CN107194625B - Wind power plant wind curtailment electric quantity evaluation method based on neural network - Google Patents

Wind power plant wind curtailment electric quantity evaluation method based on neural network Download PDF

Info

Publication number
CN107194625B
CN107194625B CN201710613966.6A CN201710613966A CN107194625B CN 107194625 B CN107194625 B CN 107194625B CN 201710613966 A CN201710613966 A CN 201710613966A CN 107194625 B CN107194625 B CN 107194625B
Authority
CN
China
Prior art keywords
wind
layer
fan
neural network
output
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN201710613966.6A
Other languages
Chinese (zh)
Other versions
CN107194625A (en
Inventor
孙荣富
宁文元
王靖然
王若阳
徐海翔
张昊
李胜
钱苏晋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing E Techstar Co ltd
State Grid Corp of China SGCC
State Grid Jibei Electric Power Co Ltd
Original Assignee
Beijing E Techstar Co ltd
State Grid Corp of China SGCC
State Grid Jibei Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing E Techstar Co ltd, State Grid Corp of China SGCC, State Grid Jibei Electric Power Co Ltd filed Critical Beijing E Techstar Co ltd
Priority to CN201710613966.6A priority Critical patent/CN107194625B/en
Publication of CN107194625A publication Critical patent/CN107194625A/en
Application granted granted Critical
Publication of CN107194625B publication Critical patent/CN107194625B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • General Health & Medical Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Educational Administration (AREA)
  • General Business, Economics & Management (AREA)
  • Biophysics (AREA)
  • Operations Research (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Quality & Reliability (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Wind Motors (AREA)

Abstract

The invention provides a wind power plant abandoned wind power assessment method based on a neural network, which comprises the following steps: establishing a BP neural network model; training the BP neural network model by adopting the training sample data based on the basic parameters of the initialized BP neural network model in the step 2 to obtain the trained BP neural network model; and summing the external abandoned wind electric quantity of the whole wind field j at the moment t and the internal abandoned wind electric quantity of the whole wind field j at the moment t to obtain the whole abandoned wind electric quantity of the wind field j at the moment t. Has the advantages that: the wind power station abandoned wind power quantity can be simply, quickly and accurately evaluated, and therefore the power network dispatching center can be facilitated to better adjust the peak and the load flow.

Description

Wind power plant wind curtailment electric quantity evaluation method based on neural network
Technical Field
The invention belongs to the technical field of abandoned wind power evaluation, and particularly relates to a wind power plant abandoned wind power evaluation method based on a neural network.
Background
With the gradual maturity of wind power generation technology, wind power is highly emphasized by the development of national renewable energy, large-scale wind power plants are gradually connected to the grid, and wind power becomes the third main power source of China after thermal power and water power. However, due to the large-scale rapid unordered commissioning of wind power, the construction of a power grid architecture is relatively delayed, the coordination of a rapidly adjustable power supply in a power system is not coordinated, the wind resource is uncontrollable and intermittent, and the capacity of the rapidly adjustable power supply in the power system is limited, so that the wind power receiving capacity of the power grid is limited, and the wind curtailment situation is increased.
The difficulty in evaluating the abandoned wind power is increased due to the instability and randomness of the wind power, so that the abandoned wind power has a deviation problem. The accuracy of wind power abandonment evaluation has great influence on future wind power research and development. The abandoned wind is not beneficial to the optimal utilization of wind power resources. Therefore, accurately evaluating the amount of abandoned wind power becomes an important subject for modern power system enterprises and wind power enterprises to carry out evaluation work.
At present, relatively few researches on wind power curtailment wind power evaluation methods are carried out in China, and a great deal of research emphasis is placed on wind power consumption capacity and power grid dispatching peak regulation and load flow regulation. According to the existing domestic research results, the existing evaluation methods for the abandoned wind power quantity of the wind power plant mainly comprise 5 methods, namely a template method, a prediction curve method, a planning curve method, a power curve method and an area integral method. Then, these five methods all have the following problems: if the abandoned wind power quantity of the wind power plant needs to be evaluated, the theoretical power of the wind power at each time interval must be known firstly, but in practical application, the theoretical output data of the wind power plant is difficult to obtain due to the randomness of the wind power output, so that the calculation error of the abandoned wind power quantity is large, and therefore, an effective abandoned wind power quantity evaluation method is urgently sought.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a wind power plant abandoned wind power assessment method based on a neural network, which can effectively solve the problems.
The technical scheme adopted by the invention is as follows:
the invention provides a wind power plant abandoned wind power assessment method based on a neural network, which comprises the following steps:
step 1, establishing a BP neural network model; the topological structure of the BP neural network model comprises an input layer, a hidden layer and an output layer; the number of the neurons of the input layer is n, the number of the neurons of the hidden layer is l, and the number of the neurons of the output layer is m; any input layer neuron is xiI ∈ (1, 2 … n); any hidden layer neuron is hjJ ∈ (1, 2 … l); any output layer neuron is ok，k∈(1、2…m)；
Step 2, initializing basic parameters of the BP neural network model, comprising the following steps: learning rate μ, weight w of input layer to hidden layerijWeight w from hidden layer to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkAnd an excitation function t (x); wherein the weights w of the input layer to the hidden layerijWeight w from hidden layer to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkInitializing random numbers within a value of (-1, 1);
wherein: weights w for input layer to hidden layerijThe meaning is as follows: arbitrary input layer neurons xiTo arbitrary hidden layer neurons hjWeight in between; weight w from hidden layer to output layerjkThe meaning is as follows: arbitrary hidden layer neurons hjTo arbitrary output layer neuron okWeight in between; number of offsets a of input layer to hidden layerjThe meaning is as follows: input layer neurons to arbitrary hidden layer neurons hjThe offset number of (3); number of offsets b from hidden layer to output layerkThe meaning is as follows: transport of hidden layer neurons to arbitraryEpitopic neuron okOffset number of
Step 3, obtaining training sample data; the training sample data is three-dimensional data of historical operating states of a single fan, and the training sample data comprises the following steps: fan speed, fan direction and air density;
training the BP neural network model by adopting the training sample data based on the basic parameters of the initialized BP neural network model in the step 2 to obtain the trained BP neural network model;
the BP neural network model is trained by adopting the following method:
step 3.1, the input layer comprises three neurons, and each neuron of the input layer is respectively a fan wind speed, a fan wind direction and an air density of a fan in a power generation state;
hidden layer neuron h is calculated using the following formulajThe output value of (d):
Figure BDA0001360161150000031
calculating the output layer neuron o by adopting the following formulakThe output value of (d):
Figure BDA0001360161150000032
step 3.2, define the loss function as follows:
Figure BDA0001360161150000033
wherein: y iskThe expected output value of the neuron of the output layer is obtained, and the initial value is a historical actual active power value corresponding to each training sample data; e is a deviation value;
let ek＝yk-ok，ekThe deviation value corresponding to the k output layer neuron;
then E can be expressed as:
Figure BDA0001360161150000034
the output layer neuron o obtained by the calculation in the step 3.2kSubstituting the output value into a loss function, and calculating to obtain a deviation value E;
step 3.3, judging whether the deviation value E meets the requirement, and if so, turning to the step 3.10; if not, turning to step 3.4;
step 3.4, calculating the weight adjustment amount from the hidden layer to the output layer by adopting the following formula:
Δwjk(q+1)＝(1-γ)hjek+γΔwjk(q)
wherein:
Figure BDA0001360161150000041
wherein: gamma is the weight inertia coefficient, eqAnd eq-1Q and q-1 training errors, respectively; Δ wjk(q)For q training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a); Δ wjk(q+1)For the q +1 training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a);
and 3.5, calculating the weight adjustment amount from the input layer to the hidden layer by adopting the following formula:
Figure BDA0001360161150000042
wherein: gamma is a weight inertia coefficient; Δ wij(q)For input layer neuron x at qth trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a); Δ wij(q+1)For input layer neuron x at q +1 trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a);
step 3.6, calculate the offset b using the following equationkUpdate value of (d):
bk＝bk+μek
step 3.7, calculate the offset a using the following equationkUpdate value of (d):
Figure BDA0001360161150000043
step 3.8, therefore, the weight adjustment amount from the hidden layer to the output layer calculated in step 3.4, the weight adjustment amount from the input layer to the hidden layer calculated in step 3.5, and the offset number b calculated in step 3.6 are adoptedkAnd the offset number a calculated in step 3.7kThe updated value of the BP neural network model is optimized and adjusted to the corresponding parameters of the BP neural network model obtained by the previous training, so that the updated BP neural network model is obtained;
step 3.9, based on the updated BP neural network model obtained in step 3.8, returning to step 3.1,
step 3.10, obtaining a trained BP neural network model;
step 4, a plurality of fans exist in the wind field, wherein each fan comprises the following 9 states, namely a wind waiting state, a power generation state, a derating power generation state, a planned outage state, an unplanned outage state, a scheduling outage state, a communication interruption state, an on-site accumulated outage state and an off-site accumulated outage state; according to the historical operating data of each fan, a trained BP neural network model corresponding to the fan can be obtained through training;
therefore, if the single fan i is in a derating power generation state or a dispatching shutdown state, the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t are input into the trained BP neural network model corresponding to the single fan i, and the theoretical active power output P of the single fan i at the moment t is obtainedTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating and obtaining the off-site abandoned wind electric quantity Q of the single fan i at the moment t according to the following formulaOutdoors t.i：
Figure BDA0001360161150000051
Therefore, the off-site abandoned wind electric quantity Q of the whole-site abandoned wind electric quantity Q of the wind field j at the moment t is obtained by summing all the off-site abandoned wind electric quantities of the single fans in the wind field j at the moment t in the derated power generation state or the dispatching outage stateOutdoors t.j；
If the single fan i is in a wind waiting state, a power generation state, a planned outage state, an unplanned outage state, a communication interruption state, an on-site accumulated outage state and an off-site accumulated outage state, the single fan i
Inputting the wind speed, wind direction and air density of the fan at the moment t into a trained BP neural network model corresponding to the single fan i to obtain the theoretical active power P of the single fan i at the moment tTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating according to the following formula to obtain the abandoned wind electric quantity Q of the single fan i in the field at the moment tIn-site t.i：
Figure BDA0001360161150000052
Therefore, the wind curtailment electricity quantity Q in the wind field j at the time t is obtained by summing the wind curtailment electricity quantities in the field of the single fan in the wind field j at the time t, wherein the wind curtailment electricity quantity Q is in a wind waiting state, a power generation state, a planned outage state, an unplanned outage state, a communication interruption state, an accumulated outage state in the field and an accumulated outage state outside the fieldIn the ground t.j；
Abandoning the wind field j at the moment t outside the whole field by the wind power QOutdoors t.jWind field j at time t and abandoned wind power Q in whole fieldIn the ground t.jAnd summing to obtain the wind abandoning magnitude value of the whole wind field j at the moment t.
Preferably, the number of neurons in the input layer is 3; the number of neurons in the output layer is 1.
Preferably, in step 3, the training sample data is normalized training sample data,
the normalized calculation mode is as follows:
Figure BDA0001360161150000061
wherein: xmax，XminMaximum and minimum values of the actual variable, X, respectively*X is the actual value for the normalized value.
Preferably, in step 4, the fan wind speed, the fan wind direction and the air density of the single fan i at the time t are input into the trained BP neural network model corresponding to the single fan i, so as to obtain the theoretical active power output P of the single fan i at the time tTheory t.iThe method specifically comprises the following steps:
inputting the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t into a trained BP neural network model corresponding to the single fan i, and performing inverse normalization calculation on theoretical active output by the BP neural network model to obtain the theoretical active output PTheory t.i；
And (3) an inverse normalization calculation mode:
X＝X*×(Xmax-Xmin)+Xmin。
the wind power plant abandoned wind power evaluation method based on the neural network has the following advantages:
the wind power station abandoned wind power quantity can be simply, quickly and accurately evaluated, and therefore the power network dispatching center can be facilitated to better adjust the peak and the load flow.
Drawings
FIG. 1 is a schematic flow chart of a wind curtailment electric quantity evaluation method of a wind power plant based on a neural network, provided by the invention;
FIG. 2 is a relationship between an actual wind speed and a wind power of a wind farm.
Fig. 3 is a graph of an actual power curve and a theoretical output curve of a wind field.
Fig. 4 is a graph comparing the effect of the actual and theoretical power curves.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects solved by the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Interpretation of terms:
1. wind power abandon wind power:
and subtracting the actual on-grid generated energy of the fan from the theoretical generated energy of the fan at the corresponding wind speed to obtain the abandoned wind electric quantity.
2. Wind power single machine:
the wind power plant consists of a plurality of independent wind power generators, each wind power generator comprises a plurality of related information such as wind speed, active power, state and the like, and each fan and the information of the fan are regarded as a wind power single-machine object.
The invention provides a wind power plant abandoned wind power assessment method based on a neural network, which comprises the following steps:
the BP neural network provided by the invention is a multi-layer feedforward network trained according to an error inverse propagation algorithm, and the learning rule is to use a gradient descent method to continuously adjust the weight and the threshold value of the network through inverse propagation so as to minimize the sum of squares of errors of the network. The learning process of the BP algorithm (back propagation algorithm) consists of two processes of forward propagation of information and back propagation of errors. Each neuron of the input layer is responsible for receiving input information from the outside and transmitting the input information to each neuron of the middle layer; the middle layer is an internal information processing layer and is responsible for information transformation, and can be designed into a single-hidden layer or multi-hidden layer structure according to the requirement of information change capability; and the information transmitted to each neuron of the output layer by the last hidden layer is further processed to finish a forward propagation processing process of learning once, and an information processing result is output to the outside by the output layer. When the actual output does not match the desired output, the error back-propagation phase is entered. And the error passes through the output layer, the weight of each layer is corrected in a mode of error gradient reduction, and the error is reversely transmitted to the hidden layer and the input layer by layer. The repeated information forward propagation and error backward propagation process makes the weights of all layers continuously adjusted, and is also the process of neural network learning training, and the process is carried out until the error output by the network is reduced to an acceptable degree or preset learning times.
The learning process comprises the following steps: the neural network continuously changes the connection weight of the network under the stimulation of external input samples, so that the output of the network is continuously close to the expected output.
The essence of learning is as follows: and dynamically adjusting each connection weight.
And (3) learning rules: the weight value adjusting rule is a certain adjusting rule according to which the connection weight of each neuron in the network changes in the learning process.
The core idea is as follows: the output error is transmitted back to the input layer by layer through the hidden layer in a certain mode.
Step 1, establishing a BP neural network model; the topological structure of the BP neural network model comprises an input layer, a hidden layer and an output layer; the number of the neurons of the input layer is n, the number of the neurons of the hidden layer is l, and the number of the neurons of the output layer is m; any input layer neuron is xiI ∈ (1, 2 … n); any hidden layer neuron is hjJ ∈ (1, 2 … l); any output layer neuron is ok，k∈(1、2…m)；
Aiming at the invention, the topological structure of the BP neural network model comprises three layers, namely an input layer, a hidden layer and an output layer; wherein, the input layer comprises three neurons, the hidden layer comprises 15 neurons (the error of the test is the minimum), and the output layer comprises one neuron.
And testing the performance of the model by adopting a log-sigmod function and an n-fold cross validation mode in the hidden layer, setting the standard error to be 0.01 percent, setting the upper limit of the training times to be 5000 times, and if the error is less than 0.01 percent or the training times reach the upper limit of 5000 times, determining that the training is finished.
Step 2, initializing basic parameters of the BP neural network model, comprising the following steps: learning rate μ, weight w of input layer to hidden layerijWeight w from hidden layer to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkAnd an excitation function t (x); wherein the weights w of the input layer to the hidden layerijHidden layerWeight w to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkInitializing random numbers within a value of (-1, 1);
wherein: weights w for input layer to hidden layerijThe meaning is as follows: arbitrary input layer neurons xiTo arbitrary hidden layer neurons hjWeight in between; weight w from hidden layer to output layerjkThe meaning is as follows: arbitrary hidden layer neurons hjTo arbitrary output layer neuron okWeight in between; number of offsets a of input layer to hidden layerjThe meaning is as follows: input layer neurons to arbitrary hidden layer neurons hjThe offset number of (3); number of offsets b from hidden layer to output layerkThe meaning is as follows: each hidden layer neuron to an arbitrary output layer neuron okOffset number of
Step 3, obtaining training sample data; the training sample data is three-dimensional data of historical operating states of a single fan, and the training sample data comprises the following steps: fan speed, fan direction and air density;
for example, the operation data (minute level) of each time point of the whole year of a single fan can be selected as sample data, the input characteristics are wind speed, wind direction and air density, and the output characteristics are theoretical active power values.
Training the BP neural network model by adopting the training sample data based on the basic parameters of the initialized BP neural network model in the step 2 to obtain the trained BP neural network model;
the BP neural network model is trained by adopting the following method:
step 3.1, the input layer comprises three neurons, and each neuron of the input layer is respectively a fan wind speed, a fan wind direction and an air density of a fan in a power generation state;
hidden layer neuron h is calculated using the following formulajThe output value of (d):
Figure BDA0001360161150000091
calculating the output layer neuron o by adopting the following formulakThe output value of (d):
Figure BDA0001360161150000092
step 3.2, define the loss function as follows:
Figure BDA0001360161150000093
wherein: y iskThe expected output value of the neuron of the output layer is obtained, and the initial value is a historical actual active power value corresponding to each training sample data; e is a deviation value;
let ek＝yk-ok，ekThe deviation value corresponding to the k output layer neuron;
then E can be expressed as:
Figure BDA0001360161150000101
the output layer neuron o obtained by the calculation in the step 3.2kSubstituting the output value into a loss function, and calculating to obtain a deviation value E;
step 3.3, judging whether the deviation value E meets the requirement, and if so, turning to the step 3.10; if not, turning to step 3.4;
and 3.4, firstly processing weights from the hidden layer to the output layer, and solving a partial derivative of the weights of the errors by using a gradient descent method in order to minimize the errors.
Figure BDA0001360161150000102
Therefore, the weight adjustment from the hidden layer to the output layer is calculated by the following formula:
Δwjk(q+1)＝(1-γ)hjek+γΔwjk(q)
wherein:
Figure BDA0001360161150000103
wherein: gamma is the weight inertia coefficient, eqAnd eq-1Q and q-1 training errors, respectively; Δ wjk(q)For q training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a); Δ wjk(q+1)For the q +1 training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a);
and 3.5, carrying out weight processing from the input layer to the hidden layer, and solving the partial derivative of the weight.
Figure BDA0001360161150000111
In the above formula:
Figure BDA0001360161150000112
Figure BDA0001360161150000113
therefore, the weight adjustment from the input layer to the hidden layer is calculated by the following formula:
Figure BDA0001360161150000114
wherein: gamma is a weight inertia coefficient; Δ wij(q)For input layer neuron x at qth trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a); Δ wij(q+1)For input layer neuron x at q +1 trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a);
step 3.6, updating the offset number:
Figure BDA0001360161150000115
the offset b is calculated using the formulakUpdate value of (d):
bk＝bk+μek
in the step 3.7, the step of the method,
Figure BDA0001360161150000116
in the above formula:
Figure BDA0001360161150000121
Figure BDA0001360161150000122
the offset a is calculated using the following equationkUpdate value of (d):
Figure BDA0001360161150000123
step 3.8, therefore, the weight adjustment amount from the hidden layer to the output layer calculated in step 3.4, the weight adjustment amount from the input layer to the hidden layer calculated in step 3.5, and the offset number b calculated in step 3.6 are adoptedkAnd the offset number a calculated in step 3.7kThe updated value of the BP neural network model is optimized and adjusted to the corresponding parameters of the BP neural network model obtained by the previous training, so that the updated BP neural network model is obtained;
step 3.9, based on the updated BP neural network model obtained in step 3.8, returning to step 3.1,
step 3.10, obtaining a trained BP neural network model;
step 4, a plurality of fans exist in the wind field, wherein each fan comprises the following 9 states, namely a wind waiting state, a power generation state, a derating power generation state, a planned outage state, an unplanned outage state, a scheduling outage state, a communication interruption state, an on-site accumulated outage state and an off-site accumulated outage state; according to the historical operating data of each fan, a trained BP neural network model corresponding to the fan can be obtained through training;
therefore, if the single fan i is in a derating power generation state or a dispatching shutdown state, the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t are input into the trained BP neural network model corresponding to the single fan i, and the theoretical active power output P of the single fan i at the moment t is obtainedTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating and obtaining the off-site abandoned wind electric quantity Q of the single fan i at the moment t according to the following formulaOutdoors t.i：
Figure BDA0001360161150000131
Therefore, the off-site abandoned wind electric quantity Q of the whole-site abandoned wind electric quantity Q of the wind field j at the moment t is obtained by summing all the off-site abandoned wind electric quantities of the single fans in the wind field j at the moment t in the derated power generation state or the dispatching outage stateOutdoors t.j；
If the single fan i is in a wind waiting state, a power generation state, a planned outage state, an unplanned outage state, a communication interruption state, an on-site accumulated outage state and an off-site accumulated outage state, the single fan i
Inputting the wind speed, wind direction and air density of the fan at the moment t into a trained BP neural network model corresponding to the single fan i to obtain the theoretical active power P of the single fan i at the moment tTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating according to the following formula to obtain the abandoned wind electric quantity Q of the single fan i in the field at the moment tIn-site t.i：
Figure BDA0001360161150000132
Therefore, all of wind farm j at time t will beSumming the wind power quantities of the single fans in the wind waiting state, the power generation state, the planned outage state, the unplanned outage state, the communication interruption state, the accumulated outage state in the wind farm and the accumulated outage state outside the wind farm to obtain the wind farm j at the moment t and the wind power quantity Q of the wind farm in the whole wind farmIn the ground t.j；
Abandoning the wind field j at the moment t outside the whole field by the wind power QOutdoors t.jWind field j at time t and abandoned wind power Q in whole fieldIn the ground t.jAnd summing to obtain the wind abandoning magnitude value of the whole wind field j at the moment t.
In the invention, before data input is carried out on the model, reasonable normalization processing needs to be carried out on the original data, and whether the processing mode is reasonable or not directly influences the prediction effect.
Analyzing influence factors: the fan output power may be represented by:
Figure BDA0001360161150000133
in the formula: pwOutputting power for the wind wheel; ρ is the air density; a is the swept area of the wind wheel; cpIs the wind wheel power coefficient; v is wind speed; by comprehensively analyzing the magnitude of the wind power output influence factors, the wind speed, the wind direction and the air density (determined by the air pressure, the temperature and the humidity) are selected as input variables of the model.
By analyzing the relationship between the wind power of the sample board machine and the wind speed, the wind direction and the air density, the output power variation range of the wind power plant corresponding to the same wind speed is large, mainly because the wind turbines are distributed in a wide geographical range, the wind speed and the wind power have spatial correlation, and the air density and the influence factors thereof also have spatial correlation with the power.
By analyzing the graph 2, the wind speed and the wind power have nonlinear correlation, but certain probability confidence exists, so that the method is suitable for training and modeling by adopting a neural network mode.
In step 3, the training sample data is normalized training sample data,
the normalized calculation mode is as follows:
Figure BDA0001360161150000141
wherein: xmax，XminMaximum and minimum values of the actual variable, X, respectively*X is the actual value for the normalized value.
And after the basic data of the wind turbine is normalized, normalizing the basic data to be between 0 and 1.
In step 4, inputting the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t into the trained BP neural network model corresponding to the single fan i to obtain the theoretical active power output P of the single fan i at the moment tTheory.The method specifically comprises the following steps:
inputting the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t into a trained BP neural network model corresponding to the single fan i, and performing inverse normalization calculation on theoretical active output by the BP neural network model to obtain the theoretical active output PTheory t.i；
And (3) an inverse normalization calculation mode:
X＝X*×(Xmax-Xmin)+Xmin。
taking annual wind power data of all wind power plants in a certain province as an example, the annual wind power data are used as training samples, all fans of the wind power plants have 9 states, namely waiting wind, generating power, derating power generation (dispatching power limit), planned outage, unplanned outage, dispatching outage (dispatching power limit), communication interruption, accumulated outage in the wind power plants and accumulated outage outside the wind power plants, and for each fan, only the state data when the state of the fan is power generation is screened as the training sample data.
For the screened data samples, in order to improve the learning precision and efficiency of the artificial neural network and solve the problem of neuron saturation, data normalization processing is required to be carried out, the wind speed and the air density are normalized to the range of [0,1] according to the historical maximum value, and the wind direction can be normalized to [0,1] by selecting the positive selection mode and the cosine mode of the wind direction. Normalization is also performed for the target values.
Brief summary of wind abandonment procedure
The system allows for data statistics time intervals: 1 minute by one point (1440 points all day), 5 minute by one point (288 points all day), or 15 minute by one point (96 points all day), described below with 1440 points all day as a reference, the system could also be modified to 288 points or 96 points.
At present, the states of the fans are divided into 9 states of waiting for wind, generating power, scheduling derating power generation, scheduling outage, unplanned outage, scheduling outage, communication interruption and accumulated outage. And drawing theoretical output, actual output, wind abandon in the wind field and wind abandon out of the wind field curves and calculating corresponding electric quantity according to the running state of the fan by adopting a single-machine information method. The main calculation idea is as follows (taking the system time interval as a calculation unit, currently 1 minute): 1) and acquiring the active value of the fan, namely the actual output. 2) And calculating to obtain theoretical output according to the wind speed, the wind direction and the density of the engine room. 3) And calculating the electric quantity of the abandoned wind outside the field in the field according to different states of the fan.
In the above calculation content, the theoretical output value is obtained by adopting an artificial intelligent BP neural network self-learning system to analyze and fit historical operating data of each fan, and the theoretical output value under the corresponding wind speed of each fan can be obtained as long as the wind speed, the wind direction and the density value are obtained. The precision of the theoretical active relation between the wind speed given by the system and the theoretical active relation is 0.01m/s in the aspect of the wind speed, the range is 0-20 m/s, and the rated output of the fan can be achieved when the fan is 10-13 m/s generally.
The problem of wind curtailment is more and more severe due to the large-scale rapid unordered grid connection of wind power and the limitation of wind power consumption capacity, and the finding of an effective assessment method for the wind curtailment power becomes an important subject for better peak load regulation and load flow regulation of a power network dispatching center. The method of the sample board machine, the method of the prediction curve, the method of the plan curve, the method of the power curve and the method of the area integration all show certain advantages when estimating the abandoned wind electric quantity of the wind power plant, but have a plurality of disadvantages. Therefore, it is increasingly important to provide some reasonable, simple and practical methods for estimating the wind curtailment power.
Due to the fluctuation, randomness and uncontrollable property of wind resources and the complexity of objective factors of a wind field, when the evaluation work of the abandoned wind power of the wind power plant is carried out according to local conditions, the factors of the output of the wind power plant are also considered, so that the evaluation accuracy of the abandoned wind power is gradually improved. Through the fan theoretical power algorithm of the system, model training is carried out through historical mass data, the influence of physical performance and spatial factor difference existing in a single fan is deeply excavated, the size of the power to be sent under the condition of corresponding to an external complex mixed working condition can be accurately judged, and the theoretical output of each single fan is accurately restored; the mode of multi-state analysis and dynamic matching calculation of the abandoned wind power of the single machine can accurately judge the abandoned wind calculation state of the fan, well make up for the error of unified and uniform calculation training from the full field angle, and more truly restore the real abandoned wind power of the station. The actual power curve and the theoretical output curve of a certain wind field are shown in FIG. 3: the effect of the actual and theoretical power curves is shown in fig. 4.
The wind power plant abandoned wind power evaluation method based on the neural network has the following advantages:
the wind power station abandoned wind power quantity can be simply, quickly and accurately evaluated, and therefore the power network dispatching center can be facilitated to better adjust the peak and the load flow.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the principle of the present invention, and such modifications and improvements should also be considered within the scope of the present invention.

Claims (1)

1. A wind power plant curtailment electric quantity evaluation method based on a neural network is characterized by comprising the following steps:
step 1, establishing a BP neural network model; the topological structure of the BP neural network model comprises an input layer, a hidden layer and an output layer; the number of the neurons of the input layer is n, the number of the neurons of the hidden layer is l, and the number of the neurons of the output layer is m; any input layer neuron is xiI ∈ (1, 2.. n); neuron of arbitrary hidden layer ishjJ ∈ (1, 2.. l); any output layer neuron is ok，k∈(1、2...m)；
Step 2, initializing basic parameters of the BP neural network model, comprising the following steps: learning rate μ, weight w of input layer to hidden layerijWeight w from hidden layer to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkAnd an excitation function t (x); wherein the weights w of the input layer to the hidden layerijWeight w from hidden layer to output layerjkThe number of offsets a of the input layer to the hidden layerjBias number b from hidden layer to output layerkInitializing random numbers within a value of (-1, 1);
wherein: weights w for input layer to hidden layerijThe meaning is as follows: arbitrary input layer neurons xiTo arbitrary hidden layer neurons hjWeight in between; weight w from hidden layer to output layerjkThe meaning is as follows: arbitrary hidden layer neurons hjTo arbitrary output layer neuron okWeight in between; number of offsets a of input layer to hidden layerjThe meaning is as follows: input layer neurons to arbitrary hidden layer neurons hjThe offset number of (3); number of offsets b from hidden layer to output layerkThe meaning is as follows: each hidden layer neuron to an arbitrary output layer neuron okOffset number of
Step 3, obtaining training sample data; the training sample data is three-dimensional data of historical operating states of a single fan, and the training sample data comprises the following steps: fan speed, fan direction and air density;
training the BP neural network model by adopting the training sample data based on the basic parameters of the initialized BP neural network model in the step 2 to obtain the trained BP neural network model;
the BP neural network model is trained by adopting the following method:
step 3.1, the input layer comprises three neurons, and each neuron of the input layer is respectively a fan wind speed, a fan wind direction and an air density of a fan in a power generation state;
by usingHidden layer neuron h is calculated by the following formulajThe output value of (d):
Figure FDA0002543000240000021
calculating the output layer neuron o by adopting the following formulakThe output value of (d):
Figure FDA0002543000240000022
step 3.2, define the loss function as follows:
Figure FDA0002543000240000023
wherein: y iskThe expected output value of the neuron of the output layer is obtained, and the initial value is a historical actual active power value corresponding to each training sample data; e is a deviation value;
let ek＝yk-ok，ekThe deviation value corresponding to the k output layer neuron;
then E can be expressed as:
Figure FDA0002543000240000024
the output layer neuron o obtained by the calculation in the step 3.2kSubstituting the output value into a loss function, and calculating to obtain a deviation value E;
step 3.3, judging whether the deviation value E meets the requirement, and if so, turning to the step 3.10; if not, turning to step 3.4;
step 3.4, calculating the weight adjustment amount from the hidden layer to the output layer by adopting the following formula:
Δwjk(q+1)＝(1-γ)hjek+γΔwjk(q)
wherein:
Figure FDA0002543000240000031
wherein: gamma is the weight inertia coefficient, eqAnd eq-1Q and q-1 training errors, respectively; Δ wjk(q)For q training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a); Δ wjk(q+1)For the q +1 training hidden layer neuron hjTo output layer neuron okThe weight adjustment amount of (a);
and 3.5, calculating the weight adjustment amount from the input layer to the hidden layer by adopting the following formula:
Figure FDA0002543000240000032
wherein: gamma is a weight inertia coefficient; Δ wij(q)For input layer neuron x at qth trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a); Δ wij(q+1)For input layer neuron x at q +1 trainingiNeurons h to the hidden layerjThe weight adjustment amount of (a);
step 3.6, calculate the offset b using the following equationkUpdate value of (d):
bk＝bk+μek
step 3.7, calculate the offset a using the following equationkUpdate value of (d):
Figure FDA0002543000240000041
step 3.8, therefore, the weight adjustment amount from the hidden layer to the output layer calculated in step 3.4, the weight adjustment amount from the input layer to the hidden layer calculated in step 3.5, and the offset number b calculated in step 3.6 are adoptedkAnd the offset number a calculated in step 3.7kIs obtained by optimizing and adjusting the updated value of the previous trainingObtaining the updated BP neural network model according to the corresponding parameters of the BP neural network model;
step 3.9, based on the updated BP neural network model obtained in step 3.8, returning to step 3.1,
step 3.10, obtaining a trained BP neural network model;
step 4, a plurality of fans exist in the wind field, wherein each fan comprises the following 9 states, namely a wind waiting state, a power generation state, a derating power generation state, a planned outage state, an unplanned outage state, a scheduling outage state, a communication interruption state, an on-site accumulated outage state and an off-site accumulated outage state; according to the historical operating data of each fan, a trained BP neural network model corresponding to the fan can be obtained through training;
therefore, if the single fan i is in a derating power generation state or a dispatching shutdown state, the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t are input into the trained BP neural network model corresponding to the single fan i, and the theoretical active power output P of the single fan i at the moment t is obtainedTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating and obtaining the off-site abandoned wind electric quantity Q of the single fan i at the moment t according to the following formulaOutdoors t.i：
Figure FDA0002543000240000042
Therefore, the off-site abandoned wind electric quantity Q of the whole-site abandoned wind electric quantity Q of the wind field j at the moment t is obtained by summing all the off-site abandoned wind electric quantities of the single fans in the wind field j at the moment t in the derated power generation state or the dispatching outage stateOutdoors t.j；
If the single fan i is in a wind waiting state, a power generation state, a planned outage state, an unplanned outage state, a communication interruption state, an in-site accumulated outage state and an out-site accumulated outage state, inputting the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t into a trained BP neural network model corresponding to the single fan i to obtain the single fan iTheoretical active power P of fan i at time tTheory t.i(ii) a Then obtaining the actual active power output P of the single fan i at the moment tPractice t.iCalculating according to the following formula to obtain the abandoned wind electric quantity Q of the single fan i in the field at the moment tIn the ground t.i：
Figure FDA0002543000240000051
Therefore, the wind curtailment electricity quantity Q in the wind field j at the time t is obtained by summing the wind curtailment electricity quantities in the field of the single fan in the wind field j at the time t, wherein the wind curtailment electricity quantity Q is in a wind waiting state, a power generation state, a planned outage state, an unplanned outage state, a communication interruption state, an accumulated outage state in the field and an accumulated outage state outside the fieldIn the ground t.j；
Abandoning the wind field j at the moment t outside the whole field by the wind power QOutdoors t.jWind field j at time t and abandoned wind power Q in whole fieldIn the ground t.jSumming to obtain the total wind abandoning magnitude value of the wind field j at the moment t;
wherein, the number of the neurons of the input layer is 3; the number of the neurons of the output layer is 1;
wherein, in step 3, the training sample data is normalized training sample data,
the normalized calculation mode is as follows:
Figure FDA0002543000240000052
wherein: xmax，XminMaximum and minimum values of the actual variable, X, respectively*Is a normalized value, and X is an actual value;
in step 4, inputting the fan wind speed, the fan wind direction and the air density of the single fan i at the moment t into a trained BP neural network model corresponding to the single fan i to obtain the theoretical active power output P of the single fan i at the moment tTheory t.iThe method specifically comprises the following steps:
fan with single fan i at time tInputting the wind speed, the wind direction of the fan and the air density into a trained BP neural network model corresponding to a single fan i, and performing inverse normalization calculation on theoretical active output by the BP neural network model to obtain the theoretical active output PTheory t.i；
And (3) an inverse normalization calculation mode:
X＝X*×(Xmax-Xmin)+Xmin。
CN201710613966.6A 2017-07-25 2017-07-25 Wind power plant wind curtailment electric quantity evaluation method based on neural network Expired - Fee Related CN107194625B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710613966.6A CN107194625B (en) 2017-07-25 2017-07-25 Wind power plant wind curtailment electric quantity evaluation method based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710613966.6A CN107194625B (en) 2017-07-25 2017-07-25 Wind power plant wind curtailment electric quantity evaluation method based on neural network

Publications (2)

Publication Number Publication Date
CN107194625A CN107194625A (en) 2017-09-22
CN107194625B true CN107194625B (en) 2020-11-24

Family

ID=59884247

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710613966.6A Expired - Fee Related CN107194625B (en) 2017-07-25 2017-07-25 Wind power plant wind curtailment electric quantity evaluation method based on neural network

Country Status (1)

Country Link
CN (1) CN107194625B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107732962B (en) * 2017-09-29 2020-12-01 国网辽宁省电力有限公司 Abandoned wind reduction method based on ultra-short term abandoned wind curve prediction
CN108090608A (en) * 2017-12-13 2018-05-29 上海海事大学 A kind of gantry crane trend prediction method based on BP neural network
CN108718297A (en) * 2018-04-27 2018-10-30 广州西麦科技股份有限公司 Ddos attack detection method, device, controller and medium based on BP neural network
CN109146323B (en) * 2018-09-12 2021-08-03 国网辽宁省电力有限公司 Fan efficiency evaluation method and device and computer storage medium
CN109695546A (en) * 2018-12-29 2019-04-30 中国大唐集团新能源科学技术研究院有限公司 A kind of universal fan monitoring method
CN111489046A (en) * 2019-01-29 2020-08-04 广东省公共卫生研究院 Regional food safety evaluation model based on supply chain and BP neural network
CN110118926A (en) * 2019-05-27 2019-08-13 电子科技大学 PCB based on Electromagnetic Environmental Effect distorts intelligent detecting method
CN112685954B (en) * 2020-12-29 2023-10-03 中国航天空气动力技术研究院 Method and device for predicting wind speed and fan rotating speed of automobile environment wind tunnel
CN112784215B (en) * 2021-01-21 2022-04-15 中国三峡新能源(集团)股份有限公司 Wind power plant theoretical generating capacity calculation method based on real-time power remote measurement
CN113848706B (en) * 2021-09-13 2024-01-12 无锡宏源机电科技股份有限公司 Silk thread tension detection method, control method and control device
CN117115377A (en) * 2023-10-18 2023-11-24 云南滇能智慧能源有限公司 Wind farm energy model creation method, device, equipment and storage medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105205564A (en) * 2015-09-29 2015-12-30 沈阳工程学院 Wind power plant wind curtailment electric quantity statistical system and method based on anemometer tower neural network

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105205564A (en) * 2015-09-29 2015-12-30 沈阳工程学院 Wind power plant wind curtailment electric quantity statistical system and method based on anemometer tower neural network

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
BP神经网络在图像字符识别中的改进和应用;张永宏等;《南京信息工程大学学报自然科学版》;20120630;527 *
Valuation framework for large scale electricity storage in a case with wind curtailment;Loisel R et al.;《Energy policy》;20101231;7323-7337 *
基于多任务学习的自然图像分类研究;刘成等;《计算机应用研究》;20120731;2273-2274 *
弃风电量评估方法的研究现状及技术展望;钟宏宇;《电器与能效管理技术》;20160430(第4期);33—37 *
风电场弃风电量评估方法的研究;钟宏宇;《中国优秀硕士学位论文全文数据库 工程科技II辑》;20170315;C042—1202 *

Also Published As

Publication number Publication date
CN107194625A (en) 2017-09-22

Similar Documents

Publication Publication Date Title
CN107194625B (en) Wind power plant wind curtailment electric quantity evaluation method based on neural network
CN109117951B (en) BP neural network-based probability load flow online calculation method
CN102102626B (en) Method for forecasting short-term power in wind power station
Yu et al. The forecast of the electrical energy generated by photovoltaic systems using neural network method
Rizwan et al. Artificial intelligence based approach for short term load forecasting for selected feeders at madina saudi arabia
CN111680823A (en) Wind direction information prediction method and system
Alblawi et al. PV solar power forecasting based on hybrid MFFNN-ALO
Arora et al. Wind speed forecasting model for northern-western region of India using decision tree and multilayer perceptron neural network approach
CN111488974B (en) Ocean wind energy downscaling method based on deep learning neural network
Wang et al. Rolling forecast nature gas spot price with back propagation neural network
CN115713029A (en) Wind power plant stochastic model prediction optimization control method considering delay
Chen et al. Variation-cognizant probabilistic power flow analysis via multi-task learning
CN113642784B (en) Wind power ultra-short-term prediction method considering fan state
CN101929684A (en) Coal consumption calculating method of composite firing low heat value gas unit
CN108345996A (en) A kind of system and method reducing wind power checking energy
Kehe et al. Research of wind power prediction model based on RBF neural network
CN111027816B (en) Photovoltaic power generation efficiency calculation method based on data envelope analysis
CN111461297A (en) Solar irradiation quantity optimization prediction algorithm based on MPC and E L M neural network
CN109494747A (en) A kind of power grid probability load flow calculation method based on alternating gradient algorithm
Wang et al. A new improved combined model algorithm for the application of photovoltaic power prediction
Han et al. Distributed Control Strategy of Renewable Energy Clusters Considering the Uncertainty of Output
Zeng et al. Wind speed prediction based on improved grey neural network model
Mohammed et al. Ultra-short-term wind power prediction using a hybrid model
Jallal et al. Multi-Target Learning Algorithm for Solar Radiation Components Forecasting Based on the Desired Tilt Angle of a Solar Energy System
Lei et al. Research on Intelligent PID Control Algorithm Based on Neural Network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20201124

Termination date: 20210725

CF01 Termination of patent right due to non-payment of annual fee