Photovoltaic power station load prediction method

The data of photovoltaic power stations is processed through the isolated forest algorithm and the LSTM model is combined with a variety of environmental data, which solves the problem of high data interference in photovoltaic power station load prediction, achieving more accurate load demand prediction.

CN119046616APending Publication Date: 2024-11-29GUANGDONG KEYSTAR INTELLIGENCE ROBOT CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411229631.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

In the existing photovoltaic power plant load prediction technology, the data is highly disturbing and cannot effectively train historical data to obtain accurate load demand prediction.

Method used

The isolated forest algorithm is used to denoiser and normalize data, and the LSTM prediction model is used to combine solar radiation, temperature, humidity and other data to construct a one-dimensional matrix set for prediction.

Benefits of technology

The data is processed through the isolated forest algorithm, which improves the accuracy of prediction. The LSTM model can remember recent information and predicts the load demand of photovoltaic power stations more accurately in the short term.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119046616A_ABST
    Figure CN119046616A_ABST
Patent Text Reader

Abstract

The invention proposes a photovoltaic power station load prediction method, and the method comprises the steps: A, collecting the historical data of a photovoltaic power station, and building a one-dimensional matrix set based on a data set of a plurality of historical data; step B, performing data processing on the power load data and arranging the power load data into a new one-dimensional matrix set, including executing a data cleaning operation, executing a denoising operation by adopting an isolated forest algorithm, and executing a normalization operation; and step C, inputting the new one-dimensional matrix set into the LSTM prediction model for prediction, including taking the one-dimensional matrix set after data processing as an input sequence of a current time step of the LSTM prediction model for calculation so as to output an updated hidden state of the LSTM prediction model. And transmitting the updated hidden state to an output layer of the LSTM prediction model to obtain a predicted one-dimensional matrix set, namely predicted load data. According to the method, the anti-interference capability of data processing is improved through the isolated forest algorithm, and the short-term memory characteristics of the LSTM model are integrated to predict the power data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power station load prediction, and in particular relates to a photovoltaic power station load prediction method. Background Art

[0002] Photovoltaic power station is a facility that uses solar energy to generate electricity. It mainly converts solar energy into electrical energy through photovoltaic power generation systems. The load demand forecast of photovoltaic power stations is crucial to improving the efficiency, reliability and sustainability of energy management. It not only helps to make full use of renewable energy resources, but also reduces energy costs, waste, and environmental impact, thereby bringing benefits on multiple levels.

[0003] However, the current technical means for load prediction of photovoltaic power stations have too much data interference, and it is impossible to effectively train historical data to predict more accurate load demand data. Summary of the invention

[0004] In view of the problems raised by the background technology, the present invention proposes a method for optimizing energy efficiency of a photovoltaic power station.

[0005] To achieve this object, the present invention adopts the following technical solutions:

[0006] A method for predicting load of a photovoltaic power station, step A: collecting historical data of the photovoltaic power station, the historical data including solar radiation O i , Temperature T i , Photovoltaic module power generation P i 、PV panel internal temperature IT i 、Internal humidity H i , wind speed W i 、Internal power consumption I i and power load data Q i , based on multiple historical data sets, a one-dimensional matrix set X is constructed i =[O i , T i , P i , IT i , H i , W i , I i , Q i ], where i represents different time points;

[0007] Step B: Processing the power load data and arranging it into a new one-dimensional matrix set, including performing data cleaning operations, performing denoising operations using an isolation forest algorithm, and performing normalization operations;

[0008] Step C: Input the new one-dimensional matrix set into the LSTM prediction model for prediction, including using the one-dimensional matrix set after data processing as the input sequence of the current time step of the LSTM prediction model to calculate to output the updated hidden state of the LSTM prediction model, and passing the updated hidden state to the output layer of the LSTM prediction model to obtain the predicted one-dimensional matrix set, that is, the predicted load data.

[0009] Preferably, in step B, performing a data cleaning operation on the one-dimensional matrix set includes:

[0010] Power load data Q i ≤0 in the matrix.

[0011] Preferably, in step B, performing a denoising operation on the power load data using an isolation forest algorithm includes:

[0012] Step B1: Convert the power load data Q i The set where the data is located is recorded as data set X, and τ sample points are randomly selected from the data set X to form a subset X′ of the data set X, and the subset X′ is placed in the root node of the isolation tree; where:

[0013]

[0014] x i represents the subset X′, that is, the set consisting of τ sample points randomly selected from the data set X at different time points i, and the sample points selected in the set are denoted as x iτ ;

[0015] Step B2: Randomly specify a dimension q from the d dimensions, randomly generate a cutting point p in the subset X′, and let in:

[0016] x ij Represents the sample points when a dimension q is randomly specified;

[0017] Step B3: The cutting point p generates a hyperplane to divide the subset X′ into two subspaces, wherein the sample points in one subspace are sample points with a dimension smaller than the cutting point p, and the sample points with a dimension smaller than the cutting point p are placed in the left child node of the isolation tree;

[0018] The sample points in the other subspace are sample points whose dimensions are greater than or equal to the cutting point p. The sample points whose dimensions are greater than or equal to the cutting point p are placed in the right child node of the isolation tree.

[0019] Step B4: recursively perform steps B2 and B3 until all child nodes of the isolated tree have only one sample point or the isolated tree has reached a specified height;

[0020] Step B5: Loop step B1 to step B4 until a specified number of isolated trees are generated.

[0021] Preferably, after generating a specified number of isolated trees, each sample point in each isolated tree is allowed to traverse each isolated tree, the average height of each sample point in the forest composed of all isolated trees is calculated, and the outlier score of each sample point is calculated. The calculation formula of the outlier score is as follows:

[0022] s(x,a)=2E(h(x)) / c(a);

[0023]

[0024] H(c)=ln(c)+β, β=0.5772156649;

[0025] in:

[0026] s(x, a) represents the abnormal index of sample point x in the isolated tree, that is, the outlier score, and its value range is [0, 1];

[0027] a represents the threshold parameter;

[0028] E(x) represents the expected value of each sample point x;

[0029] h(x) represents the number of partitions that the sample point x goes through from the root node of the isolation tree to the child node where the sample point x is located, that is, the depth;

[0030] c(a) represents a constant used to scale the path length;

[0031] H(c) represents the expected path length when a sample set contains c samples.

[0032] Preferably, after calculating the outlier score, determining the sample abnormality includes:

[0033] When s(x, a) is equal to 1, it means that the sample point is an outlier;

[0034] When s(x, a) is equal to 0, it means that the sample point is a normal point;

[0035] When s(x, a) of the sample points exceeding the preset threshold is equal to 0.5, it means that the data set containing the historical data does not need to perform denoising operation;

[0036] Delete the samples with outliers in the data set to obtain the denoised power load data set X′ i .

[0037] Preferably, the data set X′ of the denoised power load data is iPerform normalization operations, including normalization based on the following formula:

[0038]

[0039] in:

[0040] X″ i a data set representing normalized power load data;

[0041] X′ i a data set representing denoised power load data;

[0042] mean represents X′ i The mean of

[0043] std represents the standard deviation of subset X′;

[0044] Reorganize the normalized power load data and other historical data into a new one-dimensional matrix:

[0045] Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ].

[0046] Preferably, in said step C, it includes:

[0047] Step C1: Calculate the forget gate of the LSTM prediction model:

[0048] F t =σ(W f *[h t-1 , x t ]+b f );

[0049] in:

[0050] F t represents the input of the forget gate, W f represents the weight matrix of the forget gate, b f represents the bias vector of the forget gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i, W i , I i , X″ i ];

[0051] Step C2: Calculate the update of the cell state at each time step based on the input of the forget gate:

[0052] c1 t =F t *cl t-1 ;

[0053] in:

[0054] c1 t represents the cell state of the current time step after the forget gate is updated, c1 t-1 The cell state at the previous time step;

[0055] Step C3: Calculate the input gate of the LSTM prediction model:

[0056] I t =σ(W i *[h t-1 , x t ]+b i );

[0057] in:

[0058] I t represents the input of the input gate, W i represents the weight matrix of the input gate, b i represents the bias vector of the input gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0059] Step C4: Calculate the update of the cell state at each time step based on the input of the input gate:

[0060]

[0061]

[0062] in:

[0063] c trepresents the updated cell state of the current time step of the input gate, c t-1 The cell state at the previous time step, represents the candidate updated cell state, tanh represents the hyperbolic tangent function, W c represents the weight matrix of candidate updates, b c represents the bias vector of the candidate update, h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″″ i ];

[0064] Step C5: Calculate the output gate of the LSTM prediction model:

[0065] O t =σ(W o *[h t-1 , x t ]+b o );

[0066] in:

[0067] O t represents the input of the output gate, W o represents the weight matrix of the input gate, b o represents the bias vector of the output gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0068] Step C6: Update the hidden state of the current time step according to the input of the output gate;

[0069] h t =O t *tanh(c t );

[0070] in:

[0071] h t represents the updated hidden state of the current time step, O t represents the input of the output gate, tanh represents the hyperbolic tangent function, c t represents the updated cell state at the current time step of the input gate;

[0072] Step C7: Update the hidden state h t Passed to the output layer of the LSTM prediction model to obtain a one-dimensional matrix set after prediction.

[0073] Compared with the prior art, one of the above technical solutions has the following beneficial effects:

[0074] 1. The present invention processes data through the isolation forest algorithm, which can determine the percentage of discarded data by itself, increase data diversity or concentration, and make the prediction more accurate;

[0075] 2. For photovoltaic systems, the power load has a certain periodicity. In this regard, the present invention proposes an LSTM prediction model. The LSTM prediction model can remember the characteristics of recent information. In the short term, the power load does not fluctuate much. Based on data such as solar radiation, temperature, and humidity, the load demand in the future period can be better predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a flow chart of the photovoltaic power station load prediction method of the present invention. DETAILED DESCRIPTION

[0077] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0078] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0080] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0081] This application proposes a photovoltaic power station load prediction method, such as Figure 1 As shown, step A: collecting historical data of photovoltaic power plants, the historical data includes solar radiation O i , Temperature T i , Photovoltaic module power generation P i 、PV panel internal temperature IT i 、Internal humidity H i , wind speed W i 、Internal power consumption I i and power load data Q i , based on multiple historical data sets, a one-dimensional matrix set X is constructed i =[O i , T i , P i , IT i , H i , W i , I i , Q i ], where i represents different time points;

[0082] Step B: Processing the power load data and arranging it into a new one-dimensional matrix set, including performing data cleaning operations, performing denoising operations using an isolation forest algorithm, and performing normalization operations;

[0083] Step C: Input the new one-dimensional matrix set into the LSTM prediction model for prediction, including using the one-dimensional matrix set after data processing as the input sequence of the current time step of the LSTM prediction model to calculate to output the updated hidden state of the LSTM prediction model, and passing the updated hidden state to the output layer of the LSTM prediction model to obtain the predicted one-dimensional matrix set, that is, the predicted load data.

[0084] Preferably, in step B, performing a data cleaning operation on the one-dimensional matrix set includes:

[0085] Power load data Q i ≤0 in the matrix.

[0086] Preferably, in step B, performing a denoising operation on the power load data using an isolation forest algorithm includes:

[0087] Step B1: Convert the power load data Q i The set where the data is located is recorded as data set X, and τ sample points are randomly selected from the data set X to form a subset X′ of the data set X, and the subset X′ is placed in the root node of the isolation tree; where:

[0088]

[0089] x i represents the subset X′, that is, the set consisting of τ sample points randomly selected from the data set X at different time points i, and the sample points selected in the set are denoted as x iτ ;

[0090] Step B2: Randomly specify a dimension q from the d dimensions, randomly generate a cutting point p in the subset X′, and let in:

[0091] x ij Represents the sample points when a dimension q is randomly specified;

[0092] Step B3: The cutting point p generates a hyperplane to divide the subset X′ into two subspaces, wherein the sample points in one subspace are sample points with a dimension smaller than the cutting point p, and the sample points with a dimension smaller than the cutting point p are placed in the left child node of the isolation tree;

[0093] The sample points in the other subspace are sample points whose dimensions are greater than or equal to the cutting point p. The sample points whose dimensions are greater than or equal to the cutting point p are placed in the right child node of the isolation tree.

[0094] Step B4: recursively perform steps B2 and B3 until all child nodes of the isolated tree have only one sample point or the isolated tree has reached a specified height;

[0095] Step B5: Loop step B1 to step B4 until a specified number of isolated trees are generated.

[0096] Preferably, after generating a specified number of isolated trees, each sample point in each isolated tree is allowed to traverse each isolated tree, the average height of each sample point in the forest composed of all isolated trees is calculated, and the outlier score of each sample point is calculated. The calculation formula of the outlier score is as follows:

[0097] s(x,a)=2E(h(x)) / c(a);

[0098]

[0099] H(c)=ln(c)+β, β=0.5772156649;

[0100] in:

[0101] s(x, a) represents the abnormal index of sample point x in the isolated tree, that is, the outlier score, and its value range is [0, 1];

[0102] a represents the threshold parameter;

[0103] E(x) represents the expected value of each sample point x;

[0104] h(x) represents the number of partitions that the sample point x goes through from the root node of the isolation tree to the child node where the sample point x is located, that is, the depth;

[0105] c(a) represents a constant used to scale the path length;

[0106] H(c) represents the expected path length when a sample set contains c samples.

[0107] Preferably, after calculating the outlier score, determining the sample abnormality includes:

[0108] When s(x, a) is equal to 1, it means that the sample point is an outlier;

[0109] When s(x, a) is equal to 0, it means that the sample point is a normal point;

[0110] When s(x, a) of the sample points exceeding the preset threshold is equal to 0.5, it means that the data set containing the historical data does not need to perform denoising operation;

[0111] Delete the samples with outliers in the data set to obtain the denoised power load data set X′ i .

[0112] In this application, in actual application, it can be understood that: the closer s(x, α) is to 1, the higher the possibility of being an outlier;

[0113] The closer s(x, α) is to 0, the higher the probability that it is a normal point;

[0114] If s(x, α) of most samples is close to 0.5, it means that there are no obvious outliers in the entire data set and no denoising is required.

[0115] Delete the groups with high probability of abnormal points to obtain the denoised power load data X′ i .

[0116] Preferably, the data set X′ of the denoised power load data is i Perform normalization operations, including normalization based on the following formula:

[0117]

[0118] in:

[0119] X″ i a data set representing normalized power load data;

[0120] X′ i a data set representing denoised power load data;

[0121] mean represents X i The mean of ′;

[0122] std represents the standard deviation of subset X′;

[0123] Reorganize the normalized power load data and other historical data into a new one-dimensional matrix:

[0124] Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ].

[0125] Preferably, the long short-term memory network (LSTM) is a recurrent neural network (RNN) variant commonly used to process sequence data. The LSTM layer uses the autocorrelation function ACF and the partial autocorrelation function PACF to analyze the autocorrelation between the historical value and the current value, and uses the memory of the LSTM to extract the time series characteristics of the sequence;

[0126] The information storage unit of LSTM consists of an input gate, a forget gate, and an output gate to achieve selective storage and extraction of important information in the sequence; these gated units determine the update of the cell state at each time step. t and the hidden state h t Calculation of (LSTM output);

[0127] For photovoltaic systems, the power load has a certain periodicity. The LSTM model can memorize recent information. In the short term, the power load does not fluctuate much. Based on data such as solar radiation, temperature, and humidity, the load demand in the future period can be better predicted.

[0128] In the step C, it includes:

[0129] Step C1: The forget gate determines how much information needs to be discarded from the cell state at the current time step. Calculate the forget gate of the LSTM prediction model:

[0130] F t =σ(W f *[h t-1 , x t ]+b f );

[0131] in:

[0132] F t represents the input of the forget gate, W f represents the weight matrix of the forget gate, b f represents the bias vector of the forget gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0133] Step C2: The input gate determines how many new messages need to be added to the cell state at the current time step, and calculates the update of the cell state at each time step based on the input of the forget gate:

[0134] c1 t =F t *c1 t-1 ;

[0135] in:

[0136] c1 t represents the cell state of the current time step after the forget gate is updated, c1 t-1 The cell state at the previous time step;

[0137] Step C3: The output gate determines how much cell state information needs to be output at the current time step, and calculates the input gate of the LSTM prediction model:

[0138] I t =σ(W i *[h t-1 , x t ]+b i );

[0139] in:

[0140] I t represents the input of the input gate, W i represents the weight matrix of the input gate, b i represents the bias vector of the input gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0141] Step C4: Calculate the update of the cell state at each time step based on the input of the input gate:

[0142]

[0143]

[0144] in:

[0145] c t represents the updated cell state of the current time step of the input gate, c t-1 The cell state at the previous time step, represents the candidate updated cell state, tanh represents the hyperbolic tangent function, W c represents the weight matrix of candidate updates, b c represents the bias vector of the candidate update, h t-1 represents the hidden state of the previous time step, xt Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0146] Step C5: Calculate the output gate of the LSTM prediction model:

[0147] O t =σ(W o *[h t-1 , x t ]+b o );

[0148] in:

[0149] O t represents the input of the output gate, W o represents the weight matrix of the input gate, b o represents the bias vector of the output gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ];

[0150] Step C6: Update the hidden state of the current time step according to the input of the output gate;

[0151] h t =O t *tanh(c t );

[0152] in:

[0153] h t represents the updated hidden state of the current time step, O t represents the input of the output gate, tanh represents the hyperbolic tangent function, c t represents the updated cell state at the current time step of the input gate;

[0154] Step C7: Update the hidden state h tPassed to the output layer of the LSTM prediction model to obtain a one-dimensional matrix set after prediction.

[0155] h t is the output of the LSTM model at the current time step, which can be used to make predictions for the next step or passed to other layers for further processing. t is based on the input O of the current time step t and the hidden state of the previous time step (h t-1 ) is calculated. Prediction Y i Usually, the hidden state h t Passed to the output layer for calculation.

[0156] The technical principle of the present invention is described above in conjunction with specific embodiments. These descriptions are only for explaining the principle of the present invention and cannot be interpreted as limiting the scope of protection of the present invention in any way. Based on the explanations herein, those skilled in the art can associate other specific implementations of the present invention without paying creative labor, and these methods will fall within the scope of protection of the present invention.

Claims

1. A method for predicting load of a photovoltaic power station, Features: Step A: Collect historical data of photovoltaic power plants, including solar radiation O i , Temperature T i , Photovoltaic module power generation P i 、PV panel internal temperature IT i 、Internal humidity H i , wind speed W i 、Internal power consumption I i and power load data Q i , based on multiple historical data sets, a one-dimensional matrix set X is constructed i =[O i , T i , P i , IT i , H i , W i , I i , Q i ], where i represents different time points; Step B: Processing the power load data and arranging it into a new one-dimensional matrix set, including performing data cleaning operations, performing denoising operations using an isolation forest algorithm, and performing normalization operations; Step C: Input the new one-dimensional matrix set into the LSTM prediction model for prediction, including using the one-dimensional matrix set after data processing as the input sequence of the current time step of the LSTM prediction model to calculate to output the updated hidden state of the LSTM prediction model, and passing the updated hidden state to the output layer of the LSTM prediction model to obtain the predicted one-dimensional matrix set, that is, the predicted load data.

2. According to the method for predicting load of a photovoltaic power station as claimed in claim 1, Features: In the step B, performing a data cleaning operation on a one-dimensional matrix set includes: Power load data Q i ≤0 in the matrix.

3. According to the method for predicting load of a photovoltaic power station as claimed in claim 1, Features: In step B, performing a denoising operation on the power load data using an isolation forest algorithm includes: Step B1: Convert the power load data Q i The set where the data is located is recorded as data set X, and τ sample points are randomly selected from the data set X to form a subset X′ of the data set X, and the subset X′ is placed in the root node of the isolation tree; where: x i represents the subset X′, that is, the set consisting of τ sample points randomly selected from the data set X at different time points i, and the sample points selected in the set are denoted as x iτ ; Step B2: Randomly specify a dimension q from the d dimensions, randomly generate a cutting point p in the subset X′, and let in: x ij Represents the sample points when a dimension q is randomly specified; Step B3: The cutting point p generates a hyperplane to divide the subset X′ into two subspaces, wherein the sample points in one subspace are sample points with a dimension smaller than the cutting point p, and the sample points with a dimension smaller than the cutting point p are placed in the left child node of the isolation tree; The sample points in the other subspace are sample points whose dimensions are greater than or equal to the cutting point p. The sample points whose dimensions are greater than or equal to the cutting point p are placed in the right child node of the isolation tree. Step B4: recursively perform steps B2 and B3 until all child nodes of the isolated tree have only one sample point or the isolated tree has reached a specified height; Step B5: Loop step B1 to step B4 until a specified number of isolated trees are generated.

4. A photovoltaic power station load prediction method according to claim 3, Features: After generating a specified number of isolated trees, each sample point in each isolated tree is traversed through each isolated tree, the average height of each sample point in the forest composed of all isolated trees is calculated, and the outlier score of each sample point is calculated. The calculation formula of the outlier score is as follows: s(x,a)=2E(h(x)) / c(a); H(c)=ln(c)+β, β=0.5772156649; in: s(x, a) represents the abnormal index of sample point x in the isolated tree, that is, the outlier score, and its value range is [0, 1]; a represents the threshold parameter; E(x) represents the expected value of each sample point x; h(x) represents the number of partitions that the sample point x goes through from the root node of the isolation tree to the child node where the sample point x is located, that is, the depth; c(a) represents a constant used to scale the path length; H(c) represents the expected path length when a sample set contains c samples.

5. A photovoltaic power station load prediction method according to claim 4, Features: After calculating the outlier score, the sample abnormality is judged as follows: When s(x, a) is equal to 1, it means that the sample point is an outlier; When s(x, a) is equal to 0, it means that the sample point is a normal point; When s(x, a) of the sample points exceeding the preset threshold is equal to 0.5, it means that the data set containing the historical data does not need to perform denoising operation; Delete the samples with outliers in the data set to obtain the denoised power load data set X′ i .

6. A method for predicting load of a photovoltaic power station according to claim 5, Features: The denoised power load data set X′ i Perform normalization operations, including normalization based on the following formula: in: X″ i a data set representing normalized power load data; X′ i a data set representing denoised power load data; mean represents X′ i The mean of std represents the standard deviation of subset X′; Reorganize the normalized power load data and other historical data into a new one-dimensional matrix: Y i =[O i ,T i ,P i ,IT i ,H i ,W i ,I i ,X″ i ]。 7. A method for predicting load of a photovoltaic power station according to claim 6, Features: In the step C, it includes: Step C1: Calculate the forget gate of the LSTM prediction model: F t =σ(W f *[h t-1 ,x t ]+b f ); in: F t represents the input of the forget gate, W f represents the weight matrix of the forget gate, b f represents the bias vector of the forget gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ]; Step C2: Calculate the update of the cell state at each time step based on the input of the forget gate: c1 t =F t *c1 t-1 ; in: c1 t represents the cell state of the current time step after the forget gate is updated, c1 t-1 The cell state at the previous time step; Step C3: Calculate the input gate of the LSTM prediction model: I t =σ(W i *[h t-1 ,x t ]+b i ); in: I t represents the input of the input gate, W i represents the weight matrix of the input gate, b i represents the bias vector of the input gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ]; Step C4: Calculate the update of the cell state at each time step based on the input of the input gate: in: c t represents the updated cell state of the current time step of the input gate, c t-1 The cell state at the previous time step, represents the candidate updated cell state, tanh represents the hyperbolic tangent function, W c represents the weight matrix of candidate updates, b c represents the bias vector of the candidate update, h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ]; Step C5: Calculate the output gate of the LSTM prediction model: The t =σ(W o *[h t-1 ,x t ]+b o ); in: O t represents the input of the output gate, W o represents the weight matrix of the input gate, b o represents the bias vector of the output gate, σ represents the sigmoid function, and h t-1 represents the hidden state of the previous time step, x t Represents the input sequence of the current time step, that is, Y i =[O i , T i , P i , IT i , H i , W i , I i , X″ i ]; Step C6: Update the hidden state of the current time step according to the input of the output gate; h t =O t *tanh(c t ); in: h t represents the updated hidden state of the current time step, O t represents the input of the output gate, tanh represents the hyperbolic tangent function, c t represents the updated cell state at the current time step of the input gate; Step C7: Update the hidden state h t Passed to the output layer of the LSTM prediction model to obtain a one-dimensional matrix set after prediction.

Citation Information

Patent Citations

  • Short-term photovoltaic generation power prediction method and system based on LSTM

    CN108734331A

  • Stacker track damage positioning method based on isolated forest algorithm

    CN113011325A

  • Generation power prediction method based on photovoltaic unit output abnormal data cleaning

    CN118137482A