Distributed photovoltaic output intelligent estimation method
By combining the methods of fuzzy clustering, gated cycle units and Harris Eagle algorithm, the accuracy and complexity problems of existing photovoltaic output prediction methods in capturing the relationship between environmental factors and photovoltaic output changes and large-scale distributed photovoltaic scenarios are achieved, and high-precision and low-complexity distributed photovoltaic output estimation is achieved.
Patent Information
- Application Number
- CN202510104771.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2025-03-18
AI Technical Summary
The existing photovoltaic output prediction methods are difficult to effectively capture the deep relationship between complex environmental factors and photovoltaic output changes. In large-scale distributed photovoltaic scenarios, the prediction accuracy is insufficient and the calculation complexity is high, and the correlation and regional characteristics between different photovoltaic units are not fully considered.
A distributed photovoltaic output intelligent estimation method is adopted to establish a distributed photovoltaic output cluster based on fuzzy clustering, gated cycle units and Harris Eagle algorithm, to capture the dynamic characteristics of the time series, and optimize the model parameters through the Harris Eagle algorithm.
It significantly improves the estimation accuracy and generalization ability of distributed photovoltaic output, reduces the modeling complexity, realizes high-precision and low-complexity estimation of large-scale distributed photovoltaic output, and supports the flexible regulation of modern distribution networks and the efficient utilization of new energy.
Smart Images
Figure CN119651608A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of photovoltaic output estimation, and in particular to a distributed photovoltaic output intelligent estimation method. Background Art
[0002] The penetration rate of distributed photovoltaics in the power system continues to increase, and it has become an important part of the modern distribution network. Distributed photovoltaics are clean, efficient, and flexible, and are of great significance for alleviating the energy crisis and reducing carbon emissions. However, photovoltaic power generation is greatly affected by environmental conditions and has significant intermittent and random characteristics. This uncertainty has brought great challenges to the stable operation, optimal scheduling, and security of the power system. Therefore, conducting high-precision distributed photovoltaic output estimation research has important strategic value for improving the flexibility and reliability of distribution network operation.
[0003] Although some progress has been made in the field of photovoltaic output prediction in recent years, existing research still has many limitations. On the one hand, traditional statistical modeling methods do not adequately describe the nonlinear characteristics of photovoltaic output and have difficulty capturing the deep relationship between complex environmental factors and output changes; on the other hand, when dealing with large-scale distributed photovoltaic scenarios, prediction models based on machine learning often show problems of insufficient prediction accuracy and high computational complexity due to the lack of effective feature clustering and model parameter optimization mechanisms. In addition, most existing methods fail to fully consider the correlation and regional characteristics between different photovoltaic units, making it difficult to achieve coordinated estimation of large-scale photovoltaic systems. Summary of the invention
[0004] In order to overcome the deficiencies in the prior art, the present invention proposes a distributed photovoltaic output intelligent estimation method, which aims to achieve intelligent estimation of distributed photovoltaic output by integrating fuzzy clustering, gated recurrent unit and Harris Eagle algorithm.
[0005] In order to achieve the above object, the present invention proposes a distributed photovoltaic output intelligent estimation method, comprising the following steps:
[0006] S1: Establish a distributed photovoltaic output cluster based on fuzzy clustering and output historical output characteristics;
[0007] S2: Taking the historical output characteristics of step S1 as input, a distributed photovoltaic output estimation model based on the gated cyclic unit algorithm is established to capture the dynamic characteristics of the distributed photovoltaic output cluster time series;
[0008] S3: Use the Harris Hawk algorithm to adjust the parameters of the distributed photovoltaic output estimation model, output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be adjusted in step S2 to optimize the distributed photovoltaic output estimation model in step S2.
[0009] Preferably, the specific steps of step S1 to establish a distributed photovoltaic output cluster based on fuzzy clustering include:
[0010] S101: Distributed photovoltaic output cluster data preprocessing and input feature selection;
[0011] S102: Execute distributed photovoltaic output clustering;
[0012] S103: Determine the distributed photovoltaic output cluster generation iteration termination condition.
[0013] Preferably, the input features selected in step S101 include meteorological factors: light intensity (G), temperature (T), humidity (H); geographical characteristics: longitude and latitude (Lat, Lon) of the photovoltaic system; historical output data: historical output power (P) of the photovoltaic system. The characteristic matrix of the distributed photovoltaic output cluster can be expressed as:
[0014]
[0015] Preferably, in step S102,
[0016] The membership matrix of the initial cluster center of the distributed photovoltaic output cluster satisfies:
[0017]
[0018] Among them, C j is the jth cluster center; C is the number of clusters; u ij is the membership degree of distributed photovoltaic output data point i to the jth cluster, and its value range is [0,1];
[0019] Distributed photovoltaic output cluster center C j The updated model is:
[0020]
[0021] Among them, x i is the characteristic vector of the i-th distributed photovoltaic output data point, m is the fuzzy index, and N is the total number of distributed photovoltaic output data points;
[0022] The membership update model of distributed photovoltaic output data points is:
[0023]
[0024] Among them, ‖x i -c j ‖ is the data point x i To cluster center C j The Euclidean distance of .
[0025] Preferably, the iterative termination condition for determining the generation of the distributed photovoltaic output cluster in step S103 is:
[0026]
[0027] Among them, ò. is the tolerance error; Generate iterative target values for the distributed photovoltaic output clusters at the t+1th and tth times respectively;
[0028] The iterative target value generated by the distributed photovoltaic output cluster can be expressed as:
[0029]
[0030] Preferably, step S2 establishes a distributed photovoltaic output estimation model based on a gated cyclic unit algorithm, and the specific steps include:
[0031] S201: Use the historical output characteristics of the distributed photovoltaic output cluster obtained in step S1 as the input variable x of the gated cycle unit a ;
[0032] S202: The influence of the input of the current time step on the hidden state is controlled by resetting the gate and updating the gate to capture the dynamic characteristics of the time series of the distributed photovoltaic output cluster.
[0033] Preferably, in step S202,
[0034] The update gate function can be specifically expressed as:
[0035]
[0036] In the formula, is the reset gate function input vector of the distributed photovoltaic output cluster at time t; is the output of the reset gate function at time t-1; g is the activation function; W1, V1 are the parameter matrices to be adjusted;
[0037] Among them, the reset gate function can be specifically expressed as:
[0038]
[0039] Where W2, V2 are the parameter matrices to be adjusted.
[0040] Preferably, the specific steps of using the Harris Eagle algorithm to adjust the parameters of the distributed photovoltaic output estimation model in step S3 include:
[0041] S301: Using the parameter matrix to be tuned in step S2 as the solution space variable to be optimized of the Harris Eagle algorithm;
[0042] S302: Using the Harris Eagle algorithm fitness function to evaluate the quality of current parameters;
[0043] S303: Calculate the fitness of each Harris Hawk, and select the Harris Hawk with the highest fitness to enter the next round of iteration;
[0044] S304: Determine whether the maximum number of iterations has been reached. If so, output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be adjusted in step S2; otherwise, return to step S302.
[0045] Preferably, the function for evaluating the quality of the current parameters using the Harris Hawk algorithm fitness function in step S302 is:
[0046]
[0047] Among them, P(t) is the actual photovoltaic output value, is the predicted photovoltaic output value of the gated cycle unit, and T is the total number of time steps in the distributed photovoltaic cluster dataset.
[0048] Preferably, in step S303, the calculation model of Harris Hawk entering the next iteration is:
[0049] X i (t+1)=X i (t)+r1·A·|C·X * -X i (t)|-r2·|X i (t)-X j (t)|
[0050] Among them, X i (t+1), X i (t) is the current position of the ith hawk in the t+1th and tth generations; X * is the position of the current optimal solution; r1, A, r2, C are parameter variables.
[0051] The present invention has the following beneficial effects:
[0052] Distributed photovoltaic systems are grouped according to similarity through fuzzy clustering, which significantly reduces the complexity of modeling; the time series model based on gated recurrent units can deeply explore the dynamic relationship between meteorological conditions and photovoltaic output; the Harris Eagle algorithm uses efficient global optimization capabilities to accurately adjust the model parameters, further improve the estimation accuracy and generalization ability, and achieve high-precision and low-complexity estimation of large-scale distributed photovoltaic output, providing a scientific basis for the flexible regulation of modern distribution networks, and thus promoting the efficient utilization of new energy and the transformation to low-carbon energy. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] The present invention will be further described and explained below in conjunction with the accompanying drawings.
[0054] Figure 1 It is the distributed photovoltaic output error diagram obtained by different methods.
[0055] Figure 2 It is the iterative convergence graph of the Harris Hawk algorithm. DETAILED DESCRIPTION
[0056] The technical solution of the present invention will be more clearly and completely explained below through description of preferred embodiments of the present invention in combination with the accompanying drawings.
[0057] A distributed photovoltaic output intelligent estimation method comprises the following steps:
[0058] S1: Establish a distributed photovoltaic output cluster based on fuzzy clustering and output historical output characteristics.
[0059] Specifically, step S1 establishes a distributed photovoltaic output cluster based on fuzzy clustering. The specific steps include:
[0060] S101: Distributed photovoltaic output cluster data preprocessing and input feature selection.
[0061] The input features selected in step S101 include meteorological factors: light intensity (G), temperature (T), humidity (H); geographical characteristics: photovoltaic system longitude and latitude (Lat, Lon); historical output data: photovoltaic system historical output power (P), and the distributed photovoltaic output cluster feature matrix can be expressed as:
[0062]
[0063] S102: Execute distributed photovoltaic output clustering.
[0064] In step S102, the membership matrix of the cluster center of the distributed photovoltaic output cluster initialization satisfies:
[0065]
[0066] Among them, C j is the jth cluster center; C is the number of clusters; u ij is the membership degree of distributed photovoltaic output data point i to the jth cluster, and its value range is [0,1];
[0067] Distributed photovoltaic output cluster center C j The updated model is:
[0068]
[0069] Among them, x iis the characteristic vector of the i-th distributed photovoltaic output data point, m is the fuzzy index, and N is the total number of distributed photovoltaic output data points;
[0070] The membership update model of distributed photovoltaic output data points is:
[0071]
[0072] Among them, ‖x i -c j ‖ is the data point x i To cluster center C j The Euclidean distance of .
[0073] S103: Determine the distributed photovoltaic output cluster generation iteration termination condition.
[0074] In step S103, it is determined that the distributed photovoltaic output cluster generation iteration termination condition is:
[0075]
[0076] Among them, ò. is the tolerance error; Generate iterative target values for the distributed photovoltaic output clusters at the t+1th and tth times respectively;
[0077] The iterative target value generated by the distributed photovoltaic output cluster can be expressed as:
[0078]
[0079] S2: Taking the historical output characteristics of step S1 as input, a distributed photovoltaic output estimation model based on the gated recurrent unit algorithm is established to capture the dynamic characteristics of the distributed photovoltaic output cluster time series.
[0080] Step S2 establishes a distributed photovoltaic output estimation model based on a gated cyclic unit algorithm, and the specific steps include:
[0081] S201: Use the historical output characteristics of the distributed photovoltaic output cluster obtained in step S1 as the input variable x of the gated cycle unit a ;
[0082] S202: The influence of the input of the current time step on the hidden state is controlled by resetting the gate and updating the gate to capture the dynamic characteristics of the time series of the distributed photovoltaic output cluster.
[0083] In step S202,
[0084] The update gate function can be specifically expressed as:
[0085]
[0086] In the formula, is the reset gate function input vector of the distributed photovoltaic output cluster at time t; is the output of the reset gate function at time t-1; g is the activation function; W1, V1 are the parameter matrices to be adjusted;
[0087] Among them, the reset gate function can be specifically expressed as:
[0088]
[0089] Where W2, V2 are the parameter matrices to be adjusted.
[0090] S3: Use the Harris Hawk algorithm to adjust the parameters of the distributed photovoltaic output estimation model, output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be adjusted in step S2 to optimize the distributed photovoltaic output estimation model in step S2.
[0091] In step S3, the Harris Eagle algorithm is used to adjust the parameters of the distributed photovoltaic output estimation model. The specific steps include:
[0092] S301: Use the parameter matrix to be tuned in step S2 as the solution space variable to be optimized of the Harris Hawk algorithm.
[0093] S302: Use the Harris Eagle algorithm fitness function to evaluate the quality of current parameters.
[0094] In step S302, the function for evaluating the quality of the current parameters using the Harris Hawk algorithm fitness function is:
[0095]
[0096] Among them, P(t) is the actual photovoltaic output value, is the predicted photovoltaic output value of the gated cycle unit, and T is the total number of time steps in the distributed photovoltaic cluster dataset.
[0097] S303: Calculate the fitness of each Harris Hawk, and select the Harris Hawk with the highest fitness to enter the next round of iteration. In step S303, the calculation model for the Harris Hawk to enter the next round of iteration is:
[0098] X i (t+1)=X i (t)+r1·A·|C·X * -X i (t)|-r2·|X i (t)-X j (t)|
[0099] Among them, X i (t+1), X i(t) is the current position of the ith hawk in the t+1th and tth generations; X * is the position of the current optimal solution; r1, A, r2, C are parameter variables.
[0100] S304: Determine whether the maximum number of iterations has been reached. If so, output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be adjusted in step S2; otherwise, return to step S302.
[0101] The maximum number of iterations needs to be set during initialization, and each round needs to be compared with the maximum number of iterations. If it is reached, the solution of the current Harris Eagle algorithm is output as the parameters of the parameter matrix to be adjusted in step S2.
[0102] In order to verify the effectiveness of the present invention, we selected distributed photovoltaic in a distribution network as the research object. Distributed photovoltaic data sampling frequency: sampling once every hour to obtain the photovoltaic output value per hour. Data cycle: use 30 days (720 hours) of historical data to train the model and predict the photovoltaic output for the next 24 hours. Meteorological data source: real-time meteorological data obtained through meteorological stations or satellite remote sensing. Collected variables: light intensity (W / m 2 ), temperature (℃), humidity (%), wind speed (m / s). Number of GRU units: set to 64 GRU units. Set the Harris Hawk group size to 50 and the maximum number of iterations to 1000.
[0103] In order to demonstrate the superiority of the method proposed in this invention, different methods are used to compare the distributed photovoltaic output estimation effects. Figure 1 As shown, Figure 1 The output errors of distributed photovoltaic power obtained by different methods are shown in Figure 2, where the blue curve is the method proposed in this paper, the red curve is the traditional method, the horizontal axis is time, and the vertical axis is the estimated error. It can be seen that the estimated error of the method proposed in this paper is at a lower level than that of the traditional method in most of the time. This is because the method proposed in this paper not only uses the GRU algorithm to mine and learn the time series data of distributed photovoltaic power generation, but also uses the Harris Eagle algorithm to optimize the parameters of the GRU.
[0104] Figure 2 This is the iterative convergence diagram of the Harris Eagle algorithm. It can be seen that the method adopted by the present invention can quickly iterate to a higher fitness value in a shorter iteration round (about 200 times), so that the optimal parameters of the GRU algorithm can be learned efficiently, so that the distributed photovoltaic output estimation model based on the GRU algorithm can have a better estimation effect.
[0105] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A distributed photovoltaic output intelligent estimation method, characterized in that: The following steps are involved: S1: Establish a distributed photovoltaic output cluster based on fuzzy clustering and output historical output characteristics; S2: Taking the historical output characteristics of step S1 as input, a distributed photovoltaic output estimation model based on the gated cyclic unit algorithm is established to capture the dynamic characteristics of the distributed photovoltaic output cluster time series; S3: Use the Harris Hawk algorithm to adjust the parameters of the distributed photovoltaic output estimation model, output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be adjusted in step S2 to optimize the distributed photovoltaic output estimation model in step S2.
2. A distributed photovoltaic output intelligent estimation method according to claim 1, characterized in that: The specific steps of establishing a distributed photovoltaic output cluster based on fuzzy clustering in step S1 include: S101: Distributed photovoltaic output cluster data preprocessing and input feature selection; S102: Execute distributed photovoltaic output clustering; S103: Determine the distributed photovoltaic output cluster generation iteration termination condition.
3. A distributed photovoltaic output intelligent estimation method according to claim 2, characterized in that: The input features selected in step S101 include meteorological factors: light intensity (G), temperature (T), humidity (H); geographical characteristics: photovoltaic system longitude and latitude (Lat, Lon); historical output data: photovoltaic system historical output power (P), and the distributed photovoltaic output cluster feature matrix can be expressed as:
4. A distributed photovoltaic output intelligent estimation method according to claim 2, characterized in that: In step S102, The membership matrix of the initial cluster center of the distributed photovoltaic output cluster satisfies: Among them, C j is the jth cluster center; C is the number of clusters; u ij is the membership degree of distributed photovoltaic output data point i to the jth cluster, and its value range is [0,1]; Distributed photovoltaic output cluster center C j The updated model is: Among them, x i is the characteristic vector of the i-th distributed photovoltaic output data point, m is the fuzzy index, and N is the total number of distributed photovoltaic output data points; The membership update model of distributed photovoltaic output data points is: Among them, ‖x i -c j ‖ is the data point x i To cluster center C j The Euclidean distance of .
5. A distributed photovoltaic output intelligent estimation method according to claim 2, characterized in that: In step S103, the iterative termination condition for determining the generation of the distributed photovoltaic output cluster is: Among them, ò. is the tolerance error; Generate iterative target values for the distributed photovoltaic output clusters at the t+1th and tth times respectively; The iterative target value generated by the distributed photovoltaic output cluster can be expressed as:
6. A distributed photovoltaic output intelligent estimation method according to claim 1, characterized in that: The step S2 establishes a distributed photovoltaic output estimation model based on a gated cyclic unit algorithm, and the specific steps include: S201: Use the historical output characteristics of the distributed photovoltaic output cluster obtained in step S1 as the input variable x of the gated cycle unit a ; S202: The influence of the input of the current time step on the hidden state is controlled by resetting the gate and updating the gate to capture the dynamic characteristics of the time series of the distributed photovoltaic output cluster.
7. A distributed photovoltaic output intelligent estimation method according to claim 6, characterized in that: In step S202, The update gate function can be specifically expressed as: In the formula, is the reset gate function input vector of the distributed photovoltaic output cluster at time t; is the output of the reset gate function at time t-1; g is the activation function; W1, V1 are the parameter matrices to be adjusted; Among them, the reset gate function can be specifically expressed as: Where W2, V2 are the parameter matrices to be adjusted.
8. A distributed photovoltaic output intelligent estimation method according to claim 1, characterized in that: The specific steps of using the Harris Eagle algorithm to adjust the parameters of the distributed photovoltaic output estimation model in step S3 include: S301: Use the parameter matrix to be tuned in step S2 as the variable of the solution space to be optimized of the Harris Hawk algorithm; initialize the maximum iteration number N of the Harris Hawk algorithm h ; S302: Using the Harris Eagle algorithm fitness function to evaluate the quality of current parameters; S303: Calculate the fitness of each Harris Hawk, and select the Harris Hawk with the highest fitness to enter the next round of iteration; S304: Determine whether the current round has reached the maximum iteration round N set in step S301 h , if so, then output the solution of the current Harris Hawk algorithm as the parameters of the parameter matrix to be tuned in step S2, otherwise, return to step S302.
9. A distributed photovoltaic output intelligent estimation method according to claim 8, characterized in that: The function of evaluating the quality of the current parameters using the Harris Hawk algorithm fitness function in step S302 is: Among them, P(t) is the actual photovoltaic output value, is the predicted photovoltaic output value of the gated cycle unit, and T is the total number of time steps in the distributed photovoltaic cluster dataset.
10. A distributed photovoltaic output intelligent estimation method according to claim 8, characterized in that: In step S303, the calculation model of Harris Hawk entering the next iteration is: X i (t+1)=X i (t)+r1·A·|C·X * -X i (t)|-r2·|X i (t)-X j (t)| Among them, X i (t+1), X i (t) is the current position of the ith hawk in the t+1th and tth generations; X * is the position of the current optimal solution; r1, A, r2, C are parameter variables.