885B - Situation awareness-based active power grid reactive power prediction and control method

Info

Publication number: CN113964885B
Application number: CN202111015446.8A
Authority: CN
Inventors: 司君诚; 王元元; 孙名妤; 蔡言斌; 张双乐; 刘航航; 刘剑宁; 王银忠; 苏小向; 吕风磊; 张丹; 任敬刚
Original assignee: Dongying Power Industry Bureau Of State Grid Shandong Electric Power Co
Current assignee: Dongying Power Industry Bureau Of State Grid Shandong Electric Power Co
Priority date: 2021-08-31
Filing date: 2021-08-31
Publication date: 2023-09-22
Anticipated expiration: 2041-08-31
Also published as: CN113964885A

Abstract

The invention discloses a situation awareness-based active power grid reactive power prediction and control method, and relates to the field of power distribution network reactive power optimization. The method comprises the following steps: the method comprises the steps of (1) collecting situation elements of power grid operation data; (2) Carrying out situation understanding on the current system according to situation elements, calculating data measured by system nodes to obtain weaknesses quantization indexes of the nodes, and providing reactive compensation weaknesses; (3) According to situation elements, predicting distributed power sources and loads by using BLSTM to obtain future situations of the system; (4) And in the situation management and control stage, based on the reactive compensation weak area and the predicted characteristic data, the established optimization model is solved by utilizing an improved particle swarm algorithm, and a reactive configuration scheme is obtained. The advantages are that: the situation awareness is introduced into the reactive power optimization field, a reactive power control scheme of the system in the future situation is obtained through a situation awareness framework, a closed loop system is formed in the process, and network loss and running cost can be reduced while intervention is performed in advance.

Description

Situation awareness-based active power grid reactive power prediction and control method

Technical Field

The invention relates to a situation awareness-based active power grid reactive power prediction and control method, and belongs to the field of power distribution network reactive power optimization.

Background

With the rapid development of national economy, the scale of the power distribution network is continuously enlarged, and the distributed power generation (Distributed Generation, DG) is largely connected into the power distribution network, so that the characteristics of uncertainty and randomness of the power distribution network also provide new challenges for the safe, economic and stable operation of the power distribution network. The difficulty and complexity of the operation control of the power system are greatly increased, and the situation awareness technology of the power system is proposed based on domestic and foreign specialists.

The situation awareness technology has various expression forms in a power grid, for example, by comparing with situation awareness contents of a power transmission network, the intelligent power distribution network is systematically analyzed, and the core for developing situation awareness and situation profit is the change of focusing uncertainty factors, so that a five-step progressive key technical architecture for situation awareness, understanding, forecasting and presenting the final situation profit is provided; the other expression form is to consider the changes of an active power distribution network in terms of constituent elements, topological structures and control means, provide an active power distribution network situation awareness framework in combination with a hardware system, provide specific application scenes of multiple situation awareness for safety verification of overvoltage management, blocking management and scheduling plans from the viewpoint of active management service, and illustrate research bottlenecks in the situation awareness framework specific implementation process from the viewpoints of optimizing distribution points of synchronous measuring devices, integrating and mining mass data and on-line safety verification.

Under the situation awareness framework of the existing intelligent power distribution network, the method is applied to building a prediction framework in the active power distribution network situation prediction stage and providing a specific technical means for realizing the framework, and has important significance for fully exerting the considerable and controllable level of the power distribution network and really realizing risk early warning and active management and control.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides a situation awareness-based active power grid reactive power prediction and control method

In order to solve the technical problems, the invention provides a situation awareness-based active power grid reactive power prediction and control method, which is characterized by comprising the following steps:

(1) The method comprises the steps of acquiring state element acquisition, namely acquiring power grid steady-state operation data, transient operation data, power grid fault data and power equipment operation state data;

(2) In the situation understanding stage, a voltage amplitude value and a phase angle sample sequence set are obtained according to situation elements, the node complex voltage change rate is calculated according to the set, so that a node weakness quantification index VCR is obtained, and a power distribution network is subjected to weakness area division according to the number of reactive devices, the node space position and the node weakness index, so that reactive compensation weakness areas are obtained; the node complex voltage change rate is the weighted summation of the change rate of the voltage and the change rate of the phase angle;

(3) In the situation prediction stage, a multi-layer bidirectional long-short-term memory neural network algorithm is adopted to predict distributed power supplies and loads according to power grid historical data acquired by situation elements, so that a system state in a future situation is obtained;

(4) And in a future situation management and control stage, the reactive capacity of the weak subareas of the power distribution network is optimized by using an improved particle swarm algorithm, so that a reactive configuration scheme is obtained.

Furthermore, the situation elements in the step (1) are collected, and because the wide area measurement system and the phasor measurement unit are put into use in the power grid, the power grid steady state operation data, the transient state operation data, the power grid fault data and the power equipment operation state data can be collected in real time. This stage is mainly for the understanding and analysis of the later grid situation, and for the preparation of predictions. And reasonably configuring measurement according to the analysis and control requirements of the power distribution system so as to acquire the required data.

Further, the situation factor understanding in the step (2) is mainly to identify, count and extract a set Z formed by voltage amplitude values and phase angle sample sequences of all nodes in a static scene of the power grid for reactive compensation weak nodes of the current system ₁ ＝{(v ₁ ,θ ₁ ),(v ₂ ,θ ₂ ),…,(v _n ,θ _n ) -sampling the state data of each node in the set (v _i ,θ _i I=1, …, n) are taken as nodes in the network, in the state of normal operation of the power system, the load active power value on the bus is increased according to a certain specific proportion, the node which is crashed first is the weakest node in the distribution network, and the expansion power flow equation is adopted in the growth mode:

f(θ,V)+λb＝0

wherein θ, V are node voltage phase angle and voltage amplitude, λ is load and power generation growth parameter, and b is system node load growth mode constant.

The load active and reactive power injected by the node i and the generator injection active power can be decomposed into two parts, wherein P _Li0 ,Q _Li0 Representing the level of the initial load node i, P _Gi0 Representing the initial active output of the generator node i. The other part corresponds to the load variation and the active output variation represented by the load parameter λ, resulting in:

wherein P is _Li ,Q _Li Respectively the active level and the reactive level of the load node i, P _Gi For the active power output of the power plant node i,active power of load nodes i respectivelyGrowth mode, reactive growth mode, < >>The method is an active output increasing mode of the power generation equipment node i. The amplitude value of each node after the load is increased proportionally and the set Z formed by the phase angle sample sequence ₂ ＝{(v ₁ ,θ ₁ ),(v ₂ ,θ ₂ ),…,(v _n ,θ _n ) By Z ₁ And Z ₂ As a set of samples that identify weak nodes.

Furthermore, an index for quantifying node weakness is provided, the node weakness node where voltage is easy to collapse is obtained by quantitatively increasing node load and analyzing a PV curve, when the load is increased, voltage and phase angle change are sensitive, and the node weakness is measured by using the node complex voltage change rate. The complex voltage change rate is defined herein as the weighted sum of the change rate of the voltage and the change rate of the phase angle, which is calculated as VCR when the load of node j increases by a certain proportion of active power before the load active reaches a maximum, as follows:

wherein V is _j Represents the voltage value at the j node in normal operation, V' _j Representing the voltage value at the j node after the load is changed according to a given proportion. θ _j Representing the voltage phase angle value at the j node in normal operation, theta' _j The voltage phase angle value at the j node after the load changes according to a given proportion is shown, and the model shows that the larger the complex voltage weak index of the node is, the larger the weak of the node is.

Further, the division of the reactive compensation weak area in the step (2) is performed according to the number of reactive devices, the node space position and the node weakness index, the power distribution network is divided in a partition mode, the concept of weak nodes is expanded to weak areas, each weak area corresponds to a plurality of nodes, a concept of weak area index is provided, the weak area index is defined as VCRS, and the following calculation is performed:

Q<＝N (3)

Wherein VCRS _Q Is the Q-th weak area, i<＝j<K is the number of weak nodes in the weak area Q, V _j Representing the voltage value at the j node in the normal operation time zone, V' _j Representing the voltage value at the j node after the load is changed according to a given proportion. θ _j Representing the voltage phase angle value at the j node in normal operation, theta' _j The voltage phase angle value at the j node after the load is changed according to a given proportion is represented, N is the number of reactive compensation device groups, and the model can be used for dividing the power distribution network into different weak areas so as to ensure that each weak area contains the reactive compensation device, and the greater the weak area index is, the greater the weak property of the weak area is. The number of the weak areas is smaller than that of the reactive compensation device groups, the weak areas are ordered according to the sizes of the weak area indexes, the reactive equipment with strong supplement capability is placed in the areas, and the reactive equipment is placed in the areas, and the weak indexes of all the nodes are ordered according to the sizes of the weak areas, so that the nodes with strong weakness are preferentially considered when the reactive compensation equipment is assembled.

Further, the situation prediction in the step (3) is mainly divided into two modules, wherein the first module is mainly used for predicting the power generation of the distributed power supply, and the second module is mainly used for predicting the load. The active power distribution network is mainly connected with a large amount of distributed energy, wherein the prediction of uncontrollable distributed energy has a great influence on the perception of system situation. Intermittent energy sources in the active power distribution network mainly comprise photovoltaic power generation and wind power generation.

Under the condition of not considering atmospheric influence, the radiation illuminance of the photovoltaic panel installation place is:

g is the total radiation illuminance received by the photovoltaic panel, and the unit is W/m2; g _b The direct solar illuminance is radiated to the earth in vacuum to be a constant value; θ is the angle between the incident solar ray and the normal line of the photovoltaic panel, and changes with time in one day.

The active output of the distributed photovoltaic is related to illuminance and temperature weather conditions received by the photovoltaic panel, and an expression of the maximum value of the output active power can be obtained according to a physical model of the photovoltaic cell:

wherein U is _OC Is the photovoltaic open-circuit voltage under certain weather conditions, I _SC The photovoltaic short-circuit current is the photovoltaic short-circuit current under certain weather conditions; u (u) _OC Is the standard open circuit voltage, r _s Is a standard series resistance. In the case of weather data acquisition, the maximum photovoltaic output can be determined from the corresponding expression.

According to the circuit relation, the maximum power of photovoltaic power generation can be calculated by deducing (6) from the formula (5):

P _max ＝U _M ·I _M ＝a·U _OC ·I _SC (6)

wherein a is a parameter related to the photovoltaic device, U _OC 、I _SC The open-circuit voltage and the short-circuit current of the photovoltaic module are respectively calculated according to the formula (7) under specific weather conditions.

Wherein: g ^R Irradiance that is the standard test state; k is the temperature coefficient of the open-circuit voltage of the photovoltaic system; t (T) ^R Temperature for standard test conditions; t is the operating temperature of the photovoltaic system at the irradiance G.

As can be seen from the formula, the maximum value of the photovoltaic power generation power varies approximately as a sine function in one-day variation without taking into consideration the influence of atmospheric factors.

In actual photovoltaic power generation, fluctuation of power generation amount is caused due to atmospheric refraction, cloud load and rainfall. Therefore, the photovoltaic power generation amount and the power generation amount influenced by uncertain weather factors have larger fluctuation, and the correlation is calculated through the Pearson correlation coefficient, as shown in the formula (8):

as shown by calculation, the radiation illuminance, the temperature and the humidity have strong correlation, but the general weather forecast does not predict the radiation illuminance, the difficulty of acquiring the data is high, and the practical application is considered, the invention adopts the temperature and the humidity as weather elements for generating capacity prediction, and the input format of a sample is that

X _i ＝{T _i ,H _i ,t _i ,s _i }。 (9)

Wherein t is the sample acquisition date, s is the weather condition and is generally classified as 0, 1-1; respectively represent sunny, cloudy and rainy days.

The basic structure of the wind power generation set is a wind turbine, a doubly-fed induction generator and a converter. Wind turbine converts wind energy into mechanical energy and transmits the mechanical energy to doubly-fed induction generator, generator converts mechanical energy into electric energy with amplitude and frequency change, and in normal operation of wind turbine, neglecting wake loss, hub loss and blade tip loss, the maximum power obtainable is:

Wherein: c (C) _P.MAX For the maximum value of the power coefficient, ρ is the air density, A is the fan sector area, and v is the wind speed. From the model it can be seen that the factor that wind power generation is most affected is wind speed. Considering practical application, the wind speed is used as a weather element for generating capacity prediction, and the input format of a sample is as follows:

X _i ＝{v _i ,t _i ,s _i } (11)

Aiming at load prediction, social factors and weather factors have great influence on power load, and the influence degree of temperature, weather conditions, electricity price and social development on load data is analyzed to give specific influence factors. In the concrete analysis, the spearman correlation coefficient is shown as a formula (8), the correlation between the temperature and humidity weather factors and the load is analyzed, and the sample input format of the load is as follows in combination with the correlation factors in the traditional load prediction method:

X _i ＝{t _i ,P _i ,T _i ,H _i ,s _i } (12)

wherein t is _i For sample acquisition date, P _i T is the active load _i To average temperature of the day, H _i For the average humidity s of the day _i Is a weather condition, and is generally classified as 0, 1-1; respectively represent sunny, cloudy and rainy days.

Furthermore, the LSTM neural network is extended from the RNN neural network by selecting the prediction method, so that the problems of gradient explosion and gradient disappearance which can occur on the RNN structure are solved, and the problem of error accumulation caused by the unidirectional structure of the LSTM in the prediction is also solved. Analyzing the structure and principle of the LSTM neural network load prediction model, and providing a specific improvement method, namely the BLSTM neural network, aiming at the problem of error accumulation.

Firstly, BLSTM replaces traditional LSTM prediction, and BLSTM is a BLSTM neural network combining positive phase and negative phase by adding a reverse prediction process on the structure of LSTM, so that the authenticity of load data can be kept as far as possible in theory.

Each layer of BLSTM neural network of the prediction model consists of a forward LSTM network and a reverse LSTM network, the input of the next layer of network is jointly determined by the output results of the forward LSTM and the reverse LSTM of the upper layer, the result of the last layer of network model is simultaneously determined by the calculation results of each layer of forward and reverse, and s is the formula _t ⁽ⁱ⁾ 、s _t ⁽ -i ₁ ⁾ The values of the ith hidden layer at the time t-1 and the time t are respectively represented, and the weight is not shared due to forward calculation and reverse calculation, wherein V is calculated in the forward calculation ⁽ⁱ⁾ 、U ⁽ⁱ⁾ 、W ⁽ⁱ⁾ Representing the weight matrix from the ith hidden layer to the output layer, from the input layer to the hidden layer and between the hidden layers, and V 'during the reverse calculation' ⁽ⁱ⁾ 、U ^′(i) 、W ^′(i) Representing the corresponding inverse weight matrix. i represents the number of BLSTM layers, and i=0, 1,2 … ≡, represents the value of the output layer, the prediction model is as follows:

where the variables appearing in the formula are weights in the BLSTM neuron that are not expanding. According to different prediction objects, different input and output are required to be constructed on a prediction model, the accuracy degree of a prediction result is evaluated by four evaluation indexes of mean square error, root mean square error, mean absolute error and decision coefficient, and the prediction value is The true value is y= { y ₁ ,y ₂ ，y ₃ ，...，y _n }。

(1) Mean Square Error (MSE)

MSE (Mean Squared Error) is called mean square error, also called variance, and MSE is defined as:

the value range is [0, + ] and is equal to 0 when the predicted value is completely consistent with the true value, namely the perfect model; the larger the error, the larger the value.

(2) Root Mean Square Error (RMSE)

RMSE (Root Mean Squard Error) root mean square error, also known as standard error, is in fact the MSE added with a root number, which is relatively intuitive in magnitude, for example rmse=10, and the regression effect can be considered to differ by 10 from the true value average, where RMSE is defined as follows:

(3) Mean Absolute Error (MAE)

Average absolute error (Mean Absolute Deviation), also known as average absolute dispersion. The average absolute error can avoid the problem of mutual offset of errors, so that the magnitude of the actual prediction error can be accurately reflected, and the MAE is defined as follows:

(4) Determining coefficient (R) ² )

Determining the coefficient R ² (R-Square), also known as the decision coefficient or goodness of fit, is typically used in regression models to evaluate the degree of agreement between predicted and actual values, R ² Is defined as follows:

the value range is [0,1], in general, the larger R-Squared is, the better the model fitting effect is, and if the result is 0, the model fitting effect is poor; if the result is 1, it is stated that the model is error free.

Furthermore, in the future situation management and control stage in the step (4), the key of situation management and control is to select a voltage regulation mode.

The control mode of the distributed power supply comprises a variable speed double-fed induction generator set and photovoltaic power generation.

For a variable speed doubly-fed induction generator set, doubly-fed generator control depends on a vector control technology, and PQ decoupling is performed by using the technology, so that a doubly-fed fan can optimally control reactive voltage, wherein a stator flux orientation control technology is applied to a rotor side frequency converter, namely a coordinate axis rotating at a synchronous speed is arranged on a stator flux, and the power of a rotor side is as follows:

wherein u is _s Representing the stator voltage amplitude; l (L) _s Representing stator self-inductance; l (L) _m Representing the mutual inductance between the stator and the rotor; i.e _rq ，i _rd Respectively representing the rotor currents of d axis and q axis under dq0 coordinate system; psi phi type _s Representing the stator flux linkage.

As can be seen from equation (18), controlling the rotor current can achieve control of the rotor-side active and reactive power, respectively. The rotor side frequency converter is used for control, the rotating speed controller provides an active power reference value, and the reactive power reference value is determined according to the requirement of fan voltage control. The network side frequency converter is used for active and reactive decoupling, and generally adopts a voltage directional control scheme. The voltage directional control of the network side frequency converter is that the voltage vector is arranged on the shaft, and under the voltage reference coordinate system of the network side frequency converter, the power of the network side is as follows:

wherein i is _d ，i _q Representing the active current and the reactive current of the network side under the dq0 coordinate system respectively, it is known from the formula (19) that the decoupling control of PQ can be realized by controlling i _d ，i _q To complete.

According to the analysis, the doubly-fed generator set can output reactive power to the power grid through the stator side frequency converter and the grid side frequency converter at the same time, and can participate in reactive compensation of the power grid, so that voltage control is realized.

The basic principle of photovoltaic power generation in a distributed power supply is the conversion of light energy into electric energy, wherein a voltage type inverter is adopted as an inverter. The grid connection of the photovoltaic power generation is completed, the photovoltaic array is required to realize the conversion from direct current to alternating current through the inverter, and the grid connection mode only has an inversion link, so that a control system of the inverter plays a very important role

The power delivered to the grid side by the photovoltaic system can be obtained by the following formula:

wherein Z represents the reactance between the grid and the inverter;representing a line impedance angle; u represents the voltage amplitude of the power grid side; u (U) _I Representing the magnitude of the inverter output voltage; phi represents U and U _I Phase difference between them.

As can be seen from formula (20), by adjusting U _I And phi can realize the regulation of system power, so that the photovoltaic power generation system also has reactive power output capability, can play a role in reactive power compensation of a power grid, and realizes the regulation of voltage.

For voltage regulation control of power grid equipment, mainly a parallel capacitor, a static var compensator SVC and an on-load voltage regulating transformer are adopted, and the voltage regulation mode of the power grid equipment is fixed and is not described in the expansion.

Furthermore, after the future situation management and control and the reactive voltage regulation mode in the step (4) are selected, the key is that the optimal compensation capacity is calculated through an optimization algorithm, the optimal reactive compensation capacity in each reactive compensation weak partition is calculated through an improved particle swarm algorithm, gradient information is not needed for the improved particle swarm algorithm (Particle Swarm Optimization, PSO), parameters are few, coding is not needed in practical application, and the reactive compensation weak partition can be directly used. Each particle will update the flight speed and the position at that moment according to the current extremum Pbest and the global extremum Gbest, as shown in the formulas (21), (22):

Wherein V is _i Is the speed; x is X _i Is the position; k is the number of iterations; individual learning factor c ₁ And social learning factor c ₂ The value is generally 2; r is (r) ₁ And r ₂ Is located at [0,1 ]]Random numbers within the interval.

And the optimal reactive compensation scheme of the future situation of the system can be finally obtained by taking the node voltage optimum and the network loss minimum as objective functions. Minimum system network loss

Wherein P is _loss (t) is the total network loss of the system at the moment t, r _ij For the resistance value of branch ij, I _ij And (t) is the current of the branch ij at the moment t, and N is the distribution network line set. Node voltage deviation is minimal:

wherein: the voltage deviation of the DeltaU (t) node at the time t. And (3) load flow constraint:

wherein P is _i ，Q _i Representing the active power and the reactive power at the node i respectively; v (V) _i ,V _j Representing the voltages at nodes i, j, respectively; g _ij ，B _ij ，θ _ij Representing the conductance, susceptance and phase difference between nodes i, j, respectively.

Voltage constraint:

U _min ≤U _i ≤U _max (26)

reactive compensation device constraint:

wherein Q is _DG.i.min 、Q _DG.i.max For the upper and lower limits of reactive power output of the distributed power supply at node i, Q _c.i.min 、Q _c.i.max And the upper and lower limits of the capacity of the reactive compensation equipment at the node i are set.

Based on the reactive power weak subareas and situation prediction, the whole network optimization is carried out through an improved particle swarm optimization algorithm, and the optimal compensation capacity of each weak subarea can be determined, so that an optimal reactive power compensation scheme is provided.

Further, the situation awareness accuracy is considered from three aspects, namely voltage awareness accuracy, network loss awareness accuracy and reactive compensation scheme awareness accuracy, system node voltage, system network loss and prediction management and control scheme are obtained according to the test set data, and a prediction situation assessment set is formed:

W＝{V ₁ ,V ₂ ,…,V _m ,P _loss ,Q ₁ ,Q ₂ ,…,Q _N } (28)

where V is the voltage amplitude in the future situation, P _loss And Q is reactive compensation capacity in the future situation, wherein the system loss in the future situation is the system network loss in the future situation.

The actual situation assessment set is

W _TRUTH ＝{V _T1 ,V _T2 ,…,V _Tm ,P _Tloss ,Q _T1 ,Q _T2 ,…,Q _TN } (29)

Wherein V is the voltage amplitude in the actual situation, P _loss And Q is reactive compensation capacity in the actual situation, wherein the reactive compensation capacity is the system network loss in the actual situation.

By fitting goodness:

to determine the accuracy of situational awareness.

The beneficial effects of the invention are as follows:

according to the reactive power compensation method, situation awareness technology is introduced into the reactive power optimization field, and reactive power compensation weak areas are divided through a situation awareness framework firstly through system node weakness indexes; in a situation prediction stage, predicting characteristic data by using a BLSTM to obtain a system running state in a future situation, and in a situation management and control stage, performing reactive power optimization on the whole network through an improved PSO algorithm to obtain a control scheme in the future situation; the whole process involves the current moment and the future moment to form a closed loop system, and the system loss and the running cost can be reduced while the intervention is advanced.

Drawings

FIG. 1A basic frame diagram of situation awareness

Fig. 2: PV curve

Fig. 3: BLSTM model diagram

Fig. 4: improved particle swarm algorithm flow chart

Fig. 5: basic flow chart

Detailed Description

The invention is further illustrated, but not limited, by the following examples in connection with the accompanying drawings. As shown in fig. 1, the active power grid reactive power prediction and control method based on situation awareness comprises the following steps:

1) And in the situation element acquisition stage, power grid steady-state operation data, transient operation data, power grid fault data and power equipment operation state data are acquired.

2) In the situation understanding stage, a set Z formed by voltage amplitude and phase angle sample sequences of all nodes in a static scene of a power grid is counted and extracted ₁ ＝{(v ₁ ,θ ₁ ),(v ₂ ,θ ₂ ),…,(v _n ,θ _n ) -sampling the state data of each node in the set (v _i ,θ _i I=1, …, n) as a node in the network, in a state where the power system is operating normally, the load active power value on the bus is increased by adopting an extended power flow equation according to a certain specific proportion. And judging the weak node according to the node collapse sequence.

3) And taking the node complex voltage change rate VCR as a node weakness quantization index, and dividing a power distribution network into weak areas according to the number of reactive devices, the node space position and the node weakness index.

4) And in the situation prediction stage, the distributed energy power generation is predicted, and widely used photovoltaic power generation and wind power generation are considered. Weather element X adopting temperature and humidity as photovoltaic power generation quantity prediction _i ＝{T _i ,H _i ,t _i ,s _i Wind speed is used as weather element X for wind power generation prediction _i ＝{v _i ,t _i ,s _i Adopting temperature, weather condition, electricity price and social development as factors X of load prediction _i ＝{t _i ,P _i ,T _i ,H _i ,s _i }. And predicting the distributed power supply and the load through a multi-layer bidirectional long-short-term memory neural network algorithm to obtain the system state in the future situation. Each layer of BLSTM neural network of the prediction model consists of a forward LSTM network and a reverse LSTM network, the input of the next layer of network is jointly determined by the output results of the forward LSTM and the reverse LSTM of the upper layer, the result of the last layer of network model is simultaneously determined by the calculation results of each layer of forward LSTM and reverse LSTM, and the method comprises the following steps ofThe values of the ith hidden layer at the time t-1 and the time t are respectively represented, and the weight is not shared due to forward calculation and reverse calculation, wherein V is calculated in the forward calculation ⁽ⁱ⁾ 、U ⁽ⁱ⁾ 、W ⁽ⁱ⁾ Representing the weight matrix from the ith hidden layer to the output layer, from the input layer to the hidden layer and between the hidden layers, and V 'during the reverse calculation' ⁽ⁱ⁾ 、U ^′(i) 、W ^′(i) Representing the corresponding inverse weight matrix. i represents the number of BLSTM layers, and i=0, 1,2 … ≡, represents the value of the output layer, the prediction model is as follows:

Wherein the variables appearing in the formula are BLSTMWeights in neurons are not spreading. According to different prediction objects, different input and output are required to be constructed on a prediction model, the accuracy degree of a prediction result is evaluated by four evaluation indexes of mean square error, root mean square error, mean absolute error and decision coefficient, and the prediction value isThe true value is y= { y ₁ ,y ₂ ，y ₃ ，...，y _n }。

(2) Mean Square Error (MSE)

(2) Root Mean Square Error (RMSE)

(3) Mean Absolute Error (MAE)

(4) Determining coefficient (R) ² )

5) In the future situation management and control stage, distributed power supply and power grid equipment are adopted for voltage regulation. The power delivered to the grid side by the photovoltaic system can be obtained by the following formula:

As can be seen from formula (6), by adjusting U _I And phi can realize the regulation of system power, so that the photovoltaic power generation system also has reactive power output capability, can play a role in reactive power compensation of a power grid, and realizes the regulation of voltage.

6) And (3) performing reactive power optimization on the power distribution network by using an improved particle swarm algorithm, and calculating the optimal reactive power compensation capacity in each reactive power compensation weak partition. And the optimal reactive compensation scheme of the future situation of the system can be finally obtained by taking the node voltage optimum and the network loss minimum as objective functions. Minimum system network loss

Voltage constraint:

U _min ≤U _i ≤U _max (10)

reactive compensation device constraint:

wherein Q is _DG.i.min 、Q _DG.i.max Reactive power for distributed power supply at node iUpper and lower limits of output, Q _c.i.min 、Q _c.i.max And the upper and lower limits of the capacity of the reactive compensation equipment at the node i are set. Based on the reactive power weak subareas and situation prediction, the whole network optimization is carried out through an improved particle swarm optimization algorithm, and the optimal compensation capacity of each weak subarea can be determined, so that an optimal reactive power compensation scheme is provided.

7) Obtaining system node voltage, system network loss and a prediction management and control scheme according to the test set data to form a prediction situation assessment set:

W＝{V ₁ ,V ₂ ,…,V _m ,P _loss ,Q ₁ ,Q ₂ ,…,Q _N } (12)

The actual situation assessment set is

W _TRUTH ＝{V _T1 ,V _T2 ,…,V _Tm ,P _Tloss ,Q _T1 ,Q _T2 ,…,Q _TN } (13)

By fitting goodness:

to determine the accuracy of situational awareness. Reactive active prediction and control scheme based on situation awareness aiming at the current power grid running state can be obtained, accuracy is judged, and effectiveness of the method is verified

Claims

1. A situation awareness-based active power grid reactive power prediction and control method is characterized by comprising the following steps:

(2) In the situation understanding stage, a voltage amplitude value and a phase angle sample sequence set are obtained according to situation elements, the node complex voltage change rate is calculated according to the set, so that a node weaknesses quantization index VCR is obtained, and the weak area of the power distribution network is divided according to the number of reactive devices, the node space position and the node weaknesses quantization index VCR, so that reactive compensation weak areas are obtained; the node complex voltage change rate is the weighted summation of the change rate of the voltage and the change rate of the phase angle;

2. The situation awareness-based active prediction and control method for the power grid is characterized in that situation elements in the step (1) are collected, and as a wide area measurement system and a phasor measurement unit are used in the power grid, steady-state operation data, transient operation data, power grid fault data and power equipment operation state data of the power grid can be collected in real time; the method mainly prepares for understanding, analyzing and predicting the situation of the power grid at the back; and reasonably configuring measurement according to the analysis and control requirements of the power distribution system so as to acquire the required data.

3. The situation awareness based active power grid reactive power prediction and control method according to claim 1, wherein the situation awareness stage in the step (2) identifies reactive power compensation weak nodes of the current power grid state, and counts and extracts a set Z formed by voltage amplitude and phase angle sample sequences of each node in a static scene of the power grid ₁ ＝{(v ₁ ,θ ₁ ),(v ₂ ,θ ₂ ),…,(v _n ,θ _n ) }, each node in the setStatus data sample (v) _i ,θ _i I=1, …, n) are taken as nodes in the network, in the state of normal operation of the power system, the load active power value on the bus is increased according to a certain specific proportion, the node which is crashed first is the weakest node in the distribution network, and the expansion power flow equation is adopted in the growth mode:

f(θ,V)+λb＝0 (1)

wherein θ, V are node voltage phase angle and voltage amplitude, λ is load and power generation growth parameter, and b is system node load growth mode constant;

the load active and reactive power injected by the node i and the generator injection active power can be decomposed into two parts, wherein P _Li0 ,Q _Li0 Representing the level of the initial load node i, P _Gi0 Representing the initial active output of the generator node i; the other part corresponds to the load variation and the active output variation represented by the load parameter λ, resulting in:

wherein P is _Li ,Q _Li Respectively the active level and the reactive level of the load node i, P _Gi For the active power output of the power plant node i,active and reactive growth of load node i, respectively, < >>An active output increasing mode for the node i of the power generation equipment; the amplitude value of each node after the load is increased proportionally and the set Z formed by the phase angle sample sequence ₂ ＝{(v ₁ ,θ ₁ ),(v ₂ ,θ ₂ ),…,(v _n ,θ _n ) By Z ₁ And Z ₂ As a set of samples that identify weak nodes.

4. The reactive power compensation weak node identification method based on situation awareness in the power grid reactive power active prediction and control method according to claim 1 is characterized in that an index for quantifying node weakness is provided, through quantitative increase of node load, according to analysis of a PV curve, it is known that nodes which are prone to collapse in voltage are weak nodes, when load is increased, voltage and phase angle change are sensitive, the node weakness is measured by using node complex voltage change rate, the complex voltage change rate is defined as weighted summation of voltage change rate and phase angle change rate, before load active power reaches maximum, when load of a node j is increased by a certain proportion of active power, the node complex voltage change rate (Voltage Change Rate) is VCR, and the calculation is as follows:

wherein V is _j Represents the voltage value at the j node in normal operation, V' _j The voltage value at the j node after the load changes according to a given proportion is represented; θ _j Representing the voltage phase angle value at the j node in normal operation, theta' _j The voltage phase angle value at the j node after the load changes according to a given proportion is shown, and the larger the complex voltage weak index of the node is, the larger the weak of the node is shown by the formula (3).

5. The situation awareness-based power grid reactive power active prediction and control method is characterized in that the division of reactive power compensation weak areas in the step (2) is performed according to the number of reactive power devices, the spatial positions of nodes and the node weakness indexes, the power distribution network is divided in a partition mode, the concept of weak nodes is expanded to weak areas, each weak area corresponds to a plurality of nodes, the concept of a weak area index is provided, the weak area index is defined as VCRS, and the method is calculated as follows:

Q<＝N (5)

wherein VCRS _Q Is the Q-th weak area, i<＝j<K is the number of weak nodes in the weak area Q, V _j Representing the voltage value at the j node in the normal operation time zone, V' _j The voltage value at the j node after the load changes according to a given proportion is represented; θ _j Representing the voltage phase angle value at the j node in normal operation, theta' _j The voltage phase angle value at the j node after the load is changed according to a given proportion is represented, N is the number of reactive compensation device groups, and the formula (5) shows that the power distribution network is divided into different weak areas, so that each weak area is ensured to contain a reactive compensation device, and the greater the weak area index is, the greater the weak property of the weak area is; the number of the weak areas is smaller than that of the reactive compensation device groups, the weak areas are ordered according to the sizes of the weak area indexes, the reactive equipment with strong supplement capability is placed in the areas, and the reactive equipment is placed in the areas, and the weak indexes of all the nodes are ordered according to the sizes of the weak areas, so that the nodes with strong weakness are preferentially considered when the reactive compensation equipment is assembled.

6. The situation awareness based active prediction and control method for the power grid, according to claim 1, is characterized in that the situation prediction in the step (3) is mainly divided into two modules, wherein the first module mainly predicts the power generation of the distributed power supply, and the second module mainly predicts the load; the active power distribution network is mainly connected with a large amount of distributed energy, wherein the prediction of uncontrollable distributed energy has a great influence on the perception of system situation, and the intermittent energy in the active power distribution network mainly comprises photovoltaic power generation and wind power generation;

G＝G _b *cosθ (6)

wherein G is the total radiation received by the photovoltaic panelDegree, its unit is W/m2; g _b The direct solar illuminance is radiated to the earth in vacuum to be a constant value; θ is the angle between the incident solar ray and the normal line of the photovoltaic panel, and changes with time in one day;

the active output of the distributed photovoltaic is related to the illuminance and temperature conditions received by the photovoltaic panel, and the expression of the maximum value of the output active power can be obtained according to the physical model of the photovoltaic cell:

wherein U is _OC Is the photovoltaic open-circuit voltage under certain weather conditions, I _SC The photovoltaic short-circuit current is the photovoltaic short-circuit current under certain weather conditions; u (u) _OC Is the standard open circuit voltage, r _s Is a standard series resistor; under the condition of acquiring weather data, the maximum photovoltaic output power can be obtained according to the corresponding expression;

according to the circuit relation, the maximum power of photovoltaic power generation can be calculated by the formula (5) and the formula (6):

P _max ＝U _M ·I _M ＝a·U _OC ·I _SC (8)

wherein a is a parameter related to the photovoltaic device, U _OC 、I _SC The open-circuit voltage and the short-circuit current of the photovoltaic module are respectively calculated according to the formula (7);

wherein: g ^R Irradiance that is the standard test state; k is the temperature coefficient of the open-circuit voltage of the photovoltaic system; t (T) ^R Temperature for standard test conditions; t is the operating temperature of the photovoltaic system under the irradiation illuminance G;

as can be seen from the formula, the maximum value of the photovoltaic power generation power approximately varies as a sine function in one-day variation without considering the influence of atmospheric factors; in the actual power generation of the photovoltaic, fluctuation of the generated energy can be caused due to atmospheric refraction, cloud cover and rainfall; therefore, the photovoltaic power generation amount and the power generation amount influenced by uncertain weather factors have larger fluctuation, and the correlation is calculated through the Pearson correlation coefficient, as shown in the formula (8):

As shown by calculation, the radiation illuminance, the temperature and the humidity have strong correlation, but the general weather forecast does not predict the radiation illuminance, the difficulty of acquiring the data is high, the actual application is considered, the temperature and the humidity are adopted as weather elements for generating capacity prediction, and the input format of a sample is as follows

X _i ＝{T _i ,H _i ,t _i ,s _i } (11)

Wherein T is the surface temperature, H is the humidity, T is the sample collection date, s is the weather condition, and the temperature is 0, 1-1; respectively represents sunny, cloudy and rainy;

the basic structure of the wind power generation set is a wind machine, a doubly-fed induction generator and a converter; wind turbine converts wind energy into mechanical energy and transmits the mechanical energy to doubly-fed induction generator, generator converts mechanical energy into electric energy with amplitude and frequency change, and in normal operation of wind turbine, neglecting wake loss, hub loss and blade tip loss, the maximum power obtainable is:

wherein: c (C) _P.MAX The power coefficient is the maximum value, ρ is the air density, A is the fan sector area, and v is the wind speed; from the model, it can be seen that the factor of the wind power generation, which is most affected, is wind speed; considering practical application, wind speed is used as a weather element for generating capacity prediction, and the input format of a sample is as follows:

X _i ＝{v _i ,t _i ,s _i } (13)

wherein t is the sample acquisition date, s is the weather condition and is divided into 0,1 and-1; respectively represents sunny, cloudy and rainy;

Aiming at load prediction, social factors and weather factors have great influence on power load, analyzing the influence degree of temperature, weather conditions, electricity price and social development on load data, and giving specific influence factors; in the concrete analysis, the spearman correlation coefficient is shown as a formula (8), the correlation between temperature and humidity factors and the load is analyzed, and the sample input format of the load is as follows in combination with the correlation factors in the traditional load prediction method:

X _i ＝{t _i ,P _i ,T _i ,H _i ,s _i } (14)

wherein t is _i For sample acquisition date, P _i T is the active load _i To average temperature of the day, H _i For the average humidity s of the day _i Is weather condition, and is divided into 0,1 and-1; respectively represent sunny, cloudy and rainy days.

7. The situation awareness based active prediction and control method for power grid reactive power according to claim 6, wherein the prediction method is selected, the LSTM neural network is extended from the RNN neural network, the problems of gradient explosion and gradient disappearance which can occur on the RNN structure are solved, and the problem of error accumulation caused by the unidirectional structure of the LSTM in prediction is solved; analyzing the structure and principle of the LSTM neural network load prediction model, and giving a specific improvement method, namely a BLSTM neural network, aiming at the problem of error accumulation;

Firstly, replacing the traditional LSTM prediction by using a BLSTM, wherein the BLSTM is a BLSTM neural network combining positive phase and negative phase by adding a reverse prediction process on the structure of the LSTM, so that the authenticity of load data can be kept as far as possible in theory;

each layer of BLSTM neural network of the prediction model consists of a forward LSTM network and a reverse LSTM network, the input of the next layer of network is determined by the output results of the forward LSTM and the reverse LSTM of the upper layer, and the last layer of network modelThe result is determined by the forward and backward calculation results of each layer simultaneously, in the formulaThe values of the ith hidden layer at the time t-1 and the time t are respectively represented, and the weight is not shared due to forward calculation and reverse calculation, wherein V is calculated in the forward calculation ⁽ⁱ⁾ 、U ⁽ⁱ⁾ 、W ⁽ⁱ⁾ Representing the weight matrix from the ith hidden layer to the output layer, from the input layer to the hidden layer and between the hidden layers, and V 'during the reverse calculation' ⁽ⁱ⁾ 、U ^′(i) 、W ^′(i) Representing a corresponding inverse weight matrix; i represents the number of BLSTM layers, and i=0, 1,2 … ≡, represents the value of the output layer, the prediction model is as follows:

wherein the variables appearing in the formula are weights in BLSTM neurons, different input and output are constructed on the prediction model according to different prediction objects, the accuracy degree of the prediction result is evaluated by four evaluation indexes of mean square error, root mean square error, average absolute error and determination coefficient, and the prediction value is The true value is y= { y ₁ ,y ₂ ，y ₃ ，...，y _n -a }; wherein (1)>Representing a set of predicted values, y _i Representing a set of real values;

(1) mean Square Error (MSE)

the value range is [0, + ] and is equal to 0 when the predicted value is completely consistent with the true value, namely the perfect model; the larger the error, the larger the MSE value;

(2) root Mean Square Error (RMSE)

the value range is [0, + ] and is equal to 0 when the predicted value is completely consistent with the true value, namely the perfect model; the larger the error, the larger the value;

(3) mean Absolute Error (MAE)

Average absolute error (Mean Absolute Deviation), also called average absolute dispersion; the average absolute error can avoid the problem of mutual offset of errors, so that the magnitude of the actual prediction error can be accurately reflected, and the MAE is defined as follows:

the value range is [0, + ] and is equal to 0 when the predicted value is completely consistent with the true value, namely the perfect model; the larger the error, the larger the MAE value;

(4) Determining coefficient (R) ² )

8. The situation awareness-based power grid reactive power active prediction and control method is characterized in that in a future situation management and control stage in the step (4), the key of situation management and control is selection of a voltage regulation mode, and voltage regulation is performed by adopting a distributed power supply and power grid equipment;

wherein u is _s Representing the stator voltage amplitude; l (L) _s Representing stator self-inductance; l (L) _m Representing the mutual inductance between the stator and the rotor; i.e _rq ，i _rd Respectively representing the rotor currents of d axis and q axis under dq0 coordinate system; psi phi type _s Represents the stator flux linkage;

as can be seen from the formula (18), the control of the rotor current can realize the control of the active power and the reactive power of the rotor side respectively; the rotor side frequency converter is utilized for control, the rotating speed controller provides an active power reference value, and a reactive power reference value is determined according to the requirement of fan voltage control; the network side frequency converter is used for active and reactive decoupling, and generally adopts a voltage directional control scheme; the voltage directional control of the network side frequency converter is that the voltage vector is arranged on the shaft, and under the voltage reference coordinate system of the network side frequency converter, the power of the network side is as follows:

wherein i is _d ，i _q Representing the active current and the reactive current of the network side under the dq0 coordinate system respectively, it can be known from the formula (21) that the decoupling control of PQ can be realized by controlling i _d ，i _q To complete;

therefore, the doubly-fed generator set can output reactive power to the power grid through the stator side frequency converter and the grid side frequency converter at the same time, namely, the doubly-fed generator set can participate in reactive compensation of the power grid, and voltage control is realized;

for photovoltaic power generation in a distributed power supply, the basic principle is conversion from light energy to electric energy, wherein a voltage type inverter is adopted as an inverter; the grid connection of the photovoltaic power generation is completed, the photovoltaic array is required to realize the conversion from direct current to alternating current through the inverter, and the grid connection mode only has an inversion link, so that a control system of the inverter plays a very important role;

wherein Z represents the reactance between the grid and the inverter;representing a line impedance angle; u represents the voltage amplitude of the power grid side; u (U) _I Representing the magnitude of the inverter output voltage; phi represents U and U _I A phase difference between them;

as can be seen from formula (20), by adjusting U _I And phi can realize the adjustment of system power, so that the photovoltaic power generation system also has reactive power output capability, can play a role in reactive power compensation of a power grid, and realizes the adjustment of voltage;

for the voltage regulation control of the power grid equipment, a parallel capacitor, a Static Var Compensator (SVC) and an on-load voltage regulation transformer can be adopted, and the voltage regulation mode of the power grid equipment is fixed.

9. The situation awareness-based power grid reactive power active prediction and control method according to claim 1 is characterized in that future situation management and control in the step (4) is performed, after a reactive power voltage regulation mode is selected, the key is that an optimal compensation capacity is calculated through an optimization algorithm, the optimal reactive power compensation capacity in each reactive power compensation weak partition is calculated through an improved particle swarm algorithm, gradient information is not needed for the improved particle swarm algorithm (Particle Swarm Optimization, PSO), parameters are few, coding is not needed in practical application, and the method can be directly used; each particle will update the flight speed and the position at that moment based on the current extremum Pbest and the global extremum Gbest, as shown in equations (23), (24):

Wherein V is _i Is the speed; x is X _i Is the position; k is the number of iterations; individual learning factor c ₁ And social learning factor c ₂ The value is 2; r is (r) ₁ And r ₂ Is located at [0,1 ]]Random numbers within the interval;

the optimal reactive compensation scheme of the future situation of the system can be finally obtained by taking the node voltage optimum and the minimum network loss as objective functions;

the system has the minimum network loss:

wherein P is _loss (t) is the total network loss of the system at the moment t, r _ij For the resistance value of branch ij, I _ij (t) is the current of the branch ij at the moment t, and N is the distribution network line set;

node voltage deviation is minimal:

wherein the voltage deviation of the DeltaU (t) node at the time t;

and (3) load flow constraint:

wherein P is _i ，Q _i Representing the active power and the reactive power at the node i respectively; v (V) _i ,V _j Representing the voltages at nodes i, j, respectively; g _ij ，B _ij ，θ _ij Respectively representing the conductance, susceptance and phase difference between the nodes i and j;

voltage constraint:

U _min ≤U _i ≤U _max (28)

reactive compensation device constraint:

wherein Q is _DG.i.min 、Q _DG.i.max For the upper and lower limits of reactive power output of the distributed power supply at node i, Q _c.i.min 、Q _c.i.max The upper limit and the lower limit of the capacity of the reactive compensation equipment at the node i are set;

10. The method for actively predicting and controlling reactive power of a power grid based on situation awareness according to claim 9, wherein the situation awareness accuracy is considered from three aspects, namely voltage awareness accuracy, network loss awareness accuracy and reactive power compensation scheme awareness accuracy, and system node voltage, system network loss and prediction control scheme are obtained according to test set data to form a prediction situation assessment set:

W＝{V ₁ ,V ₂ ,…,V _m ,P _loss ,Q ₁ ,Q ₂ ,…,Q _N } (30)

where V is the voltage amplitude in the future situation, P _loss The system network loss in the future situation is represented by Q, and the reactive compensation capacity in the future situation is represented by Q;

the actual situation assessment set is

W _TRUTH ＝{V _T1 ,V _T2 ,…,V _Tm ,P _Tloss ,Q _T1 ,Q _T2 ,…,Q _TN } (31)

Wherein V is the voltage amplitude in the actual situation, P _loss The system network loss in the actual situation is adopted, and Q is the reactive compensation capacity in the actual situation;

by fitting goodness:

to determine the accuracy of situational awareness.